coreboot-gerrit November 2015

coreboot-gerrit@coreboot.org

1 participants
1544 discussions

Patch set updated for coreboot: northbridge/amd/amdfam10: Fix invalid NUMA table
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12035 -gerrit commit d42834b4854b112f56c6826e7fee5110577c74ab Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Sat Aug 8 02:40:58 2015 -0500 northbridge/amd/amdfam10: Fix invalid NUMA table The existing code generated an invalid NUMA table that was rejected by Linux, leading to poor resource allocation. This was due to system I/O resources being inserted into the table when the table should only contain DRAM resources. Do not include system I/O resources (i.e. resources with an index less than 0x10) in the NUMA table. Change-Id: I99c200382b52a99687daf266a84873d9ae2df025 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/northbridge/amd/amdfam10/acpi.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/src/northbridge/amd/amdfam10/acpi.c b/src/northbridge/amd/amdfam10/acpi.c index ad54abd..25df6b3 100644 --- a/src/northbridge/amd/amdfam10/acpi.c +++ b/src/northbridge/amd/amdfam10/acpi.c @@ -102,7 +102,8 @@ static void set_srat_mem(void *gp, struct device *dev, struct resource *res) } // need to figure out NV - state->current += acpi_create_srat_mem((acpi_srat_mem_t *)state->current, (res->index & 0xf), basek, sizek, 1); + if (res->index > 0xf) + state->current += acpi_create_srat_mem((acpi_srat_mem_t *)state->current, (res->index & 0xf), basek, sizek, 1); } static unsigned long acpi_fill_srat(unsigned long current)

1 0

Patch set updated for coreboot: northbridge/amd/amdht: Trivial update to comment
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12030 -gerrit commit a66332fc59f8a4048650b02baf49b45ec8324017 Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 19:05:45 2015 -0500 northbridge/amd/amdht: Trivial update to comment The BKDG is not correct regarding HT Freq write ordering; indicate this in a comment to avoid confusion. Change-Id: I37db191c144c81aba5d4a1e6291db5669a35a31a Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/northbridge/amd/amdht/h3ncmn.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/src/northbridge/amd/amdht/h3ncmn.c b/src/northbridge/amd/amdht/h3ncmn.c index 29524af..c97d592 100644 --- a/src/northbridge/amd/amdht/h3ncmn.c +++ b/src/northbridge/amd/amdht/h3ncmn.c @@ -1555,6 +1555,10 @@ static void setLinkData(sMainData *pDat, cNorthBridge *nb) } else { temp2 = 0x0; } + /* NOTE + * The Family 15h BKDG Rev. 3.14 is wrong + * Freq[4] must be set before Freq[3:0], otherwise the register writes will be ignored! + */ if (is_gt_rev_d()) AmdPCIWriteBits(linkBase + HTHOST_FREQ_REV_REG_2, 0, 0, &temp2); AmdPCIWriteBits(linkBase + HTHOST_FREQ_REV_REG, 11, 8, &temp);

1 0

Patch set updated for coreboot: nb/amd/amdmct/mct_ddr3: Force DRAM retraining on every boot
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12034 -gerrit commit 74490d92f56f5e14170ee2401fa66ba96bdc420b Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 23:59:33 2015 -0500 nb/amd/amdmct/mct_ddr3: Force DRAM retraining on every boot Stability issues have arisen on multiple Family 15h systems when configuration restoration is enabled. In all cases these stability issues resolved by allowing the RAM to go through a full training cycle. Change-Id: I017e0dd5120110124d5b5d5276befef6f7740614 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/northbridge/amd/amdmct/mct_ddr3/mct_d.c | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c index b80918c..1dce138 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c @@ -1742,6 +1742,16 @@ restartinit: allow_config_restore = 0; } + /* FIXME + * Stability issues have arisen on multiple Family 15h systems + * when configuration restoration is enabled. In all cases these + * stability issues resolved by allowing the RAM to go through a + * full training cycle. + * + * Debug and reenable this! + */ + allow_config_restore = 0; + for (Node = 0; Node < MAX_NODES_SUPPORTED; Node++) { struct DCTStatStruc *pDCTstat; pDCTstat = pDCTstatA + Node;

