diff options
| author | Yuval Adam <_@yuv.al> | 2017-08-30 08:25:59 +0000 |
|---|---|---|
| committer | Yuval Adam <_@yuv.al> | 2017-08-30 08:25:59 +0000 |
| commit | 8e4bff4cab0ddac6060645b0715210484d02ff40 (patch) | |
| tree | 3a5c3024371a04692a3a6d9974d001cdff8ebf84 /drivers/staging/wlags49_h2/hcf.c | |
Diffstat (limited to 'drivers/staging/wlags49_h2/hcf.c')
| -rw-r--r-- | drivers/staging/wlags49_h2/hcf.c | 4748 |
1 files changed, 4748 insertions, 0 deletions
diff --git a/drivers/staging/wlags49_h2/hcf.c b/drivers/staging/wlags49_h2/hcf.c new file mode 100644 index 00000000..4aac26d4 --- /dev/null +++ b/drivers/staging/wlags49_h2/hcf.c @@ -0,0 +1,4748 @@ +/************************************************************************************************************ + * + * FILE : HCF.C + * + * DATE : $Date: 2004/08/05 11:47:10 $ $Revision: 1.10 $ + * Original: 2004/06/02 10:22:22 Revision: 1.85 Tag: hcf7_t20040602_01 + * Original: 2004/04/15 09:24:41 Revision: 1.63 Tag: hcf7_t7_20040415_01 + * Original: 2004/04/13 14:22:44 Revision: 1.62 Tag: t7_20040413_01 + * Original: 2004/04/01 15:32:55 Revision: 1.59 Tag: t7_20040401_01 + * Original: 2004/03/10 15:39:27 Revision: 1.55 Tag: t20040310_01 + * Original: 2004/03/04 11:03:37 Revision: 1.53 Tag: t20040304_01 + * Original: 2004/03/02 14:51:21 Revision: 1.50 Tag: t20040302_03 + * Original: 2004/02/24 13:00:27 Revision: 1.43 Tag: t20040224_01 + * Original: 2004/02/19 10:57:25 Revision: 1.39 Tag: t20040219_01 + * + * AUTHOR : Nico Valster + * + * SPECIFICATION: ........ + * + * DESCRIPTION : HCF Routines for Hermes-II (callable via the Wireless Connection I/F or WCI) + * Local Support Routines for above procedures + * + * Customizable via HCFCFG.H, which is included by HCF.H + * + ************************************************************************************************************* + * + * + * SOFTWARE LICENSE + * + * This software is provided subject to the following terms and conditions, + * which you should read carefully before using the software. Using this + * software indicates your acceptance of these terms and conditions. If you do + * not agree with these terms and conditions, do not use the software. + * + * COPYRIGHT © 1994 - 1995 by AT&T. All Rights Reserved + * COPYRIGHT © 1996 - 2000 by Lucent Technologies. All Rights Reserved + * COPYRIGHT © 2001 - 2004 by Agere Systems Inc. All Rights Reserved + * All rights reserved. + * + * Redistribution and use in source or binary forms, with or without + * modifications, are permitted provided that the following conditions are met: + * + * . Redistributions of source code must retain the above copyright notice, this + * list of conditions and the following Disclaimer as comments in the code as + * well as in the documentation and/or other materials provided with the + * distribution. + * + * . Redistributions in binary form must reproduce the above copyright notice, + * this list of conditions and the following Disclaimer in the documentation + * and/or other materials provided with the distribution. + * + * . Neither the name of Agere Systems Inc. nor the names of the contributors + * may be used to endorse or promote products derived from this software + * without specific prior written permission. + * + * Disclaimer + * + * THIS SOFTWARE IS PROVIDED "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, + * INCLUDING, BUT NOT LIMITED TO, INFRINGEMENT AND THE IMPLIED WARRANTIES OF + * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. ANY + * USE, MODIFICATION OR DISTRIBUTION OF THIS SOFTWARE IS SOLELY AT THE USERS OWN + * RISK. IN NO EVENT SHALL AGERE SYSTEMS INC. OR CONTRIBUTORS BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, INCLUDING, BUT NOT LIMITED TO, CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT + * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH + * DAMAGE. + * + * + ************************************************************************************************************/ + + +/************************************************************************************************************ + ** + ** Implementation Notes + ** + * - a leading marker of //! is used. The purpose of such a sequence is to help to understand the flow + * An example is: //!rc = HCF_SUCCESS; + * if this is superfluous because rc is already guaranteed to be 0 but it shows to the (maintenance) + * programmer it is an intentional omission at the place where someone could consider it most appropriate at + * first glance + * - using near pointers in a model where ss!=ds is an invitation for disaster, so be aware of how you specify + * your model and how you define variables which are used at interrupt time + * - remember that sign extension on 32 bit platforms may cause problems unless code is carefully constructed, + * e.g. use "(hcf_16)~foo" rather than "~foo" + * + ************************************************************************************************************/ + +#include "hcf.h" // HCF and MSF common include file +#include "hcfdef.h" // HCF specific include file +#include "mmd.h" // MoreModularDriver common include file +#include <linux/bug.h> +#include <linux/kernel.h> + +#if ! defined offsetof +#define offsetof(s,m) ((unsigned int)&(((s *)0)->m)) +#endif // offsetof + + +/***********************************************************************************************************/ +/*************************************** PROTOTYPES ******************************************************/ +/***********************************************************************************************************/ +HCF_STATIC int cmd_exe( IFBP ifbp, hcf_16 cmd_code, hcf_16 par_0 ); +HCF_STATIC int init( IFBP ifbp ); +HCF_STATIC int put_info( IFBP ifbp, LTVP ltvp ); +HCF_STATIC int put_info_mb( IFBP ifbp, CFG_MB_INFO_STRCT FAR * ltvp ); +#if (HCF_TYPE) & HCF_TYPE_WPA +HCF_STATIC void calc_mic( hcf_32* p, hcf_32 M ); +void calc_mic_rx_frag( IFBP ifbp, wci_bufp p, int len ); +void calc_mic_tx_frag( IFBP ifbp, wci_bufp p, int len ); +HCF_STATIC int check_mic( IFBP ifbp ); +#endif // HCF_TYPE_WPA + +HCF_STATIC void calibrate( IFBP ifbp ); +HCF_STATIC int cmd_cmpl( IFBP ifbp ); +HCF_STATIC hcf_16 get_fid( IFBP ifbp ); +HCF_STATIC void isr_info( IFBP ifbp ); +#if HCF_DMA +HCF_STATIC DESC_STRCT* get_frame_lst(IFBP ifbp, int tx_rx_flag); +#endif // HCF_DMA +HCF_STATIC void get_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ); //char*, byte count (usually even) +#if HCF_DMA +HCF_STATIC void put_frame_lst( IFBP ifbp, DESC_STRCT *descp, int tx_rx_flag ); +#endif // HCF_DMA +HCF_STATIC void put_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ); +HCF_STATIC void put_frag_finalize( IFBP ifbp ); +HCF_STATIC int setup_bap( IFBP ifbp, hcf_16 fid, int offset, int type ); +#if (HCF_ASSERT) & HCF_ASSERT_PRINTF +static int fw_printf(IFBP ifbp, CFG_FW_PRINTF_STRCT FAR *ltvp); +#endif // HCF_ASSERT_PRINTF + +HCF_STATIC int download( IFBP ifbp, CFG_PROG_STRCT FAR *ltvp ); +HCF_STATIC hcf_8 hcf_encap( wci_bufp type ); +HCF_STATIC hcf_8 null_addr[4] = { 0, 0, 0, 0 }; +#if ! defined IN_PORT_WORD //replace I/O Macros with logging facility +extern FILE *log_file; + +#define IN_PORT_WORD(port) in_port_word( (hcf_io)(port) ) + +static hcf_16 in_port_word( hcf_io port ) { + hcf_16 i = (hcf_16)_inpw( port ); + if ( log_file ) { + fprintf( log_file, "\nR %2.2x %4.4x", (port)&0xFF, i); + } + return i; +} // in_port_word + +#define OUT_PORT_WORD(port, value) out_port_word( (hcf_io)(port), (hcf_16)(value) ) + +static void out_port_word( hcf_io port, hcf_16 value ) { + _outpw( port, value ); + if ( log_file ) { + fprintf( log_file, "\nW %2.02x %4.04x", (port)&0xFF, value ); + } +} + +void IN_PORT_STRING_32( hcf_io prt, hcf_32 FAR * dst, int n) { + int i = 0; + hcf_16 FAR * p; + if ( log_file ) { + fprintf( log_file, "\nread string_32 length %04x (%04d) at port %02.2x to addr %lp", + (hcf_16)n, (hcf_16)n, (hcf_16)(prt)&0xFF, dst); + } + while ( n-- ) { + p = (hcf_16 FAR *)dst; + *p++ = (hcf_16)_inpw( prt ); + *p = (hcf_16)_inpw( prt ); + if ( log_file ) { + fprintf( log_file, "%s%08lx ", i++ % 0x08 ? " " : "\n", *dst); + } + dst++; + } +} // IN_PORT_STRING_32 + +void IN_PORT_STRING_8_16( hcf_io prt, hcf_8 FAR * dst, int n) { //also handles byte alignment problems + hcf_16 FAR * p = (hcf_16 FAR *)dst; //this needs more elaborate code in non-x86 platforms + int i = 0; + if ( log_file ) { + fprintf( log_file, "\nread string_16 length %04x (%04d) at port %02.2x to addr %lp", + (hcf_16)n, (hcf_16)n, (hcf_16)(prt)&0xFF, dst ); + } + while ( n-- ) { + *p =(hcf_16)_inpw( prt); + if ( log_file ) { + if ( i++ % 0x10 ) { + fprintf( log_file, "%04x ", *p); + } else { + fprintf( log_file, "\n%04x ", *p); + } + } + p++; + } +} // IN_PORT_STRING_8_16 + +void OUT_PORT_STRING_32( hcf_io prt, hcf_32 FAR * src, int n) { + int i = 0; + hcf_16 FAR * p; + if ( log_file ) { + fprintf( log_file, "\nwrite string_32 length %04x (%04d) at port %02.2x", + (hcf_16)n, (hcf_16)n, (hcf_16)(prt)&0xFF); + } + while ( n-- ) { + p = (hcf_16 FAR *)src; + _outpw( prt, *p++ ); + _outpw( prt, *p ); + if ( log_file ) { + fprintf( log_file, "%s%08lx ", i++ % 0x08 ? " " : "\n", *src); + } + src++; + } +} // OUT_PORT_STRING_32 + +void OUT_PORT_STRING_8_16( hcf_io prt, hcf_8 FAR * src, int n) { //also handles byte alignment problems + hcf_16 FAR * p = (hcf_16 FAR *)src; //this needs more elaborate code in non-x86 platforms + int i = 0; + if ( log_file ) { + fprintf( log_file, "\nwrite string_16 length %04x (%04d) at port %04x", n, n, (hcf_16)prt); + } + while ( n-- ) { + (void)_outpw( prt, *p); + if ( log_file ) { + if ( i++ % 0x10 ) { + fprintf( log_file, "%04x ", *p); + } else { + fprintf( log_file, "\n%04x ", *p); + } + } + p++; + } +} // OUT_PORT_STRING_8_16 + +#endif // IN_PORT_WORD + +/************************************************************************************************************ + ******************************* D A T A D E F I N I T I O N S ******************************************** + ************************************************************************************************************/ + +#if HCF_ASSERT +IFBP BASED assert_ifbp = NULL; //to make asserts easily work under MMD and DHF +#endif // HCF_ASSERT + +/* SNAP header to be inserted in Ethernet-II frames */ +HCF_STATIC hcf_8 BASED snap_header[] = { 0xAA, 0xAA, 0x03, 0x00, 0x00, //5 bytes signature + + 0 }; //1 byte protocol identifier + +#if (HCF_TYPE) & HCF_TYPE_WPA +HCF_STATIC hcf_8 BASED mic_pad[8] = { 0x5A, 0, 0, 0, 0, 0, 0, 0 }; //MIC padding of message +#endif // HCF_TYPE_WPA + +#if defined MSF_COMPONENT_ID +CFG_IDENTITY_STRCT BASED cfg_drv_identity = { + sizeof(cfg_drv_identity)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_IDENTITY, // (0x0826) + MSF_COMPONENT_ID, + MSF_COMPONENT_VAR, + MSF_COMPONENT_MAJOR_VER, + MSF_COMPONENT_MINOR_VER +} ; + +CFG_RANGES_STRCT BASED cfg_drv_sup_range = { + sizeof(cfg_drv_sup_range)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_SUP_RANGE, // (0x0827) + + COMP_ROLE_SUPL, + COMP_ID_DUI, + {{ DUI_COMPAT_VAR, + DUI_COMPAT_BOT, + DUI_COMPAT_TOP + }} +} ; + +struct CFG_RANGE3_STRCT BASED cfg_drv_act_ranges_pri = { + sizeof(cfg_drv_act_ranges_pri)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_ACT_RANGES_PRI, // (0x0828) + + COMP_ROLE_ACT, + COMP_ID_PRI, + { + { 0, 0, 0 }, // HCF_PRI_VAR_1 not supported by HCF 7 + { 0, 0, 0 }, // HCF_PRI_VAR_2 not supported by HCF 7 + { 3, //var_rec[2] - Variant number + CFG_DRV_ACT_RANGES_PRI_3_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_PRI_3_TOP // - Top Compatibility + } + } +} ; + + +struct CFG_RANGE4_STRCT BASED cfg_drv_act_ranges_sta = { + sizeof(cfg_drv_act_ranges_sta)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_ACT_RANGES_STA, // (0x0829) + + COMP_ROLE_ACT, + COMP_ID_STA, + { +#if defined HCF_STA_VAR_1 + { 1, //var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_STA_1_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_STA_1_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_STA_VAR_1 +#if defined HCF_STA_VAR_2 + { 2, //var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_STA_2_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_STA_2_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_STA_VAR_2 +// For Native_USB (Not used!) +#if defined HCF_STA_VAR_3 + { 3, //var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_STA_3_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_STA_3_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_STA_VAR_3 +// Warp +#if defined HCF_STA_VAR_4 + { 4, //var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_STA_4_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_STA_4_TOP // - Top Compatibility + } +#else + { 0, 0, 0 } +#endif // HCF_STA_VAR_4 + } +} ; + + +struct CFG_RANGE6_STRCT BASED cfg_drv_act_ranges_hsi = { + sizeof(cfg_drv_act_ranges_hsi)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_ACT_RANGES_HSI, // (0x082A) + COMP_ROLE_ACT, + COMP_ID_HSI, + { +#if defined HCF_HSI_VAR_0 // Controlled deployment + { 0, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_HSI_0_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_HSI_0_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_HSI_VAR_0 + { 0, 0, 0 }, // HCF_HSI_VAR_1 not supported by HCF 7 + { 0, 0, 0 }, // HCF_HSI_VAR_2 not supported by HCF 7 + { 0, 0, 0 }, // HCF_HSI_VAR_3 not supported by HCF 7 +#if defined HCF_HSI_VAR_4 // Hermes-II all types + { 4, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_HSI_4_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_HSI_4_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_HSI_VAR_4 +#if defined HCF_HSI_VAR_5 // WARP Hermes-2.5 + { 5, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_HSI_5_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_HSI_5_TOP // - Top Compatibility + } +#else + { 0, 0, 0 } +#endif // HCF_HSI_VAR_5 + } +} ; + + +CFG_RANGE4_STRCT BASED cfg_drv_act_ranges_apf = { + sizeof(cfg_drv_act_ranges_apf)/sizeof(hcf_16) - 1, //length of RID + CFG_DRV_ACT_RANGES_APF, // (0x082B) + + COMP_ROLE_ACT, + COMP_ID_APF, + { +#if defined HCF_APF_VAR_1 //(Fake) Hermes-I + { 1, //var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_APF_1_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_APF_1_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_APF_VAR_1 +#if defined HCF_APF_VAR_2 //Hermes-II + { 2, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_APF_2_BOTTOM, // - Bottom Compatibility + CFG_DRV_ACT_RANGES_APF_2_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_APF_VAR_2 +#if defined HCF_APF_VAR_3 // Native_USB + { 3, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_APF_3_BOTTOM, // - Bottom Compatibility !!!!!see note below!!!!!!! + CFG_DRV_ACT_RANGES_APF_3_TOP // - Top Compatibility + }, +#else + { 0, 0, 0 }, +#endif // HCF_APF_VAR_3 +#if defined HCF_APF_VAR_4 // WARP Hermes 2.5 + { 4, // var_rec[1] - Variant number + CFG_DRV_ACT_RANGES_APF_4_BOTTOM, // - Bottom Compatibility !!!!!see note below!!!!!!! + CFG_DRV_ACT_RANGES_APF_4_TOP // - Top Compatibility + } +#else + { 0, 0, 0 } +#endif // HCF_APF_VAR_4 + } +} ; +#define HCF_VERSION TEXT( "HCF$Revision: 1.10 $" ) + +static struct /*CFG_HCF_OPT_STRCT*/ { + hcf_16 len; //length of cfg_hcf_opt struct + hcf_16 typ; //type 0x082C + hcf_16 v0; //offset HCF_VERSION + hcf_16 v1; // MSF_COMPONENT_ID + hcf_16 v2; // HCF_ALIGN + hcf_16 v3; // HCF_ASSERT + hcf_16 v4; // HCF_BIG_ENDIAN + hcf_16 v5; // /* HCF_DLV | HCF_DLNV */ + hcf_16 v6; // HCF_DMA + hcf_16 v7; // HCF_ENCAP + hcf_16 v8; // HCF_EXT + hcf_16 v9; // HCF_INT_ON + hcf_16 v10; // HCF_IO + hcf_16 v11; // HCF_LEGACY + hcf_16 v12; // HCF_MAX_LTV + hcf_16 v13; // HCF_PROT_TIME + hcf_16 v14; // HCF_SLEEP + hcf_16 v15; // HCF_TALLIES + hcf_16 v16; // HCF_TYPE + hcf_16 v17; // HCF_NIC_TAL_CNT + hcf_16 v18; // HCF_HCF_TAL_CNT + hcf_16 v19; // offset tallies + char val[sizeof(HCF_VERSION)]; +} BASED cfg_hcf_opt = { + sizeof(cfg_hcf_opt)/sizeof(hcf_16) -1, + CFG_HCF_OPT, // (0x082C) + ( sizeof(cfg_hcf_opt) - sizeof(HCF_VERSION) - 4 )/sizeof(hcf_16), +#if defined MSF_COMPONENT_ID + MSF_COMPONENT_ID, +#else + 0, +#endif // MSF_COMPONENT_ID + HCF_ALIGN, + HCF_ASSERT, + HCF_BIG_ENDIAN, + 0, // /* HCF_DLV | HCF_DLNV*/, + HCF_DMA, + HCF_ENCAP, + HCF_EXT, + HCF_INT_ON, + HCF_IO, + HCF_LEGACY, + HCF_MAX_LTV, + HCF_PROT_TIME, + HCF_SLEEP, + HCF_TALLIES, + HCF_TYPE, +#if (HCF_TALLIES) & ( HCF_TALLIES_NIC | HCF_TALLIES_HCF ) + HCF_NIC_TAL_CNT, + HCF_HCF_TAL_CNT, + offsetof(IFB_STRCT, IFB_TallyLen ), +#else + 0, 0, 0, +#endif // HCF_TALLIES_NIC / HCF_TALLIES_HCF + HCF_VERSION +}; // cfg_hcf_opt +#endif // MSF_COMPONENT_ID + +HCF_STATIC LTV_STRCT BASED cfg_null = { 1, CFG_NULL, {0} }; + +HCF_STATIC hcf_16* BASED xxxx[ ] = { + &cfg_null.len, //CFG_NULL 0x0820 +#if defined MSF_COMPONENT_ID + &cfg_drv_identity.len, //CFG_DRV_IDENTITY 0x0826 + &cfg_drv_sup_range.len, //CFG_DRV_SUP_RANGE 0x0827 + &cfg_drv_act_ranges_pri.len, //CFG_DRV_ACT_RANGES_PRI 0x0828 + &cfg_drv_act_ranges_sta.len, //CFG_DRV_ACT_RANGES_STA 0x0829 + &cfg_drv_act_ranges_hsi.len, //CFG_DRV_ACT_RANGES_HSI 0x082A + &cfg_drv_act_ranges_apf.len, //CFG_DRV_ACT_RANGES_APF 0x082B + &cfg_hcf_opt.len, //CFG_HCF_OPT 0x082C + NULL, //IFB_PRIIdentity placeholder 0xFD02 + NULL, //IFB_PRISup placeholder 0xFD03 +#endif // MSF_COMPONENT_ID + NULL //endsentinel +}; +#define xxxx_PRI_IDENTITY_OFFSET (ARRAY_SIZE(xxxx) - 3) + + +/************************************************************************************************************ + ************************** T O P L E V E L H C F R O U T I N E S ************************************** + ************************************************************************************************************/ + +/************************************************************************************************************ + * + *.MODULE int hcf_action( IFBP ifbp, hcf_16 action ) + *.PURPOSE Changes the run-time Card behavior. + * Performs Miscellanuous actions. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * action number identifying the type of change + * - HCF_ACT_INT_FORCE_ON enable interrupt generation by WaveLAN NIC + * - HCF_ACT_INT_OFF disable interrupt generation by WaveLAN NIC + * - HCF_ACT_INT_ON compensate 1 HCF_ACT_INT_OFF, enable interrupt generation if balance reached + * - HCF_ACT_PRS_SCAN Hermes Probe Response Scan (F102) command + * - HCF_ACT_RX_ACK acknowledge non-DMA receiver to Hermes + * - HCF_ACT_SCAN Hermes Inquire Scan (F101) command (non-WARP only) + * - HCF_ACT_SLEEP DDS Sleep request + * - HCF_ACT_TALLIES Hermes Inquire Tallies (F100) command + * + *.RETURNS + * HCF_SUCCESS all (including invalid) + * HCF_INT_PENDING HCF_ACT_INT_OFF, interrupt pending + * HCF_ERR_NO_NIC HCF_ACT_INT_OFF, NIC presence check fails + * + *.CONDITIONS + * Except for hcf_action with HCF_ACT_INT_FORCE_ON or HCF_ACT_INT_OFF as parameter or hcf_connect with an I/O + * address (i.e. not HCF_DISCONNECT), all hcf-function calls MUST be preceded by a call of hcf_action with + * HCF_ACT_INT_OFF as parameter. + * Note that hcf_connect defaults to NIC interrupt disabled mode, i.e. as if hcf_action( HCF_ACT_INT_OFF ) + * was called. + * + *.DESCRIPTION + * hcf_action supports the following mode changing action-code pairs that are antonyms + * - HCF_ACT_INT_[FORCE_]ON / HCF_ACT_INT_OFF + * + * Additionally hcf_action can start the following actions in the NIC: + * - HCF_ACT_PRS_SCAN + * - HCF_ACT_RX_ACK + * - HCF_ACT_SCAN + * - HCF_ACT_SLEEP + * - HCF_ACT_TALLIES + * + * o HCF_ACT_INT_OFF: Sets NIC Interrupts mode Disabled. + * This command, and the associated [Force] Enable NIC interrupts command, are only available if the HCF_INT_ON + * compile time option is not set at 0x0000. + * + * o HCF_ACT_INT_ON: Sets NIC Interrupts mode Enabled. + * Enable NIC Interrupts, depending on the number of preceding Disable NIC Interrupt calls. + * + * o HCF_ACT_INT_FORCE_ON: Force NIC Interrupts mode Enabled. + * Sets NIC Interrupts mode Enabled, regardless off the number of preceding Disable NIC Interrupt calls. + * + * The disabling and enabling of interrupts are antonyms. + * These actions must be balanced. + * For each "disable interrupts" there must be a matching "enable interrupts". + * The disable interrupts may be executed multiple times in a row without intervening enable interrupts, in + * other words, the disable interrupts may be nested. + * The interrupt generation mechanism is disabled at the first call with HCF_ACT_INT_OFF. + * The interrupt generation mechanism is re-enabled when the number of calls with HCF_ACT_INT_ON matches the + * number of calls with INT_OFF. + * + * It is not allowed to have more Enable NIC Interrupts calls than Disable NIC Interrupts calls. + * The interrupt generation mechanism is initially (i.e. after hcf_connect) disabled. + * An MSF based on a interrupt strategy must call hcf_action with INT_ON in its initialization logic. + * + *! The INT_OFF/INT_ON housekeeping is initialized at 0x0000 by hcf_connect, causing the interrupt generation + * mechanism to be disabled at first. This suits MSF implementation based on a polling strategy. + * + * o HCF_ACT_SLEEP: Initiates the Disconnected DeepSleep process + * This command is only available if the HCF_DDS compile time option is set. It triggers the F/W to start the + * sleep handshaking. Regardless whether the Host initiates a Disconnected DeepSleep (DDS) or the F/W initiates + * a Connected DeepSleep (CDS), the Host-F/W sleep handshaking is completed when the NIC Interrupts mode is + * enabled (by means of the balancing HCF_ACT_INT_ON), i.e. at that moment the F/W really goes into sleep mode. + * The F/W is wokenup by the HCF when the NIC Interrupts mode are disabled, i.e. at the first HCF_ACT_INT_OFF + * after going into sleep. + * + * The following Miscellaneous actions are defined: + * + * o HCF_ACT_RX_ACK: Receiver Acknowledgement (non-DMA, non-USB mode only) + * Acking the receiver, frees the NIC memory used to hold the Rx frame and allows the F/W to + * report the existence of the next Rx frame. + * If the MSF does not need access (any longer) to the current frame, e.g. because it is rejected based on the + * look ahead or copied to another buffer, the receiver may be acked. Acking earlier is assumed to have the + * potential of improving the performance. + * If the MSF does not explicitly ack the receiver, the acking is done implicitly if: + * - the received frame fits in the look ahead buffer, by the hcf_service_nic call that reported the Rx frame + * - if not in the above step, by hcf_rcv_msg (assuming hcf_rcv_msg is called) + * - if neither of the above implicit acks nor an explicit ack by the MSF, by the first hcf_service_nic after + * the hcf_service_nic that reported the Rx frame. + * Note: If an Rx frame is already acked, an explicit ACK by the MSF acts as a NoOperation. + * + * o HCF_ACT_TALLIES: Inquire Tallies command + * This command is only operational if the F/W is enabled. + * The Inquire Tallies command requests the F/W to provide its current set of tallies. + * See also hcf_get_info with CFG_TALLIES as parameter. + * + * o HCF_ACT_PRS_SCAN: Inquire Probe Response Scan command + * This command is only operational if the F/W is enabled. + * The Probe Response Scan command starts a scan sequence. + * The HCF puts the result of this action in an MSF defined buffer (see CFG_RID_LOG_STRCT). + * + * o HCF_ACT_SCAN: Inquire Scan command + * This command is only supported for HII F/W (i.e. pre-WARP) and it is operational if the F/W is enabled. + * The Inquire Scan command starts a scan sequence. + * The HCF puts the result of this action in an MSF defined buffer (see CFG_RID_LOG_STRCT). + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - NIC interrupts are not disabled while required by parameter action. + * - an invalid code is specified in parameter action. + * - HCF_ACT_INT_ON commands outnumber the HCF_ACT_INT_OFF commands. + * - reentrancy, may be caused by calling hcf_functions without adequate protection against NIC interrupts or + * multi-threading + * + * - Since the HCF does not maintain status information relative to the F/W enabled state, it is not asserted + * whether HCF_ACT_SCAN, HCF_ACT_PRS_SCAN or HCF_ACT_TALLIES are only used while F/W is enabled. + * + *.DIAGRAM + * 0: The assert embedded in HCFLOGENTRY checks against re-entrancy. Re-entrancy could be caused by a MSF logic + * at task-level calling hcf_functions without shielding with HCF_ACT_ON/_OFF. However the HCF_ACT_INT_OFF + * action itself can per definition not be protected this way. Based on code inspection, it can be concluded, + * that there is no re-entrancy PROBLEM in this particular flow. It does not seem worth the trouble to + * explicitly check for this condition (although there was a report of an MSF which ran into this assert. + * 2:IFB_IntOffCnt is used to balance the INT_OFF and INT_ON calls. Disabling of the interrupts is achieved by + * writing a zero to the Hermes IntEn register. In a shared interrupt environment (e.g. the mini-PCI NDIS + * driver) it is considered more correct to return the status HCF_INT_PENDING if and only if, the current + * invocation of hcf_service_nic is (apparently) called in the ISR when the ISR was activated as result of a + * change in HREG_EV_STAT matching a bit in HREG_INT_EN, i.e. not if invoked as result of another device + * generating an interrupt on the shared interrupt line. + * Note 1: it has been observed that under certain adverse conditions on certain platforms the writing of + * HREG_INT_EN can apparently fail, therefore it is paramount that HREG_INT_EN is written again with 0 for + * each and every call to HCF_ACT_INT_OFF. + * Note 2: it has been observed that under certain H/W & S/W architectures this logic is called when there is + * no NIC at all. To cater for this, the value of HREG_INT_EN is validated. If the unused bit 0x0100 is set, + * it is assumed there is no NIC. + * Note 3: During the download process, some versions of the F/W reset HREG_SW_0, hence checking this + * register for HCF_MAGIC (the classical NIC presence test) when HCF_ACT_INT_OFF is called due to another + * card interrupting via a shared IRQ during a download, fails. + *4: The construction "if ( ifbp->IFB_IntOffCnt-- == 0 )" is optimal (in the sense of shortest/quickest + * path in error free flows) but NOT fail safe in case of too many INT_ON invocations compared to INT_OFF). + * Enabling of the interrupts is achieved by writing the Hermes IntEn register. + * - If the HCF is in Defunct mode, the interrupts stay disabled. + * - Under "normal" conditions, the