diff options
author | Yinghai Lu <yinghailu@gmail.com> | 2006-10-04 20:46:15 +0000 |
---|---|---|
committer | Yinghai Lu <yinghailu@gmail.com> | 2006-10-04 20:46:15 +0000 |
commit | d4b278c02c1da92219ebeb34204b9768934aeca3 (patch) | |
tree | 488d097cac9744cfc9b8ff7c89ce69bcb21370cb /src/northbridge/amd/amdk8/raminit_f_dqs.c | |
parent | 2e3757d11c565a8fe68dc2a2c34975e98304533c (diff) | |
download | coreboot-d4b278c02c1da92219ebeb34204b9768934aeca3.tar.xz |
AMD Rev F support
git-svn-id: svn://svn.coreboot.org/coreboot/trunk@2435 2b7e53f0-3cfb-0310-b3e9-8179ed1497e1
Diffstat (limited to 'src/northbridge/amd/amdk8/raminit_f_dqs.c')
-rw-r--r-- | src/northbridge/amd/amdk8/raminit_f_dqs.c | 2036 |
1 files changed, 2036 insertions, 0 deletions
diff --git a/src/northbridge/amd/amdk8/raminit_f_dqs.c b/src/northbridge/amd/amdk8/raminit_f_dqs.c new file mode 100644 index 0000000000..f77b6d6c86 --- /dev/null +++ b/src/northbridge/amd/amdk8/raminit_f_dqs.c @@ -0,0 +1,2036 @@ +/* + yhlu 2005.10 dqs training +*/ +//0: mean no debug info +#define DQS_TRAIN_DEBUG 0 + +static inline void print_debug_dqs(const char *str, unsigned val, unsigned level) +{ +#if DQS_TRAIN_DEBUG > 0 + if(DQS_TRAIN_DEBUG > level) { + #if CONFIG_USE_INIT == 1 + printk_debug("%s%x\r\n", str, val); + #else + print_debug(str); print_debug_hex32(val); print_debug("\r\n"); + #endif + } +#endif +} + +static inline void print_debug_dqs_pair(const char *str, unsigned val, const char *str2, unsigned val2, unsigned level) +{ +#if DQS_TRAIN_DEBUG > 0 + if(DQS_TRAIN_DEBUG > level) { + #if CONFIG_USE_INIT == 1 + printk_debug("%s%08x%s%08x\r\n", str, val, str2, val2); + #else + print_debug(str); print_debug_hex32(val); print_debug(str2); print_debug_hex32(val2); print_debug("\r\n"); + #endif + } +#endif +} + +static inline void print_debug_dqs_tsc(const char *str, unsigned i, unsigned val, unsigned val2, unsigned level) +{ +#if DQS_TRAIN_DEBUG > 0 + if(DQS_TRAIN_DEBUG > level) { + #if CONFIG_USE_INIT == 1 + printk_debug("%s[%02x]=%08x%08x\r\n", str, i, val, val2); + #else + print_debug(str); print_debug("["); print_debug_hex8(i); print_debug("]="); print_debug_hex32(val); print_debug_hex32(val2); print_debug("\r\n"); + #endif + } +#endif +} + +static inline void print_debug_dqs_tsc_x(const char *str, unsigned i, unsigned val, unsigned val2) +{ + #if CONFIG_USE_INIT == 1 + printk_debug("%s[%02x]=%08x%08x\r\n", str, i, val, val2); + #else + print_debug(str); print_debug("["); print_debug_hex8(i); print_debug("]="); print_debug_hex32(val); print_debug_hex32(val2); print_debug("\r\n"); + #endif + +} + +static void fill_mem_cs_sysinfo(unsigned nodeid, const struct mem_controller *ctrl, struct sys_info *sysinfo) +{ + + int i; + sysinfo->mem_base[nodeid] = pci_read_config32(ctrl->f1, 0x40 + (nodeid<<3)); + + for(i=0;i<8; i++) { + sysinfo->cs_base[nodeid*8+i] = pci_read_config32(ctrl->f2, 0x40 + (i<<2)); + } + + sysinfo->hole_reg[nodeid] = pci_read_config32(ctrl->f1, 0xf0); + +} +static unsigned Get_MCTSysAddr(const struct mem_controller *ctrl, unsigned cs_idx, struct sys_info *sysinfo) +{ + uint32_t dword; + uint32_t mem_base; + unsigned nodeid = ctrl->node_id; + +#if HW_MEM_HOLE_SIZEK != 0 + uint32_t hole_reg; +#endif + + //get the local base addr of the chipselect + dword = sysinfo->cs_base[nodeid * 8 + cs_idx]; + dword &= 0xfffffff0; + + //sys addr= node base + local cs base + mem_base = sysinfo->mem_base[nodeid]; + mem_base &= 0xffff0000; + + dword += mem_base; +#if HW_MEM_HOLE_SIZEK != 0 + hole_reg = sysinfo->hole_reg[nodeid]; + if(hole_reg & 1) { + unsigned hole_startk; + hole_startk = (hole_reg & (0xff<<24)) >> 10; + if( (dword >= (hole_startk<<2)) && (dword < ((4*1024*1024)<<2))) { + dword += ((4*1024*1024 - hole_startk)<<2); + } + } +#endif + + //add 1MB offset to avoid compat area + dword += (1<<(20-8)); + + //So final result is upper 32 bit addr + + return dword; + +} + +static unsigned Get_RcvrSysAddr(const struct mem_controller * ctrl, unsigned channel, unsigned cs_idx, struct sys_info *sysinfo) +{ +#if 0 + //get SB_64MuxedMode + uint32_t dword; + dword = pci_read_config32(ctrl->f2, DRAM_CTRL_MISC); + if((dword & DCM_Mode64BitMux) == DCM_Mode64BitMux) { + if(channel) cs_idx += 4; // translate Receiver number to Chipsel + } +#endif + + return Get_MCTSysAddr(ctrl, cs_idx, sysinfo); + +} + +static inline unsigned long read_cr4(void) +{ + unsigned long cr4; + asm volatile ("movl %%cr4, %0" : "=r" (cr4)); + return cr4; +} + +static inline void write_cr4(unsigned long cr4) +{ + asm volatile ("movl %0, %%cr4" : : "r" (cr4)); +} + + +static inline void enable_sse2() +{ + unsigned long cr4; + cr4 = read_cr4(); + cr4 |= (1<<9); + write_cr4(cr4); +} + +static inline void disable_sse2() +{ + unsigned long cr4; + cr4 = read_cr4(); + cr4 &= ~(1<<9); + write_cr4(cr4); +} + + +static void set_wrap32dis(void) { + msr_t msr; + + msr = rdmsr(0xc0010015); + msr.lo |= (1<<17); + + wrmsr(0xc0010015, msr); + +} + +static void clear_wrap32dis(void) { + msr_t msr; + + msr = rdmsr(0xc0010015); + msr.lo &= ~(1<<17); + + wrmsr(0xc0010015, msr); + +} + +static void set_FSBASE(uint32_t addr_hi) +{ + msr_t msr; + + //set fs and use fs prefix to access the mem + msr.hi = addr_hi; + msr.lo = 0; + wrmsr(0xc0000100, msr); //FS_BASE + +} + +#if 0 +static void write_mem(uint32_t addr_hi, uint32_t addr_lo, uint32_t value) +{ + if(addr_hi == 0) { + *((uint32_t *)addr_lo) = value; + return; + } + + set_FSBASE(addr_hi); + + __asm__ volatile ( + "movl %1, %%fs:(%0)\n\t" + :: "a" (addr_lo), "b" (value) + ); + +} + +static uint32_t read_mem(uint32_t addr_hi, uint32_t addr_lo) +{ + unsigned value; + if(addr_hi == 0) { + value = *((uint32_t *)addr_lo); + return value; + } + + set_FSBASE(addr_hi); + + __asm__ volatile ( + "movl %%fs:(%1), %0\n\t" + :"=b"(value): "a" (addr_lo) + ); + + return value; + +} +#endif + +static unsigned ChipSelPresent(const struct mem_controller *ctrl, unsigned cs_idx, struct sys_info *sysinfo) +{ + unsigned enabled; + unsigned nodeid = ctrl->node_id; + + + enabled = sysinfo->cs_base[nodeid * 8 + cs_idx]; + enabled &= 1; + + return enabled; + +} + +static unsigned RcvrRankEnabled(const struct mem_controller *ctrl, int channel, int cs_idx, unsigned is_Width128, struct sys_info *sysinfo) +{ + if(!is_Width128) { + if(channel) return 0; // no channel b + } + + return ChipSelPresent(ctrl, cs_idx, sysinfo); +} + +static void WriteLNTestPattern(unsigned addr_lo, uint8_t *buf_a, unsigned line_num) +{ + __asm__ volatile ( + "1:\n\t" + "movdqa (%3), %%xmm0\n\t" + "movntdq %%xmm0, %%fs:(%0)\n\t" /* xmm0 is 128 bit */ + "addl %1, %0\n\t" + "addl %1, %3\n\t" + "loop 1b\n\t" + + :: "a" (addr_lo), "d" (16), "c" (line_num * 4), "b"(buf_a) + ); + + +} + +static void Write1LTestPattern(unsigned addr, unsigned p, uint8_t *buf_a, uint8_t *buf_b) +{ + uint8_t *buf; + if(p==1) { buf = buf_b; } + else { buf = buf_a; } + + set_FSBASE (addr>>24); + + WriteLNTestPattern(addr<<8, buf, 1); +} + +static void Read1LTestPattern(unsigned addr) +{ + unsigned value; + + set_FSBASE(addr>>24); + + /* 1st move causes read fill (to exclusive or shared)*/ + __asm__ volatile ( + "movl %%fs:(%1), %0\n\t" + :"=b"(value): "a" (addr<<8) + ); + +} + +#define DQS_PASS 0 +#define DQS_FAIL 1 + +#define DQS_FIRST_PASS 1 +#define DQS_SECOND_PASS 2 + +#define SB_NORCVREN 11 +#define RCVREN_MARGIN 6 +#define SB_SmallRCVR 13 +#define SB_CHA2BRCVREN 12 +#define SB_NODQSPOS 14 +#define MIN_DQS_WNDW 3 +#define SB_SMALLDQS 15 + + +static unsigned CompareTestPatternQW0(unsigned channel, unsigned addr, unsigned pattern, const uint32_t *TestPattern0, const uint32_t *TestPattern1, const uint32_t *TestPattern2, unsigned Pass, unsigned is_Width128) +{ + uint32_t addr_lo; + uint32_t *test_buf; + uint32_t value; + uint32_t value_test; + unsigned result = DQS_FAIL; + + if(Pass == DQS_FIRST_PASS) { + if(pattern==1) { + test_buf = (uint32_t *)TestPattern1; + } + else { + test_buf = (uint32_t *)TestPattern0; + } + } + else { + test_buf = (uint32_t *)TestPattern2; + } + + set_FSBASE(addr>>24); + + addr_lo = addr<<8; + + if(is_Width128 && (channel == 1)) { + addr_lo += 8; //second channel + test_buf += 2; + } + + __asm__ volatile ( + "movl %%fs:(%1), %0\n\t" + :"=b"(value): "a" (addr_lo) + ); + + value_test = *test_buf; + + + print_debug_dqs_pair("\t\t\t\t\t\tQW0.lo : test_buf= ", (unsigned)test_buf, " value = ", value_test, 4); + print_debug_dqs_pair("\t\t\t\t\t\tQW0.lo : addr_lo = ", addr_lo, " value = ", value, 4); + + if(value == value_test) { + addr_lo += 4; + test_buf++; + __asm__ volatile ( + "movl %%fs:(%1), %0\n\t" + :"=b"(value): "a" (addr_lo) + ); + value_test = *test_buf; + print_debug_dqs_pair("\t\t\t\t\t\tQW0.hi : test_buf= ", (unsigned)test_buf, " value = ", value_test, 4); + print_debug_dqs_pair("\t\t\t\t\t\tQW0.hi : addr_lo = ", addr_lo, " value = ", value, 4); + + if(value == value_test){ + result = DQS_PASS; + } + } + + if(Pass == DQS_SECOND_PASS) { // second pass need to be inverted + if(result==DQS_PASS) { + result = DQS_FAIL; + } + else { + result = DQS_PASS; + } + } + + return result; + +} + +static void SetMaxAL_RcvrDly(const struct mem_controller *ctrl, unsigned dly) +{ + uint32_t reg; + + dly += (20-1); // round it + dly /= 20; // convert from unit 50ps to 1ns + + dly += 6; + + + reg = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH); + reg &= ~(DCH_MaxAsyncLat_MASK <<DCH_MaxAsyncLat_SHIFT); + reg |= ((dly - DCH_MaxAsyncLat_BASE) << DCH_MaxAsyncLat_SHIFT); + pci_write_config32(ctrl->f2, DRAM_CONFIG_HIGH, reg); + +} + +/* + Set the Target range to WT IO (using an IORR overlapping the already existing + WB dram type). Use IORR0 +*/ +static void SetTargetWTIO(unsigned addr) +{ + msr_t msr; + msr.hi = addr>>24; + msr.lo = addr<<8; + wrmsr(0xc0010016, msr); //IORR0 BASE + + msr.hi = 0xff; + msr.lo = 0xfc000800; // 64MB Mask + wrmsr(0xc0010017, msr); // IORR0 Mask +} + +static void ResetTargetWTIO(void) +{ + msr_t msr; + + msr.hi = 0; + msr.lo = 0; + wrmsr(0xc0010017, msr); // IORR0 Mask +} + +static void proc_CLFLUSH(unsigned addr) +{ + + set_FSBASE(addr>>24); + + /* 1st move causes read fill (to exclusive or shared)*/ + __asm__ volatile ( + /* clflush fs:[eax] */ + "clflush %%fs:(%0)\n\t" + ::"a" (addr<<8) + ); + +} +static void proc_IOCLFLUSH(unsigned addr) +{ + SetTargetWTIO(addr); + proc_CLFLUSH(addr); + ResetTargetWTIO(); +} + +static void ResetDCTWrPtr(const struct mem_controller *ctrl) +{ + uint32_t dword; + unsigned index = 0x10; + + dword = pci_read_config32_index_wait(ctrl->f2, 0x98, index); + pci_write_config32_index_wait(ctrl->f2, 0x98, index, dword); + +} + + +static uint16_t get_exact_T1000(unsigned i) +{ + // 200 266, 333, 400 + const static uint16_t T1000_a[]= { 5000, 3759, 3003, 2500 }; + + static const uint16_t TT_a[] = { + /*200 266 333 400 */ + /*4 */ 6250, 6250, 6250, 6250, + /*5 */ 5000, 5000, 5000, 2500, + /*6 */ 5000, 4166, 4166, 2500, + /*7 */ 5000, 4285, 3571, 2500, + + /*8 */ 5000, 3750, 3125, 2500, + /*9 */ 5000, 3888, 3333, 2500, + /*10*/ 5000, 4000, 3000, 2500, + /*11*/ 5000, 4090, 3181, 2500, + + /*12*/ 5000, 3750, 3333, 2500, + /*13*/ 5000, 3846, 3076, 2500, + /*14*/ 5000, 3928, 3214, 2500, + /*15*/ 5000, 4000, 3000, 2500, + }; + + unsigned fid_cur; + int index; + + msr_t msr; + msr = rdmsr(0xc0010042); + fid_cur = msr.lo & 0x3f; + + index = fid_cur>>1; + + if(index>12) return T1000_a[i]; + + return TT_a[index * 4+i]; + +} + +static void InitDQSPos4RcvrEn(const struct mem_controller *ctrl) +{ + int i; + uint32_t dword; + + dword = 0x00000000; + for(i=1; i<=3; i++) { + /* Program the DQS Write Timing Control Registers (Function 2:Offset 0x9c, index 0x01-0x03, 0x21-0x23) to 0x00 for all bytes */ + pci_write_config32_index_wait(ctrl->f2, 0x98, i, dword); + pci_write_config32_index_wait(ctrl->f2, 0x98, i+0x20, dword); + } + + dword = 0x2f2f2f2f; + for(i=5; i<=7; i++) { + /* Program the DQS Write Timing Control Registers (Function 2:Offset 0x9c, index 0x05-0x07, 0x25-0x27) to 0x2f for all bytes */ + pci_write_config32_index_wait(ctrl->f2, 0x98, i, dword); + pci_write_config32_index_wait(ctrl->f2, 0x98, i+0x20, dword); + } + + +} +#ifndef K8_REV_F_SUPPORT_F0_F1_WORKAROUND +#define K8_REV_F_SUPPORT_F0_F1_WORKAROUND 1 +#endif + +static void TrainRcvrEn(const struct mem_controller *ctrl, unsigned Pass, struct sys_info *sysinfo) +{ + + const static uint32_t TestPattern0[] = { + 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, + 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, + 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, + 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, 0xaaaaaaaa, + }; + const static uint32_t TestPattern1[] = { + 0x55555555, 0x55555555, 0x55555555, 0x55555555, + 0x55555555, 0x55555555, 0x55555555, 0x55555555, + 0x55555555, 0x55555555, 0x55555555, 0x55555555, + 0x55555555, 0x55555555, 0x55555555, 0x55555555, + }; + const static uint32_t TestPattern2[] = { + 0x12345678, 0x87654321, 0x23456789, 0x98765432, + 0x59385824, 0x30496724, 0x24490795, 0x99938733, + 0x40385642, 0x38465245, 0x29432163, 0x05067894, + 0x12349045, 0x98723467, 0x12387634, 0x34587623, + }; + + uint8_t pattern_buf_x[64 * 4 + 16]; // We need to two cache line So have more 16 bytes to keep 16 byte alignment */ + uint8_t *buf_a, *buf_b; + uint32_t ecc_bit; + uint32_t dword; + uint8_t *dqs_rcvr_dly_a = &sysinfo->dqs_rcvr_dly_a[ctrl->node_id * 2* 8] ; //8 node, channel 2, receiver 8 + + int i; + + unsigned channel, receiver; + + unsigned Errors; + unsigned CTLRMaxDelay; + unsigned T1000; + + unsigned LastTest; + unsigned CurrTest; + unsigned Test0, Test1; + + unsigned RcvrEnDlyRmin; + + unsigned two_ranks; + unsigned RcvrEnDly; + + unsigned PatternA; + unsigned PatternB; + + unsigned TestAddr0, TestAddr0B, TestAddr1, TestAddr1B; + + unsigned CurrRcvrCHADelay; + + unsigned tmp; + + unsigned is_Width128 = sysinfo->meminfo[ctrl->node_id].is_Width128; + + unsigned cpu_f0_f1; + + if(Pass == DQS_FIRST_PASS) { + InitDQSPos4RcvrEn(ctrl); + } + + //enable SSE2 + enable_sse2(); + + //wrap32dis + set_wrap32dis(); + + //disable ECC temp + dword = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW); + ecc_bit = dword & DCL_DimmEccEn; + dword &= ~(DCL_DimmEccEn); + pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dword); + + + if(Pass == DQS_FIRST_PASS) { +#if K8_REV_F_SUPPORT_F0_F1_WORKAROUND == 1 + cpu_f0_f1 = is_cpu_pre_f2_in_bsp(ctrl->node_id); + if(!cpu_f0_f1) +#endif + { +#if 1 + /* Set the DqsRcvEnTrain bit */ + dword = pci_read_config32(ctrl->f2, DRAM_CTRL); + dword |= DC_DqsRcvEnTrain; + pci_write_config32(ctrl->f2, DRAM_CTRL, dword); +#endif + } + } + + //get T1000 figures (cycle time (ns)) * 1K + dword = pci_read_config32(ctrl->f2, DRAM_CONFIG_HIGH); + dword &= DCH_MemClkFreq_MASK; + + T1000 = get_exact_T1000(dword); + + // SetupRcvrPattern + buf_a = (uint8_t *)(((uint32_t)(&pattern_buf_x[0]) + 0x10) & (0xfffffff0)); + buf_b = buf_a + 128; //?? + if(Pass==DQS_FIRST_PASS) { + for(i=0;i<16;i++) { + *((uint32_t *)(buf_a + i*4)) = TestPattern0[i]; + *((uint32_t *)(buf_b + i*4)) = TestPattern1[i]; + } + } + else { + for(i=0;i<16;i++) { + *((uint32_t *)(buf_a + i*4)) = TestPattern2[i]; + *((uint32_t *)(buf_b + i*4)) = TestPattern2[i]; + } + } + + print_debug_dqs("\r\nTrainRcvEn: 0 ctrl", ctrl->node_id, 0); + + print_debug_addr("TrainRcvEn: buf_a:", buf_a); + + Errors = 0; + /* for each channel */ + CTLRMaxDelay = 0; + for(channel = 0; (channel < 2) && (!Errors); channel++) + { + print_debug_dqs("\tTrainRcvEn51: channel ",channel, 1); + + /* for each rank */ + /* there are four recriver pairs, loosely associated with CS */ + for( receiver = 0; (receiver < 8) && (!Errors); receiver+=2) + { + + unsigned index=(receiver>>1) * 3 + 0x10; + + print_debug_dqs("\t\tTrainRcvEn52: index ", index, 2); + + if(is_Width128) { + if(channel) { + dword = pci_read_config32_index_wait(ctrl->f2, 0x98, index); + CurrRcvrCHADelay= dword & 0xff; + } + } + else { + if(channel) { + index += 0x20; + } + } + + LastTest = DQS_FAIL; + RcvrEnDlyRmin = 0xaf; + + if(!RcvrRankEnabled(ctrl, channel, receiver, is_Width128, sysinfo)) continue; + + /* for each DQS receiver enable setting */ + + TestAddr0 = Get_RcvrSysAddr(ctrl, channel, receiver, sysinfo); + + TestAddr0B = TestAddr0 + (1<<(20+2-8)); // 4MB + + if(RcvrRankEnabled(ctrl, channel, receiver+1, is_Width128, sysinfo)) { + TestAddr1 = Get_RcvrSysAddr(ctrl, channel, receiver+1, sysinfo); + TestAddr1B = TestAddr1 + (1<<(20+2-8)); //4MB + two_ranks = 1; + } + else { + two_ranks = 0; + } + + print_debug_dqs("\t\tTrainRcvEn53: TestAddr0B ", TestAddr0B, 2); + + Write1LTestPattern(TestAddr0, 0, buf_a, buf_b); // rank0 of dimm, test p0 + Write1LTestPattern(TestAddr0B, 1, buf_a, buf_b); //rank0 of dimm, test p1 + + if(two_ranks == 1) { + Write1LTestPattern(TestAddr1, 0, buf_a, buf_b); //rank 1 of dimm + Write1LTestPattern(TestAddr1B, 1, buf_a, buf_b);//rank 1 of dimm + } + + if(Pass == DQS_FIRST_PASS) { + RcvrEnDly = 0; + } else { + RcvrEnDly = dqs_rcvr_dly_a[channel * 8 + receiver]; + } + + while ( RcvrEnDly < 0xaf) { // Sweep Delay value here + print_debug_dqs("\t\t\tTrainRcvEn541: RcvrEnDly ", RcvrEnDly, 3); + + if(RcvrEnDly & 1) { + /* Odd steps get another pattern such that even + and odd steps alternate. + The pointers to the patterns will be swapped + at the end of the loop so they are correspond + */ + PatternA = 1; + PatternB = 0; + } + else { + /* Even step */ + PatternA = 0; + PatternB = 1; + } + + /* Program current Receiver enable delay */ + pci_write_config32_index_wait(ctrl->f2, 0x98, index, RcvrEnDly); + + if(is_Width128) { + /* Program current Receiver enable delay chaannel b */ + pci_write_config32_index_wait(ctrl->f2, 0x98, index+ 0x20, RcvrEnDly); + } + + /* Program the MaxAsyncLat filed with the + current DQS receiver enable setting plus 6ns + */ + /*Porgram MaxAsyncLat to correspond with current delay */ + SetMaxAL_RcvrDly(ctrl, RcvrEnDly); + + CurrTest = DQS_FAIL; + + Read1LTestPattern(TestAddr0); //Cache Fill + /* ROM vs cache compare */ + Test0 = CompareTestPatternQW0(channel, TestAddr0, PatternA, TestPattern0, TestPattern1, TestPattern2, Pass, is_Width128); + proc_IOCLFLUSH(TestAddr0); + + ResetDCTWrPtr(ctrl); + + print_debug_dqs("\t\t\tTrainRcvEn542: Test0 ", Test0, 3); + + if(Test0 == DQS_PASS) { + + Read1LTestPattern(TestAddr0B); + Test1 = CompareTestPatternQW0(channel, TestAddr0B, PatternB, TestPattern0, TestPattern1, TestPattern2, Pass, is_Width128); + proc_IOCLFLUSH(TestAddr0B); + + ResetDCTWrPtr(ctrl); + + print_debug_dqs("\t\t\tTrainRcvEn543: Test1 ", Test1, 3); + + if(Test1 == DQS_PASS) { + if(two_ranks) { + Read1LTestPattern(TestAddr1); + Test0 = CompareTestPatternQW0(channel, TestAddr1, PatternA, TestPattern0, TestPattern1, TestPattern2, Pass, is_Width128); + proc_IOCLFLUSH(TestAddr1); + ResetDCTWrPtr(ctrl); + + if(Test0 == DQS_PASS) { + Read1LTestPattern(TestAddr1B); + Test1 = CompareTestPatternQW0(channel, TestAddr1B, PatternB, TestPattern0, TestPattern1, TestPattern2, Pass, is_Width128); + proc_IOCLFLUSH(TestAddr1B); + ResetDCTWrPtr(ctrl); + + if(Test1 == DQS_PASS) { + CurrTest = DQS_PASS; + } + } + print_debug_dqs("\t\t\tTrainRcvEn544: Test0 ", Test0, 3); + } + else { + CurrTest = DQS_PASS; + } + } + } + + print_debug_dqs("\t\t\tTrainRcvEn55: RcvrEnDly ", RcvrEnDly, 3); + + if(CurrTest == DQS_PASS) { + if(LastTest == DQS_FAIL) { + RcvrEnDlyRmin = RcvrEnDly; + break; + } + } + + LastTest = CurrTest; + + /* swap the rank 0 pointers */ + tmp = TestAddr0; + TestAddr0 = TestAddr0B; + TestAddr0B = tmp; + + /* swap the rank 1 pointers */ + tmp = TestAddr1; + TestAddr1 = TestAddr1B; + TestAddr1B = tmp; + + print_debug_dqs("\t\t\tTrainRcvEn56: RcvrEnDly ", RcvrEnDly, 3); + + RcvrEnDly++; + + } // while RcvrEnDly + + print_debug_dqs("\t\tTrainRcvEn61: RcvrEnDly ", RcvrEnDly, 2); + + if(RcvrEnDlyRmin == 0xaf) { + //no passing window + Errors |= SB_NORCVREN; + } + + if(Pass == DQS_FIRST_PASS) { + // We need a better value for DQSPos trainning + RcvrEnDly = RcvrEnDlyRmin /* + RCVREN_MARGIN * T1000/64/50 */; + } else { + RcvrEnDly = RcvrEnDlyRmin; + } + + if(RcvrEnDly > 0xae) { + //passing window too narrow, too far delayed + Errors |= SB_SmallRCVR; + RcvrEnDly = 0xae; + } + + if(Pass == DQS_SECOND_PASS) { //second pass must average vales + RcvrEnDly += dqs_rcvr_dly_a[channel * 8 + receiver] /* - (RCVREN_MARGIN * T1000/64/50)*/; + RcvrEnDly >>= 1; + } + + dqs_rcvr_dly_a[channel * 8 + receiver] = RcvrEnDly; + + //Set final RcvrEnDly for this DIMM and Channel + pci_write_config32_index_wait(ctrl->f2, 0x98, index, RcvrEnDly); + + if(is_Width128) { + pci_write_config32_index_wait(ctrl->f2, 0x98, index+0x20, RcvrEnDly); // channel B + if(channel) { + pci_write_config32_index_wait(ctrl->f2, 0x98, index, CurrRcvrCHADelay); + if(RcvrEnDly > CurrRcvrCHADelay) { + dword = RcvrEnDly - CurrRcvrCHADelay; + } + else { + dword = CurrRcvrCHADelay - RcvrEnDly; + } + dword *= 50; + if(dword > T1000) { + Errors |= SB_CHA2BRCVREN; + } + } + } + + print_debug_dqs("\t\tTrainRcvEn63: RcvrEnDly ", RcvrEnDly, 2); + + if(RcvrEnDly > CTLRMaxDelay) { + CTLRMaxDelay = RcvrEnDly; + } + + print_debug_dqs("\t\tTrainRcvEn64: CTLRMaxDelay ", CTLRMaxDelay, 2); + + } /* receiver */ + } /* channel */ + + print_debug_dqs("\tTrainRcvEn65: CTLRMaxDelay ", CTLRMaxDelay, 1); + + /* Program the MaxAsysncLat field with the largest DQS Receiver Enable setting */ + SetMaxAL_RcvrDly(ctrl, CTLRMaxDelay); + ResetDCTWrPtr(ctrl); + + //Enable ECC again + dword = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW); + dword &= ~(DCL_DimmEccEn); + dword |= ecc_bit; + pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dword); + + if(Pass == DQS_FIRST_PASS) { +#if K8_REV_F_SUPPORT_F0_F1_WORKAROUND == 1 + if(!cpu_f0_f1) +#endif + { + dword = pci_read_config32(ctrl->f2, DRAM_CTRL); + dword &= ~DC_DqsRcvEnTrain; + pci_write_config32(ctrl->f2, DRAM_CTRL, dword); + } + } + + //Clear wrap32dis + + clear_wrap32dis(); + + //restore SSE2 setting + disable_sse2(); + +#if MEM_TRAIN_SEQ != 1 + /* We need tidy output for type 1 */ + #if CONFIG_USE_INIT == 1 + printk_debug(" CTLRMaxDelay=%02x", CTLRMaxDelay); + #else + print_debug(" CTLRMaxDelay="); print_debug_hex8(CTLRMaxDelay); + #endif +#endif + + if(CTLRMaxDelay==0xae) { + soft_reset(); // try more or downgrade? + } + +} + +#define DQS_READDIR 1 +#define DQS_WRITEDIR 0 + + +static void SetDQSDelayCSR(const struct mem_controller *ctrl, unsigned channel, unsigned bytelane, unsigned direction, unsigned dqs_delay) +{ //ByteLane could be 0-8, last is for ECC + unsigned index; + uint32_t dword; + unsigned shift; + + dqs_delay &= 0xff; + + index = (bytelane>>2) + 1 + channel * 0x20 + (direction << 2); + shift = bytelane; + while(shift>3) { + shift-=4; + } + shift <<= 3; // 8 bit + + dword = pci_read_config32_index_wait(ctrl->f2, 0x98, index); + dword &= ~(0x3f<<shift); + dword |= (dqs_delay<<shift); + pci_write_config32_index_wait(ctrl->f2, 0x98, index, dword); + +} + +static void SetDQSDelayAllCSR(const struct mem_controller *ctrl, unsigned channel, unsigned direction, unsigned dqs_delay) +{ + unsigned index; + uint32_t dword; + int i; + + dword = 0; + dqs_delay &= 0xff; + for(i=0;i<4;i++) { + dword |= dqs_delay<<(i*8); + } + + index = 1 + channel * 0x20 + direction * 4; + + for(i=0; i<2; i++) { + pci_write_config32_index_wait(ctrl->f2, 0x98, index + i, dword); + } + +} + +static unsigned MiddleDQS(unsigned min_d, unsigned max_d) +{ + unsigned size_d; + size_d = max_d-min_d; + if(size_d & 1) { //need round up + min_d++; + } + return ( min_d + (size_d>>1)); +} + +static inline void save_dqs_delay(unsigned channel, unsigned bytelane, unsigned direction, uint8_t *dqs_delay_a, uint8_t dqs_delay) +{ + dqs_delay_a[channel * 2*9 + direction * 9 + bytelane] = dqs_delay; +} + +static void WriteDQSTestPattern(unsigned addr_lo, unsigned pattern , uint8_t *buf_a) +{ + WriteLNTestPattern(addr_lo, buf_a, (pattern+1) * 9); +} + +static void ReadL18TestPattern(unsigned addr_lo) +{ + //set fs and use fs prefix to access the mem + __asm__ volatile ( + "movl %%fs:-128(%%esi), %%eax\n\t" //TestAddr cache line + "movl %%fs:-64(%%esi), %%eax\n\t" //+1 + "movl %%fs:(%%esi), %%eax\n\t" //+2 + "movl %%fs:64(%%esi), %%eax\n\t" //+3 + + "movl %%fs:-128(%%edi), %%eax\n\t" //+4 + "movl %%fs:-64(%%edi), %%eax\n\t" //+5 + "movl %%fs:(%%edi), %%eax\n\t" //+6 + "movl %%fs:64(%%edi), %%eax\n\t" //+7 + + "movl %%fs:-128(%%ebx), %%eax\n\t" //+8 + "movl %%fs:-64(%%ebx), %%eax\n\t" //+9 + "movl %%fs:(%%ebx), %%eax\n\t" //+10 + "movl %%fs:64(%%ebx), %%eax\n\t" //+11 + + "movl %%fs:-128(%%ecx), %%eax\n\t" //+12 + "movl %%fs:-64(%%ecx), %%eax\n\t" //+13 + "movl %%fs:(%%ecx), %%eax\n\t" //+14 + "movl %%fs:64(%%ecx), %%eax\n\t" //+15 + + "movl %%fs:-128(%%edx), %%eax\n\t" //+16 + "movl %%fs:-64(%%edx), %%eax\n\t" //+17 + + :: "a"(0), "b" (addr_lo+128+8*64), "c" (addr_lo+128+12*64), "d" (addr_lo +128+16*64), "S"(addr_lo+128), "D"(addr_lo+128+4*64) + ); + +} + +static void ReadL9TestPattern(unsigned addr_lo) +{ + + //set fs and use fs prefix to access the mem + __asm__ volatile ( + + "movl %%fs:-128(%%ecx), %%eax\n\t" //TestAddr cache line + "movl %%fs:-64(%%ecx), %%eax\n\t" //+1 + "movl %%fs:(%%ecx), %%eax\n\t" //+2 + "movl %%fs:64(%%ecx), %%eax\n\t" //+3 + + "movl %%fs:-128(%%edx), %%eax\n\t" //+4 + "movl %%fs:-64(%%edx), %%eax\n\t" //+5 + "movl %%fs:(%%edx), %%eax\n\t" //+6 + "movl %%fs:64(%%edx), %%eax\n\t" //+7 + + "movl %%fs:-128(%%ebx), %%eax\n\t" //+8 + + :: "a"(0), "b" (addr_lo+128+8*64), "c"(addr_lo+128), "d"(addr_lo+128+4*64) + ); + +} + + +static void ReadDQSTestPattern(unsigned addr_lo, unsigned pattern) +{ + if(pattern == 0) { + ReadL9TestPattern(addr_lo); + } + else { + ReadL18TestPattern(addr_lo); + } +} + +static void FlushDQSTestPattern_L9(unsigned addr_lo) +{ + __asm__ volatile ( + "clflush %%fs:-128(%%ecx)\n\t" + "clflush %%fs:-64(%%ecx)\n\t" + "clflush %%fs:(%%ecx)\n\t" + "clflush %%fs:64(%%ecx)\n\t" + + "clflush %%fs:-128(%%eax)\n\t" + "clflush %%fs:-64(%%eax)\n\t" + "clflush %%fs:(%%eax)\n\t" + "clflush %%fs:64(%%eax)\n\t" + + "clflush %%fs:-128(%%ebx)\n\t" + + :: "b" (addr_lo+128+8*64), "c"(addr_lo+128), "a"(addr_lo+128+4*64) + ); + +} +static __attribute__((noinline)) void FlushDQSTestPattern_L18(unsigned addr_lo) +{ + __asm__ volatile ( + "clflush %%fs:-128(%%eax)\n\t" + "clflush %%fs:-64(%%eax)\n\t" + "clflush %%fs:(%%eax)\n\t" + "clflush %%fs:64(%%eax)\n\t" + + "clflush %%fs:-128(%%edi)\n\t" + "clflush %%fs:-64(%%edi)\n\t" + "clflush %%fs:(%%edi)\n\t" + "clflush %%fs:64(%%edi)\n\t" + + "clflush %%fs:-128(%%ebx)\n\t" + "clflush %%fs:-64(%%ebx)\n\t" + "clflush %%fs:(%%ebx)\n\t" + "clflush %%fs:64(%%ebx)\n\t" + + "clflush %%fs:-128(%%ecx)\n\t" + "clflush %%fs:-64(%%ecx)\n\t" + "clflush %%fs:(%%ecx)\n\t" + "clflush %%fs:64(%%ecx)\n\t" + + "clflush %%fs:-128(%%edx)\n\t" + "clflush %%fs:-64(%%edx)\n\t" + + :: "b" (addr_lo+128+8*64), "c" (addr_lo+128+12*64), "d" (addr_lo +128+16*64), "a"(addr_lo+128), "D"(addr_lo+128+4*64) + ); +} + +static void FlushDQSTestPattern(unsigned addr_lo, unsigned pattern ) +{ + + if(pattern == 0){ + FlushDQSTestPattern_L9(addr_lo); + } + else { + FlushDQSTestPattern_L18(addr_lo); + } +} + +static unsigned CompareDQSTestPattern(unsigned channel, unsigned addr_lo, unsigned pattern, uint8_t *buf_a) +{ + uint32_t *test_buf; + unsigned bitmap = 0xff; + unsigned bytelane; + int i; + uint32_t value; + int j; + uint32_t value_test; + + test_buf = (uint32_t *)buf_a; + + + if(pattern && channel) { + addr_lo += 8; //second channel + test_buf+= 2; + } + + bytelane = 0; + for(i=0;i<9*64/4;i++) { + __asm__ volatile ( + "movl %%fs:(%1), %0\n\t" + :"=b"(value): "a" (addr_lo) + ); + value_test = *test_buf; + + print_debug_dqs_pair("\t\t\t\t\t\ttest_buf= ", (unsigned)test_buf, " value = ", value_test, 7); + print_debug_dqs_pair("\t\t\t\t\t\ttaddr_lo = ",addr_lo, " value = ", value, 7); + + for(j=0;j<4*8;j+=8) { + if(((value>>j)&0xff) != ((value_test>>j)& 0xff)) { + bitmap &= ~(1<<bytelane); + } + + bytelane++; + bytelane &= 0x7; + } + print_debug_dqs("\t\t\t\t\t\tbitmap = ", bitmap, 7); + + if(bytelane == 0) { + if(pattern == 1) { //dual channel + addr_lo += 8; //skip over other channel's data + test_buf += 2; + } + } + addr_lo += 4; + test_buf +=1; + + } + + + return bitmap; + +} + +static unsigned TrainDQSPos(const struct mem_controller *ctrl, unsigned channel, unsigned Direction, unsigned Pattern, uint8_t *buf_a, uint8_t *dqs_delay_a, struct sys_info *sysinfo) +{ + unsigned ByteLane; + unsigned Errors; + unsigned