1 0

Patch set updated for coreboot: amd/amdmct/mct_ddr3: Fix poor performance on Family 15h CPUs
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12029 -gerrit commit c299d9f1b064542af6603c9e6909f1047c1b1b98 Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 19:05:29 2015 -0500 amd/amdmct/mct_ddr3: Fix poor performance on Family 15h CPUs Change-Id: Ib6bc197e43e40ba2b923b1eb1229bacafc8be360 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/northbridge/amd/amdmct/mct_ddr3/mct_d.c | 370 +++++++++++++++++++++---- src/northbridge/amd/amdmct/mct_ddr3/mct_d.h | 1 + src/northbridge/amd/amdmct/mct_ddr3/mctdqs_d.c | 65 ++++- src/northbridge/amd/amdmct/mct_ddr3/mctproc.c | 49 +++- src/northbridge/amd/amdmct/mct_ddr3/mctsrc.c | 195 ++++++++++++- src/northbridge/amd/amdmct/mct_ddr3/mctwl.c | 4 + 6 files changed, 604 insertions(+), 80 deletions(-) diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c index 283d897..b80918c 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.c @@ -32,6 +32,8 @@ * supported. */ +// #define DEBUG_DIMM_SPD 1 + static u8 ReconfigureDIMMspare_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); static void DQSTiming_D(struct MCTStatStruc *pMCTstat, @@ -168,7 +170,8 @@ static u32 mct_MR1Odt_RDimm(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat, u8 dct, u32 MrsChipSel); static u32 mct_DramTermDyn_RDimm(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat, u8 dimm); -static u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, u8 dct, u32 misc2); +static u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, + uint8_t dct, uint32_t misc2, uint32_t DramControl); static void mct_BeforeDQSTrainSamp(struct DCTStatStruc *pDCTstat); static void mct_WriteLevelization_HW(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA, uint8_t Pass); @@ -1366,6 +1369,8 @@ static uint8_t fam15h_slow_access_mode(struct DCTStatStruc *pDCTstat, uint8_t dc static void set_2t_configuration(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat, u8 dct) { + printk(BIOS_DEBUG, "%s: Start\n", __func__); + uint32_t dev; uint32_t reg; uint32_t dword; @@ -1388,6 +1393,8 @@ static void set_2t_configuration(struct MCTStatStruc *pMCTstat, else dword &= ~(0x1 << 20); /* Clear 2T CMD mode */ Set_NB32_DCT(dev, dct, reg, dword); + + printk(BIOS_DEBUG, "%s: Done\n", __func__); } static void precise_ndelay_fam15(struct MCTStatStruc *pMCTstat, uint32_t nanoseconds) { @@ -2009,6 +2016,7 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, /* Disable training mode */ uint8_t lane; uint8_t dimm; + uint16_t sword; uint8_t receiver; uint8_t max_lane; uint8_t ecc_enabled; @@ -2023,21 +2031,37 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, uint16_t twrwrdd; uint16_t cdd_twrwrdd; uint16_t twrrd; + uint16_t cdd_twrrd; + uint16_t cdd_trwtto; uint16_t trwtto; uint8_t first_dimm; uint16_t delay; uint16_t delay2; + uint8_t min_value; + uint8_t write_early; uint8_t read_odt_delay; uint8_t write_odt_delay; + uint8_t buffer_data_delay; + int16_t latency_difference; uint16_t difference; uint16_t current_total_delay_1[MAX_BYTE_LANES]; uint16_t current_total_delay_2[MAX_BYTE_LANES]; + uint8_t ddr_voltage_index; + uint8_t max_dimms_installable; /* FIXME * This should be platform configurable */ uint8_t dimm_event_l_pin_support = 0; + if (pDCTstat->DIMMValidDCT[dct] == 0) + ddr_voltage_index = 1; + else + ddr_voltage_index = dct_ddr_voltage_index(pDCTstat, dct); + + ddr_voltage_index = dct_ddr_voltage_index(pDCTstat, dct); + max_dimms_installable = mctGet_NVbits(NV_MAX_DIMMS_PER_CH); + ecc_enabled = !!(pMCTstat->GStatus & 1 << GSB_ECCDIMMs); if (ecc_enabled) max_lane = 9; @@ -2071,6 +2095,24 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, else write_odt_delay = 0; + dword = (Get_NB32_DCT(dev, dct, 0xa8) >> 24) & 0x3; + write_early = dword / 2; + + latency_difference = Get_NB32_DCT(dev, dct, 0x200) & 0x1f; + dword = Get_NB32_DCT(dev, dct, 0x20c) & 0x1f; + latency_difference -= dword; + + if (pDCTstat->Status & (1 << SB_LoadReduced)) { + /* LRDIMM */ + + /* TODO + * Implement LRDIMM support + * See Fam15h BKDG Rev. 3.14 section 2.10.5.5 + */ + } else { + buffer_data_delay = 0; + } + /* TODO: * Adjust trdrdsddc if four-rank DIMMs are installed per * section 2.10.5.5.1 of the Family 15h BKDG. @@ -2106,7 +2148,7 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, } /* Convert the difference to MEMCLKs */ - cdd_trdrddd = (((cdd_trdrddd >> 5) & 0x1f) + 1) / 2; + cdd_trdrddd = (((cdd_trdrddd + (1 << 6) - 1) >> 6) & 0xf); /* Calculate Trdrddd */ delay = (read_odt_delay + 3) * 2; @@ -2152,7 +2194,7 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, } /* Convert the difference to MEMCLKs */ - cdd_twrwrdd = (((cdd_twrwrdd >> 5) & 0x1f) + 1) / 2; + cdd_twrwrdd = (((cdd_twrwrdd + (1 << 6) - 1) >> 6) & 0xf); /* Calculate Twrwrdd */ delay = (write_odt_delay + 3) * 2; @@ -2171,6 +2213,107 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, dword &= ~(0x1 << 18); /* DisAutoRefresh = 0 */ Set_NB32_DCT(dev, dct, 0x8c, dword); /* DRAM Timing High */ + /* Configure power saving options */ + dword = Get_NB32_DCT(dev, dct, 0xa8); /* Dram Miscellaneous 2 */ + dword |= (0x1 << 22); /* PrtlChPDEnhEn = 0x1 */ + dword |= (0x1 << 21); /* AggrPDEn = 0x1 */ + Set_NB32_DCT(dev, dct, 0xa8, dword); /* Dram Miscellaneous 2 */ + + /* Configure partial power down delay */ + dword = Get_NB32(dev, 0x244); /* DRAM Controller Miscellaneous 3 */ + dword &= ~0xf; /* PrtlChPDDynDly = 0x2 */ + dword |= 0x2; + Set_NB32(dev, 0x244, dword); /* DRAM Controller Miscellaneous 3 */ + + /* Configure power save delays */ + delay = 0xa; + delay2 = 0x3; + + /* Family 15h BKDG Table 214 */ + if ((pDCTstat->Status & (1 << SB_Registered)) + || (pDCTstat->Status & (1 << SB_LoadReduced))) { + if (memclk_index <= 0x6) { + if (ddr_voltage_index < 0x4) + /* 1.5 or 1.35V */ + delay2 = 0x3; + else + /* 1.25V */ + delay2 = 0x4; + } + else if ((memclk_index == 0xa) + || (memclk_index == 0xe)) + delay2 = 0x4; + else if (memclk_index == 0x12) + delay2 = 0x5; + else if (memclk_index == 0x16) + delay2 = 0x6; + } else { + if (memclk_index <= 0x6) + delay2 = 0x3; + else if ((memclk_index == 0xa) + || (memclk_index == 0xe)) + delay2 = 0x4; + else if (memclk_index == 0x12) + delay2 = 0x5; + else if (memclk_index == 0x16) + delay2 = 0x6; + } + + /* Family 15h BKDG Table 215 */ + if (memclk_index <= 0x6) + delay = 0xa; + else if (memclk_index == 0xa) + delay = 0xd; + else if (memclk_index == 0xe) + delay = 0x10; + else if (memclk_index == 0x12) + delay = 0x14; + else if (memclk_index == 0x16) + delay = 0x17; + + dword = Get_NB32_DCT(dev, dct, 0x248); /* Dram Power Management 0 */ + dword &= ~(0x3f << 24); /* AggrPDDelay = 0x0 */ + dword &= ~(0x3f << 16); /* PchgPDEnDelay = 0x1 */ + dword |= (0x1 << 16); + dword &= ~(0x1f << 8); /* Txpdll = delay */ + dword |= ((delay & 0x1f) << 8); + dword &= ~0xf; /* Txp = delay2 */ + dword |= delay2 & 0xf; + Set_NB32_DCT(dev, dct, 0x248, dword); /* Dram Power Management 0 */ + + /* Family 15h BKDG Table 216 */ + if (memclk_index <= 0x6) { + delay = 0x5; + delay2 = 0x3; + } + else if (memclk_index == 0xa) { + delay = 0x6; + delay2 = 0x3; + } + else if (memclk_index == 0xe) { + delay = 0x7; + delay2 = 0x4; + } + else if (memclk_index == 0x12) { + delay = 0x8; + delay2 = 0x4; + } + else if (memclk_index == 0x16) { + delay = 0xa; + delay2 = 0x5; + } + + dword = Get_NB32_DCT(dev, dct, 0x24c); /* Dram Power Management 1 */ + dword &= ~(0x3f << 24); /* Tcksrx = delay */ + dword |= ((delay & 0x3f) << 24); + dword &= ~(0x3f << 16); /* Tcksre = delay */ + dword |= ((delay & 0x3f) << 16); + dword &= ~(0x3f << 8); /* Tckesr = delay2 + 1 */ + dword |= (((delay2 + 1) & 0x3f) << 8); + dword &= ~0xf; /* Tpd = delay2 */ + dword |= delay2 & 0xf; + Set_NB32_DCT(dev, dct, 0x24c, dword); /* Dram Power Management 1 */ + dword = Get_NB32_DCT(dev, dct, 0x94); /* DRAM Configuration High */ dword |= (0xf << 24); /* DcqBypassMax = 0xf */ dword |= (0x1 << 22); /* BankSwizzleMode = 1 */ @@ -2223,15 +2366,98 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, } } - /* TODO - * Calculate Twrrd per section 2.10.5.5.3 of the Family 15h BKDG - */ - twrrd = 0xb; + /* Calculate the Critical Delay Difference for Twrrd */ + cdd_twrrd = 0; + for (receiver = 0; receiver < 8; receiver += 2) { + dimm = (receiver >> 1); - /* TODO - * Calculate TrwtTO per section 2.10.5.5.4 of the Family 15h BKDG - */ - trwtto = 0x16; + if (!mct_RcvrRankEnabled_D(pMCTstat, pDCTstat, dct, receiver)) + continue; + + read_dqs_write_timing_control_registers(current_total_delay_1, dev, dct, dimm, index_reg); + read_dqs_receiver_enable_control_registers(current_total_delay_2, dev, dct, dimm, index_reg); + + for (lane = 0; lane < max_lane; lane++) { + if (current_total_delay_1[lane] > current_total_delay_2[lane]) + difference = current_total_delay_1[lane] - current_total_delay_2[lane]; + else + difference = current_total_delay_2[lane] - current_total_delay_1[lane]; + + if (difference > cdd_twrrd) + cdd_twrrd = difference; + } + } + + /* Convert the difference to MEMCLKs */ + cdd_twrrd = (((cdd_twrrd + (1 << 6) - 1) >> 6) & 