HCF is only interested in Info Events, Rx Events and Notify Events. + * - When the HCF is out of Tx/Notify resources, the HCF is also interested in Alloc Events. + * - via HCF_EXT, the MSF programmer can also request HREG_EV_TICK and/or HREG_EV_TX_EXC interrupts. + * For DMA operation, the DMA hardware handles the alloc events. The DMA engine will generate a 'TxDmaDone' + * event as soon as it has pumped a frame from host ram into NIC-RAM (note that the frame does not have to be + * transmitted then), and a 'RxDmaDone' event as soon as a received frame has been pumped from NIC-RAM into + * host ram. Note that the 'alloc' event has been removed from the event-mask, because the DMA engine will + * react to and acknowledge this event. + *6: ack the "old" Rx-event. See "Rx Buffer free strategy" in hcf_service_nic above for more explanation. + * IFB_RxFID and IFB_RxLen must be cleared to bring both the internal HCF house keeping and the information + * supplied to the MSF in the state "no frame received". + *8: The HCF_ACT_SCAN, HCF_ACT_PRS_SCAN and HCF_ACT_TALLIES activity are merged by "clever" algebraic + * manipulations of the RID-values and action codes, so foregoing robustness against migration problems for + * ease of implementation. The assumptions about numerical relationships between CFG_TALLIES etc and + * HCF_ACT_TALLIES etc are checked by the "#if" statements just prior to the body of this routine, resulting + * in: err "maintenance" during compilation if the assumptions are no longer met. The writing of HREG_PARAM_1 + * with 0x3FFF in case of an PRS scan, is a kludge to get around lack of specification, hence different + * implementation in F/W and Host. + * When there is no NIC RAM available, some versions of the Hermes F/W do report 0x7F00 as error in the + * Result field of the Status register and some F/W versions don't. To mask this difference to the MSF all + * return codes of the Hermes are ignored ("best" and "most simple" solution to these types of analomies with + * an acceptable loss due to ignoring all error situations as well). + * The "No inquire space" is reported via the Hermes tallies. + *30: do not HCFASSERT( rc, rc ) since rc == HCF_INT_PENDING is no error + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +#if ( (HCF_TYPE) & HCF_TYPE_HII5 ) == 0 +#if CFG_SCAN != CFG_TALLIES - HCF_ACT_TALLIES + HCF_ACT_SCAN +err: "maintenance" apparently inviolated the underlying assumption about the numerical values of these macros +#endif +#endif // HCF_TYPE_HII5 +#if CFG_PRS_SCAN != CFG_TALLIES - HCF_ACT_TALLIES + HCF_ACT_PRS_SCAN +err: "maintenance" apparently inviolated the underlying assumption about the numerical values of these macros +#endif +int +hcf_action( IFBP ifbp, hcf_16 action ) +{ + int rc = HCF_SUCCESS; + + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); +#if HCF_INT_ON + HCFLOGENTRY( action == HCF_ACT_INT_FORCE_ON ? HCF_TRACE_ACTION_KLUDGE : HCF_TRACE_ACTION, action ); /* 0 */ +#if (HCF_SLEEP) + HCFASSERT( ifbp->IFB_IntOffCnt != 0xFFFE || action == HCF_ACT_INT_OFF, + MERGE_2( action, ifbp->IFB_IntOffCnt ) ); +#else + HCFASSERT( ifbp->IFB_IntOffCnt != 0xFFFE, action ); +#endif // HCF_SLEEP + HCFASSERT( ifbp->IFB_IntOffCnt != 0xFFFF || + action == HCF_ACT_INT_OFF || action == HCF_ACT_INT_FORCE_ON, action ); + HCFASSERT( ifbp->IFB_IntOffCnt <= 16 || ifbp->IFB_IntOffCnt >= 0xFFFE, + MERGE_2( action, ifbp->IFB_IntOffCnt ) ); //nesting more than 16 deep seems unreasonable +#endif // HCF_INT_ON + + switch (action) { +#if HCF_INT_ON + hcf_16 i; + case HCF_ACT_INT_OFF: // Disable Interrupt generation +#if HCF_SLEEP + if ( ifbp->IFB_IntOffCnt == 0xFFFE ) { // WakeUp test ;?tie this to the "new" super-LinkStat + ifbp->IFB_IntOffCnt++; // restore conventional I/F + OPW(HREG_IO, HREG_IO_WAKEUP_ASYNC ); // set wakeup bit + OPW(HREG_IO, HREG_IO_WAKEUP_ASYNC ); // set wakeup bit to counteract the clearing by F/W + // 800 us latency before FW switches to high power + MSF_WAIT(800); // MSF-defined function to wait n microseconds. +//OOR if ( ifbp->IFB_DSLinkStat & CFG_LINK_STAT_DS_OOR ) { // OutOfRange +// printk(KERN_NOTICE "ACT_INT_OFF: Deepsleep phase terminated, enable and go to AwaitConnection\n" ); //;?remove me 1 day +// hcf_cntl( ifbp, HCF_CNTL_ENABLE ); +// } +// ifbp->IFB_DSLinkStat &= ~( CFG_LINK_STAT_DS_IR | CFG_LINK_STAT_DS_OOR); //clear IR/OOR state + } +#endif // HCF_SLEEP + /*2*/ ifbp->IFB_IntOffCnt++; +//! rc = 0; + i = IPW( HREG_INT_EN ); + OPW( HREG_INT_EN, 0 ); + if ( i & 0x1000 ) { + rc = HCF_ERR_NO_NIC; + } else { + if ( i & IPW( HREG_EV_STAT ) ) { + rc = HCF_INT_PENDING; + } + } + break; + + case HCF_ACT_INT_FORCE_ON: // Enforce Enable Interrupt generation + ifbp->IFB_IntOffCnt = 0; + //Fall through in HCF_ACT_INT_ON + + case HCF_ACT_INT_ON: // Enable Interrupt generation + /*4*/ if ( ifbp->IFB_IntOffCnt-- == 0 && ifbp->IFB_CardStat == 0 ) { + //determine Interrupt Event mask +#if HCF_DMA + if ( ifbp->IFB_CntlOpt & USE_DMA ) { + i = HREG_EV_INFO | HREG_EV_RDMAD | HREG_EV_TDMAD | HREG_EV_TX_EXT; //mask when DMA active + } else +#endif // HCF_DMA + { + i = HREG_EV_INFO | HREG_EV_RX | HREG_EV_TX_EXT; //mask when DMA not active + if ( ifbp->IFB_RscInd == 0 ) { + i |= HREG_EV_ALLOC; //mask when no TxFID available + } + } +#if HCF_SLEEP + if ( ( IPW(HREG_EV_STAT) & ( i | HREG_EV_SLEEP_REQ ) ) == HREG_EV_SLEEP_REQ ) { + // firmware indicates it would like to go into sleep modus + // only acknowledge this request if no other events that can cause an interrupt are pending + ifbp->IFB_IntOffCnt--; //becomes 0xFFFE + OPW( HREG_INT_EN, i | HREG_EV_TICK ); + OPW( HREG_EV_ACK, HREG_EV_SLEEP_REQ | HREG_EV_TICK | HREG_EV_ACK_REG_READY ); + } else +#endif // HCF_SLEEP + { + OPW( HREG_INT_EN, i | HREG_EV_SLEEP_REQ ); + } + } + break; +#endif // HCF_INT_ON + +#if (HCF_SLEEP) & HCF_DDS + case HCF_ACT_SLEEP: // DDS Sleep request + hcf_cntl( ifbp, HCF_CNTL_DISABLE ); + cmd_exe( ifbp, HCMD_SLEEP, 0 ); + break; +// case HCF_ACT_WAKEUP: // DDS Wakeup request +// HCFASSERT( ifbp->IFB_IntOffCnt == 0xFFFE, ifbp->IFB_IntOffCnt ); +// ifbp->IFB_IntOffCnt++; // restore conventional I/F +// OPW( HREG_IO, HREG_IO_WAKEUP_ASYNC ); +// MSF_WAIT(800); // MSF-defined function to wait n microseconds. +// rc = hcf_action( ifbp, HCF_ACT_INT_OFF ); /*bogus, IFB_IntOffCnt == 0xFFFF, so if you carefully look +// *at the #if HCF_DDS statements, HCF_ACT_INT_OFF is empty +// *for DDS. "Much" better would be to merge the flows for +// *DDS and DEEP_SLEEP +// */ +// break; +#endif // HCF_DDS + + case HCF_ACT_RX_ACK: //Receiver ACK + /*6*/ if ( ifbp->IFB_RxFID ) { + DAWA_ACK( HREG_EV_RX ); + } + ifbp->IFB_RxFID = ifbp->IFB_RxLen = 0; + break; + + /*8*/ case HCF_ACT_PRS_SCAN: // Hermes PRS Scan (F102) + OPW( HREG_PARAM_1, 0x3FFF ); + //Fall through in HCF_ACT_TALLIES + case HCF_ACT_TALLIES: // Hermes Inquire Tallies (F100) +#if ( (HCF_TYPE) & HCF_TYPE_HII5 ) == 0 + case HCF_ACT_SCAN: // Hermes Inquire Scan (F101) +#endif // HCF_TYPE_HII5 + /*!! the assumptions about numerical relationships between CFG_TALLIES etc and HCF_ACT_TALLIES etc + * are checked by #if statements just prior to this routine resulting in: err "maintenance" */ + cmd_exe( ifbp, HCMD_INQUIRE, action - HCF_ACT_TALLIES + CFG_TALLIES ); + break; + + default: + HCFASSERT( DO_ASSERT, action ); + break; + } + //! do not HCFASSERT( rc == HCF_SUCCESS, rc ) /* 30*/ + HCFLOGEXIT( HCF_TRACE_ACTION ); + return rc; +} // hcf_action + + +/************************************************************************************************************ + * + *.MODULE int hcf_cntl( IFBP ifbp, hcf_16 cmd ) + *.PURPOSE Connect or disconnect a specific port to a specific network. + *!! ;???????????????? continue needs more explanation + * recovers by means of "continue" when the connect process in CCX mode fails + * Enables or disables data transmission and reception for the NIC. + * Activates static NIC configuration for a specific port at connect. + * Activates static configuration for all ports at enable. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * cmd 0x001F: Hermes command (disable, enable, connect, disconnect, continue) + * HCF_CNTL_ENABLE Enable + * HCF_CNTL_DISABLE Disable + * HCF_CNTL_CONTINUE Continue + * HCF_CNTL_CONNECT Connect + * HCF_CNTL_DISCONNECT Disconnect + * 0x0100: command qualifier (continue) + * HCMD_RETRY retry flag + * 0x0700: port number (connect/disconnect) + * HCF_PORT_0 MAC Port 0 + * HCF_PORT_1 MAC Port 1 + * HCF_PORT_2 MAC Port 2 + * HCF_PORT_3 MAC Port 3 + * HCF_PORT_4 MAC Port 4 + * HCF_PORT_5 MAC Port 5 + * HCF_PORT_6 MAC Port 6 + * + *.RETURNS + * HCF_SUCCESS + *!! via cmd_exe + * HCF_ERR_NO_NIC + * HCF_ERR_DEFUNCT_... + * HCF_ERR_TIME_OUT + * + *.DESCRIPTION + * The parameter cmd contains a number of subfields. + * The actual value for cmd is created by logical or-ing the appropriate mnemonics for the subfields. + * The field 0x001F contains the command code + * - HCF_CNTL_ENABLE + * - HCF_CNTL_DISABLE + * - HCF_CNTL_CONNECT + * - HCF_CNTL_DISCONNECT + * - HCF_CNTL_CONTINUE + * + * For HCF_CNTL_CONTINUE, the field 0x0100 contains the retry flag HCMD_RETRY. + * For HCF_CNTL_CONNECT and HCF_CNTL_DISCONNECT, the field 0x0700 contains the port number as HCF_PORT_#. + * For Station as well as AccessPoint F/W, MAC Port 0 is the "normal" communication channel. + * For AccessPoint F/W, MAC Port 1 through 6 control the WDS links. + * + * Note that despite the names HCF_CNTL_DISABLE and HCF_CNTL_ENABLE, hcf_cntl does not influence the NIC + * Interrupts mode. + * + * The Connect is used by the MSF to bring a particular port in an inactive state as far as data transmission + * and reception are concerned. + * When a particular port is disconnected: + * - the F/W disables the receiver for that port. + * - the F/W ignores send commands for that port. + * - all frames (Receive as well as pending Transmit) for that port on the NIC are discarded. + * + * When the NIC is disabled, above list applies to all ports, i.e. the result is like all ports are + * disconnected. + * + * When a particular port is connected: + * - the F/W effectuates the static configuration for that port. + * - enables the receiver for that port. + * - accepts send commands for that port. + * + * Enabling has the following effects: + * - the F/W effectuates the static configuration for all ports. + * The F/W only updates its static configuration at a transition from disabled to enabled or from + * disconnected to connected. + * In order to enforce the static configuration, the MSF must assure that such a transition takes place. + * Due to such a disable/enable or disconnect/connect sequence, Rx/Tx frames may be lost, in other words, + * configuration may impact communication. + * - The DMA Engine (if applicable) is enabled. + * Note that the Enable Function by itself only enables data transmission and reception, it + * does not enable the Interrupt Generation mechanism. This is done by hcf_action. + * + * Disabling has the following effects: + *!! ;?????is the following statement really true + * - it acts as a disconnect on all ports. + * - The DMA Engine (if applicable) is disabled. + * + * For impact of the disable command on the behavior of hcf_dma_tx/rx_get see the appropriate sections. + * + * Although the Enable/Disable and Connect/Disconnect are antonyms, there is no restriction on their sequencing, + * in other words, they may be called multiple times in arbitrary sequence without being paired or balanced. + * Each time one of these functions is called, the effects of the preceding calls cease. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - NIC interrupts are not disabled. + * - A command other than Continue, Enable, Disable, Connect or Disconnect is given. + * - An invalid combination of the subfields is given or a bit outside the subfields is given. + * - any return code besides HCF_SUCCESS. + * - reentrancy, may be caused by calling a hcf_function without adequate protection against NIC interrupts or + * multi-threading + * + *.DIAGRAM + * hcf_cntl takes successively the following actions: + *2: If the HCF is in Defunct mode or incompatible with the Primary or Station Supplier in the Hermes, + * hcf_cntl() returns immediately with HCF_ERR_NO_NIC;? as status. + *8: when the port is disabled, the DMA engine needs to be de-activated, so the host can safely reclaim tx + * packets from the tx descriptor chain. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +hcf_cntl( IFBP ifbp, hcf_16 cmd ) +{ + int rc = HCF_ERR_INCOMP_FW; +#if HCF_ASSERT + { int x = cmd & HCMD_CMD_CODE; + if ( x == HCF_CNTL_CONTINUE ) x &= ~HCMD_RETRY; + else if ( (x == HCMD_DISABLE || x == HCMD_ENABLE) && ifbp->IFB_FWIdentity.comp_id == COMP_ID_FW_AP ) { + x &= ~HFS_TX_CNTL_PORT; + } + HCFASSERT( x==HCF_CNTL_ENABLE || x==HCF_CNTL_DISABLE || HCF_CNTL_CONTINUE || + x==HCF_CNTL_CONNECT || x==HCF_CNTL_DISCONNECT, cmd ); + } +#endif // HCF_ASSERT +// #if (HCF_SLEEP) & HCF_DDS +// HCFASSERT( ifbp->IFB_IntOffCnt != 0xFFFE, cmd ); +// #endif // HCF_DDS + HCFLOGENTRY( HCF_TRACE_CNTL, cmd ); + if ( ifbp->IFB_CardStat == 0 ) { /*2*/ + /*6*/ rc = cmd_exe( ifbp, cmd, 0 ); +#if (HCF_SLEEP) & HCF_DDS + ifbp->IFB_TickCnt = 0; //start 2 second period (with 1 tick uncertanty) +#endif // HCF_DDS + } +#if HCF_DMA + //!rlav : note that this piece of code is always executed, regardless of the DEFUNCT bit in IFB_CardStat. + // The reason behind this is that the MSF should be able to get all its DMA resources back from the HCF, + // even if the hardware is disfunctional. Practical example under Windows : surprise removal. + if ( ifbp->IFB_CntlOpt & USE_DMA ) { + hcf_io io_port = ifbp->IFB_IOBase; + DESC_STRCT *p; + if ( cmd == HCF_CNTL_DISABLE || cmd == HCF_CNTL_ENABLE ) { + OUT_PORT_DWORD( (io_port + HREG_DMA_CTRL), DMA_CTRLSTAT_RESET); /*8*/ + ifbp->IFB_CntlOpt &= ~DMA_ENABLED; + } + if ( cmd == HCF_CNTL_ENABLE ) { + OUT_PORT_DWORD( (io_port + HREG_DMA_CTRL), DMA_CTRLSTAT_GO); + /* ;? by rewriting hcf_dma_rx_put you can probably just call hcf_dma_rx_put( ifbp->IFB_FirstDesc[DMA_RX] ) + * as additional beneficiary side effect, the SOP and EOP bits will also be cleared + */ + ifbp->IFB_CntlOpt |= DMA_ENABLED; + HCFASSERT( NT_ASSERT, NEVER_TESTED ); + // make the entire rx descriptor chain DMA-owned, so the DMA engine can (re-)use it. + p = ifbp->IFB_FirstDesc[DMA_RX]; + if (p != NULL) { //;? Think this over again in the light of the new chaining strategy + if ( 1 ) { //begin alternative + HCFASSERT( NT_ASSERT, NEVER_TESTED ); + put_frame_lst( ifbp, ifbp->IFB_FirstDesc[DMA_RX], DMA_RX ); + if ( ifbp->IFB_FirstDesc[DMA_RX] ) { + put_frame_lst( ifbp, ifbp->IFB_FirstDesc[DMA_RX]->next_desc_addr, DMA_RX ); + } + } else { + while ( p ) { + //p->buf_cntl.cntl_stat |= DESC_DMA_OWNED; + p->BUF_CNT |= DESC_DMA_OWNED; + p = p->next_desc_addr; + } + // a rx chain is available so hand it over to the DMA engine + p = ifbp->IFB_FirstDesc[DMA_RX]; + OUT_PORT_DWORD( (io_port + HREG_RXDMA_PTR32), p->desc_phys_addr); + } //end alternative + } + } + } +#endif // HCF_DMA + HCFASSERT( rc == HCF_SUCCESS, rc ); + HCFLOGEXIT( HCF_TRACE_CNTL ); + return rc; +} // hcf_cntl + + +/************************************************************************************************************ + * + *.MODULE int hcf_connect( IFBP ifbp, hcf_io io_base ) + *.PURPOSE Grants access right for the HCF to the IFB. + * Initializes Card and HCF housekeeping. + * + *.ARGUMENTS + * ifbp (near) address of the Interface Block + * io_base non-USB: I/O Base address of the NIC (connect) + * non-USB: HCF_DISCONNECT + * USB: HCF_CONNECT, HCF_DISCONNECT + * + *.RETURNS + * HCF_SUCCESS + * HCF_ERR_INCOMP_PRI + * HCF_ERR_INCOMP_FW + * HCF_ERR_DEFUNCT_CMD_SEQ + *!! HCF_ERR_NO_NIC really returned ;? + * HCF_ERR_NO_NIC + * HCF_ERR_TIME_OUT + * + * MSF-accessible fields of Result Block: + * IFB_IOBase entry parameter io_base + * IFB_IORange HREG_IO_RANGE (0x40/0x80) + * IFB_Version version of the IFB layout + * IFB_FWIdentity CFG_FW_IDENTITY_STRCT, specifies the identity of the + * "running" F/W, i.e. tertiary F/W under normal conditions + * IFB_FWSup CFG_SUP_RANGE_STRCT, specifies the supplier range of + * the "running" F/W, i.e. tertiary F/W under normal conditions + * IFB_HSISup CFG_SUP_RANGE_STRCT, specifies the HW/SW I/F range of the NIC + * IFB_PRIIdentity CFG_PRI_IDENTITY_STRCT, specifies the Identity of the Primary F/W + * IFB_PRISup CFG_SUP_RANGE_STRCT, specifies the supplier range of the Primary F/W + * all other all MSF accessible fields, which are not specified above, are zero-filled + * + *.CONDITIONS + * It is the responsibility of the MSF to assure the correctness of the I/O Base address. + * + * Note: hcf_connect defaults to NIC interrupt disabled mode, i.e. as if hcf_action( HCF_ACT_INT_OFF ) + * was called. + * + *.DESCRIPTION + * hcf_connect passes the MSF-defined location of the IFB to the HCF and grants or revokes access right for the + * HCF to the IFB. Revoking is done by specifying HCF_DISCONNECT rather than an I/O address for the parameter + * io_base. Every call of hcf_connect in "connect" mode, must eventually be followed by a call of hcf_connect + * in "disconnect" mode. Calling hcf_connect in "connect"/"disconnect" mode can not be nested. + * The IFB address must be used as a handle with all subsequent HCF-function calls and the HCF uses the IFB + * address as a handle when it performs a call(back) of an MSF-function (i.e. msf_assert). + * + * Note that not only the MSF accessible fields are cleared, but also all internal housekeeping + * information is re-initialized. + * This implies that all settings which are done via hcf_action and hcf_put_info (e.g. CFG_MB_ASSERT, CFG_REG_MB, + * CFG_REG_INFO_LOG) must be done again. The only field which is not cleared, is IFB_MSFSup. + * + * If HCF_INT_ON is selected as compile option, NIC interrupts are disabled. + * + * Assert fails if + * - ifbp is not properly aligned ( ref chapter HCF_ALIGN in 4.1.1) + * - I/O Base Address is not a multiple of 0x40 (note: 0x0000 is explicitly allowed). + * + *.DIAGRAM + * + *0: Throughout hcf_connect you need to distinguish the connect from the disconnect case, which requires + * some attention about what to use as "I/O" address when for which purpose. + *2: + *2a: Reset H-II by toggling reset bit in IO-register on and off. + * The HCF_TYPE_PRELOADED caters for the DOS environment where H-II is loaded by a separate program to + * overcome the 64k size limit posed on DOS drivers. + * The macro OPW is not yet useable because the IFB_IOBase field is not set. + * Note 1: hopefully the clearing and initializing of the IFB (see below) acts as a delay which meets the + * specification for S/W reset + * Note 2: it turns out that on some H/W constellations, the clock to access the EEProm is not lowered + * to an appropriate frequency by HREG_IO_SRESET. By giving an HCMD_INI first, this problem is worked around. + *2b: Experimentally it is determined over a wide range of F/W versions that are waiting for the for Cmd bit in + * Ev register gives a workable strategy. The available documentation does not give much clues. + *4: clear and initialize the IFB + * The HCF house keeping info is designed such that zero is the appropriate initial value for as much as + * feasible IFB-items. + * The readable fields mentioned in the description section and some HCF specific fields are given their + * actual value. + * IFB_TickIni is initialized at best guess before calibration + * Hcf_connect defaults to "no interrupt generation" (implicitly achieved by the zero-filling). + *6: Register compile-time linked MSF Routine and set default filter level + * cast needed to get around the "near" problem in DOS COM model + * er C2446: no conversion from void (__near __cdecl *)(unsigned char __far *,unsigned int,unsigned short,int) + * to void (__far __cdecl *)(unsigned char __far *,unsigned int,unsigned short,int) + *8: If a command is apparently still active (as indicated by the Busy bit in Cmd register) this may indicate a + * blocked cmd pipe line. To unblock the following actions are done: + * - Ack everything + * - Wait for Busy bit drop in Cmd register + * - Wait for Cmd bit raise in Ev register + * The two waits are combined in a single HCF_WAIT_WHILE to optimize memory size. If either of these waits + * fail (prot_cnt becomes 0), then something is serious wrong. Rather than PANICK, the assumption is that the + * next cmd_exe will fail, causing the HCF to go into DEFUNCT mode + *10: Ack everything to unblock a (possibly blocked) cmd pipe line + * Note 1: it is very likely that an Alloc event is pending and very well possible that a (Send) Cmd event is + * pending on non-initial calls + * Note 2: it is assumed that this strategy takes away the need to ack every conceivable event after an + * Hermes Initialize + *12: Only H-II NEEDS the Hermes Initialize command. Due to the different semantics for H-I and H-II + * Initialize command, init() does not (and can not, since it is called e.g. after a download) execute the + * Hermes Initialize command. Executing the Hermes Initialize command for H-I would not harm but not do + * anything useful either, so it is skipped. + * The return status of cmd_exe is ignored. It is assumed that if cmd_exe fails, init fails too + *14: use io_base as a flag to merge hcf_connect and hcf_disconnect into 1 routine + * the call to init and its subsequent call of cmd_exe will return HCF_ERR_NO_NIC if appropriate. This status + * is (badly) needed by some legacy combination of NT4 and card services which do not yield an I/O address in + * time. + * + *.NOTICE + * On platforms where the NULL-pointer is not a bit-pattern of all zeros, the zero-filling of the IFB results + * in an incorrect initialization of pointers. + * The implementation of the MailBox manipulation in put_mb_info protects against the absence of a MailBox + * based on IFB_MBSize, IFB_MBWp and ifbp->IFB_MBRp. This has ramifications on the initialization of the + * MailBox via hcf_put_info with the CFG_REG_MB type, but it prevents dependency on the "NULL-"ness of + * IFB_MBp. + * + *.NOTICE + * There are a number of problems when asserting and logging hcf_connect, e.g. + * - Asserting on re-entrancy of hcf_connect by means of + * "HCFASSERT( (ifbp->IFB_AssertTrace & HCF_ASSERT_CONNECT) == 0, 0 )" is not useful because IFB contents + * are undefined + * - Asserting before the IFB is cleared will cause mdd_assert() to interpret the garbage in IFB_AssertRtn + * as a routine address + * Therefore HCFTRACE nor HCFLOGENTRY is called by hcf_connect. + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +hcf_connect( IFBP ifbp, hcf_io io_base ) +{ + int rc = HCF_SUCCESS; + hcf_io io_addr; + hcf_32 prot_cnt; + hcf_8 *q; + LTV_STRCT x; +#if HCF_ASSERT + hcf_16 xa = ifbp->IFB_FWIdentity.typ; + /* is assumed to cause an assert later on if hcf_connect is called without intervening hcf_disconnect. + * xa == CFG_FW_IDENTITY in subsequent calls without preceding hcf_disconnect, + * xa == 0 in subsequent calls with preceding hcf_disconnect, + * xa == "garbage" (any value except CFG_FW_IDENTITY is acceptable) in the initial call + */ +#endif // HCF_ASSERT + + if ( io_base == HCF_DISCONNECT ) { //disconnect + io_addr = ifbp->IFB_IOBase; + OPW( HREG_INT_EN, 0 ); //;?workaround against dying F/W on subsequent hcf_connect calls + } else { //connect /* 0 */ + io_addr = io_base; + } + +#if 0 //;? if a subsequent hcf_connect is preceded by an hcf_disconnect the wakeup is not needed !! +#if HCF_SLEEP + OUT_PORT_WORD( .....