BanksPresent; + + unsigned MutualCSPassW[48]; + + unsigned ChipSel; + unsigned DQSDelay; + + unsigned TestAddr; + + unsigned LastTest; + unsigned RnkDlyFilterMax, RnkDlyFilterMin; + unsigned RnkDlySeqPassMax, RnkDlySeqPassMin; + + Errors = 0; + BanksPresent = 0; + + print_debug_dqs("\t\t\tTrainDQSPos begin ", 0, 3); + + print_debug_addr("TrainDQSPos: MutualCSPassW[48] :", MutualCSPassW); + + for(DQSDelay=0; DQSDelay<48; DQSDelay++) { + MutualCSPassW[DQSDelay] = 0xff; // Bitmapped status per delay setting, 0xff=All positions passing (1= PASS) + } + + for(ChipSel = 0; ChipSel < 8; ChipSel++) { //logical register chipselects 0..7 + print_debug_dqs("\t\t\t\tTrainDQSPos: 11 ChipSel ", ChipSel, 4); + if(!ChipSelPresent(ctrl, ChipSel, sysinfo)) continue; + BanksPresent = 1; + + TestAddr = Get_MCTSysAddr(ctrl, ChipSel, sysinfo); + + print_debug_dqs("\t\t\t\tTrainDQSPos: 12 TestAddr ", TestAddr, 4); + + //set fs and use fs prefix to access the mem + set_FSBASE(TestAddr>>24); + + if(Direction == DQS_READDIR) { + print_debug_dqs("\t\t\t\tTrainDQSPos: 13 for read so write at first", 0, 4); + WriteDQSTestPattern(TestAddr<<8, Pattern, buf_a); + } + + for(DQSDelay = 0; DQSDelay < 48; DQSDelay++ ){ + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 141 DQSDelay ", DQSDelay, 5); + if(MutualCSPassW[DQSDelay] == 0) continue; //skip current delay value if other chipselects have failed all 8 bytelanes + SetDQSDelayAllCSR(ctrl, channel, Direction, DQSDelay); + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 142 MutualCSPassW ", MutualCSPassW[DQSDelay], 5); + if(Direction == DQS_WRITEDIR) { + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 143 for write", 0, 5); + WriteDQSTestPattern(TestAddr<<8, Pattern, buf_a); + } + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 144 Pattern ", Pattern, 5); + ReadDQSTestPattern(TestAddr<<8, Pattern); + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 145 MutualCSPassW ", MutualCSPassW[DQSDelay], 5); + MutualCSPassW[DQSDelay] &= CompareDQSTestPattern(channel, TestAddr<<8, Pattern, buf_a); //0: fail, 1=pass + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 146 MutualCSPassW ", MutualCSPassW[DQSDelay], 5); + SetTargetWTIO(TestAddr); + FlushDQSTestPattern(TestAddr<<8, Pattern); + ResetTargetWTIO(); + } + } + + if(BanksPresent) + for(ByteLane = 0; ByteLane < 8; ByteLane++) { + print_debug_dqs("\t\t\t\tTrainDQSPos: 31 ByteLane ",ByteLane, 4); + + LastTest = DQS_FAIL; + RnkDlySeqPassMax = 0; + RnkDlyFilterMax = 0; + RnkDlyFilterMin = 0; + for(DQSDelay=0; DQSDelay<48; DQSDelay++) { + if(MutualCSPassW[DQSDelay] & (1<<ByteLane)) { + + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 321 DQSDelay ", DQSDelay, 5); + print_debug_dqs("\t\t\t\t\tTrainDQSPos: 322 MutualCSPassW ", MutualCSPassW[DQSDelay], 5); + + RnkDlySeqPassMax = DQSDelay; + if(LastTest == DQS_FAIL) { + RnkDlySeqPassMin = DQSDelay; //start sequential run + } + if((RnkDlySeqPassMax - RnkDlySeqPassMin)>(RnkDlyFilterMax-RnkDlyFilterMin)){ + RnkDlyFilterMin = RnkDlySeqPassMin; + RnkDlyFilterMax = RnkDlySeqPassMax; + } + LastTest = DQS_PASS; + } + else { + LastTest = DQS_FAIL; + } + } + print_debug_dqs("\t\t\t\tTrainDQSPos: 33 RnkDlySeqPassMax ", RnkDlySeqPassMax, 4); + + if(RnkDlySeqPassMax == 0) { + Errors |= SB_NODQSPOS; // no passing window + } + else { + print_debug_dqs("\t\t\t\tTrainDQSPos: 34 RnkDlyFilterMax ", RnkDlyFilterMax, 4); + print_debug_dqs("\t\t\t\tTrainDQSPos: 34 RnkDlyFilterMin ", RnkDlyFilterMin, 4); + if((RnkDlyFilterMax - RnkDlyFilterMin)< MIN_DQS_WNDW){ + Errors |= SB_SMALLDQS; + } + else { + unsigned middle_dqs; + middle_dqs = MiddleDQS(RnkDlyFilterMin, RnkDlyFilterMax); + print_debug_dqs("\t\t\t\tTrainDQSPos: 35 middle_dqs ",middle_dqs, 4); + SetDQSDelayCSR(ctrl, channel, ByteLane, Direction, middle_dqs); + save_dqs_delay(channel, ByteLane, Direction, dqs_delay_a, middle_dqs); + } + } + + } + + print_debug_dqs("\t\t\tTrainDQSPos: end", 0xff, 3); + + return Errors; + + +} + +static unsigned TrainReadDQS(const struct mem_controller *ctrl, unsigned channel, unsigned pattern, uint8_t *buf_a, uint8_t *dqs_delay_a, struct sys_info *sysinfo) +{ + print_debug_dqs("\t\tTrainReadPos", 0, 2); + return TrainDQSPos(ctrl, channel, DQS_READDIR, pattern, buf_a, dqs_delay_a, sysinfo); +} + +static unsigned TrainWriteDQS(const struct mem_controller *ctrl, unsigned channel, unsigned pattern, uint8_t *buf_a, uint8_t *dqs_delay_a, struct sys_info *sysinfo) +{ + print_debug_dqs("\t\tTrainWritePos", 0, 2); + return TrainDQSPos(ctrl, channel, DQS_WRITEDIR, pattern, buf_a, dqs_delay_a, sysinfo); +} + + + +static unsigned TrainDQSRdWrPos(const struct mem_controller *ctrl, struct sys_info *sysinfo) +{ + const static uint32_t TestPatternJD1a[] = { + 0x00000000,0x00000000,0xFFFFFFFF,0xFFFFFFFF, // QW0-1, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW2-3, ALL-EVEN + 0x00000000,0x00000000,0xFFFFFFFF,0xFFFFFFFF, // QW4-5, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW6-7, ALL-EVEN + 0xFeFeFeFe,0xFeFeFeFe,0x01010101,0x01010101, // QW0-1, DQ0-ODD + 0xFeFeFeFe,0xFeFeFeFe,0x01010101,0x01010101, // QW2-3, DQ0-ODD + 0x01010101,0x01010101,0xFeFeFeFe,0xFeFeFeFe, // QW4-5, DQ0-ODD + 0xFeFeFeFe,0xFeFeFeFe,0x01010101,0x01010101, // QW6-7, DQ0-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW0-1, DQ1-ODD + 0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd, // QW2-3, DQ1-ODD + 0xFdFdFdFd,0xFdFdFdFd,0x02020202,0x02020202, // QW4-5, DQ1-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW6-7, DQ1-ODD + 0x04040404,0x04040404,0xfBfBfBfB,0xfBfBfBfB, // QW0-1, DQ2-ODD + 0x04040404,0x04040404,0x04040404,0x04040404, // QW2-3, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW4-5, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW6-7, DQ2-ODD + 0x08080808,0x08080808,0xF7F7F7F7,0xF7F7F7F7, // QW0-1, DQ3-ODD + 0x08080808,0x08080808,0x08080808,0x08080808, // QW2-3, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0x08080808,0x08080808, // QW4-5, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7, // QW6-7, DQ3-ODD + 0x10101010,0x10101010,0x10101010,0x10101010, // QW0-1, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0x10101010,0x10101010, // QW2-3, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF, // QW4-5, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0x10101010,0x10101010, // QW6-7, DQ4-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW0-1, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0x20202020,0x20202020, // QW2-3, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW4-5, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW6-7, DQ5-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW0-1, DQ6-ODD + 0x40404040,0x40404040,0xBfBfBfBf,0xBfBfBfBf, // QW2-3, DQ6-ODD + 0x40404040,0x40404040,0xBfBfBfBf,0xBfBfBfBf, // QW4-5, DQ6-ODD + 0x40404040,0x40404040,0xBfBfBfBf,0xBfBfBfBf, // QW6-7, DQ6-ODD + 0x80808080,0x80808080,0x7F7F7F7F,0x7F7F7F7F, // QW0-1, DQ7-ODD + 0x80808080,0x80808080,0x7F7F7F7F,0x7F7F7F7F, // QW2-3, DQ7-ODD + 0x80808080,0x80808080,0x7F7F7F7F,0x7F7F7F7F, // QW4-5, DQ7-ODD + 0x80808080,0x80808080,0x80808080,0x80808080 // QW6-7, DQ7-ODD + }; + const static uint32_t TestPatternJD1b[] = { + 0x00000000,0x00000000,0x00000000,0x00000000, // QW0,CHA-B, ALL-EVEN + 0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF, // QW1,CHA-B, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW2,CHA-B, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW3,CHA-B, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW4,CHA-B, ALL-EVEN + 0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF,0xFFFFFFFF, // QW5,CHA-B, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW6,CHA-B, ALL-EVEN + 0x00000000,0x00000000,0x00000000,0x00000000, // QW7,CHA-B, ALL-EVEN + 0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe, // QW0,CHA-B, DQ0-ODD + 0x01010101,0x01010101,0x01010101,0x01010101, // QW1,CHA-B, DQ0-ODD + 0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe, // QW2,CHA-B, DQ0-ODD + 0x01010101,0x01010101,0x01010101,0x01010101, // QW3,CHA-B, DQ0-ODD + 0x01010101,0x01010101,0x01010101,0x01010101, // QW4,CHA-B, DQ0-ODD + 0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe, // QW5,CHA-B, DQ0-ODD + 0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe,0xFeFeFeFe, // QW6,CHA-B, DQ0-ODD + 0x01010101,0x01010101,0x01010101,0x01010101, // QW7,CHA-B, DQ0-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW0,CHA-B, DQ1-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW1,CHA-B, DQ1-ODD + 0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd, // QW2,CHA-B, DQ1-ODD + 0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd, // QW3,CHA-B, DQ1-ODD + 0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd,0xFdFdFdFd, // QW4,CHA-B, DQ1-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW5,CHA-B, DQ1-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW6,CHA-B, DQ1-ODD + 0x02020202,0x02020202,0x02020202,0x02020202, // QW7,CHA-B, DQ1-ODD + 0x04040404,0x04040404,0x04040404,0x04040404, // QW0,CHA-B, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW1,CHA-B, DQ2-ODD + 0x04040404,0x04040404,0x04040404,0x04040404, // QW2,CHA-B, DQ2-ODD + 0x04040404,0x04040404,0x04040404,0x04040404, // QW3,CHA-B, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW4,CHA-B, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW5,CHA-B, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW6,CHA-B, DQ2-ODD + 0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB,0xfBfBfBfB, // QW7,CHA-B, DQ2-ODD + 0x08080808,0x08080808,0x08080808,0x08080808, // QW0,CHA-B, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7, // QW1,CHA-B, DQ3-ODD + 0x08080808,0x08080808,0x08080808,0x08080808, // QW2,CHA-B, DQ3-ODD + 0x08080808,0x08080808,0x08080808,0x08080808, // QW3,CHA-B, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7, // QW4,CHA-B, DQ3-ODD + 0x08080808,0x08080808,0x08080808,0x08080808, // QW5,CHA-B, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7, // QW6,CHA-B, DQ3-ODD + 0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7,0xF7F7F7F7, // QW7,CHA-B, DQ3-ODD + 0x10101010,0x10101010,0x10101010,0x10101010, // QW0,CHA-B, DQ4-ODD + 0x10101010,0x10101010,0x10101010,0x10101010, // QW1,CHA-B, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF, // QW2,CHA-B, DQ4-ODD + 0x10101010,0x10101010,0x10101010,0x10101010, // QW3,CHA-B, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF, // QW4,CHA-B, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF, // QW5,CHA-B, DQ4-ODD + 0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF,0xeFeFeFeF, // QW6,CHA-B, DQ4-ODD + 0x10101010,0x10101010,0x10101010,0x10101010, // QW7,CHA-B, DQ4-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW0,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW1,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW2,CHA-B, DQ5-ODD + 0x20202020,0x20202020,0x20202020,0x20202020, // QW3,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW4,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW5,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW6,CHA-B, DQ5-ODD + 0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF,0xdFdFdFdF, // QW7,CHA-B, DQ5-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW0,CHA-B, DQ6-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW1,CHA-B, DQ6-ODD + 0x40404040,0x40404040,0x40404040,0x40404040, // QW2,CHA-B, DQ6-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW3,CHA-B, DQ6-ODD + 0x40404040,0x40404040,0x40404040,0x40404040, // QW4,CHA-B, DQ6-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW5,CHA-B, DQ6-ODD + 0x40404040,0x40404040,0x40404040,0x40404040, // QW6,CHA-B, DQ6-ODD + 0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf,0xBfBfBfBf, // QW7,CHA-B, DQ6-ODD + 0x80808080,0x80808080,0x80808080,0x80808080, // QW0,CHA-B, DQ7-ODD + 0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F, // QW1,CHA-B, DQ7-ODD + 0x80808080,0x80808080,0x80808080,0x80808080, // QW2,CHA-B, DQ7-ODD + 0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F, // QW3,CHA-B, DQ7-ODD + 0x80808080,0x80808080,0x80808080,0x80808080, // QW4,CHA-B, DQ7-ODD + 0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F,0x7F7F7F7F, // QW5,CHA-B, DQ7-ODD + 0x80808080,0x80808080,0x80808080,0x80808080, // QW6,CHA-B, DQ7-ODD + 0x80808080,0x80808080,0x80808080,0x80808080 // QW7,CHA-B, DQ7-ODD + }; + uint8_t pattern_buf_x[64 * 18 + 16]; // We need to two cache line So have more 16 bytes to keep 16 byte alignment */ + uint8_t *buf_a; + + unsigned pattern; + uint32_t dword; + uint32_t ecc_bit; + unsigned Errors; + unsigned channel; + int i; + unsigned DQSWrDelay; + unsigned is_Width128 = sysinfo->meminfo[ctrl->node_id].is_Width128; + uint8_t *dqs_delay_a = &sysinfo->dqs_delay_a[ctrl->node_id * 2*2*9]; //channel 2, direction 2 , bytelane *9 + + //enable SSE2 + enable_sse2(); + + //wrap32dis + set_wrap32dis(); + + //disable ECC temp + dword = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW); + ecc_bit = dword & DCL_DimmEccEn; + dword &= ~(DCL_DimmEccEn); + pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dword); + + //SetupDqsPattern + buf_a = (uint8_t *)(((uint32_t)(&pattern_buf_x[0]) + 0x10) & (~0xf)); + + if(is_Width128){ + pattern = 1; + for(i=0;i<16*18;i++) { + *((uint32_t *)(buf_a + i*4)) = TestPatternJD1b[i]; + } + } + else { + pattern = 0; + for(i=0; i<16*9;i++) { + *((uint32_t *)(buf_a + i*4)) = TestPatternJD1a[i]; + } + + } + + print_debug_dqs("\r\nTrainDQSRdWrPos: 0 ctrl ", ctrl->node_id, 0); + + print_debug_addr("TrainDQSRdWrPos: buf_a:", buf_a); + + Errors = 0; + + channel = 0; + while( (channel<2) && (!Errors)) { + print_debug_dqs("\tTrainDQSRdWrPos: 1 channel ",channel, 1); + for(DQSWrDelay = 0; DQSWrDelay < 48; DQSWrDelay++) { + unsigned err; + SetDQSDelayAllCSR(ctrl, channel, DQS_WRITEDIR, DQSWrDelay); + print_debug_dqs("\t\tTrainDQSRdWrPos: 21 DQSWrDelay ", DQSWrDelay, 2); + err= TrainReadDQS(ctrl, channel, pattern, buf_a, dqs_delay_a, sysinfo); + print_debug_dqs("\t\tTrainDQSRdWrPos: 22 err ",err, 2); + if(err == 0) break; + Errors |= err; + } + + print_debug_dqs("\tTrainDQSRdWrPos: 3 DQSWrDelay ", DQSWrDelay, 1); + + if(DQSWrDelay < 48) { + Errors = TrainWriteDQS(ctrl, channel, pattern, buf_a, dqs_delay_a, sysinfo); + print_debug_dqs("\tTrainDQSRdWrPos: 