0xf); + + /* Fam15h BKDG section 2.10.5.5.3 */ + if (pDCTstat->Status & (1 << SB_LoadReduced)) { + /* LRDIMM */ + + /* TODO + * Implement LRDIMM support + * See Fam15h BKDG Rev. 3.14 section 2.10.5.5 + */ + twrrd = 0xb; + } else { + sword = (((int16_t)cdd_twrrd + 1 - ((int16_t)write_early * 2)) + 1) / 2; + if (sword < 0) + sword = 0; + if (((uint16_t)sword) > write_odt_delay) + dword = sword; + else + dword = write_odt_delay; + dword += 3; + if (latency_difference < dword) { + dword -= latency_difference; + if (dword < 1) + twrrd = 1; + else + twrrd = dword; + } else { + twrrd = 1; + } + } + + /* Calculate the Critical Delay Difference for TrwtTO */ + cdd_trwtto = 0; + for (receiver = 0; receiver < 8; receiver += 2) { + dimm = (receiver >> 1); + + if (!mct_RcvrRankEnabled_D(pMCTstat, pDCTstat, dct, receiver)) + continue; + + read_dqs_receiver_enable_control_registers(current_total_delay_1, dev, dct, dimm, index_reg); + read_dqs_write_timing_control_registers(current_total_delay_2, dev, dct, dimm, index_reg); + + for (lane = 0; lane < max_lane; lane++) { + if (current_total_delay_1[lane] > current_total_delay_2[lane]) + difference = current_total_delay_1[lane] - current_total_delay_2[lane]; + else + difference = current_total_delay_2[lane] - current_total_delay_1[lane]; + + if (difference > cdd_trwtto) + cdd_trwtto = difference; + } + } + + /* Convert the difference to MEMCLKs */ + cdd_trwtto = (((cdd_trwtto + (1 << 6) - 1) >> 6) & 0xf); + + /* Fam15h BKDG section 2.10.5.5.4 */ + if (max_dimms_installable == 1) + min_value = 0; + else + min_value = read_odt_delay + buffer_data_delay; + sword = (((int16_t)cdd_trwtto - 1 + ((int16_t)write_early * 2)) + 1) / 2; + sword += latency_difference + 3; + if (sword < 0) + sword = 0; + if (((uint16_t)sword) > min_value) + trwtto = (uint16_t)sword; + else + trwtto = min_value; dword = Get_NB32_DCT(dev, dct, 0xa4); /* DRAM Controller Temperature Throttle */ dword &= ~(0x1 << 11); /* BwCapEn = 0 */ @@ -2242,6 +2468,7 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(dev, dct, 0x110); /* DRAM Controller Select Low */ dword &= ~(0x1 << 2); /* DctSelIntLvEn = interleave_channels */ dword |= (interleave_channels & 0x1) << 2; + dword |= (0x3 << 6); /* DctSelIntLvAddr = 0x3 */ Set_NB32_DCT(dev, dct, 0x110, dword); /* DRAM Controller Select Low */ /* NOTE @@ -2249,22 +2476,6 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, * otherwise semi-random lockups will occur due to misconfigured scrubbing hardware! */ - /* FIXME - * The BKDG-recommended settings cause memory corruption on the ASUS KGPE-D16. - * Investigate and fix... - */ -#if 0 - /* Fam15h BKDG section 2.10.5.5.1 */ - dword = Get_NB32_DCT(dev, dct, 0x218); /* DRAM Timing 5 */ - dword &= ~(0xf << 24); /* TrdrdSdSc = 0x1 */ - dword |= (0x1 << 24); - dword &= ~(0xf << 16); /* TrdrdSdDc = trdrdsddc */ - dword |= ((trdrdsddc & 0xf) << 16); - dword &= ~(0xf); /* TrdrdDd = trdrddd */ - dword |= (trdrddd & 0xf); - Set_NB32_DCT(dev, dct, 0x218, dword); /* DRAM Timing 5 */ -#endif - /* Fam15h BKDG section 2.10.5.5.2 */ dword = Get_NB32_DCT(dev, dct, 0x214); /* DRAM Timing 4 */ dword &= ~(0xf << 16); /* TwrwrSdSc = 0x1 */ @@ -2277,8 +2488,14 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, /* Fam15h BKDG section 2.10.5.5.3 */ dword = Get_NB32_DCT(dev, dct, 0x218); /* DRAM Timing 5 */ + dword &= ~(0xf << 24); /* TrdrdSdSc = 0x1 */ + dword |= (0x1 << 24); + dword &= ~(0xf << 16); /* TrdrdSdDc = trdrdsddc */ + dword |= ((trdrdsddc & 0xf) << 16); dword &= ~(0xf << 8); /* Twrrd = twrrd */ dword |= ((twrrd & 0xf) << 8); + dword &= ~(0xf); /* TrdrdDd = trdrddd */ + dword |= (trdrddd & 0xf); Set_NB32_DCT(dev, dct, 0x218, dword); /* DRAM Timing 5 */ /* Fam15h BKDG section 2.10.5.5.4 */ @@ -2289,12 +2506,6 @@ static void fam15EnableTrainingMode(struct MCTStatStruc *pMCTstat, dword |= ((((dword >> 8) & 0x1f) + 1) << 16); Set_NB32_DCT(dev, dct, 0x21c, dword); /* DRAM Timing 6 */ - /* Configure partial power down delay */ - dword = Get_NB32(dev, 0x244); /* DRAM Controller Miscellaneous 3 */ - dword &= ~0xf; /* PrtlChPDDynDly = 0x2 */ - dword |= 0x2; - Set_NB32(dev, 0x244, dword); /* DRAM Controller Miscellaneous 3 */ - /* Enable prefetchers */ dword = Get_NB32(dev, 0x11c); /* Memory Controller Configuration High */ dword &= ~(0x1 << 13); /* PrefIoDis = 0 */ @@ -2383,6 +2594,8 @@ static void DQSTiming_D(struct MCTStatStruc *pMCTstat, mct_TrainDQSPos_D(pMCTstat, pDCTstatA); + TrainMaxRdLatency_En_D(pMCTstat, pDCTstatA); + if (is_fam15h()) exit_training_mode_fam15(pMCTstat, pDCTstatA); else @@ -2961,6 +3174,13 @@ static void ClearDCT_D(struct MCTStatStruc *pMCTstat, } while(reg < reg_end) { + if ((reg & 0xFF) == 0x84) { + if (is_fam15h()) { + val = Get_NB32_DCT(dev, dct, reg); + val &= ~(0x1 << 23); /* Clear PchgPDModeSel */ + val &= ~0x3; /* Clear BurstCtrl */ + } + } if ((reg & 0xFF) == 0x90) { if (pDCTstat->LogicalCPUID & AMD_DR_Dx) { val = Get_NB32_DCT(dev, dct, reg); /* get DRAMConfigLow */ @@ -3079,14 +3299,30 @@ static void SPD2ndTiming(struct MCTStatStruc *pMCTstat, /* Convert DRAM CycleTiming values and store into DCT structure */ byte = pDCTstat->DIMMAutoSpeed; - if (byte == 7) - tCK16x = 20; - else if (byte == 6) - tCK16x = 24; - else if (byte == 5) - tCK16x = 30; - else - tCK16x = 40; + if (is_fam15h()) { + if (byte == 0x16) + tCK16x = 17; + else if (byte == 0x12) + tCK16x = 20; + else if (byte == 0xe) + tCK16x = 24; + else if (byte == 0xa) + tCK16x = 30; + else if (byte == 0x6) + tCK16x = 40; + else + tCK16x = 48; + } + else { + if (byte == 7) + tCK16x = 20; + else if (byte == 6) + tCK16x = 24; + else if (byte == 5) + tCK16x = 30; + else + tCK16x = 40; + } /* Notes: 1. All secondary time values given in SPDs are in binary with units of ns. @@ -3119,7 +3355,7 @@ static void SPD2ndTiming(struct MCTStatStruc *pMCTstat, val = Max_TrpT; pDCTstat->Trp = val; - /*Trrd*/ + /* Trrd */ pDCTstat->DIMMTrrd = Trrd; val = Trrd / tCK16x; if (Trrd % tCK16x) { /* round up number of busclocks */ @@ -3237,21 +3473,31 @@ static void SPD2ndTiming(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(dev, dct, 0x200); /* DRAM Timing 0 */ dword &= ~(0x3f1f1f1f); - dword |= ((pDCTstat->Tras + 0xf) & 0x3f) << 24; /* Tras */ - dword |= ((pDCTstat->Trp + 0x5) & 0x1f) << 16; /* Trp */ - dword |= ((pDCTstat->Trcd + 0x5) & 0x1f) << 8; /* Trcd */ + dword |= (pDCTstat->Tras & 0x3f) << 24; /* Tras */ + val = pDCTstat->Trp; + val = mct_AdjustSPDTimings(pMCTstat, pDCTstat, val); + dword |= (val & 0x1f) << 16; /* Trp */ + dword |= (pDCTstat->Trcd & 0x1f) << 8; /* Trcd */ dword |= (pDCTstat->CASL & 0x1f); /* Tcl */ Set_NB32_DCT(dev, dct, 0x200, dword); /* DRAM Timing 0 */ dword = Get_NB32_DCT(dev, dct, 0x204); /* DRAM Timing 1 */ dword &= ~(0x0f3f0f3f); - dword |= ((pDCTstat->Trtp + 0x4) & 0xf) << 24; /* Trtp */ - if (pDCTstat->Tfaw != 0) - dword |= ((((pDCTstat->Tfaw - 0x1) * 2) + 0x10) & 0x3f) << 16; /* FourActWindow */ - dword |= ((pDCTstat->Trrd + 0x4) & 0xf) << 8; /* Trrd */ - dword |= ((pDCTstat->Trc + 0xb) & 0x3f); /* Trc */ + dword |= (pDCTstat->Trtp & 0xf) << 24; /* Trtp */ + if (pDCTstat->Tfaw != 0) { + val = pDCTstat->Tfaw; + val = mct_AdjustSPDTimings(pMCTstat, pDCTstat, val); + if ((val > 0x5) && (val < 0x2b)) + dword |= (val & 0x3f) << 16; /* FourActWindow */ + } + dword |= (pDCTstat->Trrd & 0xf) << 8; /* Trrd */ + dword |= (pDCTstat->Trc & 0x3f); /* Trc */ Set_NB32_DCT(dev, dct, 0x204, dword); /* DRAM Timing 1 */ + /* Trfc0-Trfc3 */ + for (i=0; i<4; i++) + if (pDCTstat->Trfc[i] == 0x0) + pDCTstat->Trfc[i] = 0x4; dword = Get_NB32_DCT(dev, dct, 0x208); /* DRAM Timing 2 */ dword &= ~(0x07070707); dword |= (pDCTstat->Trfc[3] & 0x7) << 24; /* Trfc3 */ @@ -3262,14 +3508,14 @@ static void SPD2ndTiming(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(dev, dct, 0x20c); /* DRAM Timing 3 */ dword &= ~(0x00000f00); - dword |= ((pDCTstat->Twtr + 0x4) & 0xf) << 8; /* Twtr */ + dword |= (pDCTstat->Twtr & 0xf) << 8; /* Twtr */ dword &= ~(0x0000001f); dword |= (Tcwl & 0x1f); /* Tcwl */ Set_NB32_DCT(dev, dct, 0x20c, dword); /* DRAM Timing 3 */ dword = Get_NB32_DCT(dev, dct, 0x22c); /* DRAM Timing 10 */ dword &= ~(0x0000001f); - dword |= ((pDCTstat->Twr + 0x4) & 0x1f); /* Twr */ + dword |= (pDCTstat->Twr & 0x1f); /* Twr */ Set_NB32_DCT(dev, dct, 0x22c, dword); /* DRAM Timing 10 */ if (pDCTstat->Speed > mhz_to_memclk_config(mctGet_NVbits(NV_MIN_MEMCLK))) { @@ -3865,6 +4111,8 @@ static u8 AutoConfig_D(struct MCTStatStruc *pMCTstat, } } + DramConfigMisc2 = mct_SetDramConfigMisc2(pDCTstat, dct, DramConfigMisc2, DramControl); + printk(BIOS_DEBUG, "AutoConfig_D: DramControl: %08x\n", DramControl); printk(BIOS_DEBUG, "AutoConfig_D: DramTimingLo: %08x\n", DramTimingLo); printk(BIOS_DEBUG, "AutoConfig_D: DramConfigMisc: %08x\n", DramConfigMisc); @@ -3876,7 +4124,6 @@ static u8 AutoConfig_D(struct MCTStatStruc *pMCTstat, Set_NB32_DCT(dev, dct, 0x78, DramControl); Set_NB32_DCT(dev, dct, 0x88, DramTimingLo); Set_NB32_DCT(dev, dct, 0xa0, DramConfigMisc); - DramConfigMisc2 = mct_SetDramConfigMisc2(pDCTstat, dct, DramConfigMisc2); Set_NB32_DCT(dev, dct, 0xa8, DramConfigMisc2); Set_NB32_DCT(dev, dct, 