+HREG_IO, HREG_IO_WAKEUP_ASYNC ); //OPW not yet useable + MSF_WAIT(800); // MSF-defined function to wait n microseconds. + note that MSF_WAIT uses not yet defined!!!! IFB_IOBase and IFB_TickIni (via PROT_CNT_INI) + so be careful if this code is restored +#endif // HCF_SLEEP +#endif // 0 + +#if ( (HCF_TYPE) & HCF_TYPE_PRELOADED ) == 0 //switch clock back for SEEPROM access !!! + OUT_PORT_WORD( io_addr + HREG_CMD, HCMD_INI ); //OPW not yet useable + prot_cnt = INI_TICK_INI; + HCF_WAIT_WHILE( (IN_PORT_WORD( io_addr + HREG_EV_STAT) & HREG_EV_CMD) == 0 ); + OUT_PORT_WORD( (io_addr + HREG_IO), HREG_IO_SRESET ); //OPW not yet useable /* 2a*/ +#endif // HCF_TYPE_PRELOADED + for ( q = (hcf_8*)(&ifbp->IFB_Magic); q > (hcf_8*)ifbp; *--q = 0 ) /*NOP*/; /* 4 */ + ifbp->IFB_Magic = HCF_MAGIC; + ifbp->IFB_Version = IFB_VERSION; +#if defined MSF_COMPONENT_ID //a new IFB demonstrates how dirty the solution is + xxxx[xxxx_PRI_IDENTITY_OFFSET] = NULL; //IFB_PRIIdentity placeholder 0xFD02 + xxxx[xxxx_PRI_IDENTITY_OFFSET+1] = NULL; //IFB_PRISup placeholder 0xFD03 +#endif // MSF_COMPONENT_ID +#if (HCF_TALLIES) & ( HCF_TALLIES_NIC | HCF_TALLIES_HCF ) + ifbp->IFB_TallyLen = 1 + 2 * (HCF_NIC_TAL_CNT + HCF_HCF_TAL_CNT); //convert # of Tallies to L value for LTV + ifbp->IFB_TallyTyp = CFG_TALLIES; //IFB_TallyTyp: set T value +#endif // HCF_TALLIES_NIC / HCF_TALLIES_HCF + ifbp->IFB_IOBase = io_addr; //set IO_Base asap, so asserts via HREG_SW_2 don't harm + ifbp->IFB_IORange = HREG_IO_RANGE; + ifbp->IFB_CntlOpt = USE_16BIT; +#if HCF_ASSERT + assert_ifbp = ifbp; + ifbp->IFB_AssertLvl = 1; +#if (HCF_ASSERT) & HCF_ASSERT_LNK_MSF_RTN + if ( io_base != HCF_DISCONNECT ) { + ifbp->IFB_AssertRtn = (MSF_ASSERT_RTNP)msf_assert; /* 6 */ + } +#endif // HCF_ASSERT_LNK_MSF_RTN +#if (HCF_ASSERT) & HCF_ASSERT_MB //build the structure to pass the assert info to hcf_put_info + ifbp->IFB_AssertStrct.len = sizeof(ifbp->IFB_AssertStrct)/sizeof(hcf_16) - 1; + ifbp->IFB_AssertStrct.typ = CFG_MB_INFO; + ifbp->IFB_AssertStrct.base_typ = CFG_MB_ASSERT; + ifbp->IFB_AssertStrct.frag_cnt = 1; + ifbp->IFB_AssertStrct.frag_buf[0].frag_len = + ( offsetof(IFB_STRCT, IFB_AssertLvl) - offsetof(IFB_STRCT, IFB_AssertLine) ) / sizeof(hcf_16); + ifbp->IFB_AssertStrct.frag_buf[0].frag_addr = &ifbp->IFB_AssertLine; +#endif // HCF_ASSERT_MB +#endif // HCF_ASSERT + IF_PROT_TIME( prot_cnt = ifbp->IFB_TickIni = INI_TICK_INI ); +#if ( (HCF_TYPE) & HCF_TYPE_PRELOADED ) == 0 + //!! No asserts before Reset-bit in HREG_IO is cleared + OPW( HREG_IO, 0x0000 ); //OPW useable /* 2b*/ + HCF_WAIT_WHILE( (IPW( HREG_EV_STAT) & HREG_EV_CMD) == 0 ); + IF_PROT_TIME( HCFASSERT( prot_cnt, IPW( HREG_EV_STAT) ) ); + IF_PROT_TIME( if ( prot_cnt ) prot_cnt = ifbp->IFB_TickIni ); +#endif // HCF_TYPE_PRELOADED + //!! No asserts before Reset-bit in HREG_IO is cleared + HCFASSERT( DO_ASSERT, MERGE_2( HCF_ASSERT, 0xCAF0 ) ); //just to proof that the complete assert machinery is working + HCFASSERT( xa != CFG_FW_IDENTITY, 0 ); // assert if hcf_connect is called without intervening hcf_disconnect. + HCFASSERT( ((hcf_32)(void*)ifbp & (HCF_ALIGN-1) ) == 0, (hcf_32)(void*)ifbp ); + HCFASSERT( (io_addr & 0x003F) == 0, io_addr ); + //if Busy bit in Cmd register + if (IPW( HREG_CMD ) & HCMD_BUSY ) { /* 8 */ + //. Ack all to unblock a (possibly) blocked cmd pipe line + OPW( HREG_EV_ACK, ~HREG_EV_SLEEP_REQ ); + //. Wait for Busy bit drop in Cmd register + //. Wait for Cmd bit raise in Ev register + HCF_WAIT_WHILE( ( IPW( HREG_CMD ) & HCMD_BUSY ) && (IPW( HREG_EV_STAT) & HREG_EV_CMD) == 0 ); + IF_PROT_TIME( HCFASSERT( prot_cnt, IPW( HREG_EV_STAT) ) ); /* if prot_cnt == 0, cmd_exe will fail, causing DEFUNCT */ + } + OPW( HREG_EV_ACK, ~HREG_EV_SLEEP_REQ ); +#if ( (HCF_TYPE) & HCF_TYPE_PRELOADED ) == 0 /*12*/ + (void)cmd_exe( ifbp, HCMD_INI, 0 ); +#endif // HCF_TYPE_PRELOADED + if ( io_base != HCF_DISCONNECT ) { + rc = init( ifbp ); /*14*/ + if ( rc == HCF_SUCCESS ) { + x.len = 2; + x.typ = CFG_NIC_BUS_TYPE; + (void)hcf_get_info( ifbp, &x ); + ifbp->IFB_BusType = x.val[0]; + //CFG_NIC_BUS_TYPE not supported -> default 32 bits/DMA, MSF has to overrule via CFG_CNTL_OPT + if ( x.len == 0 || x.val[0] == 0x0002 || x.val[0] == 0x0003 ) { +#if (HCF_IO) & HCF_IO_32BITS + ifbp->IFB_CntlOpt &= ~USE_16BIT; //reset USE_16BIT +#endif // HCF_IO_32BITS +#if HCF_DMA + ifbp->IFB_CntlOpt |= USE_DMA; //SET DMA +#else + ifbp->IFB_IORange = 0x40 /*i.s.o. HREG_IO_RANGE*/; +#endif // HCF_DMA + } + } + } else HCFASSERT( ( ifbp->IFB_Magic ^= HCF_MAGIC ) == 0, ifbp->IFB_Magic ) /*NOP*/; + /* of above HCFASSERT only the side effect is needed, NOP in case HCFASSERT is dummy */ + ifbp->IFB_IOBase = io_base; /* 0*/ + return rc; +} // hcf_connect + +#if HCF_DMA +/************************************************************************************************************ + * Function get_frame_lst + * - resolve the "last host-owned descriptor" problems when a descriptor list is reclaimed by the MSF. + * + * The FrameList to be reclaimed as well as the DescriptorList always start in IFB_FirstDesc[tx_rx_flag] + * and this is always the "current" DELWA Descriptor. + * + * If a FrameList is available, the last descriptor of the FrameList to turned into a new DELWA Descriptor: + * - a copy is made from the information in the last descriptor of the FrameList into the current + * DELWA Descriptor + * - the remainder of the DescriptorList is detached from the copy by setting the next_desc_addr at NULL + * - the DMA control bits of the copy are cleared to do not confuse the MSF + * - the copy of the last descriptor (i.e. the "old" DELWA Descriptor) is chained to the prev Descriptor + * of the FrameList, thus replacing the original last Descriptor of the FrameList. + * - IFB_FirstDesc is changed to the address of that replaced (original) last descriptor of the FrameList, + * i.e. the "new" DELWA Descriptor. + * + * This function makes a copy of that last host-owned descriptor, so the MSF will get a copy of the descriptor. + * On top of that, it adjusts DMA related fields in the IFB structure. + // perform a copying-scheme to circumvent the 'last host owned descriptor cannot be reclaimed' limitation imposed by H2.5's DMA hardware design + // a 'reclaim descriptor' should be available in the HCF: + * + * Returns: address of the first descriptor of the FrameList + * + 8: Be careful once you start re-ordering the steps in the copy process, that it still works for cases + * of FrameLists of 1, 2 and more than 2 descriptors + * + * Input parameters: + * tx_rx_flag : specifies 'transmit' or 'receive' descriptor. + * + ************************************************************************************************************/ +HCF_STATIC DESC_STRCT* +get_frame_lst( IFBP ifbp, int tx_rx_flag ) +{ + + DESC_STRCT *head = ifbp->IFB_FirstDesc[tx_rx_flag]; + DESC_STRCT *copy, *p, *prev; + + HCFASSERT( tx_rx_flag == DMA_RX || tx_rx_flag == DMA_TX, tx_rx_flag ); + //if FrameList + if ( head ) { + //. search for last descriptor of first FrameList + p = prev = head; + while ( ( p->BUF_SIZE & DESC_EOP ) == 0 && p->next_desc_addr ) { + if ( ( ifbp->IFB_CntlOpt & DMA_ENABLED ) == 0 ) { //clear control bits when disabled + p->BUF_CNT &= DESC_CNT_MASK; + } + prev = p; + p = p->next_desc_addr; + } + //. if DMA enabled + if ( ifbp->IFB_CntlOpt & DMA_ENABLED ) { + //. . if last descriptor of FrameList is DMA owned + //. . or if FrameList is single (DELWA) Descriptor + if ( p->BUF_CNT & DESC_DMA_OWNED || head->next_desc_addr == NULL ) { + //. . . refuse to return FrameList to caller + head = NULL; + } + } + } + //if returnable FrameList found + if ( head ) { + //. if FrameList is single (DELWA) Descriptor (implies DMA disabled) + if ( head->next_desc_addr == NULL ) { + //. . clear DescriptorList + /*;?ifbp->IFB_LastDesc[tx_rx_flag] =*/ ifbp->IFB_FirstDesc[tx_rx_flag] = NULL; + //. else + } else { + //. . strip hardware-related bits from last descriptor + //. . remove DELWA Descriptor from head of DescriptorList + copy = head; + head = head->next_desc_addr; + //. . exchange first (Confined) and last (possibly imprisoned) Descriptor + copy->buf_phys_addr = p->buf_phys_addr; + copy->buf_addr = p->buf_addr; + copy->BUF_SIZE = p->BUF_SIZE &= DESC_CNT_MASK; //get rid of DESC_EOP and possibly DESC_SOP + copy->BUF_CNT = p->BUF_CNT &= DESC_CNT_MASK; //get rid of DESC_DMA_OWNED +#if (HCF_EXT) & HCF_DESC_STRCT_EXT + copy->DESC_MSFSup = p->DESC_MSFSup; +#endif // HCF_DESC_STRCT_EXT + //. . turn into a DELWA Descriptor + p->buf_addr = NULL; + //. . chain copy to prev /* 8*/ + prev->next_desc_addr = copy; + //. . detach remainder of the DescriptorList from FrameList + copy->next_desc_addr = NULL; + copy->next_desc_phys_addr = 0xDEAD0000; //! just to be nice, not really needed + //. . save the new start (i.e. DELWA Descriptor) in IFB_FirstDesc + ifbp->IFB_FirstDesc[tx_rx_flag] = p; + } + //. strip DESC_SOP from first descriptor + head->BUF_SIZE &= DESC_CNT_MASK; + //head->BUF_CNT &= DESC_CNT_MASK; get rid of DESC_DMA_OWNED + head->next_desc_phys_addr = 0xDEAD0000; //! just to be nice, not really needed + } + //return the just detached FrameList (if any) + return head; +} // get_frame_lst + + +/************************************************************************************************************ + * Function put_frame_lst + * + * This function + * + * Returns: address of the first descriptor of the FrameList + * + * Input parameters: + * tx_rx_flag : specifies 'transmit' or 'receive' descriptor. + * + * The following list should be kept in sync with hcf_dma_tx/rx_put, in order to get them in the WCI-spec !!!! + * Assert fails if + * - DMA is not enabled + * - descriptor list is NULL + * - a descriptor in the descriptor list is not double word aligned + * - a count of size field of a descriptor contains control bits, i.e. bits in the high order nibble. + * - the DELWA descriptor is not a "singleton" DescriptorList. + * - the DELWA descriptor is not the first Descriptor supplied + * - a non_DMA descriptor is supplied before the DELWA Descriptor is supplied + * - Possibly more checks could be added !!!!!!!!!!!!! + + *.NOTICE + * The asserts marked with *sc* are really sanity checks for the HCF, they can (supposedly) not be influenced + * by incorrect MSF behavior + + // The MSF is required to supply the HCF with a single descriptor for MSF tx reclaim purposes. + // This 'reclaim descriptor' can be recognized by the fact that its buf_addr field is zero. + ********************************************************************************************* + * Although not required from a hardware perspective: + * - make each descriptor in this rx-chain DMA-owned. + * - Also set the count to zero. EOP and SOP bits are also cleared. + *********************************************************************************************/ +HCF_STATIC void +put_frame_lst( IFBP ifbp, DESC_STRCT *descp, int tx_rx_flag ) +{ + DESC_STRCT *p = descp; + hcf_16 port; + + HCFASSERT( ifbp->IFB_CntlOpt & USE_DMA, ifbp->IFB_CntlOpt); //only hcf_dma_tx_put must also be DMA_ENABLED + HCFASSERT( tx_rx_flag == DMA_RX || tx_rx_flag == DMA_TX, tx_rx_flag ); + HCFASSERT( p , 0 ); + + while ( p ) { + HCFASSERT( ((hcf_32)p & 3 ) == 0, (hcf_32)p ); + HCFASSERT( (p->BUF_CNT & ~DESC_CNT_MASK) == 0, p->BUF_CNT ); + HCFASSERT( (p->BUF_SIZE & ~DESC_CNT_MASK) == 0, p->BUF_SIZE ); + p->BUF_SIZE &= DESC_CNT_MASK; //!!this SHOULD be superfluous in case of correct MSF + p->BUF_CNT &= tx_rx_flag == DMA_RX ? 0 : DESC_CNT_MASK; //!!this SHOULD be superfluous in case of correct MSF + p->BUF_CNT |= DESC_DMA_OWNED; + if ( p->next_desc_addr ) { +// HCFASSERT( p->buf_addr && p->buf_phys_addr && p->BUF_SIZE && +/- p->BUF_SIZE, ... ); + HCFASSERT( p->next_desc_addr->desc_phys_addr, (hcf_32)p->next_desc_addr ); + p->next_desc_phys_addr = p->next_desc_addr->desc_phys_addr; + } else { // + p->next_desc_phys_addr = 0; + if ( p->buf_addr == NULL ) { // DELWA Descriptor + HCFASSERT( descp == p, (hcf_32)descp ); //singleton DescriptorList + HCFASSERT( ifbp->IFB_FirstDesc[tx_rx_flag] == NULL, (hcf_32)ifbp->IFB_FirstDesc[tx_rx_flag]); + HCFASSERT( ifbp->IFB_LastDesc[tx_rx_flag] == NULL, (hcf_32)ifbp->IFB_LastDesc[tx_rx_flag]); + descp->BUF_CNT = 0; //&= ~DESC_DMA_OWNED; + ifbp->IFB_FirstDesc[tx_rx_flag] = descp; +// part of alternative ifbp->IFB_LastDesc[tx_rx_flag] = ifbp->IFB_FirstDesc[tx_rx_flag] = descp; + // if "recycling" a FrameList + // (e.g. called from hcf_cntl( HCF_CNTL_ENABLE ) + // . prepare for activation DMA controller +// part of alternative descp = descp->next_desc_addr; + } else { //a "real" FrameList, hand it over to the DMA engine + HCFASSERT( ifbp->IFB_FirstDesc[tx_rx_flag], (hcf_32)descp ); + HCFASSERT( ifbp->IFB_LastDesc[tx_rx_flag], (hcf_32)descp ); + HCFASSERT( ifbp->IFB_LastDesc[tx_rx_flag]->next_desc_addr == NULL, + (hcf_32)ifbp->IFB_LastDesc[tx_rx_flag]->next_desc_addr); +// p->buf_cntl.cntl_stat |= DESC_DMA_OWNED; + ifbp->IFB_LastDesc[tx_rx_flag]->next_desc_addr = descp; + ifbp->IFB_LastDesc[tx_rx_flag]->next_desc_phys_addr = descp->desc_phys_addr; + port = HREG_RXDMA_PTR32; + if ( tx_rx_flag ) { + p->BUF_SIZE |= DESC_EOP; // p points at the last descriptor in the caller-supplied descriptor chain + descp->BUF_SIZE |= DESC_SOP; + port = HREG_TXDMA_PTR32; + } + OUT_PORT_DWORD( (ifbp->IFB_IOBase + port), descp->desc_phys_addr ); + } + ifbp->IFB_LastDesc[tx_rx_flag] = p; + } + p = p->next_desc_addr; + } +} // put_frame_lst + + +/************************************************************************************************************ + * + *.MODULE DESC_STRCT* hcf_dma_rx_get( IFBP ifbp ) + *.PURPOSE decapsulate a message and provides that message to the MSF. + * reclaim all descriptors in the rx descriptor chain. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * pointer to a FrameList + * + *.DESCRIPTION + * hcf_dma_rx_get is intended to return a received frame when such a frame is deposited in Host memory by the + * DMA engine. In addition hcf_dma_rx_get can be used to reclaim all descriptors in the rx descriptor chain + * when the DMA Engine is disabled, e.g. as part of a driver unloading strategy. + * hcf_dma_rx_get must be called repeatedly by the MSF when hcf_service_nic signals availability of a rx frame + * through the HREG_EV_RDMAD flag of IFB_DmaPackets. The calling must stop when a NULL pointer is returned, at + * which time the HREG_EV_RDMAD flag is also cleared by the HCF to arm the mechanism for the next frame + * reception. + * Regardless whether the DMA Engine is currently enabled (as controlled via hcf_cntl), if the DMA controller + * deposited an Rx-frame in the Rx-DescriptorList, this frame is detached from the Rx-DescriptorList, + * transformed into a FrameList (i.e. updating the housekeeping fields in the descriptors) and returned to the + * caller. + * If no such Rx-frame is available in the Rx-DescriptorList, the behavior of hcf_dma_rx_get depends on the + * status of the DMA Engine. + * If the DMA Engine is enabled, a NULL pointer is returned. + * If the DMA Engine is disabled, the following strategy is used: + * - the complete Rx-DescriptorList is returned. The DELWA Descriptor is not part of the Rx-DescriptorList. + * - If there is no Rx-DescriptorList, the DELWA Descriptor is returned. + * - If there is no DELWA Descriptor, a NULL pointer is returned. + * + * If the MSF performs an disable/enable sequence without exhausting the Rx-DescriptorList as described above, + * the enable command will reset all house keeping information, i.e. already received but not yet by the MSF + * retrieved frames are lost and the next frame will be received starting with the oldest descriptor. + * + * The HCF can be used in 2 fashions: with and without decapsulation for data transfer. + * This is controlled at compile time by the HCF_ENC bit of the HCF_ENCAP system constant. + * If appropriate, decapsulation is done by moving some data inside the buffers and updating the descriptors + * accordingly. + *!! ;?????where did I describe why a simple manipulation with the count values does not suffice? + * + *.DIAGRAM + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ + +DESC_STRCT* +hcf_dma_rx_get (IFBP ifbp) +{ + DESC_STRCT *descp; // pointer to start of FrameList + + descp = get_frame_lst( ifbp, DMA_RX ); + if ( descp && descp->buf_addr ) { + + //skip decapsulation at confined descriptor +#if (HCF_ENCAP) == HCF_ENC + int i; + DESC_STRCT *p = descp->next_desc_addr; //pointer to 2nd descriptor of frame + HCFASSERT(p, 0); + // The 2nd descriptor contains (maybe) a SNAP header plus part or whole of the payload. + //determine decapsulation sub-flag in RxFS + i = *(wci_recordp)&descp->buf_addr[HFS_STAT] & ( HFS_STAT_MSG_TYPE | HFS_STAT_ERR ); + if ( i == HFS_STAT_TUNNEL || + ( i == HFS_STAT_1042 && hcf_encap( (wci_bufp)&p->buf_addr[HCF_DASA_SIZE] ) != ENC_TUNNEL )) { + // The 2nd descriptor contains a SNAP header plus part or whole of the payload. + HCFASSERT( p->BUF_CNT == (p->buf_addr[5] + (p->buf_addr[4]<<8) + 2*6 + 2 - 8), p->BUF_CNT ); + // perform decapsulation + HCFASSERT(p->BUF_SIZE >=8, p->BUF_SIZE); + // move SA[2:5] in the second buffer to replace part of the SNAP header + for ( i=3; i >= 0; i--) p->buf_addr[i+8] = p->buf_addr[i]; + // copy DA[0:5], SA[0:1] from first buffer to second buffer + for ( i=0; i<8; i++) p->buf_addr[i] = descp->buf_addr[HFS_ADDR_DEST + i]; + // make first buffer shorter in count + descp->BUF_CNT = HFS_ADDR_DEST; + } + } +#endif // HCF_ENC + if ( descp == NULL ) ifbp->IFB_DmaPackets &= (hcf_16)~HREG_EV_RDMAD; //;?could be integrated into get_frame_lst + HCFLOGEXIT( HCF_TRACE_DMA_RX_GET ); + return descp; +} // hcf_dma_rx_get + + +/************************************************************************************************************ + * + *.MODULE void hcf_dma_rx_put( IFBP ifbp, DESC_STRCT *descp ) + *.PURPOSE supply buffers for receive purposes. + * supply the Rx-DELWA descriptor. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * descp address of a DescriptorList + * + *.RETURNS N.A. + * + *.DESCRIPTION + * This function is called by the MSF to supply the HCF with new/more buffers for receive purposes. + * The HCF can be used in 2 fashions: with and without encapsulation for data transfer. + * This is controlled at compile time by the HCF_ENC bit of the HCF_ENCAP system constant. + * As a consequence, some additional constraints apply to the number of descriptor and the buffers associated + * with the first 2 descriptors. Independent of the encapsulation feature, the COUNT fields are ignored. + * A special case is the supplying of the DELWA descriptor, which must be supplied as the first descriptor. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - NIC interrupts are not disabled while required by parameter action. + * - in case decapsulation by the HCF is selected: + * - The first databuffer does not have the exact size corresponding with the RxFS up to the 802.3 DestAddr + * field (== 29 words). + * - The FrameList does not consists of at least 2 Descriptors. + * - The second databuffer does not have the minimum size of 8 bytes. + *!! The 2nd part of the list of asserts should be kept in sync with put_frame_lst, in order to get + *!! them in the WCI-spec !!!! + * - DMA is not enabled + * - descriptor list is NULL + * - a descriptor in the descriptor list is not double word aligned + * - a count of size field of a descriptor contains control bits, i.e. bits in the high order nibble. + * - the DELWA descriptor is not a "singleton" DescriptorList. + * - the DELWA descriptor is not the first Descriptor supplied + * - a non_DMA descriptor is supplied before the DELWA Descriptor is supplied + *!! - Possibly more checks could be added !!!!!!!!!!!!! + * + *.DIAGRAM + * + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +void +hcf_dma_rx_put( IFBP ifbp, DESC_STRCT *descp ) +{ + + HCFLOGENTRY( HCF_TRACE_DMA_RX_PUT, 0xDA01 ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + + put_frame_lst( ifbp, descp, DMA_RX ); +#if HCF_ASSERT && (HCF_ENCAP) == HCF_ENC + if ( descp->buf_addr ) { + HCFASSERT( descp->BUF_SIZE == HCF_DMA_RX_BUF1_SIZE, descp->BUF_SIZE ); + HCFASSERT( descp->next_desc_addr, 0 ); // first descriptor should be followed by another descriptor + // The second DB is for SNAP and payload purposes. It should be a minimum of 12 bytes in size. + HCFASSERT( descp->next_desc_addr->BUF_SIZE >= 12, descp->next_desc_addr->BUF_SIZE ); + } +#endif // HCFASSERT / HCF_ENC + HCFLOGEXIT( HCF_TRACE_DMA_RX_PUT ); +} // hcf_dma_rx_put + + +/************************************************************************************************************ + * + *.MODULE DESC_STRCT* hcf_dma_tx_get( IFBP ifbp ) + *.PURPOSE DMA mode: reclaims and decapsulates packets in the tx descriptor chain if: + * - A Tx packet has been copied from host-RAM into NIC-RAM by the DMA engine + * - The Hermes/DMAengine have been disabled + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * pointer to a reclaimed Tx packet. + * + *.DESCRIPTION + * impact of the disable command: + * When a non-empty pool of Tx descriptors exists (created by means of hcf_dma_put_tx), the MSF + * is supposed to empty that pool by means of hcf_dma_tx_get calls after the disable in an + * disable/enable sequence. + * + *.DIAGRAM + * + *.NOTICE + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +DESC_STRCT* +hcf_dma_tx_get( IFBP ifbp ) +{ + DESC_STRCT *descp; // pointer to start of FrameList + + descp = get_frame_lst( ifbp, DMA_TX ); + if ( descp && descp->buf_addr ) { + //skip decapsulation at confined descriptor +#if (HCF_ENCAP) == HCF_ENC + if ( ( descp->BUF_CNT == HFS_TYPE )) { + // perform decapsulation if needed + descp->next_desc_addr->buf_phys_addr -= HCF_DASA_SIZE; + descp->next_desc_addr->BUF_CNT += HCF_DASA_SIZE; + } +#endif // HCF_ENC + } + if ( descp == NULL ) { //;?could be integrated into get_frame_lst + ifbp->IFB_DmaPackets &= (hcf_16)~HREG_EV_TDMAD; + } + HCFLOGEXIT( HCF_TRACE_DMA_TX_GET ); + return descp; +} // hcf_dma_tx_get + + +/************************************************************************************************************ + * + *.MODULE void hcf_dma_tx_put( IFBP ifbp, DESC_STRCT *descp, hcf_16 tx_cntl ) + *.PURPOSE puts a packet in the Tx DMA queue in host ram and kicks off the TxDma engine. + * supply the Tx-DELWA descriptor. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * descp address of Tx Descriptor Chain (i.e. a single Tx frame) + * tx_cntl indicates MAC-port and (Hermes) options + * + *.RETURNS N.A. + * + *.DESCRIPTION + * The HCF can be used in 2 fashions: with and without encapsulation for data transfer. + * This is controlled at compile time by the HCF_ENC bit of the HCF_ENCAP system constant. + * + * Regardless of the HCF_ENCAP system constant, the descriptor list created to describe the frame to be + * transmitted, must supply space to contain the 802.11 header, preceding the actual frame to be transmitted. + * Basically, this only supplies working storage to the HCF which passes this on to the DMA engine. + * As a consequence the contents of this space do not matter. + * Nevertheless BUF_CNT must take in account this storage. + * This working space to contain the 802.11 header may not be fragmented, the first buffer must be + * sufficiently large to contain at least the 802.11 header, i.e. HFS_ADDR_DEST (29 words or 0x3A bytes). + * This way, the HCF can simply, regardless whether or not the HCF encapsulates the frame, write the parameter + * tx_cntl at offset 0x36 (HFS_TX_CNTL) in the first buffer. + * Note that it is allowed to have part or all of the actual frame represented by the first descriptor as long + * as the requirement for storage for the 802.11 header is met, i.e. the 802.3 frame starts at offset + * HFS_ADDR_DEST. + * Except for the Assert on the 1st buffer in case of Encapsualtion, the SIZE fields are ignored. + * + * In case the encapsulation feature is compiled in, there are the following additional requirements. + * o The BUF_CNT of the first buffer changes from a minimum of 0x3A bytes to exactly 0x3A, i.e. the workspace + * to store the 802.11 header + * o The BUF_SIZE of the first buffer is at least the space needed to store the + * - 802.11 header (29 words) + * - 802.3 header, i.e. 12 bytes addressing information and 2 bytes length field + * - 6 bytes SNAP-header + * This results in 39 words or 0x4E bytes or HFS_TYPE. + * Note that if the BUF_SIZE is larger than 0x4E, this surplus is not used. + * o The actual frame begins in the 2nd descriptor (which is already implied by the BUF_CNT == 0x3A requirement) and the associated buffer contains at least the 802.3 header, i.e. the 14 bytes representing addressing information and length/type field + * + * When the HCF does not encapsulates (i.e. length/type field <= 1500), no changes are made to descriptors + * or buffers. + * + * When the HCF actually encapsulates (i.e. length/type field > 1500), it successively writes, starting at + * offset HFS_ADDR_DEST (0x3A) in the first buffer: + * - the 802.3 addressing information, copied from the begin of the second buffer + * - the frame length, derived from the total length of the individual fragments, corrected for the SNAP + * header length and Type field and ignoring the Destination Address, Source Address and Length field + * - the appropriate snap header (Tunnel or 1042, depending on the value of the type field). + * + * The information in the first two descriptors is adjusted accordingly: + * - the first descriptor count is changed from 0x3A to 0x4E (HFS_TYPE), which matches 0x3A + 12 + 2 + 6 + * - the second descriptor count is decreased by 12, being the moved addressing information + * - the second descriptor (physical) buffer address is increased by 12. + * + * When the descriptors are returned by hcf_dma_tx_get, the transformation of the first two descriptors is + * undone. + * + * Under any of the above scenarios, the assert BUF_CNT <= BUF_SIZE must be true for all descriptors + * In case of encapsulation, BUF_SIZE of the 1st descriptor is asserted to be at least HFS_TYPE (0x4E), so it is NOT tested. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - tx_cntl has a recognizable out-of-range value. + * - NIC interrupts are not disabled while required by parameter action. + * - in case encapsulation by the HCF is selected: + * - The FrameList does not consists of at least 2 Descriptors. + * - The first databuffer does not contain exactly the (space for) the 802.11 header (== 28 words) + * - The first databuffer does not have a size to additionally accommodate the 802.3 header and the + * SNAP header of the frame after encapsulation (== 39 words). + * - The second databuffer does not contain at least DA, SA and 'type/length' (==14 bytes or 7 words) + *!! The 2nd part of the list of asserts should be kept in sync with put_frame_lst, in order to get + *!! them in the WCI-spec !!!! + * - DMA is not enabled + * - descriptor list is NULL + * - a descriptor in the descriptor list is not double word aligned + * - a count of size field of a descriptor contains control bits, i.e. bits in the high order nibble. + * - the DELWA descriptor is not a "singleton" DescriptorList. + * - the DELWA descriptor is not the first Descriptor supplied + * - a non_DMA descriptor is supplied before the DELWA Descriptor is supplied + *!! - Possibly more checks could be added !!!!!!!!!!!!! + *.DIAGRAM + * + *.NOTICE + * + *.ENDDOC END DOCUMENTATION + * + * + *1: Write tx_cntl parameter to HFS_TX_CNTL field into the Hermes-specific header in buffer 1 + *4: determine whether encapsulation is needed and write the type (tunnel or 1042) already at the appropriate + * offset in the 1st buffer + *6: Build the encapsualtion enveloppe in the free space at the end of the 1st buffer + * - Copy DA/SA fields from the 2nd buffer + * - Calculate total length of the message (snap-header + type-field + the length of all buffer fragments + * associated with the 802.3 frame (i.e all descriptors except the first), but not the DestinationAddress, + * SourceAddress and length-field) + * Assert the message length + * Write length. Note that the message is in BE format, hence on LE platforms the length must be converted + * ;? THIS IS NOT WHAT CURRENTLY IS IMPLEMENTED + * - Write snap header. Note that the last byte of the snap header is NOT copied, that byte is already in + * place as result of the call to hcf_encap. + * Note that there are many ways to skin a cat. To express the offsets in the 1st buffer while writing + * the snap header, HFS_TYPE is chosen as a reference point to make it easier to grasp that the snap header + * and encapsualtion type are at least relative in the right. + *8: modify 1st descriptor to reflect moved part of the 802.3 header + Snap-header + * modify 2nd descriptor to skip the moved part of the 802.3 header (DA/SA + *10: set each descriptor to 'DMA owned', clear all other control bits. + * Set SOP bit on first descriptor. Set EOP bit on last descriptor. + *12: Either append the current frame to an existing descriptor list or + *14: create a list beginning with the current frame + *16: remember the new end of the list + *20: hand the frame over to the DMA engine + ************************************************************************************************************/ +void +hcf_dma_tx_put( IFBP ifbp, DESC_STRCT *descp, hcf_16 tx_cntl ) +{ + DESC_STRCT *p = descp->next_desc_addr; + int i; + +#if HCF_ASSERT + int x = ifbp->IFB_FWIdentity.comp_id == COMP_ID_FW_AP ? tx_cntl & ~HFS_TX_CNTL_PORT : tx_cntl; + HCFASSERT( (x & ~HCF_TX_CNTL_MASK ) == 0, tx_cntl ); +#endif // HCF_ASSERT + HCFLOGENTRY( HCF_TRACE_DMA_TX_PUT, 0xDA03 ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + HCFASSERT( ( ifbp->IFB_CntlOpt & (USE_DMA|DMA_ENABLED) ) == (USE_DMA|DMA_ENABLED), ifbp->IFB_CntlOpt); + + if ( descp->buf_addr ) { + *(hcf_16*)(descp->buf_addr + HFS_TX_CNTL) = tx_cntl; /*1*/ +#if (HCF_ENCAP) == HCF_ENC + HCFASSERT( descp->next_desc_addr, 0 ); //at least 2 descripors + HCFASSERT( descp->BUF_CNT == HFS_ADDR_DEST, descp->BUF_CNT ); //exact length required for 1st buffer + HCFASSERT( descp->BUF_SIZE >= HCF_DMA_TX_BUF1_SIZE, descp->BUF_SIZE ); //minimal storage for encapsulation + HCFASSERT( p->BUF_CNT >= 14, p->BUF_CNT ); //at least DA, SA and 'type' in 2nd buffer + + descp->buf_addr[HFS_TYPE-1] = hcf_encap(&descp->next_desc_addr->buf_addr[HCF_DASA_SIZE]); /*4*/ + if ( descp->buf_addr[HFS_TYPE-1] != ENC_NONE ) { + for ( i=0; i < HCF_DASA_SIZE; i++ ) { /*6*/ + descp->buf_addr[i + HFS_ADDR_DEST] = descp->next_desc_addr->buf_addr[i]; + } + i = sizeof(snap_header) + 2 - ( 2*6 + 2 ); + do { i += p->BUF_CNT; } while ( ( p = p->next_desc_addr ) != NULL ); + *(hcf_16*)(&descp->buf_addr[HFS_LEN]) = CNV_END_SHORT(i); //!! this converts on ALL platforms, how does that relate to the CCX code + for ( i=0; i < sizeof(snap_header) - 1; i++) { + descp->buf_addr[HFS_TYPE - sizeof(snap_header) + i] = snap_header[i]; + } + descp->BUF_CNT = HFS_TYPE; /*8*/ + descp->next_desc_addr->buf_phys_addr += HCF_DASA_SIZE; + descp->next_desc_addr->BUF_CNT -= HCF_DASA_SIZE; + } +#endif // HCF_ENC + } + put_frame_lst( ifbp, descp, DMA_TX ); + HCFLOGEXIT( HCF_TRACE_DMA_TX_PUT ); +} // hcf_dma_tx_put + +#endif // HCF_DMA + +/************************************************************************************************************ + * + *.MODULE hcf_8 hcf_encap( wci_bufp type ) + *.PURPOSE test whether RFC1042 or Bridge-Tunnel encapsulation is needed. + * + *.ARGUMENTS + * type (Far) pointer to the (Big Endian) Type/Length field in the message + * + *.RETURNS + * ENC_NONE len/type is "len" ( (BIG_ENDIAN)type <= 1500 ) + * ENC_TUNNEL len/type is "type" and 0x80F3 or 0x8137 + * ENC_1042 len/type is "type" but not 0x80F3 or 0x8137 + * + *.CONDITIONS + * NIC Interrupts d.c + * + *.DESCRIPTION + * Type must point to the Len/Type field of the message, this is the 2-byte field immediately after the 6 byte + * Destination Address and 6 byte Source Address. The 2 successive bytes addressed by type are interpreted as + * a Big Endian value. If that value is less than or equal to 1500, the message is assumed to be in 802.3 + * format. Otherwise the message is assumed to be in Ethernet-II format. Depending on the value of Len/Typ, + * Bridge Tunnel or RFC1042 encapsulation is needed. + * + *.DIAGRAM + * + * 1: presume 802.3, hence preset return value at ENC_NONE + * 2: convert type from "network" Endian format to native Endian + * 4: the litmus test to distinguish type and len. + * The hard code "magic" value of 1500 is intentional and should NOT be replaced by a mnemonic because it is + * not related at all to the maximum frame size supported by the Hermes. + * 6: check type against: + * 0x80F3 //AppleTalk Address Resolution Protocol (AARP) + * 0x8137 //IPX + * to determine the type of encapsulation + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC hcf_8 +hcf_encap( wci_bufp type ) +{ + + hcf_8 rc = ENC_NONE; /* 1 */ + hcf_16 t = (hcf_16)(*type<<8) + *(type+1); /* 2 */ + + if ( t > 1500 ) { /* 4 */ + if ( t == 0x8137 || t == 0x80F3 ) { + rc = ENC_TUNNEL; /* 6 */ + } else { + rc = ENC_1042; + } + } + return rc; +} // hcf_encap + + +/************************************************************************************************************ + * + *.MODULE int hcf_get_info( IFBP ifbp, LTVP ltvp ) + *.PURPOSE Obtains transient and persistent configuration information from the Card and from the HCF. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * ltvp address of LengthTypeValue structure specifying the "what" and the "how much" of the + * information to be collected from the HCF or from the Hermes + * + *.RETURNS + * HCF_ERR_LEN The provided buffer was too small + * HCF_SUCCESS Success + *!! via cmd_exe ( type >= CFG_RID_FW_MIN ) + * HCF_ERR_NO_NIC NIC removed during retrieval + * HCF_ERR_TIME_OUT Expected Hermes event did not occur in expected time + *!! via cmd_exe and setup_bap (type >= CFG_RID_FW_MIN ) + * HCF_ERR_DEFUNCT_... HCF is in defunct mode (bits 0x7F reflect cause) + * + *.DESCRIPTION + * The T-field of the LTV-record (provided by the MSF in parameter ltvp) specifies the RID wanted. The RID + * information identified by the T-field is copied into the V-field. + * On entry, the L-field specifies the size of the buffer, also called the "Initial DataLength". The L-value + * includes the size of the T-field, but not the size of the L-field itself. + * On return, the L-field indicates the number of words actually contained by the Type and Value fields. + * As the size of the Type field in the LTV-record is included in the "Initial DataLength" of the record, the + * V-field can contain at most "Initial DataLength" - 1 words of data. + * Copying stops if either the complete Information is copied or if the number of words indicated by the + * "Initial DataLength" were copied. The "Initial DataLength" acts as a safe guard against Configuration + * Information blocks that have different sizes for different F/W versions, e.g. when later versions support + * more tallies than earlier versions. + * If the size of Value field of the RID exceeds the size of the "Initial DataLength" -1, as much data + * as fits is copied, and an error status of HCF_ERR_LEN is returned. + * + * It is the responsibility of the MSF to detect card removal and re-insertion and not call the HCF when the + * NIC is absent. The MSF cannot, however, timely detect a Card removal if the Card is removed while + * hcf_get_info is in progress. Therefore, the HCF performs its own check on Card presence after the read + * operation of the NIC data. If the Card is not present or removed during the execution of hcf_get_info, + * HCF_ERR_NO_NIC is returned and the content of the Data Buffer is unpredictable. This check is not performed + * in case of the "HCF embedded" pseudo RIDs like CFG_TALLIES. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - reentrancy, may be caused by calling hcf_functions without adequate protection + * against NIC interrupts or multi-threading. + * - ltvp is a NULL pointer. + * - length field of the LTV-record at entry is 0 or 1 or has an excessive value (i.e. exceeds HCF_MAX_LTV). + * - type field of the LTV-record is invalid. + * + *.DIAGRAM + * Hcf_get_mb_info copies the contents of the oldest MailBox Info block in the MailBox to PC RAM. If len is + * less than the size of the MailBox Info block, only as much as fits in the PC RAM buffer is copied. After + * the copying the MailBox Read pointer is updated to point to the next MailBox Info block, hence the + * remainder of an "oversized" MailBox Info block is lost. The truncation of the MailBox Info block is NOT + * reflected in the return status. Note that hcf_get_info guarantees the length of the PC RAM buffer meets + * the minimum requirements of at least 2, so no PC RAM buffer overrun. + * + * Calling hcf_get_mb_info when their is no MailBox Info block available or when there is no MailBox at all, + * results in a "NULL" MailBox Info block. + * + *12: see NOTICE + *17: The return status of cmd_wait and the first hcfio_in_string can be ignored, because when one fails, the + * other fails via the IFB_DefunctStat mechanism + *20: "HCFASSERT( rc == HCF_SUCCESS, rc )" is not suitable because this will always trigger as side effect of + * the HCFASSERT in hcf_put_info which calls hcf_get_info to figure out whether the RID exists at all. + + *.NOTICE + * + * "HCF embedded" pseudo RIDs: + * CFG_MB_INFO, CFG_TALLIES, CFG_DRV_IDENTITY, CFG_DRV_SUP_RANGE, CFG_DRV_ACT_RANGES_PRI, + * CFG_DRV_ACT_RANGES_STA, CFG_DRV_ACT_RANGES_HSI + * Note the HCF_ERR_LEN is NOT adequately set, when L >= 2 but less than needed + * + * Remarks: Transfers operation information and transient and persistent configuration information from the + * Card and from the HCF to the MSF. + * The exact layout of the provided data structure depends on the action code. Copying stops if either the + * complete Configuration Information is copied or if the number of bytes indicated by len is copied. Len + * acts as a safe guard against Configuration Information blocks which have different sizes for different + * Hermes versions, e.g. when later versions support more tallies than earlier versions. It is a conscious + * decision that unused parts of the PC RAM buffer are not cleared. + * + * Remarks: The only error against which is protected is the "Read error" as result of Card removal. Only the + * last hcf_io_string need to be protected because if the first fails the second will fail as well. Checking + * for cmd_exe errors is supposed superfluous because problems in cmd_exe are already caught or will be + * caught by hcf_enable. + * + * CFG_MB_INFO: copy the oldest MailBox Info Block or the "null" block if none available. + * + * The mechanism to HCF_ASSERT on invalid typ-codes in the LTV record is based on the following strategy: + * - during the pseudo-asynchronous Hermes commands (diagnose, download) only CFG_MB_INFO is acceptable + * - some codes (e.g. CFG_TALLIES) are explicitly handled by the HCF which implies that these codes + * are valid + * - all other codes in the range 0xFC00 through 0xFFFF are passed to the Hermes. The Hermes returns an + * LTV record with a zero value in the L-field for all Typ-codes it does not recognize. This is + * defined and intended behavior, so HCF_ASSERT does not catch on this phenomena. + * - all remaining codes are invalid and cause an ASSERT. + * + *.CONDITIONS + * In case of USB, HCF_MAX_MSG ;?USED;? to limit the amount of data that can be retrieved via hcf_get_info. + * + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +hcf_get_info( IFBP ifbp, LTVP ltvp ) +{ + + int rc = HCF_SUCCESS; + hcf_16 len = ltvp->len; + hcf_16 type = ltvp->typ; + wci_recordp p = <vp->len; //destination word pointer (in LTV record) + hcf_16 *q = NULL; /* source word pointer Note!! DOS COM can't cope with FAR + * as a consequence MailBox must be near which is usually true anyway + */ + int i; + + HCFLOGENTRY( HCF_TRACE_GET_INFO, ltvp->typ ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + HCFASSERT( ltvp, 0 ); + HCFASSERT( 1 < ltvp->len && ltvp->len <= HCF_MAX_LTV + 1, MERGE_2( ltvp->typ, ltvp->len ) ); + + ltvp->len = 0; //default to: No Info Available + //filter out all specials + for ( i = 0; ( q = xxxx[i] ) != NULL && q[1] != type; i++ ) /*NOP*/; + +#if HCF_TALLIES + if ( type == CFG_TALLIES ) { /*3*/ + (void)hcf_action( ifbp, HCF_ACT_TALLIES ); + q = (hcf_16*)&ifbp->IFB_TallyLen; + } +#endif // HCF_TALLIES + + if ( type == CFG_MB_INFO ) { + if ( ifbp->IFB_MBInfoLen ) { + if ( ifbp->IFB_MBp[ifbp->IFB_MBRp] == 0xFFFF ) { + ifbp->IFB_MBRp = 0; //;?Probably superfluous + } + q = &ifbp->IFB_MBp[ifbp->IFB_MBRp]; + ifbp->IFB_MBRp += *q + 1; //update read pointer + if ( ifbp->IFB_MBp[ifbp->IFB_MBRp] == 0xFFFF ) { + ifbp->IFB_MBRp = 0; + } + ifbp->IFB_MBInfoLen = ifbp->IFB_MBp[ifbp->IFB_MBRp]; + } + } + + if ( q != NULL ) { //a special or CFG_TALLIES or CFG_MB_INFO + i = min( len, *q ) + 1; //total size of destination (including T-field) + while ( i-- ) { + *p++ = *q; +#if (HCF_TALLIES) & HCF_TALLIES_RESET + if ( q > &ifbp->IFB_TallyTyp && type == CFG_TALLIES ) { + *q = 0; + } +#endif // HCF_TALLIES_RESET + q++; + } + } else { // not a special nor CFG_TALLIES nor CFG_MB_INFO + if ( type == CFG_CNTL_OPT ) { //read back effective options + ltvp->len = 2; + ltvp->val[0] = ifbp->IFB_CntlOpt; +#if (HCF_EXT) & HCF_EXT_NIC_ACCESS + } else if ( type == CFG_PROD_DATA ) { //only needed for some test tool on top of H-II NDIS driver + hcf_io io_port; + wci_bufp pt; //pointer with the "right" type, just to help ease writing macros with embedded assembly + OPW( HREG_AUX_PAGE, (hcf_16)(PLUG_DATA_OFFSET >> 7) ); + OPW( HREG_AUX_OFFSET, (hcf_16)(PLUG_DATA_OFFSET & 0x7E) ); + io_port = ifbp->IFB_IOBase + HREG_AUX_DATA; //to prevent side effects of the MSF-defined macro + p = ltvp->val; //destination char pointer (in LTV record) + i = len - 1; + if (i > 0 ) { + pt = (wci_bufp)p; //just to help ease writing macros with embedded assembly + IN_PORT_STRING_8_16( io_port, pt, i ); //space used by T: -1 + } + } else if ( type == CFG_CMD_HCF ) { +#define P ((CFG_CMD_HCF_STRCT FAR *)ltvp) + HCFASSERT( P->cmd == CFG_CMD_HCF_REG_ACCESS, P->cmd ); //only Hermes register access supported + if ( P->cmd == CFG_CMD_HCF_REG_ACCESS ) { + HCFASSERT( P->mode < ifbp->IFB_IOBase, P->mode ); //Check Register space + ltvp->len = min( len, 4 ); //RESTORE ltv length + P->add_info = IPW( P->mode ); + } +#undef P +#endif // HCF_EXT_NIC_ACCESS +#if (HCF_ASSERT) & HCF_ASSERT_PRINTF + } else if (type == CFG_FW_PRINTF) { + rc = fw_printf(ifbp, (CFG_FW_PRINTF_STRCT*)ltvp); +#endif // HCF_ASSERT_PRINTF + } else if ( type >= CFG_RID_FW_MIN ) { +//;? by using HCMD_BUSY option when calling cmd_exe, using a get_frag with length 0 just to set up the +//;? BAP and calling cmd_cmpl, you could merge the 2 Busy waits. Whether this really helps (and what +//;? would be the optimal sequence in cmd_exe and get_frag) would have to be MEASURED + /*17*/ if ( ( rc = cmd_exe( ifbp, HCMD_ACCESS, type ) ) == HCF_SUCCESS && + ( rc = setup_bap( ifbp, type, 0, IO_IN ) ) == HCF_SUCCESS ) { + get_frag( ifbp, (wci_bufp)<vp->len, 2*len+2 BE_PAR(2) ); + if ( IPW( HREG_STAT ) == 0xFFFF ) { //NIC removal test + ltvp->len = 0; + HCFASSERT( DO_ASSERT, type ); + } + } + /*12*/ } else HCFASSERT( DO_ASSERT, type ) /*NOP*/; //NOP in case HCFASSERT is dummy + } + if ( len < ltvp->len ) { + ltvp->len = len; + if ( rc == HCF_SUCCESS ) { + rc = HCF_ERR_LEN; + } + } + HCFASSERT( rc == HCF_SUCCESS || ( rc == HCF_ERR_LEN && ifbp->IFB_AssertTrace & 1<<HCF_TRACE_PUT_INFO ), + MERGE_2( type, rc ) ); /*20*/ + HCFLOGEXIT( HCF_TRACE_GET_INFO ); + return rc; +} // hcf_get_info + + +/************************************************************************************************************ + * + *.MODULE int hcf_put_info( IFBP ifbp, LTVP ltvp ) + *.PURPOSE Transfers operation and configuration information to the Card and to the HCF. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * ltvp specifies the RID (as defined by Hermes I/F) or pseudo-RID (as defined by WCI) + * + *.RETURNS + * HCF_SUCCESS + *!! via cmd_exe + * HCF_ERR_NO_NIC NIC removed during data retrieval + * HCF_ERR_TIME_OUT Expected F/W event did not occur in time + * HCF_ERR_DEFUNCT_... + *!! via download CFG_DLNV_START <= type <= CFG_DL_STOP + *!! via put_info CFG_RID_CFG_MIN <= type <= CFG_RID_CFG_MAX + *!! via put_frag + * + *.DESCRIPTION + * The L-field of the LTV-record (provided by the MSF in parameter ltvp) specifies the size of the buffer. + * The L-value includes the size of the T-field, but not the size of the L-field. + * The T- field specifies the RID placed in the V-field by the MSF. + * + * Not all CFG-codes can be used for hcf_put_info. The following CFG-codes are valid for hcf_put_info: + * o One of the CFG-codes in the group "Network Parameters, Static Configuration Entities" + * Changes made by hcf_put_info to CFG_codes in this group will not affect the F/W + * and HCF behavior until hcf_cntl_port( HCF_PORT_ENABLE) is called. + * o One of the CFG-codes in the group "Network Parameters, Dynamic Configuration Entities" + * Changes made by hcf_put_info to CFG_codes will affect the F/W and HCF behavior immediately. + * o CFG_PROG. + * This code is used to initiate and terminate the process to download data either to + * volatile or to non-volatile RAM on the NIC as well as for the actual download. + * o CFG-codes related to the HCF behavior. + * The related CFG-codes are: + * - CFG_REG_MB + * - CFG_REG_ASSERT_RTNP + * - CFG_REG_INFO_LOG + * - CFG_CMD_NIC + * - CFG_CMD_DONGLE + * - CFG_CMD_HCF + * - CFG_NOTIFY + * + * All LTV-records "unknown" to the HCF are forwarded to the F/W. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value. + * - ltvp is a NULL pointer. + * - hcf_put_info was called without prior call to hcf_connect + * - type field of the LTV-record is invalid, i.e. neither HCF nor F/W can handle the value. + * - length field of the LTV-record at entry is less than 1 or exceeds MAX_LTV_SIZE. + * - registering a MailBox with size less than 60 or a non-aligned buffer address is used. + * - reentrancy, may be caused by calling hcf_functions without adequate protection against + * NIC interrupts or multi-threading. + * + *.DIAGRAM + * + *.NOTICE + * Remarks: In case of Hermes Configuration LTVs, the codes for the type are "cleverly" chosen to be + * identical to the RID. Hermes Configuration information is copied from the provided data structure into the + * Card. + * In case of HCF Configuration LTVs, the type values are chosen in a range which does not overlap the + * RID-range. + * + *20: + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ + +int +hcf_put_info( IFBP ifbp, LTVP ltvp ) +{ + int rc = HCF_SUCCESS; + + HCFLOGENTRY( HCF_TRACE_PUT_INFO, ltvp->typ ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + HCFASSERT( ltvp, 0 ); + HCFASSERT( 1 < ltvp->len && ltvp->len <= HCF_MAX_LTV + 1, ltvp->len ); + + //all codes between 0xFA00 and 0xFCFF are passed to Hermes +#if (HCF_TYPE) & HCF_TYPE_WPA + { + hcf_16 i; + hcf_32 FAR * key_p; + + if ( ltvp->typ == CFG_ADD_TKIP_DEFAULT_KEY || ltvp->typ == CFG_ADD_TKIP_MAPPED_KEY ) { + key_p = (hcf_32*)((CFG_ADD_TKIP_MAPPED_KEY_STRCT FAR *)ltvp)->tx_mic_key; + i = TX_KEY; //i.e. TxKeyIndicator == 1, KeyID == 0 + if ( ltvp->typ == CFG_ADD_TKIP_DEFAULT_KEY ) { + key_p = (hcf_32*)((CFG_ADD_TKIP_DEFAULT_KEY_STRCT FAR *)ltvp)->tx_mic_key; + i = CNV_LITTLE_TO_SHORT(((CFG_ADD_TKIP_DEFAULT_KEY_STRCT FAR *)ltvp)->tkip_key_id_info); + } + if ( i & TX_KEY ) { /* TxKeyIndicator == 1 + (either really set by MSF in case of DEFAULT or faked by HCF in case of MAPPED ) */ + ifbp->IFB_MICTxCntl = (hcf_16)( HFS_TX_CNTL_MIC | (i & KEY_ID )<<8 ); + ifbp->IFB_MICTxKey[0] = CNV_LONGP_TO_LITTLE( key_p ); + ifbp->IFB_MICTxKey[1] = CNV_LONGP_TO_LITTLE( (key_p+1) ); + } + i = ( i & KEY_ID ) * 2; + ifbp->IFB_MICRxKey[i] = CNV_LONGP_TO_LITTLE( (key_p+2) ); + ifbp->IFB_MICRxKey[i+1] = CNV_LONGP_TO_LITTLE( (key_p+3) ); + } +#define P ((CFG_REMOVE_TKIP_DEFAULT_KEY_STRCT FAR *)ltvp) + if ( ( ltvp->typ == CFG_REMOVE_TKIP_MAPPED_KEY ) || + ( ltvp->typ == CFG_REMOVE_TKIP_DEFAULT_KEY && + ( (ifbp->IFB_MICTxCntl >> 8) & KEY_ID ) == CNV_SHORT_TO_LITTLE(P->tkip_key_id ) + ) + ) { ifbp->IFB_MICTxCntl = 0; } //disable MIC-engine +#undef P + } +#endif // HCF_TYPE_WPA + + if ( ltvp->typ == CFG_PROG ) { + rc = download( ifbp, (CFG_PROG_STRCT FAR *)ltvp ); + } else switch (ltvp->typ) { +#if (HCF_ASSERT) & HCF_ASSERT_RT_MSF_RTN + case CFG_REG_ASSERT_RTNP: //Register MSF Routines +#define P ((CFG_REG_ASSERT_RTNP_STRCT FAR *)ltvp) + ifbp->IFB_AssertRtn = P->rtnp; +// ifbp->IFB_AssertLvl = P->lvl; //TODO not yet supported so default is set in hcf_connect + HCFASSERT( DO_ASSERT, MERGE_2( HCF_ASSERT, 0xCAF1 ) ); //just to proof that the complete assert machinery is working +#undef P + break; +#endif // HCF_ASSERT_RT_MSF_RTN +#if (HCF_EXT) & HCF_EXT_INFO_LOG + case CFG_REG_INFO_LOG: //Register Log filter + ifbp->IFB_RIDLogp = ((CFG_RID_LOG_STRCT FAR*)ltvp)->recordp; + break; +#endif // HCF_EXT_INFO_LOG + case CFG_CNTL_OPT: //overrule option + HCFASSERT( ( ltvp->val[0] & ~(USE_DMA | USE_16BIT) ) == 0, ltvp->val[0] ); + if ( ( ltvp->val[0] & USE_DMA ) == 0 ) ifbp->IFB_CntlOpt &= ~USE_DMA; + ifbp->IFB_CntlOpt |= ltvp->val[0] & USE_16BIT; + break; + + case CFG_REG_MB: //Register MailBox +#define P ((CFG_REG_MB_STRCT FAR *)ltvp) + HCFASSERT( ( (hcf_32)P->mb_addr & 0x0001 ) == 0, (hcf_32)P->mb_addr ); + HCFASSERT( (P)->mb_size >= 60, (P)->mb_size ); + ifbp->IFB_MBp = P->mb_addr; + /* if no MB present, size must be 0 for ;?the old;? put_info_mb to work correctly */ + ifbp->IFB_MBSize = ifbp->IFB_MBp == NULL ? 0 : P->mb_size; + ifbp->IFB_MBWp = ifbp->IFB_MBRp = 0; + ifbp->IFB_MBp[0] = 0; //flag the MailBox as empty + ifbp->IFB_MBInfoLen = 0; + HCFASSERT( ifbp->IFB_MBSize >= 60 || ifbp->IFB_MBp == NULL, ifbp->IFB_MBSize ); +#undef P + break; + case CFG_MB_INFO: //store MailBoxInfoBlock + rc = put_info_mb( ifbp, (CFG_MB_INFO_STRCT FAR *)ltvp ); + break; + +#if (HCF_EXT) & HCF_EXT_NIC_ACCESS + case CFG_CMD_NIC: +#define P ((CFG_CMD_NIC_STRCT FAR *)ltvp) + OPW( HREG_PARAM_2, P->parm2 ); + OPW( HREG_PARAM_1, P->parm1 ); + rc = cmd_exe( ifbp, P->cmd, P->parm0 ); + P->hcf_stat = (hcf_16)rc; + P->stat = IPW( HREG_STAT ); + P->resp0 = IPW( HREG_RESP_0 ); + P->resp1 = IPW( HREG_RESP_1 ); + P->resp2 = IPW( HREG_RESP_2 ); + P->ifb_err_cmd = ifbp->IFB_ErrCmd; + P->ifb_err_qualifier = ifbp->IFB_ErrQualifier; +#undef P + break; + case CFG_CMD_HCF: +#define P ((CFG_CMD_HCF_STRCT FAR *)ltvp) + HCFASSERT( P->cmd == CFG_CMD_HCF_REG_ACCESS, P->cmd ); //only Hermes register access supported + if ( P->cmd == CFG_CMD_HCF_REG_ACCESS ) { + HCFASSERT( P->mode < ifbp->IFB_IOBase, P->mode ); //Check Register space + OPW( P->mode, P->add_info); + } +#undef P + break; +#endif // HCF_EXT_NIC_ACCESS + +#if (HCF_ASSERT) & HCF_ASSERT_PRINTF + case CFG_FW_PRINTF_BUFFER_LOCATION: + ifbp->IFB_FwPfBuff = *(CFG_FW_PRINTF_BUFFER_LOCATION_STRCT*)ltvp; + break; +#endif // HCF_ASSERT_PRINTF + + default: //pass everything unknown above the "FID" range to the Hermes or Dongle + rc = put_info( ifbp, ltvp ); + } + //DO NOT !!! HCFASSERT( rc == HCF_SUCCESS, rc ) /* 20 */ + HCFLOGEXIT( HCF_TRACE_PUT_INFO ); + return rc; +} // hcf_put_info + + +/************************************************************************************************************ + * + *.MODULE int hcf_rcv_msg( IFBP ifbp, DESC_STRCT *descp, unsigned int offset ) + *.PURPOSE All: decapsulate a message. + * pre-HermesII.5: verify MIC. + * non-USB, non-DMA mode: Transfer a message from the NIC to the Host and acknowledge reception. + * USB: Transform a message from proprietary USB format to 802.3 format + * + *.ARGUMENTS + * ifbp address of the Interface Block + * descp Pointer to the Descriptor List location. + * offset USB: not used + * non-USB: specifies the beginning of the data to be obtained (0 corresponds with DestAddr field + * of frame). + * + *.RETURNS + * HCF_SUCCESS No WPA error ( or HCF_ERR_MIC already reported by hcf_service_nic) + * HCF_ERR_MIC message contains an erroneous MIC ( HCF_SUCCESS is reported if HCF_ERR_MIC is already + * reported by hcf_service_nic) + * HCF_ERR_NO_NIC NIC removed during data retrieval + * HCF_ERR_DEFUNCT... + * + *.DESCRIPTION + * The Receive Message Function can be executed by the MSF to obtain the Data Info fields of the message that + * is reported to be available by the Service NIC Function. + * + * The Receive Message Function copies the message data available in the Card memory into a buffer structure + * provided by the MSF. + * Only data of the message indicated by the Service NIC Function can be obtained. + * Execution of the Service NIC function may result in the availability of a new message, but it definitely + * makes the message reported by the preceding Service NIC function, unavailable. + * + * in non-USB/non-DMA mode, hcf_rcv_msg starts the copy process at the (non-negative) offset requested by the + * parameter offset, relative to HFS_ADDR_DEST, e.g offset 0 starts copying from the Destination Address, the + * very begin of the 802.3 frame message. Offset must either lay within the part of the 802.3 frame as stored + * by hcf_service_nic in the lookahead buffer or be just behind it, i.e. the first byte not yet read. + * When offset is within lookahead, data is copied from lookahead. + * When offset is beyond lookahead, data is read directly from RxFS in NIC with disregard of the actual value + * of offset + * + *.NOTICE: + * o at entry: look ahead buffer as passed with hcf_service_nic is still accessible and unchanged + * o at exit: Receive Frame in NIC memory is released + * + * Description: + * Starting at the byte indicated by the Offset value, the bytes are copied from the Data Info + * Part of the current Receive Frame Structure to the Host memory data buffer structure + * identified by descp. + * The maximum value for Offset is the number of characters of the 802.3 frame read into the + * look ahead buffer by hcf_service_nic (i.e. the look ahead buffer size minus + * Control and 802.11 fields) + * If Offset is less than the maximum value, copying starts from the look ahead buffer till the + * end of that buffer is reached + * Then (or if the maximum value is specified for Offset), the + * message is directly copied from NIC memory to Host memory. + * If an invalid (i.e. too large) offset is specified, an assert catches but the buffer contents are + * undefined. + * Copying stops if either: + * o the end of the 802.3 frame is reached + * o the Descriptor with a NULL pointer in the next_desc_addr field is reached + * + * When the copying stops, the receiver is ack'ed, thus freeing the NIC memory where the frame is stored + * As a consequence, hcf_rcv_msg can only be called once for any particular Rx frame. + * + * For the time being (PCI Bus mastering not yet supported), only the following fields of each + * of the descriptors in the descriptor list must be set by the MSF: + * o buf_cntl.buf_dim[1] + * o *next_desc_addr + * o *buf_addr + * At return from hcf_rcv_msg, the field buf_cntl.buf_dim[0] of the used Descriptors reflects + * the number of bytes in the buffer corresponding with the Descriptor. + * On the last used Descriptor, buf_cntl.buf_dim[0] is less or equal to buf_cntl.buf_dim[1]. + * On all preceding Descriptors buf_cntl.buf_dim[0] is equal to buf_cntl.buf_dim[1]. + * On all succeeding (unused) Descriptors, buf_cntl.buf_dim[0] is zero. + * Note: this I/F is based on the assumptions how the I/F needed for PCI Bus mastering will + * be, so it may change. + * + * The most likely handling of HCF_ERR_NO_NIC by the MSF is to drop the already copied + * data as elegantly as possible under the constraints and requirements posed by the (N)OS. + * If no received Frame Structure is pending, "Success" rather than "Read error" is returned. + * This error constitutes a logic flaw in the MSF + * The HCF can only catch a minority of this + * type of errors + * Based on consistency ideas, the HCF catches none of these errors. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value + * - there is no unacknowledged Rx-message available + * - offset is out of range (outside look ahead buffer) + * - descp is a NULL pointer + * - any of the descriptors is not double word aligned + * - reentrancy, may be caused by calling hcf_functions without adequate protection + * against NIC interrupts or multi-threading. + * - Interrupts are enabled. + * + *.DIAGRAM + * + *.NOTICE + * - by using unsigned int as type for offset, no need to worry about negative offsets + * - Asserting on being enabled/present is superfluous, since a non-zero IFB_lal implies that hcf_service_nic + * was called and detected a Rx-message. A zero IFB_lal will set the BUF_CNT field of at least the first + * descriptor to zero. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +hcf_rcv_msg( IFBP ifbp, DESC_STRCT *descp, unsigned int offset ) +{ + int rc = HCF_SUCCESS; + wci_bufp cp; //char oriented working pointer + hcf_16 i; + int tot_len = ifbp->IFB_RxLen - offset; //total length + wci_bufp lap = ifbp->IFB_lap + offset; //start address in LookAhead Buffer + hcf_16 lal = ifbp->IFB_lal - offset; //available data within LookAhead Buffer + hcf_16 j; + + HCFLOGENTRY( HCF_TRACE_RCV_MSG, offset ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + HCFASSERT( descp, HCF_TRACE_RCV_MSG ); + HCFASSERT( ifbp->IFB_RxLen, HCF_TRACE_RCV_MSG ); + HCFASSERT( ifbp->IFB_RxLen >= offset, MERGE_2( offset, ifbp->IFB_RxLen ) ); + HCFASSERT( ifbp->IFB_lal >= offset, offset ); + HCFASSERT( (ifbp->IFB_CntlOpt & USE_DMA) == 0, 0xDADA ); + + if ( tot_len < 0 ) { + lal = 0; tot_len = 0; //suppress all copying activity in the do--while loop + } + do { //loop over all available fragments + // obnoxious hcf.c(1480) : warning C4769: conversion of near pointer to long integer + HCFASSERT( ((hcf_32)descp & 3 ) == 0, (hcf_32)descp ); + cp = descp->buf_addr; + j = min( (hcf_16)tot_len, descp->BUF_SIZE ); //minimum of "what's` available" and fragment size + descp->BUF_CNT = j; + tot_len -= j; //adjust length still to go + if ( lal ) { //if lookahead Buffer not yet completely copied + i = min( lal, j ); //minimum of "what's available" in LookAhead and fragment size + lal -= i; //adjust length still available in LookAhead + j -= i; //adjust length still available in current fragment + /*;? while loop could be improved by moving words but that is complicated on platforms with + * alignment requirements*/ + while ( i-- ) *cp++ = *lap++; + } + if ( j ) { //if LookAhead Buffer exhausted but still space in fragment, copy directly from NIC RAM + get_frag( ifbp, cp, j BE_PAR(0) ); + CALC_RX_MIC( cp, j ); + } + } while ( ( descp = descp->next_desc_addr ) != NULL ); +#if (HCF_TYPE) & HCF_TYPE_WPA + if ( ifbp->IFB_RxFID ) { + rc = check_mic( ifbp ); //prevents MIC error report if hcf_service_nic already consumed all + } +#endif // HCF_TYPE_WPA + (void)hcf_action( ifbp, HCF_ACT_RX_ACK ); //only 1 shot to get the data, so free the resources in the NIC + HCFASSERT( rc == HCF_SUCCESS, rc ); + HCFLOGEXIT( HCF_TRACE_RCV_MSG ); + return rc; +} // hcf_rcv_msg + + +/************************************************************************************************************ + * + *.MODULE int hcf_send_msg( IFBP ifbp, DESC_STRCT *descp, hcf_16 tx_cntl ) + *.PURPOSE Encapsulate a message and append padding and MIC. + * non-USB: Transfers the resulting message from Host to NIC and initiates transmission. + * USB: Transfer resulting message into a flat buffer. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * descp pointer to the DescriptorList or NULL + * tx_cntl indicates MAC-port and (Hermes) options + * HFS_TX_CNTL_SPECTRALINK + * HFS_TX_CNTL_PRIO + * HFS_TX_CNTL_TX_OK + * HFS_TX_CNTL_TX_EX + * HFS_TX_CNTL_TX_DELAY + * HFS_TX_CNTL_TX_CONT + * HCF_PORT_0 MAC Port 0 (default) + * HCF_PORT_1 (AP only) MAC Port 1 + * HCF_PORT_2 (AP only) MAC Port 2 + * HCF_PORT_3 (AP only) MAC Port 3 + * HCF_PORT_4 (AP only) MAC Port 4 + * HCF_PORT_5 (AP only) MAC Port 5 + * HCF_PORT_6 (AP only) MAC Port 6 + * + *.RETURNS + * HCF_SUCCESS + * HCF_ERR_DEFUNCT_.. + * HCF_ERR_TIME_OUT + * + *.DESCRIPTION: + * The Send Message Function embodies 2 functions: + * o transfers a message (including MAC header) from the provided buffer structure in Host memory to the Transmit + * Frame Structure (TxFS) in NIC memory. + * o Issue a send command to the F/W to actually transmit the contents of the TxFS. + * + * Control is based on the Resource Indicator IFB_RscInd. + * The Resource Indicator is maintained by the HCF and should only be interpreted but not changed by the MSF. + * The MSF must check IFB_RscInd to be non-zero before executing the call to the Send Message Function. + * When no resources are available, the MSF must handle the queuing of the Transmit frame and check the + * Resource Indicator periodically after calling hcf_service_nic. + * + * The Send Message Functions transfers a message to NIC memory when it is called with a non-NULL descp. + * Before the Send Message Function is invoked this way, the Resource Indicator (IFB_RscInd) must be checked. + * If the Resource is not available, Send Message Function execution must be postponed until after processing of + * a next hcf_service_nic it appears that the Resource has become available. + * The message is copied from the buffer structure identified by descp to the NIC. + * Copying stops if a NULL pointer in the next_desc_addr field is reached. + * Hcf_send_msg does not check for transmit buffer overflow, because the F/W does this protection. + * In case of a transmit buffer overflow, the surplus which does not fit in the buffer is simply dropped. + * + * The Send Message Function activates the F/W to actually send the message to the medium when the + * HFS_TX_CNTL_TX_DELAY bit of the tx_cntl parameter is not set. + * If the descp parameter of the current call is non-NULL, the message as represented by descp is send. + * If the descp parameter of the current call is NULL, and if the preceding call of the Send Message Function had + * a non-NULL descp and the preceding call had the HFS_TX_CNTL_TX_DELAY bit of tx_cntl set, then the message as + * represented by the descp of the preceding call is send. + * + * Hcf_send_msg supports encapsulation (see HCF_ENCAP) of Ethernet-II frames. + * An Ethernet-II frame is transferred to the Transmit Frame structure as an 802.3 frame. + * Hcf_send_msg distinguishes between an 802.3 and an Ethernet-II frame by looking at the data length/type field + * of the frame. If this field contains a value larger than 1514, the frame is considered to be an Ethernet-II + * frame, otherwise it is treated as an 802.3 frame. + * To ease implementation of the HCF, this type/type field must be located in the first descriptor structure, + * i.e. the 1st fragment must have a size of at least 14 (to contain DestAddr, SrcAddr and Len/Type field). + * An Ethernet-II frame is encapsulated by inserting a SNAP header between the addressing information and the + * type field. This insertion is transparent for the MSF. + * The HCF contains a fixed table that stores a number of types. If the value specified by the type/type field + * occurs in this table, Bridge Tunnel Encapsulation is used, otherwise RFC1042 encapsulation is used. + * Bridge Tunnel uses AA AA 03 00 00 F8 as SNAP header, + * RFC1042 uses AA AA 03 00 00 00 as SNAP header. + * The table currently contains: + * 0 0x80F3 AppleTalk Address Resolution Protocol (AARP) + * 0 0x8137 IPX + * + * The algorithm to distinguish between 802.3 and Ethernet-II frames limits the maximum length for frames of + * 802.3 frames to 1514 bytes. + * Encapsulation can be suppressed by means of the system constant HCF_ENCAP, e.g. to support proprietary + * protocols with 802.3 like frames with a size larger than 1514 bytes. + * + * In case the HCF encapsulates the frame, the number of bytes that is actually transmitted is determined by the + * cumulative value of the buf_cntl.buf_dim[0] fields. + * In case the HCF does not encapsulate the frame, the number of bytes that is actually transmitted is not + * determined by the cumulative value of the buf_cntl.buf_dim[DESC_CNTL_CNT] fields of the desc_strct's but by + * the Length field of the 802.3 frame. + * If there is a conflict between the cumulative value of the buf_cntl.buf_dim[0] fields and the + * 802.3 Length field the 802.3 Length field determines the number of bytes actually transmitted by the NIC while + * the cumulative value of the buf_cntl.buf_dim[0] fields determines the position of the MIC, hence a mismatch + * will result in MIC errors on the Receiving side. + * Currently this problem is flagged on the Transmit side by an Assert. + * The following fields of each of the descriptors in the descriptor list must be set by the MSF: + * o buf_cntl.buf_dim[0] + * o *next_desc_addr + * o *buf_addr + * + * All bits of the tx_cntl parameter except HFS_TX_CNTL_TX_DELAY and the HCF_PORT# bits are passed to the F/W via + * the HFS_TX_CNTL field of the TxFS. + * + * Note that hcf_send_msg does not detect NIC absence. The MSF is supposed to have its own -platform dependent- + * way to recognize card removal/insertion. + * The total system must be robust against card removal and there is no principal difference between card removal + * just after hcf_send_msg returns but before the actual transmission took place or sometime earlier. + * + * Assert fails if + * - ifbp has a recognizable out-of-range value + * - descp is a NULL pointer + * - no resources for PIF available. + * - Interrupts are enabled. + * - reentrancy, may be caused by calling hcf_functions without adequate protection + * against NIC interrupts or multi-threading. + * + *.DIAGRAM + *4: for the normal case (i.e. no HFS_TX_CNTL_TX_DELAY option active), a fid is acquired via the + * routine get_fid. If no FID is acquired, the remainder is skipped without an error notification. After + * all, the MSF is not supposed to call hcf_send_msg when no Resource is available. + *7: The ControlField of the TxFS is written. Since put_frag can only return the fatal Defunct or "No NIC", the + * return status can be ignored because when it fails, cmd_wait will fail as well. (see also the note on the + * need for a return code below). + * Note that HFS_TX_CNTL has different values for H-I, H-I/WPA and H-II and HFS_ADDR_DEST has different + * values for H-I (regardless of WPA) and H-II. + * By writing 17, 1 or 2 ( implying 16, 0 or 1 garbage word after HFS_TX_CNTL) the BAP just gets to + * HFS_ADDR_DEST for H-I, H-I/WPA and H-II respectively. + *10: if neither encapsulation nor MIC calculation is needed, splitting the first fragment in two does not + * really help but it makes the flow easier to follow to do not optimize on this difference + * + * hcf_send_msg checks whether the frame is an Ethernet-II rather than an "official" 802.3 frame. + * The E-II check is based on the length/type field in the MAC header. If this field has a value larger than + * 1500, E-II is assumed. The implementation of this test fails if the length/type field is not in the first + * descriptor. If E-II is recognized, a SNAP header is inserted. This SNAP header represents either RFC1042 + * or Bridge-Tunnel encapsulation, depending on the return status of the support routine hcf_encap. + * + *.NOTICE + * hcf_send_msg leaves the responsibility to only send messages on enabled ports at the MSF level. + * This is considered the strategy which is sufficiently adequate for all "robust" MSFs, have the least + * processor utilization and being still acceptable robust at the WCI !!!!! + * + * hcf_send_msg does not NEED a return value to report NIC absence or removal during the execution of + * hcf_send_msg(), because the MSF and higher layers must be able to cope anyway with the NIC being removed + * after a successful completion of hcf_send_msg() but before the actual transmission took place. + * To accommodate user expectations the current implementation does report NIC absence. + * Defunct blocks all NIC access and will (also) be reported on a number of other calls. + * + * hcf_send_msg does not check for transmit buffer overflow because the Hermes does this protection. + * In case of a transmit buffer overflow, the surplus which does not fit in the buffer is simply dropped. + * Note that this possibly results in the transmission of incomplete frames. + * + * After some deliberation with F/W team, it is decided that - being in the twilight zone of not knowing + * whether the problem at hand is an MSF bug, HCF buf, F/W bug, H/W malfunction or even something else - there + * is no "best thing to do" in case of a failing send, hence the HCF considers the TxFID ownership to be taken + * over by the F/W and hopes for an Allocate event in due time + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +hcf_send_msg( IFBP ifbp, DESC_STRCT *descp, hcf_16 tx_cntl ) +{ + int rc = HCF_SUCCESS; + DESC_STRCT *p /* = descp*/; //working pointer + hcf_16 len; // total byte count + hcf_16 i; + + hcf_16 fid = 0; + + HCFASSERT( ifbp->IFB_RscInd || descp == NULL, ifbp->IFB_RscInd ); + HCFASSERT( (ifbp->IFB_CntlOpt & USE_DMA) == 0, 0xDADB ); + + HCFLOGENTRY( HCF_TRACE_SEND_MSG, tx_cntl ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + /* obnoxious c:/hcf/hcf.c(1480) : warning C4769: conversion of near pointer to long integer, + * so skip */ + HCFASSERT( ((hcf_32)descp & 3 ) == 0, (hcf_32)descp ); +#if HCF_ASSERT + { int x = ifbp->IFB_FWIdentity.comp_id == COMP_ID_FW_AP ? tx_cntl & ~HFS_TX_CNTL_PORT : tx_cntl; + HCFASSERT( (x & ~HCF_TX_CNTL_MASK ) == 0, tx_cntl ); + } +#endif // HCF_ASSERT + + if ( descp ) ifbp->IFB_TxFID = 0; //cancel a pre-put message + + /* the following initialization code is redundant for a pre-put message + * but moving it inside the "if fid" logic makes the merging with the + * USB flow awkward + */ +#if (HCF_TYPE) & HCF_TYPE_WPA + tx_cntl |= ifbp->IFB_MICTxCntl; +#endif // HCF_TYPE_WPA + fid = ifbp->IFB_TxFID; + if (fid == 0 && ( fid = get_fid( ifbp ) ) != 0 ) /* 4 */ + /* skip the next compound statement if: + - pre-put message or + - no fid available (which should never occur if the MSF adheres to the WCI) + */ + { // to match the closing curly bracket of above "if" in case of HCF_TYPE_USB + //calculate total length ;? superfluous unless CCX or Encapsulation + len = 0; + p = descp; + do len += p->BUF_CNT; while ( ( p = p->next_desc_addr ) != NULL ); + p = descp; +//;? HCFASSERT( len <= HCF_MAX_MSG, len ); + /*7*/ (void)setup_bap( ifbp, fid, HFS_TX_CNTL, IO_OUT ); +#if (HCF_TYPE) & HCF_TYPE_TX_DELAY + HCFASSERT( ( descp != NULL ) ^ ( tx_cntl & HFS_TX_CNTL_TX_DELAY ), tx_cntl ); + if ( tx_cntl & HFS_TX_CNTL_TX_DELAY ) { + tx_cntl &= ~HFS_TX_CNTL_TX_DELAY; //!!HFS_TX_CNTL_TX_DELAY no longer available + ifbp->IFB_TxFID = fid; + fid = 0; //!!fid no longer available, be careful when modifying code + } +#endif // HCF_TYPE_TX_DELAY + OPW( HREG_DATA_1, tx_cntl ) ; + OPW( HREG_DATA_1, 0 ); + + HCFASSERT( p->BUF_CNT >= 14, p->BUF_CNT ); + /* assume DestAddr/SrcAddr/Len/Type ALWAYS contained in 1st fragment + * otherwise life gets too cumbersome for MIC and Encapsulation !!!!!!!! + if ( p->BUF_CNT >= 14 ) { alternatively: add a safety escape !!!!!!!!!!!! } */ + + CALC_TX_MIC( NULL, -1 ); //initialize MIC + /*10*/ put_frag( ifbp, p->buf_addr, HCF_DASA_SIZE BE_PAR(0) ); //write DA, SA with MIC calculation + CALC_TX_MIC( p->buf_addr, HCF_DASA_SIZE ); //MIC over DA, SA + CALC_TX_MIC( null_addr, 4 ); //MIC over (virtual) priority field + + //if encapsulation needed +#if (HCF_ENCAP) == HCF_ENC + //write length (with SNAP-header,Type, without //DA,SA,Length ) no MIC calc. + if ( ( snap_header[sizeof(snap_header)-1] = hcf_encap( &p->buf_addr[HCF_DASA_SIZE] ) ) != ENC_NONE ) { + OPW( HREG_DATA_1, CNV_END_SHORT( len + (sizeof(snap_header) + 2) - ( 2*6 + 2 ) ) ); + //write splice with MIC calculation + put_frag( ifbp, snap_header, sizeof(snap_header) BE_PAR(0) ); + CALC_TX_MIC( snap_header, sizeof(snap_header) ); //MIC over 6 byte SNAP + i = HCF_DASA_SIZE; + } else +#endif // HCF_ENC + { + OPW( HREG_DATA_1, *(wci_recordp)&p->buf_addr[HCF_DASA_SIZE] ); + i = 14; + } + //complete 1st fragment starting with Type with MIC calculation + put_frag( ifbp, &p->buf_addr[i], p->BUF_CNT - i BE_PAR(0) ); + CALC_TX_MIC( &p->buf_addr[i], p->BUF_CNT - i ); + + //do the remaining fragments with MIC calculation + while ( ( p = p->next_desc_addr ) != NULL ) { + /* obnoxious c:/hcf/hcf.c(1480) : warning C4769: conversion of near pointer to long integer, + * so skip */ + HCFASSERT( ((hcf_32)p & 3 ) == 0, (hcf_32)p ); + put_frag( ifbp, p->buf_addr, p->BUF_CNT BE_PAR(0) ); + CALC_TX_MIC( p->buf_addr, p->BUF_CNT ); + } + //pad message, finalize MIC calculation and write MIC to NIC + put_frag_finalize( ifbp ); + } + if ( fid ) { + /*16*/ rc = cmd_exe( ifbp, HCMD_BUSY | HCMD_TX | HCMD_RECL, fid ); + ifbp->IFB_TxFID = 0; + /* probably this (i.e. no RscInd AND "HREG_EV_ALLOC") at this point in time occurs so infrequent, + * that it might just as well be acceptable to skip this + * "optimization" code and handle that additional interrupt once in a while + */ +// 180 degree error in logic ;? #if ALLOC_15 + /*20*/ if ( ifbp->IFB_RscInd == 0 ) { + ifbp->IFB_RscInd = get_fid( ifbp ); + } +// #endif // ALLOC_15 + } +// HCFASSERT( level::ifbp->IFB_RscInd, ifbp->IFB_RscInd ); + HCFLOGEXIT( HCF_TRACE_SEND_MSG ); + return rc; +} // hcf_send_msg + + +/************************************************************************************************************ + * + *.MODULE int hcf_service_nic( IFBP ifbp, wci_bufp bufp, unsigned int len ) + *.PURPOSE Services (most) NIC events. + * Provides received message + * Provides status information. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * In non-DMA mode: + * bufp address of char buffer, sufficiently large to hold the first part of the RxFS up through HFS_TYPE + * len length in bytes of buffer specified by bufp + * value between HFS_TYPE + 2 and HFS_ADDR_DEST + HCF_MAX_MSG + * + *.RETURNS + * HCF_SUCCESS + * HCF_ERR_MIC message contains an erroneous MIC (only if frame fits completely in bufp) + * + *.DESCRIPTION + * + * MSF-accessible fields of Result Block + * - IFB_RxLen 0 or Frame size. + * - IFB_MBInfoLen 0 or the L-field of the oldest MBIB. + * - IFB_RscInd + * - IFB_HCF_Tallies updated if a corresponding event occurred. + * - IFB_NIC_Tallies updated if a Tally Info frame received from the NIC. + * - IFB_DmaPackets + * - IFB_TxFsStat + * - IFB_TxFsSwSup + * - IFB_LinkStat reflects new link status or 0x0000 if no change relative to previous hcf_service_nic call. +or +* - IFB_LinkStat link status, 0x8000 reflects change relative to previous hcf_service_nic call. +* +* When IFB_MBInfoLen is non-zero, at least one MBIB is available. +* +* IFB_RxLen reflects the number of received bytes in 802.3 view (Including DestAddr, SrcAddr and Length, +* excluding MIC-padding, MIC and sum check) of active Rx Frame Structure. If no Rx Data s available, IFB_RxLen +* equals 0x0000. +* Repeated execution causes the Service NIC Function to provide information about subsequently received +* messages, irrespective whether a hcf_rcv_msg or hcf_action(HCF_ACT_RX) is performed in between. +* +* When IFB_RxLen is non-zero, a Received Frame Structure is available to be routed to the protocol stack. +* When Monitor Mode is not active, this is guaranteed to be an error-free non-WMP frame. +* In case of Monitor Mode, it may also be a frame with an error or a WMP frame. +* Erroneous frames have a non-zero error-sub field in the HFS_STAT field in the look ahead buffer. +* +* If a Receive message is available in NIC RAM, the Receive Frame Structure is (partly) copied from the NIC to +* the buffer identified by bufp. +* Copying stops either after len bytes or when the complete 802.3 frame is copied. +* During the copying the message is decapsulated (if appropriate). +* If the frame is read completely by hcf_service_nic (i.e. the frame fits completely in the lookahead buffer), +* the frame is automatically ACK'ed to the F/W and still available via the look ahead buffer and hcf_rcv_msg. +* Only if the frame is read completely by hcf_service_nic, hcf_service_nic checks the MIC and sets the return +* status accordingly. In this case, hcf_rcv_msg does not check the MIC. +* +* The MIC calculation algorithm works more efficient if the length of the look ahead buffer is +* such that it fits exactly 4 n bytes of the 802.3 frame, i.e. len == HFS_ADDR_DEST + 4*n. +* +* The Service NIC Function supports the NIC event service handling process. +* It performs the appropriate actions to service the NIC, such that the event cause is eliminated and related +* information is saved. +* The Service NIC Function is executed by the MSF ISR or polling routine as first step to determine the event +* cause(s). It is the responsibility of the MSF to perform all not directly NIC related interrupt service +* actions, e.g. in a PC environment this includes servicing the PIC, and managing the Processor Interrupt +* Enabling/Disabling. +* In case of a polled based system, the Service NIC Function must be executed "frequently". +* The Service NIC Function may have side effects related to the Mailbox and Resource Indicator (IFB_RscInd). +* +* hcf_service_nic returns: +* - The length of the data in the available MBIB (IFB_MBInfoLen) +* - Changes in the link status (IFB_LinkStat) +* - The length of the data in the available Receive Frame Structure (IFB_RxLen) +* - updated IFB_RscInd +* - Updated Tallies +* +* hcf_service_nic is presumed to neither interrupt other HCF-tasks nor to be interrupted by other HCF-tasks. +* A way to achieve this is to precede hcf_service_nic as well as all other HCF-tasks with a call to +* hcf_action to disable the card interrupts and, after all work is completed, with a call to hcf_action to +* restore (which is not necessarily the same as enabling) the card interrupts. +* In case of a polled environment, it is assumed that the MSF programmer is sufficiently familiar with the +* specific requirements of that environment to translate the interrupt strategy to a polled strategy. +* +* hcf_service_nic services the following Hermes events: +* - HREG_EV_INFO Asynchronous Information Frame +* - HREG_EV_INFO_DROP WMAC did not have sufficient RAM to build Unsolicited Information Frame +* - HREG_EV_TX_EXC (if applicable, i.e. selected via HCF_EXT_INT_TX_EX bit of HCF_EXT) +* - HREG_EV_SLEEP_REQ (if applicable, i.e. selected via HCF_DDS/HCF_CDS bit of HCF_SLEEP) +* ** in non_DMA mode +* - HREG_EV_ALLOC Asynchronous part of Allocation/Reclaim completed while out of resources at +* completion of hcf_send_msg/notify +* - HREG_EV_RX the detection of the availability of received messages +* including WaveLAN Management Protocol (WMP) message processing +* ** in DMA mode +* - HREG_EV_RDMAD +* - HREG_EV_TDMAD +*!! hcf_service_nic does not service the following Hermes events: +*!! HREG_EV_TX (the "OK" Tx Event) is no longer supported by the WCI, if it occurs it is unclear +*!! what the cause is, so no meaningful strategy is available. Not acking the bit is +*!! probably the best help that can be given to the debugger. +*!! HREG_EV_CMD handled in cmd_wait. +*!! HREG_EV_FW_DMA (i.e. HREG_EV_RXDMA, HREG_EV_TXDMA and_EV_LPESC) are either not used or used +*!! between the F/W and the DMA engine. +*!! HREG_EV_ACK_REG_READY is only applicable for H-II (i.e. not HII.5 and up, see DAWA) +* +* If, in non-DMA mode, a Rx message is available, its length is reflected by the IFB_RxLen field of the IFB. +* This length reflects the data itself and the Destination Address, Source Address and DataLength/Type field +* but not the SNAP-header in case of decapsulation by the HCF. If no message is available, IFB_RxLen is +* zero. Former versions of the HCF handled WMP messages and supported a "monitor" mode in hcf_service_nic, +* which deposited certain or all Rx messages in the MailBox. The responsibility to handle these frames is +* moved to the MSF. The HCF offers as supports hcf_put_info with CFG_MB_INFO as parameter to emulate the old +* implementation under control of the MSF. +* +* **Rx Buffer free strategy +* When hcf_service_nic reports the availability of a non-DMA message, the MSF can access that message by +* means of hcf_rcv_msg. It must be prevented that the LAN Controller writes new data in the NIC buffer +* before the MSF is finished with the current message. The NIC buffer is returned to the LAN Controller +* when: +* - the complete frame fits in the lookahead buffer or +* - hcf_rcv_msg is called or +* - hcf_action with HCF_ACT_RX is called or +* - hcf_service_nic is called again +* It can be reasoned that hcf_action( INT_ON ) should not be given before the MSF has completely processed +* a reported Rx-frame. The reason is that the INT_ON action is guaranteed to cause a (Rx-)interrupt (the +* MSF is processing a Rx-frame, hence the Rx-event bit in the Hermes register must be active). This +* interrupt will cause hcf_service_nic to be called, which will cause the ack-ing of the "last" Rx-event +* to the Hermes, causing the Hermes to discard the associated NIC RAM buffer. +* Assert fails if +* - ifbp is zero or other recognizable out-of-range value. +* - hcf_service_nic is called without a prior call to hcf_connect. +* - interrupts are enabled. +* - reentrancy, may be caused by calling hcf_functions without adequate protection +* against NIC interrupts or multi-threading. +* +* +*.DIAGRAM +*1: IFB_LinkStat is cleared, if a LinkStatus frame is received, IFB_LinkStat will be updated accordingly +* by isr_info. +or +*1: IFB_LinkStat change indication is cleared. If a LinkStatus frame is received, IFB_LinkStat will be updated +* accordingly by isr_info. +*2: IFB_RxLen must be cleared before the NIC presence check otherwise: +* - this value may stay non-zero if the NIC is pulled out at an inconvenient moment. +* - the RxAck on a zero-FID needs a zero-value for IFB_RxLen to work +* Note that as side-effect of the hcf_action call, the remainder of Rx related info is re-initialized as +* well. +*4: In case of Defunct mode, the information supplied by Hermes is unreliable, so the body of +* hcf_service_nic is skipped. Since hcf_cntl turns into a NOP if Primary or Station F/W is incompatible, +* hcf_service_nic is also skipped in those cases. +* To prevent that hcf_service_nic reports bogus information to the MSF with all - possibly difficult to +* debug - undesirable side effects, it is paramount to check the NIC presence. In former days the presence +* test was based on the Hermes register HREG_SW_0. Since in HCF_ACT_INT_OFF is chosen for strategy based on +* HREG_EV_STAT, this is now also used in hcf_service_nic. The