4 Errors ", Errors, 1); + + } + channel++; + if(!is_Width128){ + channel++; // skip channel if 64-bit mode + } + } + + //Enable ECC again + dword = pci_read_config32(ctrl->f2, DRAM_CONFIG_LOW); + dword &= ~(DCL_DimmEccEn); + dword |= ecc_bit; + pci_write_config32(ctrl->f2, DRAM_CONFIG_LOW, dword); + + //Clear wrap32dis + + clear_wrap32dis(); + + //restore SSE2 setting + disable_sse2(); + + print_debug_dqs("TrainDQSRdWrPos: ", 5, 0); + + return Errors; + +} +static inline uint8_t get_dqs_delay(unsigned channel, unsigned bytelane, unsigned direction, uint8_t *dqs_delay_a) +{ + return dqs_delay_a[channel * 2*9 + direction * 9 + bytelane]; +} + +static unsigned CalcEccDQSPos(unsigned channel,unsigned ByteLane0, unsigned ByteLane1, unsigned InterFactor, unsigned Direction, uint8_t *dqs_delay_a) +/* InterFactor: 0: 100% ByteLane 0 + 0x80: 50% between ByteLane 0 and 1 + 0xff: 99.6% ByteLane 1 and 0.4% like 0 +*/ +{ + unsigned DQSDelay0, DQSDelay1; + unsigned DQSDelay; + + DQSDelay0 = get_dqs_delay(channel, ByteLane0, Direction, dqs_delay_a); + DQSDelay1 = get_dqs_delay(channel, ByteLane1, Direction, dqs_delay_a); + + if(DQSDelay0>DQSDelay1) { + DQSDelay = DQSDelay0 - DQSDelay1; + InterFactor = 0xff - InterFactor; + } + else { + DQSDelay = DQSDelay1 - DQSDelay0; + } + + DQSDelay *= InterFactor; + + DQSDelay >>= 8; // /255 + + if(DQSDelay0>DQSDelay1) { + DQSDelay += DQSDelay1; + } + else { + DQSDelay += DQSDelay0; + } + + return DQSDelay; + +} + +static void SetEccDQSRdWrPos(const struct mem_controller *ctrl, struct sys_info *sysinfo) +{ + unsigned channel; + unsigned ByteLane; + unsigned Direction; + unsigned lane0, lane1, ratio; + unsigned dqs_delay; + + unsigned direction[] = { DQS_READDIR, DQS_WRITEDIR }; + int i; + uint8_t *dqs_delay_a = &sysinfo->dqs_delay_a[ctrl->node_id * 2*2*9]; //channel 2, direction 2 , bytelane *9 + + ByteLane = 8; + + for(channel = 0; channel < 2; channel++) { + for(i=0;i<2;i++) { + Direction = direction[i]; + lane0 = 4; lane1 = 5; ratio = 0; + dqs_delay = CalcEccDQSPos(channel, lane0, lane1, ratio, Direction, dqs_delay_a); + print_debug_dqs_pair("\t\tSetEccDQSRdWrPos: channel ", channel, Direction==DQS_READDIR? " R dqs_delay":" W dqs_delay", dqs_delay, 2); + SetDQSDelayCSR(ctrl, channel, ByteLane, Direction, dqs_delay); + save_dqs_delay(channel, ByteLane, Direction, dqs_delay_a, dqs_delay); + } + } +} + +static void train_DqsRcvrEn(const struct mem_controller *ctrl, unsigned Pass, struct sys_info *sysinfo) +{ + print_debug_dqs("\r\ntrain_DqsRcvrEn: begin ctrl ", ctrl->node_id, 0); + TrainRcvrEn(ctrl, Pass, sysinfo); + print_debug_dqs("\r\ntrain_DqsRcvrEn: end ctrl ", ctrl->node_id, 0); + +} +static void train_DqsPos(const struct mem_controller *ctrl, struct sys_info *sysinfo) +{ + print_debug_dqs("\r\ntrain_DqsPos: begin ctrl ", ctrl->node_id, 0); + if(TrainDQSRdWrPos(ctrl, sysinfo) != 0) { + print_err("\r\nDQS Training Rd Wr failed ctrl"); print_err_hex8(ctrl->node_id); print_err("\r\n"); + soft_reset(); + } + else { + SetEccDQSRdWrPos(ctrl, sysinfo); + } + print_debug_dqs("\r\ntrain_DqsPos: end ctrl ", ctrl->node_id, 0); + +} + +#if K8_REV_F_SUPPORT_F0_F1_WORKAROUND == 1 +static void f0_svm_workaround(int controllers, const struct mem_controller *ctrl, tsc_t *tsc0, struct sys_info *sysinfo) +{ + tsc_t tsc1[8]; + unsigned cpu_f0_f1[8]; + int i; + + print_debug_addr("dqs_timing: tsc1[8] :", tsc1); + + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[i]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + uint32_t dword; + + cpu_f0_f1[i] = is_cpu_pre_f2_in_bsp(i); + + if(!cpu_f0_f1[i]) continue; + + dword = pci_read_config32(ctrl[i].f2, DRAM_CTRL); + dword &= ~DC_DqsRcvEnTrain; + pci_write_config32(ctrl[i].f2, DRAM_CTRL, dword); + + dword = pci_read_config32(ctrl[i].f2, DRAM_INIT); + dword |= DI_EnDramInit; + pci_write_config32(ctrl[i].f2, DRAM_INIT, dword); + dword &= ~DI_EnDramInit; + pci_write_config32(ctrl[i].f2, DRAM_INIT, dword); + + tsc1[i] = rdtsc(); + print_debug_dqs_tsc("begin: tsc1", i, tsc1[i].hi, tsc1[i].lo, 2); + + dword = tsc1[i].lo + tsc0[i].lo; + if((dword<tsc1[i].lo) || (dword<tsc0[i].lo)) { + tsc1[i].hi++; + } + tsc1[i].lo = dword; + tsc1[i].hi+= tsc0[i].hi; + + print_debug_dqs_tsc("end : tsc1", i, tsc1[i].hi, tsc1[i].lo, 2); + + } + + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[i]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + if(!cpu_f0_f1[i]) continue; + + tsc_t tsc; + + do { + tsc = rdtsc(); + } while ((tsc1[i].hi>tsc.hi) || ((tsc1[i].hi==tsc.hi) && (tsc1[i].lo>tsc.lo))); + + print_debug_dqs_tsc("end : tsc ", i, tsc.hi, tsc.lo, 2); + } + +} + +#endif + + +/* setting variable mtrr, comes from linux kernel source */ +static void set_var_mtrr_dqs( + unsigned int reg, unsigned long basek, unsigned long sizek, + unsigned char type, unsigned address_bits) +{ + msr_t base, mask; + unsigned address_mask_high; + + address_mask_high = ((1u << (address_bits - 32u)) - 1u); + + base.hi = basek >> 22; + base.lo = basek << 10; + + if (sizek < 4*1024*1024) { + mask.hi = address_mask_high; + mask.lo = ~((sizek << 10) -1); + } + else { + mask.hi = address_mask_high & (~((sizek >> 22) -1)); + mask.lo = 0; + } + + if (reg >= 8) + return; + + if (sizek == 0) { + msr_t zero; + zero.lo = zero.hi = 0; + /* The invalid bit is kept in the mask, so we simply clear the + relevant mask register to disable a range. */ + wrmsr (MTRRphysMask_MSR(reg), zero); + } else { + /* Bit 32-35 of MTRRphysMask should be set to 1 */ + base.lo |= type; + mask.lo |= 0x800; + wrmsr (MTRRphysBase_MSR(reg), base); + wrmsr (MTRRphysMask_MSR(reg), mask); + } +} + + +/* fms: find most sigificant bit set, stolen from Linux Kernel Source. */ +static inline unsigned int fms(unsigned int x) +{ + int r; + + __asm__("bsrl %1,%0\n\t" + "jnz 1f\n\t" + "movl $0,%0\n" + "1:" : "=r" (r) : "g" (x)); + return r; +} + +/* fms: find least sigificant bit set */ +static inline unsigned int fls(unsigned int x) +{ + int r; + + __asm__("bsfl %1,%0\n\t" + "jnz 1f\n\t" + "movl $32,%0\n" + "1:" : "=r" (r) : "g" (x)); + return r; +} + +static unsigned int range_to_mtrr(unsigned int reg, + unsigned long range_startk, unsigned long range_sizek, + unsigned long next_range_startk, unsigned char type, unsigned address_bits) +{ + if (!range_sizek || (reg >= 8)) { + return reg; + } + while(range_sizek) { + unsigned long max_align, align; + unsigned long sizek; + /* Compute the maximum size I can make a range */ + max_align = fls(range_startk); + align = fms(range_sizek); + if (align > max_align) { + align = max_align; + } + sizek = 1 << align; +#if MEM_TRAIN_SEQ != 1 + #if CONFIG_USE_INIT == 1 + printk_debug("Setting variable MTRR %d, base: %4dMB, range: %4dMB, type %s\r\n", + reg, range_startk >>10, sizek >> 10, + (type==MTRR_TYPE_UNCACHEABLE)?"UC": + ((type==MTRR_TYPE_WRBACK)?"WB":"Other") + ); + #else + print_debug("Setting variable MTRR "); print_debug_hex8(reg); print_debug(", base: "); print_debug_hex16(range_startk>>10); + print_debug("MB, range: "); print_debug_hex16(sizek >> 10); print_debug("MB, type "); + print_debug( (type==MTRR_TYPE_UNCACHEABLE)?"UC\r\n": + ((type==MTRR_TYPE_WRBACK)?"WB\r\n":"Other\r\n") + ); + #endif +#endif + set_var_mtrr_dqs(reg++, range_startk, sizek, type, address_bits); + range_startk += sizek; + range_sizek -= sizek; + if (reg >= 8) + break; + } + return reg; +} + +static void set_top_mem_ap(unsigned tom_k, unsigned tom2_k) +{ + msr_t msr; + + /* Now set top of memory */ + msr.lo = (tom2_k & 0x003fffff) << 10; + msr.hi = (tom2_k & 0xffc00000) >> 22; + wrmsr(TOP_MEM2, msr); + + msr.lo = (tom_k & 0x003fffff) << 10; + msr.hi = (tom_k & 0xffc00000) >> 22; + wrmsr(TOP_MEM, msr); +} + +static void setup_mtrr_dqs(unsigned tom_k, unsigned tom2_k){ + unsigned reg; + msr_t msr; + +#if 0 + //still enable from cache_as_ram.inc + msr = rdmsr(SYSCFG_MSR); + msr.lo |= SYSCFG_MSR_MtrrFixDramModEn; + wrmsr(SYSCFG_MSR,msr); +#endif + + //[0,512k), [512k, 640k) + msr.hi = 0x1e1e1e1e; + msr.lo = msr.hi; + wrmsr(0x250, msr); + wrmsr(0x258, msr); + + //[1M, TOM) + reg = range_to_mtrr(2, 0, tom_k,4*1024*1024, MTRR_TYPE_WRBACK, 40); + + //[4G, TOM2) + if(tom2_k) { + //enable tom2 and type + msr = rdmsr(SYSCFG_MSR); + msr.lo |= (1<<21) | (1<<22); //MtrrTom2En and Tom2ForceMemTypeWB + wrmsr(SYSCFG_MSR, msr); + } + +} + +static void clear_mtrr_dqs(unsigned tom2_k){ + msr_t msr; + unsigned i; + + //still enable from cache_as_ram.inc + msr = rdmsr(SYSCFG_MSR); + msr.lo |= SYSCFG_MSR_MtrrFixDramModEn; + wrmsr(SYSCFG_MSR,msr); + + //[0,512k), [512k, 640k) + msr.hi = 0; + msr.lo = msr.hi; + wrmsr(0x250, msr); + wrmsr(0x258, msr); + + //[1M, TOM) + for(i=0x204;i<0x210;i++) { + wrmsr(i, msr); + } + + //[4G, TOM2) + if(tom2_k) { + //enable tom2 and type + msr = rdmsr(SYSCFG_MSR); + msr.lo &= ~((1<<21) | (1<<22)); //MtrrTom2En and Tom2ForceMemTypeWB + wrmsr(SYSCFG_MSR, msr); + } +} + +static void set_htic_bit(unsigned i, unsigned val, unsigned bit) +{ + uint32_t dword; + dword = pci_read_config32(PCI_DEV(0, 0x18+i, 0), HT_INIT_CONTROL); + dword &= ~(1<<bit); + dword |= ((val & 1) <<bit); + pci_write_config32(PCI_DEV(0, 0x18+i, 0), HT_INIT_CONTROL, dword); +} + + +static unsigned get_htic_bit(unsigned i, unsigned bit) +{ + uint32_t dword; + dword = pci_read_config32(PCI_DEV(0, 0x18+i, 0), HT_INIT_CONTROL); + dword &= (1<<bit); + return dword; +} + +static void wait_till_sysinfo_in_ram(void) +{ + while(1) { + if(get_htic_bit(0, 9)) return; + } +} + +static void set_sysinfo_in_ram(unsigned val) +{ + set_htic_bit(0, val, 9); +} + + +#if MEM_TRAIN_SEQ == 0 + + +#if K8_REV_F_SUPPORT_F0_F1_WORKAROUND == 1 +static void dqs_timing(int controllers, const struct mem_controller *ctrl, tsc_t *tsc0, struct sys_info *sysinfo) +#else +static void dqs_timing(int controllers, const struct mem_controller *ctrl, struct sys_info *sysinfo) +#endif +{ + int i; + + tsc_t tsc[5]; + + //need to enable mtrr, so dqs training could access the test address + setup_mtrr_dqs(sysinfo->tom_k, sysinfo->tom2_k); + + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[ i ]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + fill_mem_cs_sysinfo(i, ctrl+i, sysinfo); + } + + tsc[0] = rdtsc(); + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[ i ]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + print_debug("DQS Training:RcvrEn:Pass1: "); + print_debug_hex8(i); + train_DqsRcvrEn(ctrl+i, 1, sysinfo); + print_debug(" done\r\n"); + } + + tsc[1] = rdtsc(); +#if K8_REV_F_SUPPORT_F0_F1_WORKAROUND == 1 + f0_svm_workaround(controllers, ctrl, tsc0, sysinfo); +#endif + + tsc[2] = rdtsc(); + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[i]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + print_debug("DQS Training:DQSPos: "); + print_debug_hex8(i); + train_DqsPos(ctrl+i, sysinfo); + print_debug(" done\r\n"); + } + + tsc[3] = rdtsc(); + for(i = 0; i < controllers; i++) { + if (!sysinfo->ctrl_present[i]) + continue; + + /* Skip everything if I don't have any memory on this controller */ + if(sysinfo->meminfo[i].dimm_mask==0x00) continue; + + print_debug("DQS Training:RcvrEn:Pass2: "); + print_debug_hex8(i); + train_DqsRcvrEn(ctrl+i, 2, sysinfo); + print_debug(" done\r\n"); + sysinfo->mem_trained[i]=1; + } + + tsc[4] = rdtsc(); + clear_mtrr_dqs(sysinfo->tom2_k); + + + for(i=0;i<5;i++) { + print_debug_dqs_tsc_x("DQS Training:tsc", i, tsc[i].hi, tsc[i].lo); + } + + + +} + +#endif + + +#if MEM_TRAIN_SEQ > 0 + +static void dqs_timing(int i, const struct mem_controller *ctrl, struct sys_info *sysinfo, unsigned v) +{ + + int ii; + + tsc_t tsc[4]; + + +#if MEM_TRAIN_SEQ == 1 + if(sysinfo->mem_trained[i]) return; + //need to enable mtrr, so dqs training could access the test address + setup_mtrr_dqs(sysinfo->tom_k, sysinfo->tom2_k); +#endif + + fill_mem_cs_sysinfo(i, ctrl+i, sysinfo); + + if(v) { + tsc[0] = rdtsc(); + + print_debug("set DQS timing:RcvrEn:Pass1: "); + print_debug_hex8(i); + } + train_DqsRcvrEn(ctrl+i, 1, sysinfo); + + if(v) { + print_debug(" done\r\n"); + tsc[1] = rdtsc(); + print_debug("set DQS timing:DQSPos: "); + print_debug_hex8(i); + } + + train_DqsPos(ctrl+i, sysinfo); + + if(v) { + print_debug(" done\r\n"); + tsc[2] = rdtsc(); + + print_debug("set DQS timing:RcvrEn:Pass2: "); + print_debug_hex8(i); + } + train_DqsRcvrEn(ctrl+i, 2, sysinfo); + + if(v) { + print_debug(" done\r\n"); + + tsc[3] = rdtsc(); + } + +#if MEM_TRAIN_SEQ == 1 + clear_mtrr_dqs(sysinfo->tom2_k); +#endif + + if(v) { + for(ii=0;ii<4;ii++) { + print_debug_dqs_tsc_x("Total DQS Training : tsc ", ii, tsc[ii].hi, tsc[ii].lo); + } + } + + sysinfo->mem_trained[i]=1; + +} +#endif + +#if MEM_TRAIN_SEQ == 1 +static void train_ram(unsigned nodeid, struct sys_info *sysinfo, struct sys_info *sysinfox) +{ + dqs_timing(nodeid, sysinfo->ctrl,sysinfo, 0); // keep the output tidy +// memcpy(&sysinfox->dqs_rcvr_dly_a[nodeid * 2 * 8],&sysinfo->dqs_rcvr_dly_a[nodeid * 2 * 8], 2*8); +// memcpy(&sysinfox->dqs_delay_a[nodeid * 2 * 2 * 9], &sysinfo->dqs_delay_a[nodeid * 2 * 2 * 9], 2 * 2 * 9); + sysinfox->mem_trained[nodeid] = sysinfo->mem_trained[nodeid]; + +} +static void copy_and_run_ap_code_in_car(unsigned ret_addr); +static inline void train_ram_on_node(unsigned nodeid, unsigned coreid, struct sys_info *sysinfo, unsigned retcall) +{ + if(coreid) return; // only do it on core0 + struct sys_info *sysinfox = ((CONFIG_LB_MEM_TOPK<<10) - DCACHE_RAM_GLOBAL_VAR_SIZE); + wait_till_sysinfo_in_ram(); // use pci to get it + + if(sysinfox->mem_trained[nodeid] == 0) { + if (sysinfox->ctrl_present[ nodeid ] && sysinfox->meminfo[nodeid].dimm_mask) { + sysinfo->tom_k = sysinfox->tom_k; + sysinfo->tom2_k = sysinfox->tom2_k; + sysinfo->meminfo[nodeid].is_Width128 = sysinfox->meminfo[nodeid].is_Width128; + set_top_mem_ap(sysinfo->tom_k, sysinfo->tom2_k); // keep the ap's tom consistent with bsp's + #if CONFIG_AP_CODE_IN_CAR == 0 + print_debug("CODE IN ROM AND RUN ON NODE:"); print_debug_hex8(nodeid); print_debug("\r\n"); + train_ram(nodeid, sysinfo, sysinfox); + #else + /* Can copy dqs_timing to ap cache and run from cache? + * we need linuxbios_ap_car.rom? and treat it as linuxbios_ram.rom for ap ? + */ + copy_and_run_ap_code_in_car(retcall); + // will go back by jump + #endif + } + } +} +#endif |