0x90, DramConfigLo); ProgDramMRSReg_D(pMCTstat, pDCTstat, dct); @@ -5247,6 +5494,16 @@ static void mct_PhyController_Config(struct MCTStatStruc *pMCTstat, u32 dev = pDCTstat->dev_dct; if (pDCTstat->LogicalCPUID & (AMD_DR_DAC2_OR_C3 | AMD_RB_C3 | AMD_FAM15_ALL)) { + if (is_fam15h()) { + /* Set F2x[1, 0]98_x0D0F0F13 DllDisEarlyU and DllDisEarlyL to save power */ + for (index = 0; index < 0x9; index++) { + dword = Get_NB32_index_wait_DCT(dev, dct, index_reg, 0x0d0f0013 | (index << 8)); + dword |= (0x1 << 1); /* DllDisEarlyU = 1 */ + dword |= 0x1; /* DllDisEarlyL = 1 */ + Set_NB32_index_wait_DCT(dev, dct, index_reg, 0x0d0f0013 | (index << 8), dword); + } + } + if (pDCTstat->Dimmx4Present == 0) { /* Set bit7 RxDqsUDllPowerDown to register F2x[1, 0]98_x0D0F0F13 for * additional power saving when x4 DIMMs are not present. @@ -5291,8 +5548,9 @@ static void mct_FinalMCT_D(struct MCTStatStruc *pMCTstat, mct_ExtMCTConfig_Dx(pDCTstat); } else { /* Family 15h CPUs */ - val = 0x0ce00f00 | 0x1 << 29; /* FlushWrOnStpGnt */ - val |= 0x10 << 2; /* MctWrLimit = 16 */ + val = 0x0ce00f00; /* FlushWrOnStpGnt = 0x0 */ + val |= 0x10 << 2; /* MctWrLimit = 0x10 */ + val |= 0x1; /* DctWrLimit = 0x1 */ Set_NB32(pDCTstat->dev_dct, 0x11c, val); val = Get_NB32(pDCTstat->dev_dct, 0x1b0); @@ -6532,8 +6790,8 @@ void ProgDramMRSReg_D(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x84); if (is_fam15h()) { - dword |= DramMRS; dword &= ~0x00800003; + dword |= DramMRS; } else { dword &= ~0x00fc2f8f; dword |= DramMRS; diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.h b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.h index eb4c74e..b72b9da 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mct_d.h +++ b/src/northbridge/amd/amdmct/mct_ddr3/mct_d.h @@ -984,6 +984,7 @@ void UMAMemTyping_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat uint64_t mctGetLogicalCPUID(u32 Node); u8 ECCInit_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); void TrainReceiverEn_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA, u8 Pass); +void TrainMaxRdLatency_En_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); void mct_TrainDQSPos_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); void mctSetEccDQSRcvrEn_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); void TrainMaxReadLatency_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstatA); diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctdqs_d.c b/src/northbridge/amd/amdmct/mct_ddr3/mctdqs_d.c index 3615616..06597e2 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mctdqs_d.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mctdqs_d.c @@ -20,6 +20,9 @@ static void write_dqs_receiver_enable_control_registers(uint16_t* current_total_ static void read_read_dqs_timing_control_registers(uint16_t* current_total_delay, uint32_t dev, uint8_t dct, uint8_t dimm, uint32_t index_reg); +static void dqsTrainMaxRdLatency_SW_Fam15(struct MCTStatStruc *pMCTstat, + struct DCTStatStruc *pDCTstat); + static void CalcEccDQSPos_D(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat, u16 like, u8 scale, u8 ChipSel); @@ -214,6 +217,27 @@ void TrainReceiverEn_D(struct MCTStatStruc *pMCTstat, } } +void TrainMaxRdLatency_En_D(struct MCTStatStruc *pMCTstat, + struct DCTStatStruc *pDCTstatA) +{ + uint8_t node; + struct DCTStatStruc *pDCTstat; + + for (node = 0; node < MAX_NODES_SUPPORTED; node++) { + pDCTstat = pDCTstatA + node; + + if (pDCTstat->DCTSysLimit) { + if (is_fam15h()) { + dqsTrainMaxRdLatency_SW_Fam15(pMCTstat, pDCTstat); + } else { + /* FIXME + * Implement Family 10h MaxRdLatency training + */ + } + } + } +} + static void SetEccDQSRdWrPos_D_Fam10(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat, u8 ChipSel) { @@ -894,7 +918,7 @@ static void TrainDQSRdWrPos_D_Fam10(struct MCTStatStruc *pMCTstat, * Algorithm detailed in the Fam15h BKDG Rev. 3.14 section 2.10.5.8.5 */ static void Calc_SetMaxRdLatency_D_Fam15(struct MCTStatStruc *pMCTstat, - struct DCTStatStruc *pDCTstat, uint8_t dct) + struct DCTStatStruc *pDCTstat, uint8_t dct, uint8_t calc_min) { uint8_t dimm; uint8_t lane; @@ -938,7 +962,8 @@ static void Calc_SetMaxRdLatency_D_Fam15(struct MCTStatStruc *pMCTstat, p += (9 - dword); /* 2.10.5.8.5 (4) */ - p += 5; + if (!calc_min) + p += 5; /* 2.10.5.8.5 (5) */ dword = Get_NB32_DCT(dev, dct, 0xa8); @@ -965,7 +990,8 @@ static void Calc_SetMaxRdLatency_D_Fam15(struct MCTStatStruc *pMCTstat, p += (max_delay >> 5); /* 2.10.5.8.5 (8) */ - p += 5; + if (!calc_min) + p += 5; /* 2.10.5.8.5 (9) */ t += 800; @@ -976,13 +1002,16 @@ static void Calc_SetMaxRdLatency_D_Fam15(struct MCTStatStruc *pMCTstat, n = (((((uint64_t)p * 1000000000000ULL)/(((uint64_t)fam15h_freq_tab[mem_clk] * 1000000ULL) * 2)) + ((uint64_t)t)) * ((uint64_t)nb_clk * 1000)) / 1000000000ULL; /* 2.10.5.8.5 (11) */ - n -= 1; + if (!calc_min) + n -= 1; /* 2.10.5.8.5 (12) */ - dword = Get_NB32_DCT_NBPstate(dev, dct, nb_pstate, 0x210); - dword &= ~(0x3ff << 22); - dword |= (((n - 1) & 0x3ff) << 22); - Set_NB32_DCT_NBPstate(dev, dct, nb_pstate, 0x210, dword); + if (!calc_min) { + dword = Get_NB32_DCT_NBPstate(dev, dct, nb_pstate, 0x210); + dword &= ~(0x3ff << 22); + dword |= (((n - 1) & 0x3ff) << 22); + Set_NB32_DCT_NBPstate(dev, dct, nb_pstate, 0x210, dword); + } /* Save result for later use */ pDCTstat->CH_MaxRdLat[dct] = n - 1; @@ -1103,6 +1132,9 @@ static void read_dram_dqs_training_pattern_fam15(struct MCTStatStruc *pMCTstat, } else if (lane < 8) { Set_NB32_DCT(dev, dct, 0x274, ~0x0); Set_NB32_DCT(dev, dct, 0x278, ~(0xff << (lane * 8))); + } else if (lane == 0xff) { + Set_NB32_DCT(dev, dct, 0x274, ~0xffffffff); + Set_NB32_DCT(dev, dct, 0x278, ~0xffffffff); } else { Set_NB32_DCT(dev, dct, 0x274, ~0x0); Set_NB32_DCT(dev, dct, 0x278, ~0x0); @@ -1110,8 +1142,9 @@ static void read_dram_dqs_training_pattern_fam15(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(dev, dct, 0x27c); dword &= ~(0xff); /* EccMask = 0 */ - if ((lane != 8) || (pDCTstat->DimmECCPresent == 0)) - dword |= 0xff; /* EccMask = 0xff */ + if (lane != 0xff) + if ((lane != 8) || (pDCTstat->DimmECCPresent == 0)) + dword |= 0xff; /* EccMask = 0xff */ Set_NB32_DCT(dev, dct, 0x27c, dword); dword = Get_NB32_DCT(dev, dct, 0x270); @@ -1180,6 +1213,9 @@ static void write_dram_dqs_training_pattern_fam15(struct MCTStatStruc *pMCTstat, } else if (lane < 8) { Set_NB32_DCT(dev, dct, 0x274, ~0x0); Set_NB32_DCT(dev, dct, 0x278, ~(0xff << (lane * 8))); + } else if (lane == 0xff) { + Set_NB32_DCT(dev, dct, 0x274, ~0xffffffff); + Set_NB32_DCT(dev, dct, 0x278, ~0xffffffff); } else { Set_NB32_DCT(dev, dct, 0x274, ~0x0); Set_NB32_DCT(dev, dct, 0x278, ~0x0); @@ -1187,8 +1223,9 @@ static void write_dram_dqs_training_pattern_fam15(struct MCTStatStruc *pMCTstat, dword = Get_NB32_DCT(dev, dct, 0x27c); dword &= ~(0xff); /* EccMask = 0 */ - if ((lane != 8) || (pDCTstat->DimmECCPresent == 0)) - dword |= 0xff; /* EccMask = 0xff */ + if (lane != 0xff) + if ((lane != 8) || (pDCTstat->DimmECCPresent == 0)) + dword |= 0xff; /* EccMask = 0xff */ Set_NB32_DCT(dev, dct, 0x27c, dword); dword = Get_NB32_DCT(dev, dct, 0x270); @@ -1274,7 +1311,7 @@ static uint8_t TrainDQSRdWrPos_D_Fam15(struct MCTStatStruc *pMCTstat, uint32_t dev = pDCTstat->dev_dct; /* Calculate and program MaxRdLatency */ - Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, dct); + Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, dct, 0); Errors = 0; dual_rank = 0; @@ -1632,7 +1669,7 @@ static void TrainDQSReceiverEnCyc_D_Fam15(struct MCTStatStruc *pMCTstat, write_dqs_receiver_enable_control_registers(current_phy_phase_delay, dev, dct, dimm, index_reg); /* Calculate and program MaxRdLatency */ - Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, dct); + Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, dct, 0); /* 2.10.5.8.3 (4 B) */ dqs_results_array[current_phy_phase_delay[lane]] = TrainDQSRdWrPos_D_Fam15(pMCTstat, pDCTstat, dct, Receiver, Receiver + 2, lane, lane + 1); diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c b/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c index 0922193..cf13b40 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mctproc.c @@ -15,7 +15,8 @@ */ /* mct_SetDramConfigMisc2_Cx & mct_SetDramConfigMisc2_Dx */ -u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, u8 dct, u32 misc2) +u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, + uint8_t dct, uint32_t misc2, uint32_t DramControl) { u32 val; @@ -24,17 +25,47 @@ u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, u8 dct, u32 misc2) if (pDCTstat->LogicalCPUID & AMD_FAM15_ALL) { uint8_t cs_mux_45; uint8_t cs_mux_67; + uint32_t f2x80; - /* BKDG v3.14 Table 200 / Table 201 */ - if (MaxDimmsInstallable < 3) { - cs_mux_45 = 1; - cs_mux_67 = 1; - } else { + misc2 &= ~(0x1 << 28); /* FastSelfRefEntryDis = 0x0 */ + if (MaxDimmsInstallable == 3) { + /* FIXME 3 DIMMS per channel unimplemented */ cs_mux_45 = 0; + } else { + uint32_t f2x60 = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x60); + f2x80 = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x80); + if ((((f2x80 & 0xf) == 0x7) || ((f2x80 & 0xf) == 0x9)) + && ((f2x60 & 0x3) == 0x3)) + cs_mux_45 = 1; + else if ((((f2x80 & 0xa) == 0x7) || ((f2x80 & 0xb) == 0x9)) + && ((f2x60 & 0x3) > 0x1)) + cs_mux_45 = 1; + else + cs_mux_45 = 0; + } + + if (MaxDimmsInstallable == 1) { + cs_mux_67 = 0; + } else if (MaxDimmsInstallable == 2) { + uint32_t f2x64 = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x64); + f2x80 = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x80); + if (((((f2x80 >> 4) & 0xf) == 0x7) || (((f2x80 >> 4) & 0xf) == 0x9)) + && ((f2x64 & 0x3) == 0x3)) + cs_mux_67 = 1; + else if (((((f2x80 >> 4) & 0xa) == 0x7) || (((f2x80 >> 4) & 0xb) == 0x9)) + && ((f2x64 & 0x3) > 0x1)) + cs_mux_67 = 1; + else + cs_mux_67 = 0; + } else { + /* FIXME 3 DIMMS per channel unimplemented */ cs_mux_67 = 0; } - misc2 |= (cs_mux_45 & 0x1) << 26; - misc2 |= (cs_mux_67 & 0x1) << 27; + + misc2 &= ~(0x1 << 27); /* CsMux67 = cs_mux_67 */ + misc2 |= ((cs_mux_67 & 0x1) << 27); + misc2 &= ~(0x1 << 26); /* CsMux45 = cs_mux_45 */ + misc2 |= ((cs_mux_45 & 0x1) << 26); } else if (pDCTstat->LogicalCPUID & (AMD_DR_Dx | AMD_DR_Cx)) { if (pDCTstat->Status & (1 << SB_Registered)) { misc2 |= 1 << SubMemclkRegDly; @@ -46,8 +77,8 @@ u32 mct_SetDramConfigMisc2(struct DCTStatStruc *pDCTstat, u8 dct, u32 misc2) if (pDCTstat->LogicalCPUID & AMD_DR_Cx) misc2 |= 1 << OdtSwizzle; - val = Get_NB32_DCT(pDCTstat->dev_dct, dct, 0x78); + val = DramControl; val &= 7; val = ((~val) & 0xff) + 1; val += 6; diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctsrc.c b/src/northbridge/amd/amdmct/mct_ddr3/mctsrc.c index 19b1b8f..d69a1ee 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mctsrc.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mctsrc.c @@ -1420,7 +1420,7 @@ static void dqsTrainRcvrEn_SW_Fam15(struct MCTStatStruc *pMCTstat, } /* Calculate and program MaxRdLatency */ - Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, Channel); + Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, Channel, 0); if(_DisableDramECC) { mct_EnableDimmEccEn_D(pMCTstat, pDCTstat, _DisableDramECC); @@ -1483,6 +1483,199 @@ static void dqsTrainRcvrEn_SW_Fam15(struct MCTStatStruc *pMCTstat, printk(BIOS_DEBUG, "TrainRcvrEn: Done\n\n"); } +static void write_max_read_latency_to_registers(struct MCTStatStruc *pMCTstat, + struct DCTStatStruc *pDCTstat, uint8_t dct, uint16_t latency) +{ + uint32_t dword; + uint8_t nb_pstate; + + for (nb_pstate = 0; nb_pstate < 2; nb_pstate++) { + dword = Get_NB32_DCT_NBPstate(pDCTstat->dev_dct, dct, nb_pstate, 0x210); + dword &= ~(0x3ff << 22); + dword |= ((latency & 0x3ff) << 22); + Set_NB32_DCT_NBPstate(pDCTstat->dev_dct, dct, nb_pstate, 0x210, dword); + } +} + +/* DQS MaxRdLatency Training (Family 15h) + * Algorithm detailed in: + * The Fam15h BKDG Rev. 3.14 section 2.10.5.8.5.1 + * This algorithm runs at the highest supported MEMCLK. + */ +static void dqsTrainMaxRdLatency_SW_Fam15(struct MCTStatStruc *pMCTstat, + struct DCTStatStruc *pDCTstat) +{ + u8 Channel; + u8 Addl_Index = 0; + u8 Receiver; + u8 _DisableDramECC = 0, _Wrap32Dis = 0, _SSE2 = 0; + u32 Errors; + + u32 dev; + u32 index_reg; + u32 ch_start, ch_end; + u32 msr; + u32 cr4; + u32 lo, hi; + + uint32_t dword; + uint8_t dimm; + uint8_t lane; + uint8_t mem_clk; + uint32_t nb_clk; + uint8_t nb_pstate; + uint16_t current_total_delay[MAX_BYTE_LANES]; + uint16_t current_rdqs_total_delay[MAX_BYTE_LANES]; + uint8_t current_worst_case_total_delay_dimm; + uint16_t current_worst_case_total_delay_value; + + uint16_t fam15h_freq_tab[] = {0, 0, 0, 0, 333, 0, 400, 0, 0, 0, 533, 0, 0, 0, 667, 0, 0, 0, 800, 0, 0, 0, 933}; + + print_debug_dqs("\nTrainMaxRdLatency: Node", pDCTstat->Node_ID, 0); + + dev = pDCTstat->dev_dct; + index_reg = 0x98; + ch_start = 0; + ch_end = 2; + + cr4 = read_cr4(); + if(cr4 & ( 1 << 9)) { /* save the old value */ + _SSE2 = 1; + } + cr4 |= (1 << 9); /* OSFXSR enable SSE2 */ + write_cr4(cr4); + + msr = HWCR; + _RDMSR(msr, &lo, &hi); + /* FIXME: Why use SSEDIS */ + if(lo & (1 << 17)) { /* save the old value */ + _Wrap32Dis = 1; + } + lo |= (1 << 17); /* HWCR.wrap32dis */ + lo &= ~(1 << 15); /* SSEDIS */ + _WRMSR(msr, lo, hi); /* Setting wrap32dis allows 64-bit memory references in real mode */ + + _DisableDramECC = mct_DisableDimmEccEn_D(pMCTstat, pDCTstat); + + Errors = 0; + dev = pDCTstat->dev_dct; + + for (Channel = 0; Channel < 2; Channel++) { + print_debug_dqs("\tTrainMaxRdLatency51: Node ", pDCTstat->Node_ID, 1); + print_debug_dqs("\tTrainMaxRdLatency51: Channel ", Channel, 1); + pDCTstat->Channel = Channel; + + if (pDCTstat->DIMMValidDCT[Channel] == 0) + continue; + + mem_clk = Get_NB32_DCT(dev, Channel, 0x94) & 0x1f; + + Receiver = mct_InitReceiver_D(pDCTstat, Channel); + + /* Find DIMM with worst case receiver enable delays */ + current_worst_case_total_delay_dimm = 0; + current_worst_case_total_delay_value = 0; + + /* There are four receiver pairs, loosely associated with chipselects. + * This is essentially looping over each DIMM. + */ + for (; Receiver < 8; Receiver += 2) { + Addl_Index = (Receiver >> 1) * 3 + 0x10; + dimm = (Receiver >> 1); + + print_debug_dqs("\t\tTrainMaxRdLatency52: index ", Addl_Index, 2); + + if (!mct_RcvrRankEnabled_D(pMCTstat, pDCTstat, Channel, Receiver)) { + continue; + } + + /* Retrieve the total delay values from pass 1 of DQS receiver enable training */ + read_dqs_receiver_enable_control_registers(current_total_delay, dev, Channel, dimm, index_reg); + read_read_dqs_timing_control_registers(current_rdqs_total_delay, dev, Channel, dimm, index_reg); + + for (lane = 0; lane < 8; lane++) { + current_total_delay[lane] += current_rdqs_total_delay[lane]; + if (current_total_delay[lane] > current_worst_case_total_delay_value) { + current_worst_case_total_delay_dimm = dimm; + current_worst_case_total_delay_value = current_total_delay[lane]; + } + } + +#if DQS_TRAIN_DEBUG > 0 + for (lane = 0; lane < 8; lane++) + print_debug_dqs_pair("\t\tTrainMaxRdLatency56: Lane ", lane, " current_total_delay ", current_total_delay[lane], 2); +#endif + } + + /* 2.10.5.8.5.1.1 */ + Calc_SetMaxRdLatency_D_Fam15(pMCTstat, pDCTstat, Channel, 1); + + /* 2.10.5.8.5.1.[2,3] + * Write the DRAM training pattern to the test address + */ + write_dram_dqs_training_pattern_fam15(pMCTstat, pDCTstat, Channel, current_worst_case_total_delay_dimm << 1, 0xff); + + /* 2.10.5.8.5.1.4 + * Incrementally test each MaxRdLatency candidate + */ + for (; pDCTstat->CH_MaxRdLat[Channel] < 0x3ff; pDCTstat->CH_MaxRdLat[Channel]++) { + write_max_read_latency_to_registers(pMCTstat, pDCTstat, Channel, pDCTstat->CH_MaxRdLat[Channel]); + read_dram_dqs_training_pattern_fam15(pMCTstat, pDCTstat, Channel, current_worst_case_total_delay_dimm << 1, 0xff); + dword = Get_NB32_DCT(dev, Channel, 0x268) & 0x3ffff; + if (!dword) + break; + Set_NB32_index_wait_DCT(dev, Channel, index_reg, 0x00000050, 0x13131313); + } + + /* 2.10.5.8.5.1.5 */ + nb_pstate = 0; + mem_clk = Get_NB32_DCT(dev, Channel, 0x94) & 0x1f; + if (fam15h_freq_tab[mem_clk] == 0) { + return; + } + dword = Get_NB32(pDCTstat->dev_nbctl, (0x160 + (nb_pstate * 4))); /* Retrieve NbDid, NbFid */ + nb_clk = (200 * (((dword >> 1) & 0x1f) + 0x4)) / (((dword >> 7) & 0x1)?2:1); + + pDCTstat->CH_MaxRdLat[Channel]++; + pDCTstat->CH_MaxRdLat[Channel] += ((((uint64_t)15 * 100000000000ULL) / ((uint64_t)fam15h_freq_tab[mem_clk] * 1000000ULL)) + * ((uint64_t)nb_clk * 1000)) / 1000000000ULL; + + write_max_read_latency_to_registers(pMCTstat, pDCTstat, Channel, pDCTstat->CH_MaxRdLat[Channel]); + } + + if(_DisableDramECC) { + mct_EnableDimmEccEn_D(pMCTstat, pDCTstat, _DisableDramECC); + } + + if(!_Wrap32Dis) { + msr = HWCR; + _RDMSR(msr, &lo, &hi); + lo &= ~(1<<17); /* restore HWCR.wrap32dis */ + _WRMSR(msr, lo, hi); + } + if(!_SSE2){ + cr4 = read_cr4(); + cr4 &= ~(1<<9); /* restore cr4.OSFXSR */ + write_cr4(cr4); + } + +#if DQS_TRAIN_DEBUG > 0 + { + u8 ChannelDTD; + printk(BIOS_DEBUG, "TrainMaxRdLatency: CH_MaxRdLat:\n"); + for(ChannelDTD = 0; ChannelDTD<2; ChannelDTD++) { + printk(BIOS_DEBUG, "Channel:%x: %x\n", + ChannelDTD, pDCTstat->CH_MaxRdLat[ChannelDTD]); + } + } +#endif + + printk(BIOS_DEBUG, "TrainMaxRdLatency: Status %x\n", pDCTstat->Status); + printk(BIOS_DEBUG, "TrainMaxRdLatency: ErrStatus %x\n", pDCTstat->ErrStatus); + printk(BIOS_DEBUG, "TrainMaxRdLatency: ErrCode %x\n", pDCTstat->ErrCode); + printk(BIOS_DEBUG, "TrainMaxRdLatency: Done\n\n"); +} + u8 mct_InitReceiver_D(struct DCTStatStruc *pDCTstat, u8 dct) { if (pDCTstat->DIMMValidDCT[dct] == 0 ) { diff --git a/src/northbridge/amd/amdmct/mct_ddr3/mctwl.c b/src/northbridge/amd/amdmct/mct_ddr3/mctwl.c index 4bfcc40..b354d92 100644 --- a/src/northbridge/amd/amdmct/mct_ddr3/mctwl.c +++ b/src/northbridge/amd/amdmct/mct_ddr3/mctwl.c @@ -168,6 +168,8 @@ static void EnterSelfRefresh(struct MCTStatStruc *pMCTstat, static void ChangeMemClk(struct MCTStatStruc *pMCTstat, struct DCTStatStruc *pDCTstat) { + printk(BIOS_DEBUG, "%s: Start\n", __func__); + uint8_t DCT0Present; uint8_t DCT1Present; uint32_t dword; @@ -309,6 +311,8 @@ static void ChangeMemClk(struct MCTStatStruc *pMCTstat, mct_Wait(15000); /* Wait for 750us */ } } + + printk(BIOS_DEBUG, "%s: Done\n", __func__); } /*