motivation to change strategy is partly +* due to inconsistent F/W implementations with respect to HREG_SW_0 manipulation around reset and download. +* Note that in polled environments Card Removal is not detected by INT_OFF which makes the check in +* hcf_service_nic even more important. +*8: The event status register of the Hermes is sampled +* The assert checks for unexpected events ;?????????????????????????????????????. +* - HREG_EV_INFO_DROP is explicitly excluded from the acceptable HREG_EV_STAT bits because it indicates +* a too heavily loaded system. +* - HREG_EV_ACK_REG_READY is 0x0000 for H-I (and hopefully H-II.5) +* +* +* HREG_EV_TX_EXC is accepted (via HREG_EV_TX_EXT) if and only if HCF_EXT_INT_TX_EX set in the HCF_EXT +* definition at compile time. +* The following activities are handled: +* - Alloc events are handled by hcf_send_msg (and notify). Only if there is no "spare" resource, the +* alloc event is superficially serviced by hcf_service_nic to create a pseudo-resource with value +* 0x001. This value is recognized by get_fid (called by hcf_send_msg and notify) where the real +* TxFid is retrieved and the Hermes is acked and - hopefully - the "normal" case with a spare TxFid +* in IFB_RscInd is restored. +* - Info drop events are handled by incrementing a tally +* - LinkEvent (including solicited and unsolicited tallies) are handled by procedure isr_info. +* - TxEx (if selected at compile time) is handled by copying the significant part of the TxFS +* into the IFB for further processing by the MSF. +* Note the complication of the zero-FID protection sub-scheme in DAWA. +* Note, the Ack of all of above events is handled at the end of hcf_service_nic +*16: In case of non-DMA ( either not compiled in or due to a run-time choice): +* If an Rx-frame is available, first the FID of that frame is read, including the complication of the +* zero-FID protection sub-scheme in DAWA. Note that such a zero-FID is acknowledged at the end of +* hcf_service_nic and that this depends on the IFB_RxLen initialization in the begin of hcf_service_nic. +* The Assert validates the HCF assumption about Hermes implementation upon which the range of +* Pseudo-RIDs is based. +* Then the control fields up to the start of the 802.3 frame are read from the NIC into the lookahead buffer. +* The status field is converted to native Endianness. +* The length is, after implicit Endianness conversion if needed, and adjustment for the 14 bytes of the +* 802.3 MAC header, stored in IFB_RxLen. +* In MAC Monitor mode, 802.11 control frames with a TOTAL length of 14 are received, so without this +* length adjustment, IFB_RxLen could not be used to distinguish these frames from "no frame". +* No MIC calculation processes are associated with the reading of these Control fields. +*26: This length test feels like superfluous robustness against malformed frames, but it turned out to be +* needed in the real (hostile) world. +* The decapsulation check needs sufficient data to represent DA, SA, L, SNAP and Type which amounts to +* 22 bytes. In MAC Monitor mode, 802.11 control frames with a smaller length are received. To prevent +* that the implementation goes haywire, a check on the length is needed. +* The actual decapsulation takes place on the fly in the copying process by overwriting the SNAP header. +* Note that in case of decapsulation the SNAP header is not passed to the MSF, hence IFB_RxLen must be +* compensated for the SNAP header length. +* The 22 bytes needed for decapsulation are (more than) sufficient for the exceptional handling of the +* MIC algorithm of the L-field (replacing the 2 byte L-field with 4 0x00 bytes). +*30: The 12 in the no-WPA branch corresponds with the get_frag, the 2 with the IPW of the WPA branch +*32: If Hermes reported MIC-presence, than the MIC engine is initialized with the non-dummy MIC calculation +* routine address and appropriate key. +*34: The 8 bytes after the DA, SA, L are read and it is checked whether decapsulation is needed i.e.: +* - the Hermes reported Tunnel encapsulation or +* - the Hermes reported 1042 Encapsulation and hcf_encap reports that the HCF would not have used +* 1042 as the encapsulation mechanism +* Note that the first field of the RxFS in bufp has Native Endianness due to the conversion done by the +* BE_PAR in get_frag. +*36: The Type field is the only word kept (after moving) of the just read 8 bytes, it is moved to the +* L-field. The original L-field and 6 byte SNAP header are discarded, so IFB_RxLen and buf_addr must +* be adjusted by 8. +*40: Determine how much of the frame (starting with DA) fits in the Lookahead buffer, then read the not-yet +* read data into the lookahead buffer. +* If the lookahead buffer contains the complete message, check the MIC. The majority considered this +* I/F more appropriate then have the MSF call hcf_get_data only to check the MIC. +*44: Since the complete message is copied from NIC RAM to PC RAM, the Rx can be acknowledged to the Hermes +* to optimize the flow ( a better chance to get new Rx data in the next pass through hcf_service_nic ). +* This acknowledgement can not be done via hcf_action( HCF_ACT_RX_ACK ) because this also clears +* IFB_RxLEN thus corrupting the I/F to the MSF. +*;?: In case of DMA (compiled in and activated): + + +*54: Limiting the number of places where the F/W is acked (e.g. the merging of the Rx-ACK with the other +* ACKs), is supposed to diminish the potential of race conditions in the F/W. +* Note 1: The CMD event is acknowledged in cmd_cmpl +* Note 2: HREG_EV_ACK_REG_READY is 0x0000 for H-I (and hopefully H-II.5) +* Note 3: The ALLOC event is acknowledged in get_fid (except for the initialization flow) +* +*.NOTICE +* The Non-DMA HREG_EV_RX is handled different compared with the other F/W events. +* The HREG_EV_RX event is acknowledged by the first hcf_service_nic call after the +* hcf_service_nic call that reported the occurrence of this event. +* This acknowledgment +* makes the next Receive Frame Structure (if any) available. +* An updated IFB_RxLen +* field reflects this availability. +* +*.NOTICE +* The minimum size for Len must supply space for: +* - an F/W dependent number of bytes of Control Info field including the 802.11 Header field +* - Destination Address +* - Source Address +* - Length field +* - [ SNAP Header] +* - [ Ethernet-II Type] +* This results in 68 for Hermes-I and 80 for Hermes-II +* This way the minimum amount of information is available needed by the HCF to determine whether the frame +* must be decapsulated. +*.ENDDOC END DOCUMENTATION +* +************************************************************************************************************/ +int +hcf_service_nic( IFBP ifbp, wci_bufp bufp, unsigned int len ) +{ + + int rc = HCF_SUCCESS; + hcf_16 stat; + wci_bufp buf_addr; + hcf_16 i; + + HCFLOGENTRY( HCF_TRACE_SERVICE_NIC, ifbp->IFB_IntOffCnt ); + HCFASSERT( ifbp->IFB_Magic == HCF_MAGIC, ifbp->IFB_Magic ); + HCFASSERT_INT; + + ifbp->IFB_LinkStat = 0; // ;? to be obsoleted ASAP /* 1*/ + ifbp->IFB_DSLinkStat &= ~CFG_LINK_STAT_CHANGE; /* 1*/ + (void)hcf_action( ifbp, HCF_ACT_RX_ACK ); /* 2*/ + if ( ifbp->IFB_CardStat == 0 && ( stat = IPW( HREG_EV_STAT ) ) != 0xFFFF ) { /* 4*/ +/* IF_NOT_DMA( HCFASSERT( !( stat & ~HREG_EV_BASIC_MASK, stat ) ) + * IF_NOT_USE_DMA( HCFASSERT( !( stat & ~HREG_EV_BASIC_MASK, stat ) ) + * IF_USE_DMA( HCFASSERT( !( stat & ~( HREG_EV_BASIC_MASK ^ ( HREG_EV_...DMA.... ), stat ) ) + */ + /* 8*/ + if ( ifbp->IFB_RscInd == 0 && stat & HREG_EV_ALLOC ) { //Note: IFB_RscInd is ALWAYS 1 for DMA + ifbp->IFB_RscInd = 1; + } + IF_TALLY( if ( stat & HREG_EV_INFO_DROP ) { ifbp->IFB_HCF_Tallies.NoBufInfo++; } ); +#if (HCF_EXT) & HCF_EXT_INT_TICK + if ( stat & HREG_EV_TICK ) { + ifbp->IFB_TickCnt++; + } +#if 0 // (HCF_SLEEP) & HCF_DDS + if ( ifbp->IFB_TickCnt == 3 && ( ifbp->IFB_DSLinkStat & CFG_LINK_STAT_CONNECTED ) == 0 ) { + CFG_DDS_TICK_TIME_STRCT ltv; + // 2 second period (with 1 tick uncertanty) in not-connected mode -->go into DS_OOR + hcf_action( ifbp, HCF_ACT_SLEEP ); + ifbp->IFB_DSLinkStat |= CFG_LINK_STAT_DS_OOR; //set OutOfRange + ltv.len = 2; + ltv.typ = CFG_DDS_TICK_TIME; + ltv.tick_time = ( ( ifbp->IFB_DSLinkStat & CFG_LINK_STAT_TIMER ) + 0x10 ) *64; //78 is more right + hcf_put_info( ifbp, (LTVP)<v ); + printk(KERN_NOTICE "Preparing for sleep, link_status: %04X, timer : %d\n", + ifbp->IFB_DSLinkStat, ltv.tick_time );//;?remove me 1 day + ifbp->IFB_TickCnt++; //;?just to make sure we do not keep on printing above message + if ( ltv.tick_time < 300 * 125 ) ifbp->IFB_DSLinkStat += 0x0010; + + } +#endif // HCF_DDS +#endif // HCF_EXT_INT_TICK + if ( stat & HREG_EV_INFO ) { + isr_info( ifbp ); + } +#if (HCF_EXT) & HCF_EXT_INT_TX_EX + if ( stat & HREG_EV_TX_EXT && ( i = IPW( HREG_TX_COMPL_FID ) ) != 0 /*DAWA*/ ) { + DAWA_ZERO_FID( HREG_TX_COMPL_FID ); + (void)setup_bap( ifbp, i, 0, IO_IN ); + get_frag( ifbp, &ifbp->IFB_TxFsStat, HFS_SWSUP BE_PAR(1) ); + } +#endif // HCF_EXT_INT_TX_EX +//!rlav DMA engine will handle the rx event, not the driver +#if HCF_DMA + if ( !( ifbp->IFB_CntlOpt & USE_DMA ) ) //!! be aware of the logical indentations +#endif // HCF_DMA + /*16*/ if ( stat & HREG_EV_RX && ( ifbp->IFB_RxFID = IPW( HREG_RX_FID ) ) != 0 ) { //if 0 then DAWA_ACK + HCFASSERT( bufp, len ); + HCFASSERT( len >= HFS_DAT + 2, len ); + DAWA_ZERO_FID( HREG_RX_FID ); + HCFASSERT( ifbp->IFB_RxFID < CFG_PROD_DATA, ifbp->IFB_RxFID); + (void)setup_bap( ifbp, ifbp->IFB_RxFID, 0, IO_IN ); + get_frag( ifbp, bufp, HFS_ADDR_DEST BE_PAR(1) ); + ifbp->IFB_lap = buf_addr = bufp + HFS_ADDR_DEST; + ifbp->IFB_RxLen = (hcf_16)(bufp[HFS_DAT_LEN] + (bufp[HFS_DAT_LEN+1]<<8) + 2*6 + 2); + /*26*/ if ( ifbp->IFB_RxLen >= 22 ) { // convenient for MIC calculation (5 DWs + 1 "skipped" W) + //. get DA,SA,Len/Type and (SNAP,Type or 8 data bytes) + /*30*/ get_frag( ifbp, buf_addr, 22 BE_PAR(0) ); + /*32*/ CALC_RX_MIC( bufp, -1 ); //. initialize MIC + CALC_RX_MIC( buf_addr, HCF_DASA_SIZE ); //. MIC over DA, SA + CALC_RX_MIC( null_addr, 4 ); //. MIC over (virtual) priority field + CALC_RX_MIC( buf_addr+14, 8 ); //. skip Len, MIC over SNAP,Type or 8 data bytes) + buf_addr += 22; +#if (HCF_ENCAP) == HCF_ENC + HCFASSERT( len >= HFS_DAT + 2 + sizeof(snap_header), len ); + /*34*/ i = *(wci_recordp)&bufp[HFS_STAT] & ( HFS_STAT_MSG_TYPE | HFS_STAT_ERR ); + if ( i == HFS_STAT_TUNNEL || + ( i == HFS_STAT_1042 && hcf_encap( (wci_bufp)&bufp[HFS_TYPE] ) != ENC_TUNNEL ) ) { + //. copy E-II Type to 802.3 LEN field + /*36*/ bufp[HFS_LEN ] = bufp[HFS_TYPE ]; + bufp[HFS_LEN+1] = bufp[HFS_TYPE+1]; + //. discard Snap by overwriting with data + ifbp->IFB_RxLen -= (HFS_TYPE - HFS_LEN); + buf_addr -= ( HFS_TYPE - HFS_LEN ); // this happens to bring us at a DW boundary of 36 + } +#endif // HCF_ENC + } + /*40*/ ifbp->IFB_lal = min( (hcf_16)(len - HFS_ADDR_DEST), ifbp->IFB_RxLen ); + i = ifbp->IFB_lal - ( buf_addr - ( bufp + HFS_ADDR_DEST ) ); + get_frag( ifbp, buf_addr, i BE_PAR(0) ); + CALC_RX_MIC( buf_addr, i ); +#if (HCF_TYPE) & HCF_TYPE_WPA + if ( ifbp->IFB_lal == ifbp->IFB_RxLen ) { + rc = check_mic( ifbp ); + } +#endif // HCF_TYPE_WPA + /*44*/ if ( len - HFS_ADDR_DEST >= ifbp->IFB_RxLen ) { + ifbp->IFB_RxFID = 0; + } else { /* IFB_RxFID is cleared, so you do not get another Rx_Ack at next entry of hcf_service_nic */ + stat &= (hcf_16)~HREG_EV_RX; //don't ack Rx if processing not yet completed + } + } + // in case of DMA: signal availability of rx and/or tx packets to MSF + IF_USE_DMA( ifbp->IFB_DmaPackets |= stat & ( HREG_EV_RDMAD | HREG_EV_TDMAD ) ); + // rlav : pending HREG_EV_RDMAD or HREG_EV_TDMAD events get acknowledged here. + /*54*/ stat &= (hcf_16)~( HREG_EV_SLEEP_REQ | HREG_EV_CMD | HREG_EV_ACK_REG_READY | HREG_EV_ALLOC | HREG_EV_FW_DMA ); +//a positive mask would be easier to understand /*54*/ stat &= (hcf_16)~( HREG_EV_SLEEP_REQ | HREG_EV_CMD | HREG_EV_ACK_REG_READY | HREG_EV_ALLOC | HREG_EV_FW_DMA ); + IF_USE_DMA( stat &= (hcf_16)~HREG_EV_RX ); + if ( stat ) { + DAWA_ACK( stat ); /*DAWA*/ + } + } + HCFLOGEXIT( HCF_TRACE_SERVICE_NIC ); + return rc; +} // hcf_service_nic + + +/************************************************************************************************************ + ************************** H C F S U P P O R T R O U T I N E S ****************************************** + ************************************************************************************************************/ + + +/************************************************************************************************************ + * + *.SUBMODULE void calc_mic( hcf_32* p, hcf_32 m ) + *.PURPOSE calculate MIC on a quad byte. + * + *.ARGUMENTS + * p address of the MIC + * m 32 bit value to be processed by the MIC calculation engine + * + *.RETURNS N.A. + * + *.DESCRIPTION + * calc_mic is the implementation of the MIC algorithm. It is a monkey-see monkey-do copy of + * Michael::appendByte() + * of Appendix C of .......... + * + * + *.DIAGRAM + * + *.NOTICE + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ + +#if (HCF_TYPE) & HCF_TYPE_WPA + +#define ROL32( A, n ) ( ((A) << (n)) | ( ((A)>>(32-(n))) & ( (1UL << (n)) - 1 ) ) ) +#define ROR32( A, n ) ROL32( (A), 32-(n) ) + +#define L *p +#define R *(p+1) + +void +calc_mic( hcf_32* p, hcf_32 m ) +{ +#if HCF_BIG_ENDIAN + m = (m >> 16) | (m << 16); +#endif // HCF_BIG_ENDIAN + L ^= m; + R ^= ROL32( L, 17 ); + L += R; + R ^= ((L & 0xff00ff00) >> 8) | ((L & 0x00ff00ff) << 8); + L += R; + R ^= ROL32( L, 3 ); + L += R; + R ^= ROR32( L, 2 ); + L += R; +} // calc_mic +#undef R +#undef L +#endif // HCF_TYPE_WPA + + + +#if (HCF_TYPE) & HCF_TYPE_WPA +/************************************************************************************************************ + * + *.SUBMODULE void calc_mic_rx_frag( IFBP ifbp, wci_bufp p, int len ) + *.PURPOSE calculate MIC on a single fragment. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * bufp (byte) address of buffer + * len length in bytes of buffer specified by bufp + * + *.RETURNS N.A. + * + *.DESCRIPTION + * calc_mic_rx_frag ........ + * + * The MIC is located in the IFB. + * The MIC is separate for Tx and Rx, thus allowing hcf_send_msg to occur between hcf_service_nic and + * hcf_rcv_msg. + * + * + *.DIAGRAM + * + *.NOTICE + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +void +calc_mic_rx_frag( IFBP ifbp, wci_bufp p, int len ) +{ + static union { hcf_32 x32; hcf_16 x16[2]; hcf_8 x8[4]; } x; //* area to accumulate 4 bytes input for MIC engine + int i; + + if ( len == -1 ) { //initialize MIC housekeeping + i = *(wci_recordp)&p[HFS_STAT]; + /* i = CNV_SHORTP_TO_LITTLE(&p[HFS_STAT]); should not be neede to prevent alignment poroblems + * since len == -1 if and only if p is lookahaead buffer which MUST be word aligned + * to be re-investigated by NvR + */ + + if ( ( i & HFS_STAT_MIC ) == 0 ) { + ifbp->IFB_MICRxCarry = 0xFFFF; //suppress MIC calculation + } else { + ifbp->IFB_MICRxCarry = 0; +//* Note that "coincidentally" the bit positions used in HFS_STAT +//* correspond with the offset of the key in IFB_MICKey + i = ( i & HFS_STAT_MIC_KEY_ID ) >> 10; /* coincidentally no shift needed for i itself */ + ifbp->IFB_MICRx[0] = CNV_LONG_TO_LITTLE(ifbp->IFB_MICRxKey[i ]); + ifbp->IFB_MICRx[1] = CNV_LONG_TO_LITTLE(ifbp->IFB_MICRxKey[i+1]); + } + } else { + if ( ifbp->IFB_MICRxCarry == 0 ) { + x.x32 = CNV_LONGP_TO_LITTLE(p); + p += 4; + if ( len < 4 ) { + ifbp->IFB_MICRxCarry = (hcf_16)len; + } else { + ifbp->IFB_MICRxCarry = 4; + len -= 4; + } + } else while ( ifbp->IFB_MICRxCarry < 4 && len ) { //note for hcf_16 applies: 0xFFFF > 4 + x.x8[ifbp->IFB_MICRxCarry++] = *p++; + len--; + } + while ( ifbp->IFB_MICRxCarry == 4 ) { //contrived so we have only 1 call to calc_mic so we could bring it in-line + calc_mic( ifbp->IFB_MICRx, x.x32 ); + x.x32 = CNV_LONGP_TO_LITTLE(p); + p += 4; + if ( len < 4 ) { + ifbp->IFB_MICRxCarry = (hcf_16)len; + } + len -= 4; + } + } +} // calc_mic_rx_frag +#endif // HCF_TYPE_WPA + + +#if (HCF_TYPE) & HCF_TYPE_WPA +/************************************************************************************************************ + * + *.SUBMODULE void calc_mic_tx_frag( IFBP ifbp, wci_bufp p, int len ) + *.PURPOSE calculate MIC on a single fragment. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * bufp (byte) address of buffer + * len length in bytes of buffer specified by bufp + * + *.RETURNS N.A. + * + *.DESCRIPTION + * calc_mic_tx_frag ........ + * + * The MIC is located in the IFB. + * The MIC is separate for Tx and Rx, thus allowing hcf_send_msg to occur between hcf_service_nic and + * hcf_rcv_msg. + * + * + *.DIAGRAM + * + *.NOTICE + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +void +calc_mic_tx_frag( IFBP ifbp, wci_bufp p, int len ) +{ + static union { hcf_32 x32; hcf_16 x16[2]; hcf_8 x8[4]; } x; //* area to accumulate 4 bytes input for MIC engine + + //if initialization request + if ( len == -1 ) { + //. presume MIC calculation disabled + ifbp->IFB_MICTxCarry = 0xFFFF; + //. if MIC calculation enabled + if ( ifbp->IFB_MICTxCntl ) { + //. . clear MIC carry + ifbp->IFB_MICTxCarry = 0; + //. . initialize MIC-engine + ifbp->IFB_MICTx[0] = CNV_LONG_TO_LITTLE(ifbp->IFB_MICTxKey[0]); /*Tx always uses Key 0 */ + ifbp->IFB_MICTx[1] = CNV_LONG_TO_LITTLE(ifbp->IFB_MICTxKey[1]); + } + //else + } else { + //. if MIC enabled (Tx) / if MIC present (Rx) + //. and no carry from previous calc_mic_frag + if ( ifbp->IFB_MICTxCarry == 0 ) { + //. . preset accu with 4 bytes from buffer + x.x32 = CNV_LONGP_TO_LITTLE(p); + //. . adjust pointer accordingly + p += 4; + //. . if buffer contained less then 4 bytes + if ( len < 4 ) { + //. . . promote valid bytes in accu to carry + //. . . flag accu to contain incomplete double word + ifbp->IFB_MICTxCarry = (hcf_16)len; + //. . else + } else { + //. . . flag accu to contain complete double word + ifbp->IFB_MICTxCarry = 4; + //. . adjust remaining buffer length + len -= 4; + } + //. else if MIC enabled + //. and if carry bytes from previous calc_mic_tx_frag + //. . move (1-3) bytes from carry into accu + } else while ( ifbp->IFB_MICTxCarry < 4 && len ) { /* note for hcf_16 applies: 0xFFFF > 4 */ + x.x8[ifbp->IFB_MICTxCarry++] = *p++; + len--; + } + //. while accu contains complete double word + //. and MIC enabled + while ( ifbp->IFB_MICTxCarry == 4 ) { + //. . pass accu to MIC engine + calc_mic( ifbp->IFB_MICTx, x.x32 ); + //. . copy next 4 bytes from buffer to accu + x.x32 = CNV_LONGP_TO_LITTLE(p); + //. . adjust buffer pointer + p += 4; + //. . if buffer contained less then 4 bytes + //. . . promote valid bytes in accu to carry + //. . . flag accu to contain incomplete double word + if ( len < 4 ) { + ifbp->IFB_MICTxCarry = (hcf_16)len; + } + //. . adjust remaining buffer length + len -= 4; + } + } +} // calc_mic_tx_frag +#endif // HCF_TYPE_WPA + + +#if HCF_PROT_TIME +/************************************************************************************************************ + * + *.SUBMODULE void calibrate( IFBP ifbp ) + *.PURPOSE calibrates the S/W protection counter against the Hermes Timer tick. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS N.A. + * + *.DESCRIPTION + * calibrates the S/W protection counter against the Hermes Timer tick + * IFB_TickIni is the value used to initialize the S/W protection counter such that the expiration period + * more or less independent of the processor speed. If IFB_TickIni is not yet calibrated, it is done now. + * This calibration is "reasonably" accurate because the Hermes is in a quiet state as a result of the + * Initialize command. + * + * + *.DIAGRAM + * + *1: IFB_TickIni is initialized at INI_TICK_INI by hcf_connect. If calibrate succeeds, IFB_TickIni is + * guaranteed to be changed. As a consequence there will be only 1 shot at calibration (regardless of the + * number of init calls) under normal circumstances. + *2: Calibration is done HCF_PROT_TIME_CNT times. This diminish the effects of jitter and interference, + * especially in a pre-emptive environment. HCF_PROT_TIME_CNT is in the range of 16 through 32 and derived + * from the HCF_PROT_TIME specified by the MSF programmer. The divisor needed to scale HCF_PROT_TIME into the + * 16-32 range, is used as a multiplicator after the calibration, to scale the found value back to the + * requested range. This way a compromise is achieved between accuracy and duration of the calibration + * process. + *3: Acknowledge the Timer Tick Event. + * Each cycle is limited to at most INI_TICK_INI samples of the TimerTick status of the Hermes. + * Since the start of calibrate is unrelated to the Hermes Internal Timer, the first interval may last from 0 + * to the normal interval, all subsequent intervals should be the full length of the Hermes Tick interval. + * The Hermes Timer Tick is not reprogrammed by the HCF, hence it is running at the default of 10 k + * microseconds. + *4: If the Timer Tick Event is continuously up (prot_cnt still has the value INI_TICK_INI) or no Timer Tick + * Event occurred before the protection counter expired, reset IFB_TickIni to INI_TICK_INI, + * set the defunct bit of IFB_CardStat (thus rendering the Hermes inoperable) and exit the calibrate routine. + *8: ifbp->IFB_TickIni is multiplied to scale the found value back to the requested range as explained under 2. + * + *.NOTICE + * o Although there are a number of viewpoints possible, calibrate() uses as error strategy that a single + * failure of the Hermes TimerTick is considered fatal. + * o There is no hard and concrete time-out value defined for Hermes activities. The default 1 seconds is + * believed to be sufficiently "relaxed" for real life and to be sufficiently short to be still useful in an + * environment with humans. + * o Note that via IFB_DefunctStat time outs in cmd_wait and in hcfio_string block all Hermes access till the + * next init so functions which call a mix of cmd_wait and hcfio_string only need to check the return status + * of the last call + * o The return code is preset at Time out. + * The additional complication that no calibrated value for the protection count can be assumed since + * calibrate() does not yet have determined a calibrated value (a catch 22), is handled by setting the + * initial value at INI_TICK_INI (by hcf_connect). This approach is considered safe, because: + * - the HCF does not use the pipeline mechanism of Hermes commands. + * - the likelihood of failure (the only time when protection count is relevant) is small. + * - the time will be sufficiently large on a fast machine (busy bit drops on good NIC before counter + * expires) + * - the time will be sufficiently small on a slow machine (counter expires on bad NIC before the end user + * switches the power off in despair + * The time needed to wrap a 32 bit counter around is longer than many humans want to wait, hence the more or + * less arbitrary value of 0x40000L is chosen, assuming it does not take too long on an XT and is not too + * short on a scream-machine. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC void +calibrate( IFBP ifbp ) +{ + int cnt = HCF_PROT_TIME_CNT; + hcf_32 prot_cnt; + + HCFTRACE( ifbp, HCF_TRACE_CALIBRATE ); + if ( ifbp->IFB_TickIni == INI_TICK_INI ) { /*1*/ + ifbp->IFB_TickIni = 0; /*2*/ + while ( cnt-- ) { + prot_cnt = INI_TICK_INI; + OPW( HREG_EV_ACK, HREG_EV_TICK ); /*3*/ + while ( (IPW( HREG_EV_STAT ) & HREG_EV_TICK) == 0 && --prot_cnt ) { + ifbp->IFB_TickIni++; + } + if ( prot_cnt == 0 || prot_cnt == INI_TICK_INI ) { /*4*/ + ifbp->IFB_TickIni = INI_TICK_INI; + ifbp->IFB_DefunctStat = HCF_ERR_DEFUNCT_TIMER; + ifbp->IFB_CardStat |= CARD_STAT_DEFUNCT; + HCFASSERT( DO_ASSERT, prot_cnt ); + } + } + ifbp->IFB_TickIni <<= HCF_PROT_TIME_SHFT; /*8*/ + } + HCFTRACE( ifbp, HCF_TRACE_CALIBRATE | HCF_TRACE_EXIT ); +} // calibrate +#endif // HCF_PROT_TIME + + +#if (HCF_TYPE) & HCF_TYPE_WPA +/************************************************************************************************************ + * + *.SUBMODULE int check_mic( IFBP ifbp ) + *.PURPOSE verifies the MIC of a received non-USB frame. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * HCF_SUCCESS + * HCF_ERR_MIC + * + *.DESCRIPTION + * + * + *.DIAGRAM + * + *4: test whether or not a MIC is reported by the Hermes + *14: the calculated MIC and the received MIC are compared, the return status is set when there is a mismatch + * + *.NOTICE + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +int +check_mic( IFBP ifbp ) +{ + int rc = HCF_SUCCESS; + hcf_32 x32[2]; //* area to save rcvd 8 bytes MIC + + //if MIC present in RxFS + if ( *(wci_recordp)&ifbp->IFB_lap[-HFS_ADDR_DEST] & HFS_STAT_MIC ) { + //or if ( ifbp->IFB_MICRxCarry != 0xFFFF ) + CALC_RX_MIC( mic_pad, 8 ); //. process up to 3 remaining bytes of data and append 5 to 8 bytes of padding to MIC calculation + get_frag( ifbp, (wci_bufp)x32, 8 BE_PAR(0));//. get 8 byte MIC from NIC + //. if calculated and received MIC do not match + //. . set status at HCF_ERR_MIC + /*14*/ if ( x32[0] != CNV_LITTLE_TO_LONG(ifbp->IFB_MICRx[0]) || + x32[1] != CNV_LITTLE_TO_LONG(ifbp->IFB_MICRx[1]) ) { + rc = HCF_ERR_MIC; + } + } + //return status + return rc; +} // check_mic +#endif // HCF_TYPE_WPA + + +/************************************************************************************************************ + * + *.SUBMODULE int cmd_cmpl( IFBP ifbp ) + *.PURPOSE waits for Hermes Command Completion. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * IFB_DefunctStat + * HCF_ERR_TIME_OUT + * HCF_ERR_DEFUNCT_CMD_SEQ + * HCF_SUCCESS + * + *.DESCRIPTION + * + * + *.DIAGRAM + * + *2: Once cmd_cmpl is called, the Busy option bit in IFB_Cmd must be cleared + *4: If Status register and command code don't match either: + * - the Hermes and Host are out of sync ( a fatal error) + * - error bits are reported via the Status Register. + * Out of sync is considered fatal and brings the HCF in Defunct mode + * Errors reported via the Status Register should be caused by sequence violations in Hermes command + * sequences and hence these bugs should have been found during engineering testing. Since there is no + * strategy to cope with this problem, it might as well be ignored at run time. Note that for any particular + * situation where a strategy is formulated to handle the consequences of a particular bug causing a + * particular Error situation reported via the Status Register, the bug should be removed rather than adding + * logic to cope with the consequences of the bug. + * There have been HCF versions where an error report via the Status Register even brought the HCF in defunct + * mode (although it was not yet named like that at that time). This is particular undesirable behavior for a + * general library. + * Simply reporting the error (as "interesting") is debatable. There also have been HCF versions with this + * strategy using the "vague" HCF_FAILURE code. + * The error is reported via: + * - MiscErr tally of the HCF Tally set + * - the (informative) fields IFB_ErrCmd and IFB_ErrQualifier + * - the assert mechanism + *8: Here the Defunct case and the Status error are separately treated + * + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC int +cmd_cmpl( IFBP ifbp ) +{ + + PROT_CNT_INI; + int rc = HCF_SUCCESS; + hcf_16 stat; + + HCFLOGENTRY( HCF_TRACE_CMD_CPL, ifbp->IFB_Cmd ); + ifbp->IFB_Cmd &= ~HCMD_BUSY; /* 2 */ + HCF_WAIT_WHILE( (IPW( HREG_EV_STAT) & HREG_EV_CMD) == 0 ); /* 4 */ + stat = IPW( HREG_STAT ); +#if HCF_PROT_TIME + if ( prot_cnt == 0 ) { + IF_TALLY( ifbp->IFB_HCF_Tallies.MiscErr++ ); + rc = HCF_ERR_TIME_OUT; + HCFASSERT( DO_ASSERT, ifbp->IFB_Cmd ); + } else +#endif // HCF_PROT_TIME + { + DAWA_ACK( HREG_EV_CMD ); + /*4*/ if ( stat != (ifbp->IFB_Cmd & HCMD_CMD_CODE) ) { + /*8*/ if ( ( (stat ^ ifbp->IFB_Cmd ) & HCMD_CMD_CODE) != 0 ) { + rc = ifbp->IFB_DefunctStat = HCF_ERR_DEFUNCT_CMD_SEQ; + ifbp->IFB_CardStat |= CARD_STAT_DEFUNCT; + } + IF_TALLY( ifbp->IFB_HCF_Tallies.MiscErr++ ); + ifbp->IFB_ErrCmd = stat; + ifbp->IFB_ErrQualifier = IPW( HREG_RESP_0 ); + HCFASSERT( DO_ASSERT, MERGE_2( IPW( HREG_PARAM_0 ), ifbp->IFB_Cmd ) ); + HCFASSERT( DO_ASSERT, MERGE_2( ifbp->IFB_ErrQualifier, ifbp->IFB_ErrCmd ) ); + } + } + HCFASSERT( rc == HCF_SUCCESS, rc); + HCFLOGEXIT( HCF_TRACE_CMD_CPL ); + return rc; +} // cmd_cmpl + + +/************************************************************************************************************ + * + *.SUBMODULE int cmd_exe( IFBP ifbp, int cmd_code, int par_0 ) + *.PURPOSE Executes synchronous part of Hermes Command and - optionally - waits for Command Completion. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * cmd_code + * par_0 + * + *.RETURNS + * IFB_DefunctStat + * HCF_ERR_DEFUNCT_CMD_SEQ + * HCF_SUCCESS + * HCF_ERR_TO_BE_ADDED <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< + * + *.DESCRIPTION + * Executes synchronous Hermes Command and waits for Command Completion + * + * The general HCF strategy is to wait for command completion. As a consequence: + * - the read of the busy bit before writing the command register is superfluous + * - the Hermes requirement that no Inquiry command may be executed if there is still an unacknowledged + * Inquiry command outstanding, is automatically met. + * The Tx command uses the "Busy" bit in the cmd_code parameter to deviate from this general HCF strategy. + * The idea is that by not busy-waiting on completion of this frequently used command the processor + * utilization is diminished while using the busy-wait on all other seldom used commands the flow is kept + * simple. + * + * + * + *.DIAGRAM + * + *1: skip the body of cmd_exe when in defunct mode or when - based on the S/W Support register write and + * read back test - there is apparently no NIC. + * Note: we gave up on the "old" strategy to write the S/W Support register at magic only when needed. Due to + * the intricateness of Hermes F/W varieties ( which behave differently as far as corruption of the S/W + * Support register is involved), the increasing number of Hermes commands which do an implicit initialize + * (thus modifying the S/W Support register) and the workarounds of some OS/Support S/W induced aspects (e.g. + * the System Soft library at WinNT which postpones the actual mapping of I/O space up to 30 seconds after + * giving the go-ahead), the "magic" strategy is now reduced to a simple write and read back. This means that + * problems like a bug tramping over the memory mapped Hermes registers will no longer be noticed as side + * effect of the S/W Support register check. + *2: check whether the preceding command skipped the busy wait and if so, check for command completion + * + *.NOTICE + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ + +HCF_STATIC int +cmd_exe( IFBP ifbp, hcf_16 cmd_code, hcf_16 par_0 ) //if HCMD_BUSY of cmd_code set, then do NOT wait for completion +{ + int rc; + + HCFLOGENTRY( HCF_TRACE_CMD_EXE, cmd_code ); + HCFASSERT( (cmd_code & HCMD_CMD_CODE) != HCMD_TX || cmd_code & HCMD_BUSY, cmd_code ); //Tx must have Busy bit set + OPW( HREG_SW_0, HCF_MAGIC ); + if ( IPW( HREG_SW_0 ) == HCF_MAGIC ) { /* 1 */ + rc = ifbp->IFB_DefunctStat; + } + else rc = HCF_ERR_NO_NIC; + if ( rc == HCF_SUCCESS ) { + //;?is this a hot idea, better MEASURE performance impact + /*2*/ if ( ifbp->IFB_Cmd & HCMD_BUSY ) { + rc = cmd_cmpl( ifbp ); + } + OPW( HREG_PARAM_0, par_0 ); + OPW( HREG_CMD, cmd_code &~HCMD_BUSY ); + ifbp->IFB_Cmd = cmd_code; + if ( (cmd_code & HCMD_BUSY) == 0 ) { //;?is this a hot idea, better MEASURE performance impact + rc = cmd_cmpl( ifbp ); + } + } + HCFASSERT( rc == HCF_SUCCESS, MERGE_2( rc, cmd_code ) ); + HCFLOGEXIT( HCF_TRACE_CMD_EXE ); + return rc; +} // cmd_exe + + +/************************************************************************************************************ + * + *.SUBMODULE int download( IFBP ifbp, CFG_PROG_STRCT FAR *ltvp ) + *.PURPOSE downloads F/W image into NIC and initiates execution of the downloaded F/W. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * ltvp specifies the pseudo-RID (as defined by WCI) + * + *.RETURNS + * + *.DESCRIPTION + * + * + *.DIAGRAM + *1: First, Ack everything to unblock a (possibly) blocked cmd pipe line + * Note 1: it is very likely that an Alloc event is pending and very well possible that a (Send) Cmd event is + * pending + * Note 2: it is assumed that this strategy takes away the need to ack every conceivable event after an + * Hermes Initialize + * + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC int +download( IFBP ifbp, CFG_PROG_STRCT FAR *ltvp ) //Hermes-II download (volatile only) +{ + hcf_16 i; + int rc = HCF_SUCCESS; + wci_bufp cp; + hcf_io io_port = ifbp->IFB_IOBase + HREG_AUX_DATA; + + HCFLOGENTRY( HCF_TRACE_DL, ltvp->typ ); +#if (HCF_TYPE) & HCF_TYPE_PRELOADED + HCFASSERT( DO_ASSERT, ltvp->mode ); +#else + //if initial "program" LTV + if ( ifbp->IFB_DLMode == CFG_PROG_STOP && ltvp->mode == CFG_PROG_VOLATILE) { + //. switch Hermes to initial mode + /*1*/ OPW( HREG_EV_ACK, ~HREG_EV_SLEEP_REQ ); + rc = cmd_exe( ifbp, HCMD_INI, 0 ); /* HCMD_INI can not be part of init() because that is called on + * other occasions as well */ + rc = init( ifbp ); + } + //if final "program" LTV + if ( ltvp->mode == CFG_PROG_STOP && ifbp->IFB_DLMode == CFG_PROG_VOLATILE) { + //. start tertiary (or secondary) + OPW( HREG_PARAM_1, (hcf_16)(ltvp->nic_addr >> 16) ); + rc = cmd_exe( ifbp, HCMD_EXECUTE, (hcf_16) ltvp->nic_addr ); + if (rc == HCF_SUCCESS) { + rc = init( ifbp ); /*;? do we really want to skip init if cmd_exe failed, i.e. + * IFB_FW_Comp_Id is than possibly incorrect */ + } + //else (non-final) + } else { + //. if mode == Readback SEEPROM +#if 0 //;? as long as the next if contains a hard coded 0, might as well leave it out even more obvious + if ( 0 /*len is definitely not want we want;?*/ && ltvp->mode == CFG_PROG_SEEPROM_READBACK ) { + OPW( HREG_PARAM_1, (hcf_16)(ltvp->nic_addr >> 16) ); + OPW( HREG_PARAM_2, (hcf_16)((ltvp->len - 4) << 1) ); + //. . perform Hermes prog cmd with appropriate mode bits + rc = cmd_exe( ifbp, HCMD_PROGRAM | ltvp->mode, (hcf_16)ltvp->nic_addr ); + //. . set up NIC RAM addressability according Resp0-1 + OPW( HREG_AUX_PAGE, IPW( HREG_RESP_1) ); + OPW( HREG_AUX_OFFSET, IPW( HREG_RESP_0) ); + //. . set up L-field of LTV according Resp2 + i = ( IPW( HREG_RESP_2 ) + 1 ) / 2; // i contains max buffer size in words, a probably not very useful piece of information ;? +/*Nico's code based on i is the "real amount of data available" + if ( ltvp->len - 4 < i ) rc = HCF_ERR_LEN; + else ltvp->len = i + 4; +*/ +/* Rolands code based on the idea that a MSF should not ask for more than is available + // check if number of bytes requested exceeds max buffer size + if ( ltvp->len - 4 > i ) { + rc = HCF_ERR_LEN; + ltvp->len = i + 4; + } +*/ + //. . copy data from NIC via AUX port to LTV + cp = (wci_bufp)ltvp->host_addr; /*IN_PORT_STRING_8_16 macro may modify its parameters*/ + i = ltvp->len - 4; + IN_PORT_STRING_8_16( io_port, cp, i ); //!!!WORD length, cp MUST be a char pointer // $$ char + //. else (non-final programming) + } else +#endif //;? as long as the above if contains a hard coded 0, might as well leave it out even more obvious + { //. . get number of words to program + HCFASSERT( ltvp->segment_size, *ltvp->host_addr ); + i = ltvp->segment_size/2; + //. . copy data (words) from LTV via AUX port to NIC + cp = (wci_bufp)ltvp->host_addr; //OUT_PORT_STRING_8_16 macro may modify its parameters + //. . if mode == volatile programming + if ( ltvp->mode == CFG_PROG_VOLATILE ) { + //. . . set up NIC RAM addressability via AUX port + OPW( HREG_AUX_PAGE, (hcf_16)(ltvp->nic_addr >> 16 << 9 | (ltvp->nic_addr & 0xFFFF) >> 7 ) ); + OPW( HREG_AUX_OFFSET, (hcf_16)(ltvp->nic_addr & 0x007E) ); + OUT_PORT_STRING_8_16( io_port, cp, i ); //!!!WORD length, cp MUST be a char pointer + } + } + } + ifbp->IFB_DLMode = ltvp->mode; //save state in IFB_DLMode +#endif // HCF_TYPE_PRELOADED + HCFASSERT( rc == HCF_SUCCESS, rc ); + HCFLOGEXIT( HCF_TRACE_DL ); + return rc; +} // download + + +#if (HCF_ASSERT) & HCF_ASSERT_PRINTF +/************************************************** + * Certain Hermes-II firmware versions can generate + * debug information. This debug information is + * contained in a buffer in nic-RAM, and can be read + * via the aux port. + **************************************************/ +HCF_STATIC int +fw_printf(IFBP ifbp, CFG_FW_PRINTF_STRCT FAR *ltvp) +{ + int rc = HCF_SUCCESS; + hcf_16 fw_cnt; +// hcf_32 DbMsgBuffer = 0x29D2, DbMsgCount= 0x000029D0; +// hcf_16 DbMsgSize=0x00000080; + hcf_32 DbMsgBuffer; + CFG_FW_PRINTF_BUFFER_LOCATION_STRCT *p = &ifbp->IFB_FwPfBuff; + ltvp->len = 1; + if ( p->DbMsgSize != 0 ) { + // first, check the counter in nic-RAM and compare it to the latest counter value of the HCF + OPW( HREG_AUX_PAGE, (hcf_16)(p->DbMsgCount >> 7) ); + OPW( HREG_AUX_OFFSET, (hcf_16)(p->DbMsgCount & 0x7E) ); + fw_cnt = ((IPW( HREG_AUX_DATA) >>1 ) & ((hcf_16)p->DbMsgSize - 1)); + if ( fw_cnt != ifbp->IFB_DbgPrintF_Cnt ) { +// DbgPrint("fw_cnt=%d IFB_DbgPrintF_Cnt=%d\n", fw_cnt, ifbp->IFB_DbgPrintF_Cnt); + DbMsgBuffer = p->DbMsgBuffer + ifbp->IFB_DbgPrintF_Cnt * 6; // each entry is 3 words + OPW( HREG_AUX_PAGE, (hcf_16)(DbMsgBuffer >> 7) ); + OPW( HREG_AUX_OFFSET, (hcf_16)(DbMsgBuffer & 0x7E) ); + ltvp->msg_id = IPW(HREG_AUX_DATA); + ltvp->msg_par = IPW(HREG_AUX_DATA); + ltvp->msg_tstamp = IPW(HREG_AUX_DATA); + ltvp->len = 4; + ifbp->IFB_DbgPrintF_Cnt++; + ifbp->IFB_DbgPrintF_Cnt &= (p->DbMsgSize - 1); + } + } + return rc; +}; +#endif // HCF_ASSERT_PRINTF + + +/************************************************************************************************************ + * + *.SUBMODULE hcf_16 get_fid( IFBP ifbp ) + *.PURPOSE get allocated FID for either transmit or notify. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * 0 no FID available + * <>0 FID number + * + *.DESCRIPTION + * + * + *.DIAGRAM + * The preference is to use a "pending" alloc. If no alloc is pending, then - if available - the "spare" FID + * is used. + * If the spare FID is used, IFB_RscInd (representing the spare FID) must be cleared + * If the pending alloc is used, the alloc event must be acknowledged to the Hermes. + * In case the spare FID was depleted and the IFB_RscInd has been "faked" as pseudo resource with a 0x0001 + * value by hcf_service_nic, IFB_RscInd has to be "corrected" again to its 0x0000 value. + * + * Note that due to the Hermes-II H/W problems which are intended to be worked around by DAWA, the Alloc bit + * in the Event register is no longer a reliable indication of the presence/absence of a FID. The "Clear FID" + * part of the DAWA logic, together with the choice of the definition of the return information from get_fid, + * handle this automatically, i.e. without additional code in get_fid. + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC hcf_16 +get_fid( IFBP ifbp ) +{ + + hcf_16 fid = 0; +#if ( (HCF_TYPE) & HCF_TYPE_HII5 ) == 0 + PROT_CNT_INI; +#endif // HCF_TYPE_HII5 + + IF_DMA( HCFASSERT(!(ifbp->IFB_CntlOpt & USE_DMA), ifbp->IFB_CntlOpt) ); + + if ( IPW( HREG_EV_STAT) & HREG_EV_ALLOC) { + fid = IPW( HREG_ALLOC_FID ); + HCFASSERT( fid, ifbp->IFB_RscInd ); + DAWA_ZERO_FID( HREG_ALLOC_FID ); +#if ( (HCF_TYPE) & HCF_TYPE_HII5 ) == 0 + HCF_WAIT_WHILE( ( IPW( HREG_EV_STAT ) & HREG_EV_ACK_REG_READY ) == 0 ); + HCFASSERT( prot_cnt, IPW( HREG_EV_STAT ) ); +#endif // HCF_TYPE_HII5 + DAWA_ACK( HREG_EV_ALLOC ); //!!note that HREG_EV_ALLOC is written only once +// 180 degree error in logic ;? #if ALLOC_15 + if ( ifbp->IFB_RscInd == 1 ) { + ifbp->IFB_RscInd = 0; + } +//#endif // ALLOC_15 + } else { +// 180 degree error in logic ;? #if ALLOC_15 + fid = ifbp->IFB_RscInd; +//#endif // ALLOC_15 + ifbp->IFB_RscInd = 0; + } + return fid; +} // get_fid + + +/************************************************************************************************************ + * + *.SUBMODULE void get_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ) + *.PURPOSE reads with 16/32 bit I/O via BAP1 port from NIC RAM to Host memory. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * bufp (byte) address of buffer + * len length in bytes of buffer specified by bufp + * word_len Big Endian only: number of leading bytes to swap in pairs + * + *.RETURNS N.A. + * + *.DESCRIPTION + * process the single byte (if applicable) read by the previous get_frag and copy len (or len-1) bytes from + * NIC to bufp. + * On a Big Endian platform, the parameter word_len controls the number of leading bytes whose endianness is + * converted (i.e. byte swapped) + * + * + *.DIAGRAM + *10: The PCMCIA card can be removed in the middle of the transfer. By depositing a "magic number" in the + * HREG_SW_0 register of the Hermes at initialization time and by verifying this register, it can be + * determined whether the card is still present. The return status is set accordingly. + * Clearing the buffer is a (relative) cheap way to prevent that failing I/O results in run-away behavior + * because the garbage in the buffer is interpreted by the caller irrespective of the return status (e.g. + * hcf_service_nic has this behavior). + * + *.NOTICE + * It turns out DOS ODI uses zero length fragments. The HCF code can cope with it, but as a consequence, no + * Assert on len is possible + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC void +get_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ) +{ + hcf_io io_port = ifbp->IFB_IOBase + HREG_DATA_1; //BAP data register + wci_bufp p = bufp; //working pointer + int i; //prevent side effects from macro + int j; + + HCFASSERT( ((hcf_32)bufp & (HCF_ALIGN-1) ) == 0, (hcf_32)bufp ); + +/*1: here recovery logic for intervening BAP access between hcf_service_nic and hcf_rcv_msg COULD be added + * if current access is RxInitial + * . persistent_offset += len + */ + + i = len; + //if buffer length > 0 and carry from previous get_frag + if ( i && ifbp->IFB_CarryIn ) { + //. move carry to buffer + //. adjust buffer length and pointer accordingly + *p++ = (hcf_8)(ifbp->IFB_CarryIn>>8); + i--; + //. clear carry flag + ifbp->IFB_CarryIn = 0; + } +#if (HCF_IO) & HCF_IO_32BITS + //skip zero-length I/O, single byte I/O and I/O not worthwhile (i.e. less than 6 bytes)for DW logic + //if buffer length >= 6 and 32 bits I/O support + if ( !(ifbp->IFB_CntlOpt & USE_16BIT) && i >= 6 ) { + hcf_32 FAR *p4; //prevent side effects from macro + if ( ( (hcf_32)p & 0x1 ) == 0 ) { //. if buffer at least word aligned + if ( (hcf_32)p & 0x2 ) { //. . if buffer not double word aligned + //. . . read single word to get double word aligned + *(wci_recordp)p = IN_PORT_WORD( io_port ); + //. . . adjust buffer length and pointer accordingly + p += 2; + i -= 2; + } + //. . read as many double word as possible + p4 = (hcf_32 FAR *)p; + j = i/4; + IN_PORT_STRING_32( io_port, p4, j ); + //. . adjust buffer length and pointer accordingly + p += i & ~0x0003; + i &= 0x0003; + } + } +#endif // HCF_IO_32BITS + //if no 32-bit support OR byte aligned OR 1-3 bytes left + if ( i ) { + //. read as many word as possible in "alignment safe" way + j = i/2; + IN_PORT_STRING_8_16( io_port, p, j ); + //. if 1 byte left + if ( i & 0x0001 ) { + //. . read 1 word + ifbp->IFB_CarryIn = IN_PORT_WORD( io_port ); + //. . store LSB in last char of buffer + bufp[len-1] = (hcf_8)ifbp->IFB_CarryIn; + //. . save MSB in carry, set carry flag + ifbp->IFB_CarryIn |= 0x1; + } + } +#if HCF_BIG_ENDIAN + HCFASSERT( word_len == 0 || word_len == 2 || word_len == 4, word_len ); + HCFASSERT( word_len == 0 || ((hcf_32)bufp & 1 ) == 0, (hcf_32)bufp ); + HCFASSERT( word_len <= len, MERGE2( word_len, len ) ); + //see put_frag for an alternative implementation, but be careful about what are int's and what are + //hcf_16's + if ( word_len ) { //. if there is anything to convert + hcf_8 c; + c = bufp[1]; //. . convert the 1st hcf_16 + bufp[1] = bufp[0]; + bufp[0] = c; + if ( word_len > 1 ) { //. . if there is to convert more than 1 word ( i.e 2 ) + c = bufp[3]; //. . . convert the 2nd hcf_16 + bufp[3] = bufp[2]; + bufp[2] = c; + } + } +#endif // HCF_BIG_ENDIAN +} // get_frag + +/************************************************************************************************************ + * + *.SUBMODULE int init( IFBP ifbp ) + *.PURPOSE Handles common initialization aspects (H-I init, calibration, config.mngmt, allocation). + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS + * HCF_ERR_INCOMP_PRI + * HCF_ERR_INCOMP_FW + * HCF_ERR_TIME_OUT + * >>hcf_get_info + * HCF_ERR_NO_NIC + * HCF_ERR_LEN + * + *.DESCRIPTION + * init will successively: + * - in case of a (non-preloaded) H-I, initialize the NIC + * - calibrate the S/W protection timer against the Hermes Timer + * - collect HSI, "active" F/W Configuration Management Information + * - in case active F/W is Primary F/W: collect Primary F/W Configuration Management Information + * - check HSI and Primary F/W compatibility with the HCF + * - in case active F/W is Station or AP F/W: check Station or AP F/W compatibility with the HCF + * - in case active F/W is not Primary F/W: allocate FIDs to be used in transmit/notify process + * + * + *.DIAGRAM + *2: drop all error status bits in IFB_CardStat since they are expected to be re-evaluated. + *4: Ack everything except HREG_EV_SLEEP_REQ. It is very likely that an Alloc event is pending and + * very well possible that a Send Cmd event is pending. Acking HREG_EV_SLEEP_REQ is handled by hcf_action( + * HCF_ACT_INT_ON ) !!! + *10: Calibrate the S/W time-out protection mechanism by calling calibrate(). Note that possible errors + * in the calibration process are nor reported by init but will show up via the defunct mechanism in + * subsequent hcf-calls. + *14: usb_check_comp() is called to have the minimal visual clutter for the legacy H-I USB dongle + * compatibility check. + *16: The following configuration management related information is retrieved from the NIC: + * - HSI supplier + * - F/W Identity + * - F/W supplier + * if appropriate: + * - PRI Identity + * - PRI supplier + * appropriate means on H-I: always + * and on H-II if F/W supplier reflects a primary (i.e. only after an Hermes Reset or Init + * command). + * QUESTION ;? !!!!!! should, For each of the above RIDs the Endianness is converted to native Endianness. + * Only the return code of the first hcf_get_info is used. All hcf_get_info calls are made, regardless of + * the success or failure of the 1st hcf_get_info. The assumptions are: + * - if any call fails, they all fail, so remembering the result of the 1st call is adequate + * - a failing call will overwrite the L-field with a 0x0000 value, which services both as an + * error indication for the values cached in the IFB as making mmd_check_comp fail. + * In case of H-I, when getting the F/W identity fails, the F/W is assumed to be H-I AP F/W pre-dating + * version 9.0 and the F/W Identity and Supplier are faked accordingly. + * In case of H-II, the Primary, Station and AP Identity are merged into a single F/W Identity. + * The same applies to the Supplier information. As a consequence the PRI information can no longer be + * retrieved when a Tertiary runs. To accommodate MSFs and Utilities who depend on PRI information being + * available at any time, this information is cached in the IFB. In this cache the generic "F/W" value of + * the typ-fields is overwritten with the specific (legacy) "PRI" values. To actually re-route the (legacy) + * PRI request via hcf_get_info, the xxxx-table must be set. In case of H-I, this caching, modifying and + * re-routing is not needed because PRI information is always available directly from the NIC. For + * consistency the caching fields in the IFB are filled with the PRI information anyway. + *18: mdd_check_comp() is called to check the Supplier Variant and Range of the Host-S/W I/F (HSI) and the + * Primary Firmware Variant and Range against the Top and Bottom level supported by this HCF. If either of + * these tests fails, the CARD_STAT_INCOMP_PRI bit of IFB_CardStat is set + * Note: There should always be a primary except during production, so this makes the HCF in its current form + * unsuitable for manufacturing test systems like the FTS. This can be remedied by an adding a test like + * ifbp->IFB_PRISup.id == COMP_ID_PRI + *20: In case there is Tertiary F/W and this F/W is Station F/W, the Supplier Variant and Range of the Station + * Firmware function as retrieved from the Hermes is checked against the Top and Bottom level supported by + * this HCF. + * Note: ;? the tertiary F/W compatibility checks could be moved to the DHF, which already has checked the + * CFI and MFI compatibility of the image with the NIC before the image was downloaded. + *28: In case of non-Primary F/W: allocates and acknowledge a (TX or Notify) FID and allocates without + * acknowledge another (TX or Notify) FID (the so-called 1.5 alloc scheme) with the following steps: + * - execute the allocate command by calling cmd_exe + * - wait till either the alloc event or a time-out occurs + * - regardless whether the alloc event occurs, call get_fid to + * - read the FID and save it in IFB_RscInd to be used as "spare FID" + * - acknowledge the alloc event + * - do another "half" allocate to complete the "1.5 Alloc scheme" + * Note that above 3 steps do not harm and thus give the "cheapest" acceptable strategy. + * If a time-out occurred, then report time out status (after all) + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC int +init( IFBP ifbp ) +{ + + int rc = HCF_SUCCESS; + + HCFLOGENTRY( HCF_TRACE_INIT, 0 ); + + ifbp->IFB_CardStat = 0; /* 2*/ + OPW( HREG_EV_ACK, ~HREG_EV_SLEEP_REQ ); /* 4*/ + IF_PROT_TIME( calibrate( ifbp ) ); /*10*/ +#if 0 // OOR + ifbp->IFB_FWIdentity.len = 2; //misuse the IFB space for a put + ifbp->IFB_FWIdentity.typ = CFG_TICK_TIME; + ifbp->IFB_FWIdentity.comp_id = (1000*1000)/1024 + 1; //roughly 1 second + hcf_put_info( ifbp, (LTVP)&ifbp->IFB_FWIdentity.len ); +#endif // OOR + ifbp->IFB_FWIdentity.len = sizeof(CFG_FW_IDENTITY_STRCT)/sizeof(hcf_16) - 1; + ifbp->IFB_FWIdentity.typ = CFG_FW_IDENTITY; + rc = hcf_get_info( ifbp, (LTVP)&ifbp->IFB_FWIdentity.len ); +/* ;? conversion should not be needed for mmd_check_comp */ +#if HCF_BIG_ENDIAN + ifbp->IFB_FWIdentity.comp_id = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWIdentity.comp_id ); + ifbp->IFB_FWIdentity.variant = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWIdentity.variant ); + ifbp->IFB_FWIdentity.version_major = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWIdentity.version_major ); + ifbp->IFB_FWIdentity.version_minor = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWIdentity.version_minor ); +#endif // HCF_BIG_ENDIAN +#if defined MSF_COMPONENT_ID /*14*/ + if ( rc == HCF_SUCCESS ) { /*16*/ + ifbp->IFB_HSISup.len = sizeof(CFG_SUP_RANGE_STRCT)/sizeof(hcf_16) - 1; + ifbp->IFB_HSISup.typ = CFG_NIC_HSI_SUP_RANGE; + rc = hcf_get_info( ifbp, (LTVP)&ifbp->IFB_HSISup.len ); +/* ;? conversion should not be needed for mmd_check_comp , BUT according to a report of a BE-user it is + * should be resolved in the WARP release + * since some compilers make ugly but unnecessary code of these instructions even for LE, + * it is conditionally compiled */ +#if HCF_BIG_ENDIAN + ifbp->IFB_HSISup.role = CNV_LITTLE_TO_SHORT( ifbp->IFB_HSISup.role ); + ifbp->IFB_HSISup.id = CNV_LITTLE_TO_SHORT( ifbp->IFB_HSISup.id ); + ifbp->IFB_HSISup.variant = CNV_LITTLE_TO_SHORT( ifbp->IFB_HSISup.variant ); + ifbp->IFB_HSISup.bottom = CNV_LITTLE_TO_SHORT( ifbp->IFB_HSISup.bottom ); + ifbp->IFB_HSISup.top = CNV_LITTLE_TO_SHORT( ifbp->IFB_HSISup.top ); +#endif // HCF_BIG_ENDIAN + ifbp->IFB_FWSup.len = sizeof(CFG_SUP_RANGE_STRCT)/sizeof(hcf_16) - 1; + ifbp->IFB_FWSup.typ = CFG_FW_SUP_RANGE; + (void)hcf_get_info( ifbp, (LTVP)&ifbp->IFB_FWSup.len ); +/* ;? conversion should not be needed for mmd_check_comp */ +#if HCF_BIG_ENDIAN + ifbp->IFB_FWSup.role = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWSup.role ); + ifbp->IFB_FWSup.id = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWSup.id ); + ifbp->IFB_FWSup.variant = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWSup.variant ); + ifbp->IFB_FWSup.bottom = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWSup.bottom ); + ifbp->IFB_FWSup.top = CNV_LITTLE_TO_SHORT( ifbp->IFB_FWSup.top ); +#endif // HCF_BIG_ENDIAN + + if ( ifbp->IFB_FWSup.id == COMP_ID_PRI ) { /* 20*/ + int i = sizeof( CFG_FW_IDENTITY_STRCT) + sizeof(CFG_SUP_RANGE_STRCT ); + while ( i-- ) ((hcf_8*)(&ifbp->IFB_PRIIdentity))[i] = ((hcf_8*)(&ifbp->IFB_FWIdentity))[i]; + ifbp->IFB_PRIIdentity.typ = CFG_PRI_IDENTITY; + ifbp->IFB_PRISup.typ = CFG_PRI_SUP_RANGE; + xxxx[xxxx_PRI_IDENTITY_OFFSET] = &ifbp->IFB_PRIIdentity.len; + xxxx[xxxx_PRI_IDENTITY_OFFSET+1] = &ifbp->IFB_PRISup.len; + } + if ( !mmd_check_comp( (void*)&cfg_drv_act_ranges_hsi, &ifbp->IFB_HSISup) /* 22*/ +#if ( (HCF_TYPE) & HCF_TYPE_PRELOADED ) == 0 +//;? the PRI compatibility check is only relevant for DHF + || !mmd_check_comp( (void*)&cfg_drv_act_ranges_pri, &ifbp->IFB_PRISup) +#endif // HCF_TYPE_PRELOADED + ) { + ifbp->IFB_CardStat = CARD_STAT_INCOMP_PRI; + rc = HCF_ERR_INCOMP_PRI; + } + if ( ( ifbp->IFB_FWSup.id == COMP_ID_STA && !mmd_check_comp( (void*)&cfg_drv_act_ranges_sta, &ifbp->IFB_FWSup) ) || + ( ifbp->IFB_FWSup.id == COMP_ID_APF && !mmd_check_comp( (void*)&cfg_drv_act_ranges_apf, &ifbp->IFB_FWSup) ) + ) { /* 24 */ + ifbp->IFB_CardStat |= CARD_STAT_INCOMP_FW; + rc = HCF_ERR_INCOMP_FW; + } + } +#endif // MSF_COMPONENT_ID + + if ( rc == HCF_SUCCESS && ifbp->IFB_FWIdentity.comp_id >= COMP_ID_FW_STA ) { + PROT_CNT_INI; + /************************************************************************************** + * rlav: the DMA engine needs the host to cause a 'hanging alloc event' for it to consume. + * not sure if this is the right spot in the HCF, thinking about hcf_enable... + **************************************************************************************/ + rc = cmd_exe( ifbp, HCMD_ALLOC, 0 ); +// 180 degree error in logic ;? #if ALLOC_15 +// ifbp->IFB_RscInd = 1; //let's hope that by the time hcf_send_msg isa called, there will be a FID +//#else + if ( rc == HCF_SUCCESS ) { + HCF_WAIT_WHILE( (IPW( HREG_EV_STAT ) & HREG_EV_ALLOC) == 0 ); + IF_PROT_TIME( HCFASSERT(prot_cnt, IPW( HREG_EV_STAT )) ); +#if HCF_DMA + if ( ! ( ifbp->IFB_CntlOpt & USE_DMA ) ) +#endif // HCF_DMA + { + ifbp->IFB_RscInd = get_fid( ifbp ); + HCFASSERT( ifbp->IFB_RscInd, 0 ); + cmd_exe( ifbp, HCMD_ALLOC, 0 ); + IF_PROT_TIME( if ( prot_cnt == 0 ) rc = HCF_ERR_TIME_OUT ); + } + } +//#endif // ALLOC_15 + } + + HCFASSERT( rc == HCF_SUCCESS, rc ); + HCFLOGEXIT( HCF_TRACE_INIT ); + return rc; +} // init + +/************************************************************************************************************ + * + *.SUBMODULE void isr_info( IFBP ifbp ) + *.PURPOSE handles link events. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS N.A. + * + *.DESCRIPTION + * + * + *.DIAGRAM + *1: First the FID number corresponding with the InfoEvent is determined. + * Note the complication of the zero-FID protection sub-scheme in DAWA. + * Next the L-field and the T-field are fetched into scratch buffer info. + *2: In case of tallies, the 16 bits Hermes values are accumulated in the IFB into 32 bits values. Info[0] + * is (expected to be) HCF_NIC_TAL_CNT + 1. The contraption "while ( info[0]-- >1 )" rather than + * "while ( --info[0] )" is used because it is dangerous to determine the length of the Value field by + * decrementing info[0]. As a result of a bug in some version of the F/W, info[0] may be 0, resulting + * in a very long loop in the pre-decrement logic. + *4: In case of a link status frame, the information is copied to the IFB field IFB_linkStat + *6: All other than Tallies (including "unknown" ones) are checked against the selection set by the MSF + * via CFG_RID_LOG. If a match is found or the selection set has the wild-card type (i.e non-NULL buffer + * pointer at the terminating zero-type), the frame is copied to the (type-specific) log buffer. + * Note that to accumulate tallies into IFB AND to log them or to log a frame when a specific match occures + * AND based on the wild-card selection, you have to call setup_bap again after the 1st copy. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC void +isr_info( IFBP ifbp ) +{ + hcf_16 info[2], fid; +#if (HCF_EXT) & HCF_EXT_INFO_LOG + RID_LOGP ridp = ifbp->IFB_RIDLogp; //NULL or pointer to array of RID_LOG structures (terminated by zero typ) +#endif // HCF_EXT_INFO_LOG + + HCFTRACE( ifbp, HCF_TRACE_ISR_INFO ); /* 1 */ + fid = IPW( HREG_INFO_FID ); + DAWA_ZERO_FID( HREG_INFO_FID ); + if ( fid ) { + (void)setup_bap( ifbp, fid, 0, IO_IN ); + get_frag( ifbp, (wci_bufp)info, 4 BE_PAR(2) ); + HCFASSERT( info[0] <= HCF_MAX_LTV + 1, MERGE_2( info[1], info[0] ) ); //;? a smaller value makes more sense +#if (HCF_TALLIES) & HCF_TALLIES_NIC //Hermes tally support + if ( info[1] == CFG_TALLIES ) { + hcf_32 *p; + /*2*/ if ( info[0] > HCF_NIC_TAL_CNT ) { + info[0] = HCF_NIC_TAL_CNT + 1; + } + p = (hcf_32*)&ifbp->IFB_NIC_Tallies; + while ( info[0]-- >1 ) *p++ += IPW( HREG_DATA_1 ); //request may return zero length + } + else +#endif // HCF_TALLIES_NIC + { + /*4*/ if ( info[1] == CFG_LINK_STAT ) { + ifbp->IFB_LinkStat = IPW( HREG_DATA_1 ); + } +#if (HCF_EXT) & HCF_EXT_INFO_LOG + /*6*/ while ( 1 ) { + if ( ridp->typ == 0 || ridp->typ == info[1] ) { + if ( ridp->bufp ) { + HCFASSERT( ridp->len >= 2, ridp->typ ); + ridp->bufp[0] = min((hcf_16)(ridp->len - 1), info[0] ); //save L + ridp->bufp[1] = info[1]; //save T + get_frag( ifbp, (wci_bufp)&ridp->bufp[2], (ridp->bufp[0] - 1)*2 BE_PAR(0) ); + } + break; + } + ridp++; + } +#endif // HCF_EXT_INFO_LOG + } + HCFTRACE( ifbp, HCF_TRACE_ISR_INFO | HCF_TRACE_EXIT ); + } + return; +} // isr_info + +// +// +// #endif // HCF_TALLIES_NIC +// /*4*/ if ( info[1] == CFG_LINK_STAT ) { +// ifbp->IFB_DSLinkStat = IPW( HREG_DATA_1 ) | CFG_LINK_STAT_CHANGE; //corrupts BAP !! ;? +// ifbp->IFB_LinkStat = ifbp->IFB_DSLinkStat & CFG_LINK_STAT_FW; //;? to be obsoleted +// printk(KERN_ERR "linkstatus: %04x\n", ifbp->IFB_DSLinkStat ); //;?remove me 1 day +// #if (HCF_SLEEP) & HCF_DDS +// if ( ( ifbp->IFB_DSLinkStat & CFG_LINK_STAT_CONNECTED ) == 0 ) { //even values are disconnected etc. +// ifbp->IFB_TickCnt = 0; //start 2 second period (with 1 tick uncertanty) +// printk(KERN_NOTICE "isr_info: AwaitConnection phase started, IFB_TickCnt = 0\n" ); //;?remove me 1 day +// } +// #endif // HCF_DDS +// } +// #if (HCF_EXT) & HCF_EXT_INFO_LOG +// /*6*/ while ( 1 ) { +// if ( ridp->typ == 0 || ridp->typ == info[1] ) { +// if ( ridp->bufp ) { +// HCFASSERT( ridp->len >= 2, ridp->typ ); +// (void)setup_bap( ifbp, fid, 2, IO_IN ); //restore BAP for tallies, linkstat and specific type followed by wild card +// ridp->bufp[0] = min( ridp->len - 1, info[0] ); //save L +// get_frag( ifbp, (wci_bufp)&ridp->bufp[1], ridp->bufp[0]*2 BE_PAR(0) ); +// } +// break; //;?this break is no longer needed due to setup_bap but lets concentrate on DDS first +// } +// ridp++; +// } +// #endif // HCF_EXT_INFO_LOG +// } +// HCFTRACE( ifbp, HCF_TRACE_ISR_INFO | HCF_TRACE_EXIT ); +// +// +// +// +// return; +//} // isr_info + + +/************************************************************************************************************ + * + *.SUBMODULE void mdd_assert( IFBP ifbp, unsigned int line_number, hcf_32 q ) + *.PURPOSE filters assert on level and interfaces to the MSF supplied msf_assert routine. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * line_number line number of the line which caused the assert + * q qualifier, additional information which may give a clue about the problem + * + *.RETURNS N.A. + * + *.DESCRIPTION + * + * + *.DIAGRAM + * + *.NOTICE + * mdd_assert has been through a turmoil, renaming hcf_assert to assert and hcf_assert again and supporting off + * and on being called from the MSF level and other ( immature ) ModularDriverDevelopment modules like DHF and + * MMD. + * !!!! The assert routine is not an hcf_..... routine in the sense that it may be called by the MSF, + * however it is called from mmd.c and dhf.c, so it must be external. + * To prevent namespace pollution it needs a prefix, to prevent that MSF programmers think that + * they are allowed to call the assert logic, the prefix HCF can't be used, so MDD is selected!!!! + * + * When called from the DHF module the line number is incremented by DHF_FILE_NAME_OFFSET and when called from + * the MMD module by MMD_FILE_NAME_OFFSET. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +#if HCF_ASSERT +void +mdd_assert( IFBP ifbp, unsigned int line_number, hcf_32 q ) +{ + hcf_16 run_time_flag = ifbp->IFB_AssertLvl; + + if ( run_time_flag /* > ;?????? */ ) { //prevent recursive behavior, later to be extended to level filtering + ifbp->IFB_AssertQualifier = q; + ifbp->IFB_AssertLine = (hcf_16)line_number; +#if (HCF_ASSERT) & ( HCF_ASSERT_LNK_MSF_RTN | HCF_ASSERT_RT_MSF_RTN ) + if ( ifbp->IFB_AssertRtn ) { + ifbp->IFB_AssertRtn( line_number, ifbp->IFB_AssertTrace, q ); + } +#endif // HCF_ASSERT_LNK_MSF_RTN / HCF_ASSERT_RT_MSF_RTN +#if (HCF_ASSERT) & HCF_ASSERT_SW_SUP + OPW( HREG_SW_2, line_number ); + OPW( HREG_SW_2, ifbp->IFB_AssertTrace ); + OPW( HREG_SW_2, (hcf_16)q ); + OPW( HREG_SW_2, (hcf_16)(q >> 16 ) ); +#endif // HCF_ASSERT_SW_SUP + +#if (HCF_ASSERT) & HCF_ASSERT_MB + ifbp->IFB_AssertLvl = 0; // prevent recursive behavior + hcf_put_info( ifbp, (LTVP)&ifbp->IFB_AssertStrct ); + ifbp->IFB_AssertLvl = run_time_flag; // restore appropriate filter level +#endif // HCF_ASSERT_MB + } +} // mdd_assert +#endif // HCF_ASSERT + + +/************************************************************************************************************ + * + *.SUBMODULE void put_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ) + *.PURPOSE writes with 16/32 bit I/O via BAP1 port from Host memory to NIC RAM. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * bufp (byte) address of buffer + * len length in bytes of buffer specified by bufp + * word_len Big Endian only: number of leading bytes to swap in pairs + * + *.RETURNS N.A. + * + *.DESCRIPTION + * process the single byte (if applicable) not yet written by the previous put_frag and copy len + * (or len-1) bytes from bufp to NIC. + * + * + *.DIAGRAM + * + *.NOTICE + * It turns out DOS ODI uses zero length fragments. The HCF code can cope with it, but as a consequence, no + * Assert on len is possible + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC void +put_frag( IFBP ifbp, wci_bufp bufp, int len BE_PAR( int word_len ) ) +{ + hcf_io io_port = ifbp->IFB_IOBase + HREG_DATA_1; //BAP data register + int i; //prevent side effects from macro + hcf_16 j; + HCFASSERT( ((hcf_32)bufp & (HCF_ALIGN-1) ) == 0, (hcf_32)bufp ); +#if HCF_BIG_ENDIAN + HCFASSERT( word_len == 0 || word_len == 2 || word_len == 4, word_len ); + HCFASSERT( word_len == 0 || ((hcf_32)bufp & 1 ) == 0, (hcf_32)bufp ); + HCFASSERT( word_len <= len, MERGE_2( word_len, len ) ); + + if ( word_len ) { //if there is anything to convert + //. convert and write the 1st hcf_16 + j = bufp[1] | bufp[0]<<8; + OUT_PORT_WORD( io_port, j ); + //. update pointer and counter accordingly + len -= 2; + bufp += 2; + if ( word_len > 1 ) { //. if there is to convert more than 1 word ( i.e 2 ) + //. . convert and write the 2nd hcf_16 + j = bufp[1] | bufp[0]<<8; /*bufp is already incremented by 2*/ + OUT_PORT_WORD( io_port, j ); + //. . update pointer and counter accordingly + len -= 2; + bufp += 2; + } + } +#endif // HCF_BIG_ENDIAN + i = len; + if ( i && ifbp->IFB_CarryOut ) { //skip zero-length + j = ((*bufp)<<8) + ( ifbp->IFB_CarryOut & 0xFF ); + OUT_PORT_WORD( io_port, j ); + bufp++; i--; + ifbp->IFB_CarryOut = 0; + } +#if (HCF_IO) & HCF_IO_32BITS + //skip zero-length I/O, single byte I/O and I/O not worthwhile (i.e. less than 6 bytes)for DW logic + //if buffer length >= 6 and 32 bits I/O support + if ( !(ifbp->IFB_CntlOpt & USE_16BIT) && i >= 6 ) { + hcf_32 FAR *p4; //prevent side effects from macro + if ( ( (hcf_32)bufp & 0x1 ) == 0 ) { //. if buffer at least word aligned + if ( (hcf_32)bufp & 0x2 ) { //. . if buffer not double word aligned + //. . . write a single word to get double word aligned + j = *(wci_recordp)bufp; //just to help ease writing macros with embedded assembly + OUT_PORT_WORD( io_port, j ); + //. . . adjust buffer length and pointer accordingly + bufp += 2; i -= 2; + } + //. . write as many double word as possible + p4 = (hcf_32 FAR *)bufp; + j = (hcf_16)i/4; + OUT_PORT_STRING_32( io_port, p4, j ); + //. . adjust buffer length and pointer accordingly + bufp += i & ~0x0003; + i &= 0x0003; + } + } +#endif // HCF_IO_32BITS + //if no 32-bit support OR byte aligned OR 1 word left + if ( i ) { + //. if odd number of bytes left + if ( i & 0x0001 ) { + //. . save left over byte (before bufp is corrupted) in carry, set carry flag + ifbp->IFB_CarryOut = (hcf_16)bufp[i-1] | 0x0100; //note that i and bufp are always simultaneously modified, &bufp[i-1] is invariant + } + //. write as many word as possible in "alignment safe" way + j = (hcf_16)i/2; + OUT_PORT_STRING_8_16( io_port, bufp, j ); + } +} // put_frag + + +/************************************************************************************************************ + * + *.SUBMODULE void put_frag_finalize( IFBP ifbp ) + *.PURPOSE cleanup after put_frag for trailing odd byte and MIC transfer to NIC. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * + *.RETURNS N.A. + * + *.DESCRIPTION + * finalize the MIC calculation with the padding pattern, output the last byte (if applicable) + * of the message and the MIC to the TxFS + * + * + *.DIAGRAM + *2: 1 byte of the last put_frag may be still in IFB_CarryOut ( the put_frag carry holder ), so ........ + * 1 - 3 bytes of the last put_frag may be still in IFB_tx_32 ( the MIC engine carry holder ), so ........ + * The call to the MIC calculation routine feeds these remaining bytes (if any) of put_frag and the + * just as many bytes of the padding as needed to the MIC calculation engine. Note that the "unneeded" pad + * bytes simply end up in the MIC engine carry holder and are never used. + *8: write the remainder of the MIC and possible some garbage to NIC RAM + * Note: i is always 4 (a loop-invariant of the while in point 2) + * + *.NOTICE + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC void +put_frag_finalize( IFBP ifbp ) +{ +#if (HCF_TYPE) & HCF_TYPE_WPA + if ( ifbp->IFB_MICTxCarry != 0xFFFF) { //if MIC calculation active + CALC_TX_MIC( mic_pad, 8); //. feed (up to 8 bytes of) virtual padding to MIC engine + //. write (possibly) trailing byte + (most of) MIC + put_frag( ifbp, (wci_bufp)ifbp->IFB_MICTx, 8 BE_PAR(0) ); + } +#endif // HCF_TYPE_WPA + put_frag( ifbp, null_addr, 1 BE_PAR(0) ); //write (possibly) trailing data or MIC byte +} // put_frag_finalize + + +/************************************************************************************************************ + * + *.SUBMODULE int put_info( IFBP ifbp, LTVP ltvp ) + *.PURPOSE support routine to handle the "basic" task of hcf_put_info to pass RIDs to the NIC. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * ltvp address in NIC RAM where LVT-records are located + * + *.RETURNS + * HCF_SUCCESS + * >>put_frag + * >>cmd_wait + * + *.DESCRIPTION + * + * + *.DIAGRAM + *20: do not write RIDs to NICs which have incompatible Firmware + *24: If the RID does not exist, the L-field is set to zero. + * Note that some RIDs can not be read, e.g. the pseudo RIDs for direct Hermes commands and CFG_DEFAULT_KEYS + *28: If the RID is written successful, pass it to the NIC by means of an Access Write command + * + *.NOTICE + * The mechanism to HCF_ASSERT on invalid typ-codes in the LTV record is based on the following strategy: + * - some codes (e.g. CFG_REG_MB) are explicitly handled by the HCF which implies that these codes + * are valid. These codes are already consumed by hcf_put_info. + * - all other codes are passed to the Hermes. Before the put action is executed, hcf_get_info is called + * with an LTV record with a value of 1 in the L-field and the intended put action type in the Typ-code + * field. If the put action type is valid, it is also valid as a get action type code - except + * for CFG_DEFAULT_KEYS and CFG_ADD_TKIP_DEFAULT_KEY - so the HCF_ASSERT logic of hcf_get_info should + * not catch. + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC int +put_info( IFBP ifbp, LTVP ltvp ) +{ + + int rc = HCF_SUCCESS; + + HCFASSERT( ifbp->IFB_CardStat == 0, MERGE_2( ltvp->typ, ifbp->IFB_CardStat ) ); + HCFASSERT( CFG_RID_CFG_MIN <= ltvp->typ && ltvp->typ <= CFG_RID_CFG_MAX, ltvp->typ ); + + if ( ifbp->IFB_CardStat == 0 && /* 20*/ + ( ( CFG_RID_CFG_MIN <= ltvp->typ && ltvp->typ <= CFG_RID_CFG_MAX ) || + ( CFG_RID_ENG_MIN <= ltvp->typ /* && ltvp->typ <= 0xFFFF */ ) ) ) { +#if HCF_ASSERT //FCC8, FCB0, FCB4, FCB6, FCB7, FCB8, FCC0, FCC4, FCBC, FCBD, FCBE, FCBF + { + hcf_16 t = ltvp->typ; + LTV_STRCT x = { 2, t, {0} }; /*24*/ + hcf_get_info( ifbp, (LTVP)&x ); + if ( x.len == 0 && + ( t != CFG_DEFAULT_KEYS && t != CFG_ADD_TKIP_DEFAULT_KEY && t != CFG_REMOVE_TKIP_DEFAULT_KEY && + t != CFG_ADD_TKIP_MAPPED_KEY && t != CFG_REMOVE_TKIP_MAPPED_KEY && + t != CFG_HANDOVER_ADDR && t != CFG_DISASSOCIATE_ADDR && + t != CFG_FCBC && t != CFG_FCBD && t != CFG_FCBE && t != CFG_FCBF && + t != CFG_DEAUTHENTICATE_ADDR + ) + ) { + HCFASSERT( DO_ASSERT, ltvp->typ ); + } + } +#endif // HCF_ASSERT + + rc = setup_bap( ifbp, ltvp->typ, 0, IO_OUT ); + put_frag( ifbp, (wci_bufp)ltvp, 2*ltvp->len + 2 BE_PAR(2) ); + /*28*/ if ( rc == HCF_SUCCESS ) { + rc = cmd_exe( ifbp, HCMD_ACCESS + HCMD_ACCESS_WRITE, ltvp->typ ); + } + } + return rc; +} // put_info + + +/************************************************************************************************************ + * + *.SUBMODULE int put_info_mb( IFBP ifbp, CFG_MB_INFO_STRCT FAR * ltvp ) + *.PURPOSE accumulates a ( series of) buffers into a single Info block into the MailBox. + * + *.ARGUMENTS + * ifbp address of the Interface Block + * ltvp address of structure specifying the "type" and the fragments of the information to be synthesized + * as an LTV into the MailBox + * + *.RETURNS + * + *.DESCRIPTION + * If the data does not fit (including no MailBox is available), the IFB_MBTally is incremented and an + * error status is returned. + * HCF_ASSERT does not catch. + * Calling put_info_mb when their is no MailBox available, is considered a design error in the MSF. + * + * Note that there is always at least 1 word of unused space in the mail box. + * As a consequence: + * - no problem in pointer arithmetic (MB_RP == MB_WP means unambiguously mail box is completely empty + * - There is always free space to write an L field with a value of zero after each MB_Info block. This + * allows for an easy scan mechanism in the "get MB_Info block" logic. + * + * + *.DIAGRAM + *1: Calculate L field of the MBIB, i.e. 1 for the T-field + the cumulative length of the fragments. + *2: The free space in the MailBox is calculated (2a: free part from Write Ptr to Read Ptr, 2b: free part + * turns out to wrap around) . If this space suffices to store the number of words reflected by len (T-field + * + Value-field) plus the additional MailBox Info L-field + a trailing 0 to act as the L-field of a trailing + * dummy or empty LTV record, then a MailBox Info block is build in the MailBox consisting of + * - the value len in the first word + * - type in the second word + * - a copy of the contents of the fragments in the second and higher word + * + *4: Since put_info_mb() can more or less directly be called from the MSF level, the I/F must be robust + * against out-of-range variables. As failsafe coding, the MB update is skipped by changing tlen to 0 if + * len == 0; This will indirectly cause an assert as result of the violation of the next if clause. + *6: Check whether the free space in MailBox suffices (this covers the complete absence of the MailBox). + * Note that len is unsigned, so even MSF I/F violation works out O.K. + * The '2' in the expression "len+2" is used because 1 word is needed for L itself and 1 word is needed + * for the zero-sentinel + *8: update MailBox Info length report to MSF with "oldest" MB Info Block size. Be careful here, if you get + * here before the MailBox is registered, you can't read from the buffer addressed by IFB_MBp (it is the + * Null buffer) so don't move this code till the end of this routine but keep it where there is garuanteed + * a buffer. + * + *.NOTICE + * boundary testing depends on the fact that IFB_MBSize is guaranteed to be zero if no MailBox is present, + * and to a lesser degree, that IFB_MBWp = IFB_MBRp = 0 + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ + +HCF_STATIC int +put_info_mb( IFBP ifbp, CFG_MB_INFO_STRCT FAR * ltvp ) +{ + + int rc = HCF_SUCCESS; + hcf_16 i; //work counter + hcf_16 *dp; //destination pointer (in MailBox) + wci_recordp sp; //source pointer + hcf_16 len; //total length to copy to MailBox + hcf_16 tlen; //free length/working length/offset in WMP frame + + if ( ifbp->IFB_MBp == NULL ) return rc; //;?not sufficient + HCFASSERT( ifbp->IFB_MBp != NULL, 0 ); //!!!be careful, don't get into an endless recursion + HCFASSERT( ifbp->IFB_MBSize, 0 ); + + len = 1; /* 1 */ + for ( i = 0; i < ltvp->frag_cnt; i++ ) { + len += ltvp->frag_buf[i].frag_len; + } + if ( ifbp->IFB_MBRp > ifbp->IFB_MBWp ) { + tlen = ifbp->IFB_MBRp - ifbp->IFB_MBWp; /* 2a*/ + } else { + if ( ifbp->IFB_MBRp == ifbp->IFB_MBWp ) { + ifbp->IFB_MBRp = ifbp->IFB_MBWp = 0; // optimize Wrapping + } + tlen = ifbp->IFB_MBSize - ifbp->IFB_MBWp; /* 2b*/ + if ( ( tlen <= len + 2 ) && ( len + 2 < ifbp->IFB_MBRp ) ) { //if trailing space is too small but + // leading space is sufficiently large + ifbp->IFB_MBp[ifbp->IFB_MBWp] = 0xFFFF; //flag dummy LTV to fill the trailing space + ifbp->IFB_MBWp = 0; //reset WritePointer to begin of MailBox + tlen = ifbp->IFB_MBRp; //get new available space size + } + } + dp = &ifbp->IFB_MBp[ifbp->IFB_MBWp]; + if ( len == 0 ) { + tlen = 0; //;? what is this good for + } + if ( len + 2 >= tlen ){ /* 6 */ + //Do Not ASSERT, this is a normal condition + IF_TALLY( ifbp->IFB_HCF_Tallies.NoBufMB++ ); + rc = HCF_ERR_LEN; + } else { + *dp++ = len; //write Len (= size of T+V in words to MB_Info block + *dp++ = ltvp->base_typ; //write Type to MB_Info block + ifbp->IFB_MBWp += len + 1; //update WritePointer of MailBox + for ( i = 0; i < ltvp->frag_cnt; i++ ) { // process each of the fragments + sp = ltvp->frag_buf[i].frag_addr; + len = ltvp->frag_buf[i].frag_len; + while ( len-- ) *dp++ = *sp++; + } + ifbp->IFB_MBp[ifbp->IFB_MBWp] = 0; //to assure get_info for CFG_MB_INFO stops + ifbp->IFB_MBInfoLen = ifbp->IFB_MBp[ifbp->IFB_MBRp]; /* 8 */ + } + return rc; +} // put_info_mb + + +/************************************************************************************************************ + * + *.SUBMODULE int setup_bap( IFBP ifbp, hcf_16 fid, int offset, int type ) + *.PURPOSE set up data access to NIC RAM via BAP_1. + * + *.ARGUMENTS + * ifbp address of I/F Block + * fid FID/RID + * offset !!even!! offset in FID/RID + * type IO_IN, IO_OUT + * + *.RETURNS + * HCF_SUCCESS O.K + * HCF_ERR_NO_NIC card is removed + * HCF_ERR_DEFUNCT_TIME_OUT Fatal malfunction detected + * HCF_ERR_DEFUNCT_..... if and only if IFB_DefunctStat <> 0 + * + *.DESCRIPTION + * + * A non-zero return status indicates: + * - the NIC is considered nonoperational, e.g. due to a time-out of some Hermes activity in the past + * - BAP_1 could not properly be initialized + * - the card is removed before completion of the data transfer + * In all other cases, a zero is returned. + * BAP Initialization failure indicates an H/W error which is very likely to signal complete H/W failure. + * Once a BAP Initialization failure has occurred all subsequent interactions with the Hermes will return a + * "defunct" status till the Hermes is re-initialized by means of an hcf_connect. + * + * A BAP is a set of registers (Offset, Select and Data) offering read/write access to a particular FID or + * RID. This access is based on a auto-increment feature. + * There are two BAPs but these days the HCF uses only BAP_1 and leaves BAP_0 to the PCI Busmastering H/W. + * + * The BAP-mechanism is based on the Busy bit in the Offset register (see the Hermes definition). The waiting + * for Busy must occur between writing the Offset register and accessing the Data register. The + * implementation to wait for the Busy bit drop after each write to the Offset register, implies that the + * requirement that the Busy bit is low before the Select register is written, is automatically met. + * BAP-setup may be time consuming (e.g. 380 usec for large offsets occurs frequently). The wait for Busy bit + * drop is protected by a loop counter, which is initialized with IFB_TickIni, which is calibrated in init. + * + * The NIC I/F is optimized for word transfer and can only handle word transfer at a word boundary in NIC + * RAM. The intended solution for transfer of a single byte has multiple H/W flaws. There have been different + * S/W Workaround strategies. RID access is hcf_16 based by "nature", so no byte access problems. For Tx/Rx + * FID access, the byte logic became obsolete by absorbing it in the double word oriented nature of the MIC + * feature. + * + * + *.DIAGRAM + * + *2: the test on rc checks whether the HCF went into "defunct" mode ( e.g. BAP initialization or a call to + * cmd_wait did ever fail). + *4: the select register and offset register are set + * the offset register is monitored till a successful condition (no busy bit) is detected or till the + * (calibrated) protection counter expires + * If the counter expires, this is reflected in IFB_DefunctStat, so all subsequent calls to setup_bap fail + * immediately ( see 2) + *6: initialization of the carry as used by pet/get_frag + *8: HREG_OFFSET_ERR is ignored as error because: + * a: the Hermes is robust against it + * b: it is not known what causes it (probably a bug), hence no strategy can be specified which level is + * to handle this error in which way. In the past, it could be induced by the MSF level, e.g. by calling + * hcf_rcv_msg while there was no Rx-FID available. Since this is an MSF-error which is caught by ASSERT, + * there is no run-time action required by the HCF. + * Lumping the Offset error in with the Busy bit error, as has been done in the past turns out to be a + * disaster or a life saver, just depending on what the cause of the error is. Since no prediction can be + * done about the future, it is "felt" to be the best strategy to ignore this error. One day the code was + * accompanied by the following comment: + * // ignore HREG_OFFSET_ERR, someone, supposedly the MSF programmer ;) made a bug. Since we don't know + * // what is going on, we might as well go on - under management pressure - by ignoring it + * + *.ENDDOC END DOCUMENTATION + * + ************************************************************************************************************/ +HCF_STATIC int +setup_bap( IFBP ifbp, hcf_16 fid, int offset, int type ) +{ + PROT_CNT_INI; + int rc; + + HCFTRACE( ifbp, HCF_TRACE_STRIO ); + rc = ifbp->IFB_DefunctStat; + if (rc == HCF_SUCCESS) { /*2*/ + OPW( HREG_SELECT_1, fid ); /*4*/ + OPW( HREG_OFFSET_1, offset ); + if ( type == IO_IN ) { + ifbp->IFB_CarryIn = 0; + } + else ifbp->IFB_CarryOut = 0; + HCF_WAIT_WHILE( IPW( HREG_OFFSET_1) & HCMD_BUSY ); + HCFASSERT( !( IPW( HREG_OFFSET_1) & HREG_OFFSET_ERR ), MERGE_2( fid, offset ) ); /*8*/ + if ( prot_cnt == 0 ) { + HCFASSERT( DO_ASSERT, MERGE_2( fid, offset ) ); + rc = ifbp->IFB_DefunctStat = HCF_ERR_DEFUNCT_TIME_OUT; + ifbp->IFB_CardStat |= CARD_STAT_DEFUNCT; + } + } + HCFTRACE( ifbp, HCF_TRACE_STRIO | HCF_TRACE_EXIT ); + return rc; +} // setup_bap + |