1 0

Patch set updated for coreboot: cpu/amd/family_10h-family_15h: Set up link XCS token counts on Family 15h
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12033 -gerrit commit a0db42c8434e6852db25ef3702059dafb5dbe12b Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 23:59:17 2015 -0500 cpu/amd/family_10h-family_15h: Set up link XCS token counts on Family 15h Change-Id: I4cf6549234041c395a18a89332d95f20a596fc3e Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/init_cpus.c | 304 ++++++++++++++++++++++++++ 1 file changed, 304 insertions(+) diff --git a/src/cpu/amd/family_10h-family_15h/init_cpus.c b/src/cpu/amd/family_10h-family_15h/init_cpus.c index 5d22c00..34a67da 100644 --- a/src/cpu/amd/family_10h-family_15h/init_cpus.c +++ b/src/cpu/amd/family_10h-family_15h/init_cpus.c @@ -1054,6 +1054,12 @@ static void cpuSetAMDPCI(u8 node) uint32_t dword; uint64_t revision; + /* FIXME + * This should be configurable + */ + uint8_t sockets = 2; + uint8_t sockets_populated = 2; + printk(BIOS_DEBUG, "cpuSetAMDPCI %02d", node); revision = mctGetLogicalCPUID(node); @@ -1160,6 +1166,15 @@ static void cpuSetAMDPCI(u8 node) uint8_t compute_unit_count = 0; uint8_t compute_unit_buffer_count; + uint32_t f3xe8; + uint8_t dual_node = 0; + + f3xe8 = pci_read_config32(NODE_PCI(0, 3), 0xe8); + + /* Check for dual node capability */ + if (f3xe8 & 0x20000000) + dual_node = 1; + /* Determine the number of active compute units on this node */ f5x80 = pci_read_config32(NODE_PCI(node, 5), 0x80); cu_enabled = f5x80 & 0xf; @@ -1185,6 +1200,295 @@ static void cpuSetAMDPCI(u8 node) dword &= ~(0x1f << 4); /* L3FreeListCBC = compute_unit_buffer_count */ dword |= (compute_unit_buffer_count << 4); pci_write_config32(NODE_PCI(node, 3), 0x1a0, dword); + + /* Set up the Link to XCS Token Counts */ + uint8_t isoc_rsp_tok_1; + uint8_t isoc_preq_tok_1; + uint8_t isoc_req_tok_1; + uint8_t probe_tok_1; + uint8_t rsp_tok_1; + uint8_t preq_tok_1; + uint8_t req_tok_1; + uint8_t isoc_rsp_tok_0; + uint8_t isoc_preq_tok_0; + uint8_t isoc_req_tok_0; + uint8_t free_tokens; + uint8_t probe_tok_0; + uint8_t rsp_tok_0; + uint8_t preq_tok_0; + uint8_t req_tok_0; + + uint8_t link; + uint8_t ganged; + uint8_t iolink; + uint8_t probe_filter_enabled = !!dual_node; + for (link = 0; link < 4; link++) { + if (AMD_CpuFindCapability(node, link, &offset)) { + ganged = !!(pci_read_config32(NODE_PCI(node, 0), (link << 2) + 0x170) & 0x1); + iolink = (AMD_checkLinkType(node, link, offset) & HTPHY_LINKTYPE_NONCOHERENT); + + /* Set defaults */ + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = !ganged; + rsp_tok_1 = !ganged; + preq_tok_1 = !ganged; + req_tok_1 = !ganged; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 0; + free_tokens = 0; + probe_tok_0 = ((ganged)?2:1); + rsp_tok_0 = ((ganged)?2:1); + preq_tok_0 = ((ganged)?2:1); + req_tok_0 = ((ganged)?2:1); + + if (!iolink && ganged) { + if (!dual_node) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 3; + probe_tok_0 = 2; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } else { + if ((sockets == 1) + || ((sockets == 2) && (sockets_populated == 1))) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 2; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } else if (((sockets == 2) && (sockets_populated == 2)) + || ((sockets == 4) && (sockets_populated == 2))) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 1; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } else if ((sockets == 4) && (sockets_populated == 4)) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 2; + rsp_tok_0 = 1; + preq_tok_0 = 1; + req_tok_0 = 2; + } + } + } else if (!iolink && !ganged) { + if ((sockets == 1) + || ((sockets == 2) && (sockets_populated == 1))) { + if (probe_filter_enabled) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 1; + rsp_tok_1 = 1; + preq_tok_1 = 1; + req_tok_1 = 1; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 1; + rsp_tok_0 = 2; + preq_tok_0 = 1; + req_tok_0 = 1; + } else { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 1; + rsp_tok_1 = 1; + preq_tok_1 = 1; + req_tok_1 = 1; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 1; + rsp_tok_0 = 1; + preq_tok_0 = 1; + req_tok_0 = 1; + } + } else if ((sockets == 2) && (sockets_populated == 2)) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 1; + probe_tok_1 = 1; + rsp_tok_1 = 1; + preq_tok_1 = 1; + req_tok_1 = 1; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 2; + probe_tok_0 = 1; + rsp_tok_0 = 1; + preq_tok_0 = 1; + req_tok_0 = 1; + } else if ((sockets == 4) && (sockets_populated == 2)) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 1; + probe_tok_1 = 1; + rsp_tok_1 = 1; + preq_tok_1 = 1; + req_tok_1 = 1; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 4; + probe_tok_0 = 1; + rsp_tok_0 = 1; + preq_tok_0 = 1; + req_tok_0 = 1; + } else if ((sockets == 4) && (sockets_populated == 4)) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 1; + probe_tok_1 = 1; + rsp_tok_1 = 1; + preq_tok_1 = 1; + req_tok_1 = 1; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 1; + rsp_tok_0 = 1; + preq_tok_0 = 1; + req_tok_0 = 1; + } + } else if (iolink && ganged) { + if (!dual_node) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 3; + probe_tok_0 = 0; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } else if ((sockets == 1) + || (sockets == 2) + || ((sockets == 4) && (sockets_populated == 2))) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 1; + free_tokens = 0; + probe_tok_0 = 0; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } else if ((sockets == 4) && (sockets_populated == 4)) { + isoc_rsp_tok_1 = 0; + isoc_preq_tok_1 = 0; + isoc_req_tok_1 = 0; + probe_tok_1 = 0; + rsp_tok_1 = 0; + preq_tok_1 = 0; + req_tok_1 = 0; + isoc_rsp_tok_0 = 0; + isoc_preq_tok_0 = 0; + isoc_req_tok_0 = 2; + free_tokens = 0; + probe_tok_0 = 2; + rsp_tok_0 = 2; + preq_tok_0 = 2; + req_tok_0 = 2; + } + } + + dword = pci_read_config32(NODE_PCI(node, 3), (link << 2) + 0x148); + dword &= ~(0x3 << 30); /* FreeTok[3:2] = free_tokens[3:2] */ + dword |= (((free_tokens >> 2) & 0x3) << 30); + dword &= ~(0x1 << 28); /* IsocRspTok1 = isoc_rsp_tok_1 */ + dword |= (((isoc_rsp_tok_1) & 0x1) << 28); + dword &= ~(0x1 << 26); /* IsocPreqTok1 = isoc_preq_tok_1 */ + dword |= (((isoc_preq_tok_1) & 0x1) << 26); + dword &= ~(0x1 << 24); /* IsocReqTok1 = isoc_req_tok_1 */ + dword |= (((isoc_req_tok_1) & 0x1) << 24); + dword &= ~(0x3 << 22); /* ProbeTok1 = probe_tok_1 */ + dword |= (((probe_tok_1) & 0x3) << 22); + dword &= ~(0x3 << 20); /* RspTok1 = rsp_tok_1 */ + dword |= (((rsp_tok_1) & 0x3) << 20); + dword &= ~(0x3 << 18); /* PReqTok1 = preq_tok_1 */ + dword |= (((preq_tok_1) & 0x3) << 18); + dword &= ~(0x3 << 16); /* ReqTok1 = req_tok_1 */ + dword |= (((req_tok_1) & 0x3) << 16); + dword &= ~(0x3 << 14); /* FreeTok[1:0] = free_tokens[1:0] */ + dword |= (((free_tokens) & 0x3) << 14); + dword &= ~(0x3 << 12); /* IsocRspTok0 = isoc_rsp_tok_0 */ + dword |= (((isoc_rsp_tok_0) & 0x3) << 12); + dword &= ~(0x3 << 10); /* IsocPreqTok0 = isoc_preq_tok_0 */ + dword |= (((isoc_preq_tok_0) & 0x3) << 10); + dword &= ~(0x3 << 8); /* IsocReqTok0 = isoc_req_tok_0 */ + dword |= (((isoc_req_tok_0) & 0x3) << 8); + dword &= ~(0x3 << 6); /* ProbeTok0 = probe_tok_0 */ + dword |= (((probe_tok_0) & 0x3) << 6); + dword &= ~(0x3 << 4); /* RspTok0 = rsp_tok_0 */ + dword |= (((rsp_tok_0) & 0x3) << 4); + dword &= ~(0x3 << 2); /* PReqTok0 = preq_tok_0 */ + dword |= (((preq_tok_0) & 0x3) << 2); + dword &= ~(0x3 << 0); /* ReqTok0 = req_tok_0 */ + dword |= (((req_tok_0) & 0x3) << 0); + pci_write_config32(NODE_PCI(node, 3), (link << 2) + 0x148, dword); + } + } } printk(BIOS_DEBUG, " done\n");

1 0

Patch set updated for coreboot: amd/family_10h-family_15h: Fix poor performance on Family 15h CPUs
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12028 -gerrit commit b7b861816d08a4d03bbf6dd3323ea5c16b6f5b50 Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 19:04:49 2015 -0500 amd/family_10h-family_15h: Fix poor performance on Family 15h CPUs Change-Id: Ieb1f1fb5653651c98764de79636669802578d5f9 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/defaults.h | 13 ++++++-- src/cpu/amd/family_10h-family_15h/init_cpus.c | 45 +++++++++++++++++++++++++-- 2 files changed, 52 insertions(+), 6 deletions(-) diff --git a/src/cpu/amd/family_10h-family_15h/defaults.h b/src/cpu/amd/family_10h-family_15h/defaults.h index 3e9ec09..901b8ce 100644 --- a/src/cpu/amd/family_10h-family_15h/defaults.h +++ b/src/cpu/amd/family_10h-family_15h/defaults.h @@ -518,15 +518,15 @@ static const struct { { 3, 0x140, AMD_FAM10_ALL, AMD_PTYPE_ALL, 0x00800756, 0x00F3FFFF }, - { 3, 0x140, AMD_FAM15_ALL, AMD_PTYPE_ALL, - 0x00a11755, 0x00f3ffff }, - { 3, 0x140, AMD_FAM10_ALL, AMD_PTYPE_UMA, 0x00C37756, 0x00F3FFFF }, { 3, 0x144, AMD_FAM10_ALL, AMD_PTYPE_UMA, 0x00000036, 0x000000FF }, + { 3, 0x140, AMD_FAM15_ALL, AMD_PTYPE_ALL, + 0x00a11755, 0x00f3ffff }, + /* Errata 281 Workaround */ { 3, 0x144, ( AMD_DR_B0 | AMD_DR_B1), AMD_PTYPE_SVR, 0x00000001, 0x0000000F }, @@ -538,6 +538,13 @@ static const struct { { 3, 0x148, AMD_FAM10_ALL, AMD_PTYPE_UMA, 0x8000052A, 0xD5FFFFFF }, + /* Core Interface Buffer Count */ + { 3, 0x1a0, AMD_FAM15_ALL, AMD_PTYPE_ALL, + 0x00034004, 0x00037007 }, /* CpuToNbFreeBufCnt = 0x3, + L3ToSriReqCBC = 0x4, + L3FreeListCBC = default, + CpuCmdBufCnt = 0x4 */ + /* ACPI Power State Control Reg1 */ { 3, 0x80, AMD_FAM10_ALL, AMD_PTYPE_ALL, 0xE6002200, 0xFFFFFFFF }, diff --git a/src/cpu/amd/family_10h-family_15h/init_cpus.c b/src/cpu/amd/family_10h-family_15h/init_cpus.c index 530ffd2..5d22c00 100644 --- a/src/cpu/amd/family_10h-family_15h/init_cpus.c +++ b/src/cpu/amd/family_10h-family_15h/init_cpus.c @@ -844,8 +844,9 @@ static BOOL AMD_CpuFindCapability(u8 node, u8 cap_count, u8 * offset) */ static u32 AMD_checkLinkType(u8 node, u8 link, u8 regoff) { - u32 val; - u32 linktype = 0; + uint32_t val; + uint32_t val2; + uint32_t linktype = 0; /* Check connect, init and coherency */ val = pci_read_config32(NODE_PCI(node, 0), regoff + 0x18); @@ -860,8 +861,13 @@ static u32 AMD_checkLinkType(u8 node, u8 link, u8 regoff) if (linktype) { /* Check gen3 */ val = pci_read_config32(NODE_PCI(node, 0), regoff + 0x08); + val = (val >> 8) & 0xf; + if (is_gt_rev_d()) { + val2 = pci_read_config32(NODE_PCI(node, 0), regoff + 0x1c); + val |= (val2 & 0x1) << 4; + } - if (((val >> 8) & 0x0F) > 6) + if (val > 6) linktype |= HTPHY_LINKTYPE_HT3; else linktype |= HTPHY_LINKTYPE_HT1; @@ -1148,6 +1154,39 @@ static void cpuSetAMDPCI(u8 node) pci_write_config32(NODE_PCI(node, 3), 0xd4, dword); } + if (revision & AMD_FAM15_ALL) { + uint32_t f5x80; + uint8_t cu_enabled; + uint8_t compute_unit_count = 0; + uint8_t compute_unit_buffer_count; + + /* Determine the number of active compute units on this node */ + f5x80 = pci_read_config32(NODE_PCI(node, 5), 0x80); + cu_enabled = f5x80 & 0xf; + if (cu_enabled == 0x1) + compute_unit_count = 1; + if (cu_enabled == 0x3) + compute_unit_count = 2; + if (cu_enabled == 0x7) + compute_unit_count = 3; + if (cu_enabled == 0xf) + compute_unit_count = 4; + + if (compute_unit_count == 1) + compute_unit_buffer_count = 0x1c; + else if (compute_unit_count == 2) + compute_unit_buffer_count = 0x18; + else if (compute_unit_count == 3) + compute_unit_buffer_count = 0x14; + else + compute_unit_buffer_count = 0x10; + + dword = pci_read_config32(NODE_PCI(node, 3), 0x1a0); + dword &= ~(0x1f << 4); /* L3FreeListCBC = compute_unit_buffer_count */ + dword |= (compute_unit_buffer_count << 4); + pci_write_config32(NODE_PCI(node, 3), 0x1a0, dword); + } + printk(BIOS_DEBUG, " done\n"); }

1 0

Patch set updated for coreboot: cpu/amd/family_10h-family_15h: Configure NB register 2
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12032 -gerrit commit 1b081cac4b3c73b313d2cfa3e9cdeff2f45d63fb Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Fri Aug 7 23:58:28 2015 -0500 cpu/amd/family_10h-family_15h: Configure NB register 2 Change-Id: I55cfc96a197514212b2a4c344d3513396ebc2ad4 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/defaults.h | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/src/cpu/amd/family_10h-family_15h/defaults.h b/src/cpu/amd/family_10h-family_15h/defaults.h index 901b8ce..3fd0f4b 100644 --- a/src/cpu/amd/family_10h-family_15h/defaults.h +++ b/src/cpu/amd/family_10h-family_15h/defaults.h @@ -604,6 +604,14 @@ static const struct { [5] DisPciCfgCpuMstAbtRsp = 1, [1] SyncFloodOnUsPwDataErr = 1 */ + /* NB Configuration 2 */ + { 3, 0x188, AMD_DR_GT_B0, AMD_PTYPE_ALL, + 0x00000010, 0x00000010 }, /* EnStpGntOnFlushMaskWakeup = 0x1 */ + + /* NB Configuration 2 */ + { 3, 0x188, AMD_FAM15_ALL, AMD_PTYPE_ALL, + 0x00000200, 0x00000200 }, /* DisL3HiPriFreeListAlloc = 0x1 */ + /* errata 346 - Fam10 C2, C3 * System software should set F3x188[22] to 1b. */ { 3, 0x188, AMD_DR_Cx, AMD_PTYPE_ALL,

1 0

Patch set updated for coreboot: northbridge/amd/amdht: Add support for HT3 2.8GHz and up link frequencies
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12027 -gerrit commit 5a66c703b8504846a92f6b94a8cd1c833d051917 Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Sun Aug 2 21:36:24 2015 -0500 northbridge/amd/amdht: Add support for HT3 2.8GHz and up link frequencies Change-Id: Ifa1592d26ba7deb034046fd3f2a15149117d9a76 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/defaults.h | 8 ++- src/mainboard/asus/kgpe-d16/cmos.layout | 30 ++++----- src/northbridge/amd/amdht/h3ffeat.h | 6 +- src/northbridge/amd/amdht/h3finit.c | 93 +++++++++++++++++----------- src/northbridge/amd/amdht/h3finit.h | 18 ++++-- src/northbridge/amd/amdht/h3ncmn.c | 82 +++++++++++++++++------- src/northbridge/amd/amdht/h3ncmn.h | 3 +- 7 files changed, 155 insertions(+), 85 deletions(-) diff --git a/src/cpu/amd/family_10h-family_15h/defaults.h b/src/cpu/amd/family_10h-family_15h/defaults.h index a3c7b1f..3e9ec09 100644 --- a/src/cpu/amd/family_10h-family_15h/defaults.h +++ b/src/cpu/amd/family_10h-family_15h/defaults.h @@ -273,7 +273,13 @@ static const struct { /* Link Global Retry Control Register */ { 0, 0x150, (AMD_FAM10_ALL | AMD_FAM15_ALL), AMD_PTYPE_ALL, - 0x00073900, 0x00073F00 }, + 0x00073900, 0x00073f70 }, /* TotalRetryAttempts = 0x7, + HtRetryCrcDatInsDynEn = 0x1, + HtRetryCrcCmdPackDynEn = 0x1, + HtRetryCrcDatIns = 0x4, + HtRetryCrcCmdPack = 0x1, + ForceErrType = 0x0, + MultRetryErr = 0x0 */ /* Errata 351 * System software should program the Link Extended Control Registers[LS2En] diff --git a/src/mainboard/asus/kgpe-d16/cmos.layout b/src/mainboard/asus/kgpe-d16/cmos.layout index 21765e3..e17fec6 100644 --- a/src/mainboard/asus/kgpe-d16/cmos.layout +++ b/src/mainboard/asus/kgpe-d16/cmos.layout @@ -120,21 +120,21 @@ enumerations 10 21 42ms 10 22 84ms 11 0 Auto -11 1 2.6GHz -11 2 2.4GHz -11 3 2.2GHz -11 4 2.0GHz -11 5 1.8GHz -11 6 1.6GHz -11 7 1.4GHz -11 8 1.2GHz -11 9 1.0GHz -11 10 800MHz -11 11 600MHz -11 12 500MHz -11 13 400MHz -11 14 300MHz -11 15 200MHz +11 1 3.2GHz +11 2 3.0GHz +11 3 2.8GHz +11 4 2.6GHz +11 5 2.4GHz +11 6 2.2GHz +11 7 2.0GHz +11 8 1.8GHz +11 9 1.6GHz +11 10 1.4GHz +11 11 1.2GHz +11 12 1.0GHz +11 13 800MHz +11 14 600MHz +11 15 500MHz 12 0 1.5V 12 1 1.35V 12 2 1.25V diff --git a/src/northbridge/amd/amdht/h3ffeat.h b/src/northbridge/amd/amdht/h3ffeat.h index 52c4def..628b86e 100644 --- a/src/northbridge/amd/amdht/h3ffeat.h +++ b/src/northbridge/amd/amdht/h3ffeat.h @@ -2,6 +2,7 @@ * This file is part of the coreboot project. * * Copyright (C) 2007-2008 Advanced Micro Devices, Inc. + * Copyright (C) 2015 Timothy Pearson <tpearson(a)raptorengineeringinc.com>, Raptor Engineering * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -58,6 +59,7 @@ #define HTHOST_LINK_CAPABILITY_REG 0x00 #define HTHOST_LINK_CONTROL_REG 0x04 #define HTHOST_FREQ_REV_REG 0x08 +#define HTHOST_FREQ_REV_REG_2 0x1c #define HT_HOST_REV_REV3 0x60 #define HTHOST_FEATURE_CAP_REG 0x0C #define HTHOST_BUFFER_COUNT_REG 0x10 @@ -123,10 +125,10 @@ typedef struct /* This section is for keeping track of capabilities and possible configurations */ BOOL RegangCap; - u16 PrvFrequencyCap; + uint32_t PrvFrequencyCap; u8 PrvWidthInCap; u8 PrvWidthOutCap; - u16 CompositeFrequencyCap; + uint32_t CompositeFrequencyCap; } sPortDescriptor; diff --git a/src/northbridge/amd/amdht/h3finit.c b/src/northbridge/amd/amdht/h3finit.c index bcdcfa2..e4ce768 100644 --- a/src/northbridge/amd/amdht/h3finit.c +++ b/src/northbridge/amd/amdht/h3finit.c @@ -48,28 +48,32 @@ #define APIC_Base_BSP 8 #define APIC_Base 0x1b -#define NVRAM_LIMIT_HT_SPEED_200 0xf -#define NVRAM_LIMIT_HT_SPEED_300 0xe -#define NVRAM_LIMIT_HT_SPEED_400 0xd -#define NVRAM_LIMIT_HT_SPEED_500 0xc -#define NVRAM_LIMIT_HT_SPEED_600 0xb -#define NVRAM_LIMIT_HT_SPEED_800 0xa -#define NVRAM_LIMIT_HT_SPEED_1000 0x9 -#define NVRAM_LIMIT_HT_SPEED_1200 0x8 -#define NVRAM_LIMIT_HT_SPEED_1400 0x7 -#define NVRAM_LIMIT_HT_SPEED_1600 0x6 -#define NVRAM_LIMIT_HT_SPEED_1800 0x5 -#define NVRAM_LIMIT_HT_SPEED_2000 0x4 -#define NVRAM_LIMIT_HT_SPEED_2200 0x3 -#define NVRAM_LIMIT_HT_SPEED_2400 0x2 -#define NVRAM_LIMIT_HT_SPEED_2600 0x1 +#define NVRAM_LIMIT_HT_SPEED_200 0x12 +#define NVRAM_LIMIT_HT_SPEED_300 0x11 +#define NVRAM_LIMIT_HT_SPEED_400 0x10 +#define NVRAM_LIMIT_HT_SPEED_500 0xf +#define NVRAM_LIMIT_HT_SPEED_600 0xe +#define NVRAM_LIMIT_HT_SPEED_800 0xd +#define NVRAM_LIMIT_HT_SPEED_1000 0xc +#define NVRAM_LIMIT_HT_SPEED_1200 0xb +#define NVRAM_LIMIT_HT_SPEED_1400 0xa +#define NVRAM_LIMIT_HT_SPEED_1600 0x9 +#define NVRAM_LIMIT_HT_SPEED_1800 0x8 +#define NVRAM_LIMIT_HT_SPEED_2000 0x7 +#define NVRAM_LIMIT_HT_SPEED_2200 0x6 +#define NVRAM_LIMIT_HT_SPEED_2400 0x5 +#define NVRAM_LIMIT_HT_SPEED_2600 0x4 +#define NVRAM_LIMIT_HT_SPEED_2800 0x3 +#define NVRAM_LIMIT_HT_SPEED_3000 0x2 +#define NVRAM_LIMIT_HT_SPEED_3200 0x1 #define NVRAM_LIMIT_HT_SPEED_AUTO 0x0 -static const uint16_t ht_speed_limit[16] = - {0xFFFF, 0x7FFF, 0x3FFF, 0x1FFF, - 0x0FFF, 0x07FF, 0x03FF, 0x01FF, - 0x00FF, 0x007F, 0x003F, 0x001F, - 0x000F, 0x0007, 0x0003, 0x0001}; +static const uint32_t ht_speed_limit[20] = + {0xFFFFF, 0xFFFFF, 0x7FFFF, 0x3FFFF, + 0x0FFFF, 0x07FFF, 0x03FFF, 0x01FFF, + 0x00FFF, 0x007FF, 0x003FF, 0x001FF, + 0x000FF, 0x0007F, 0x0003F, 0x0001F, + 0x0000F, 0x00007, 0x00003, 0x00001}; static const struct ht_speed_limit_map_t { uint16_t mhz; @@ -91,9 +95,12 @@ static const struct ht_speed_limit_map_t { {2200, NVRAM_LIMIT_HT_SPEED_2200}, {2400, NVRAM_LIMIT_HT_SPEED_2400}, {2600, NVRAM_LIMIT_HT_SPEED_2600}, + {2800, NVRAM_LIMIT_HT_SPEED_2800}, + {3000, NVRAM_LIMIT_HT_SPEED_3000}, + {3200, NVRAM_LIMIT_HT_SPEED_3200}, }; -static const uint16_t ht_speed_mhz_to_hw(uint16_t mhz) +static const uint32_t ht_speed_mhz_to_hw(uint16_t mhz) { size_t i; for (i = 0; i < ARRAY_SIZE(ht_speed_limit_map); i++) @@ -452,7 +459,7 @@ static void htDiscoveryFloodFill(sMainData *pDat) /* Set currentNode's NodeID field to currentNode */ pDat->nb->writeNodeID(currentNode, currentNode, pDat->nb); - /* Enable routing tables on currentNode*/ + /* Enable routing tables on currentNode */ pDat->nb->enableRoutingTables(currentNode, pDat->nb); for (currentLinkID = 0; currentLinkID < pDat->nb->maxLinks; currentLinkID++) @@ -1427,19 +1434,30 @@ static void regangLinks(sMainData *pDat) static void selectOptimalWidthAndFrequency(sMainData *pDat) { u8 i, j; - u32 temp; - u16 cbPCBFreqLimit; - u16 cbPCBFreqLimit_NVRAM; + uint32_t temp; + uint32_t cbPCBFreqLimit; + uint32_t cbPCBFreqLimit_NVRAM; u8 cbPCBABDownstreamWidth; u8 cbPCBBAUpstreamWidth; - cbPCBFreqLimit_NVRAM = 0xFFFF; + cbPCBFreqLimit_NVRAM = 0xfffff; if (get_option(&temp, "hypertransport_speed_limit") == CB_SUCCESS) cbPCBFreqLimit_NVRAM = ht_speed_limit[temp & 0xf]; + if (!is_fam15h()) { + /* FIXME + * By default limit frequency to 2.6 GHz as there are residual + * problems with HT v3.1 implementation on at least some Socket G34 + * mainboards / Fam10h CPUs. + * Debug the issues and reenable this... + */ + if (cbPCBFreqLimit_NVRAM > 0xffff) + cbPCBFreqLimit_NVRAM = 0xffff; + } + for (i = 0; i < pDat->TotalLinks*2; i += 2) { - cbPCBFreqLimit = 0xFFFF; // Maximum allowed by autoconfiguration + cbPCBFreqLimit = 0xfffff; // Maximum allowed by autoconfiguration if (pDat->HtBlock->ht_link_configuration) cbPCBFreqLimit = ht_speed_mhz_to_hw(pDat->HtBlock->ht_link_configuration->ht_speed_limit); cbPCBFreqLimit = min(cbPCBFreqLimit, cbPCBFreqLimit_NVRAM); @@ -1484,17 +1502,18 @@ static void selectOptimalWidthAndFrequency(sMainData *pDat) } } - temp = pDat->PortList[i].PrvFrequencyCap; temp &= pDat->PortList[i+1].PrvFrequencyCap; temp &= cbPCBFreqLimit; - pDat->PortList[i].CompositeFrequencyCap = (u16)temp; - pDat->PortList[i+1].CompositeFrequencyCap = (u16)temp; + pDat->PortList[i].CompositeFrequencyCap = temp; + pDat->PortList[i+1].CompositeFrequencyCap = temp; ASSERT (temp != 0); - for (j = 15; ; j--) + for (j = 19; ; j--) { - if (temp & ((u32)1 << j)) + if ((j == 16) || (j == 15)) + continue; + if (temp & ((uint32_t)1 << j)) break; } @@ -1638,12 +1657,14 @@ static void hammerSublinkFixup(sMainData *pDat) /* Remove hiFreq from the list of valid frequencies */ temp = temp & ~((uint32)1 << hiFreq); ASSERT (temp != 0); - pDat->PortList[hiIndex].CompositeFrequencyCap = (uint16)temp; - pDat->PortList[hiIndex+1].CompositeFrequencyCap = (uint16)temp; + pDat->PortList[hiIndex].CompositeFrequencyCap = temp; + pDat->PortList[hiIndex+1].CompositeFrequencyCap = temp; - for (k = 15; ; k--) + for (k = 19; ; k--) { - if (temp & ((u32)1 << k)) + if ((j == 16) || (j == 15)) + continue; + if (temp & ((uint32_t)1 << k)) break; } diff --git a/src/northbridge/amd/amdht/h3finit.h b/src/northbridge/amd/amdht/h3finit.h index f6ab010..1bd8616 100644 --- a/src/northbridge/amd/amdht/h3finit.h +++ b/src/northbridge/amd/amdht/h3finit.h @@ -49,6 +49,9 @@ #define HT_FREQUENCY_2200M 12 #define HT_FREQUENCY_2400M 13 #define HT_FREQUENCY_2600M 14 +#define HT_FREQUENCY_2800M 17 +#define HT_FREQUENCY_3000M 18 +#define HT_FREQUENCY_3200M 19 /* Frequency Limit equates for call backs which take a frequency supported mask. */ #define HT_FREQUENCY_LIMIT_200M 1 @@ -65,6 +68,9 @@ #define HT_FREQUENCY_LIMIT_2200M 0x1FFF #define HT_FREQUENCY_LIMIT_2400M 0x3FFF #define HT_FREQUENCY_LIMIT_2600M 0x7FFF +#define HT_FREQUENCY_LIMIT_2800M 0x3FFFF +#define HT_FREQUENCY_LIMIT_3000M 0x7FFFF +#define HT_FREQUENCY_LIMIT_3200M 0xFFFFF /* * Event Notify definitions @@ -220,7 +226,7 @@ typedef struct { * @param[in] u8 Link = The Device's link number (0 or 1) * @param[in,out] u8* LinkWidthIn = modify to change the Link Witdh In * @param[in,out] u8* LinkWidthOut = modify to change the Link Witdh Out - * @param[in,out] u16* FreqCap = modify to change the link's frequency capability + * @param[in,out] u32* FreqCap = modify to change the link's frequency capability * * --------------------------------------------------------------------------------------- */ @@ -235,7 +241,7 @@ typedef struct { u8 Link, u8 *LinkWidthIn, u8 *LinkWidthOut, - u16 *FreqCap + u32 *FreqCap ); /**---------------------------------------------------------------------------------------- @@ -258,7 +264,7 @@ typedef struct { * @param[in] u8 linkB = The link on that node * @param[in,out] u8* ABLinkWidthLimit = modify to change the Link Witdh In * @param[in,out] u8* BALinkWidthLimit = modify to change the Link Witdh Out - * @param[in,out] u16* PCBFreqCap = modify to change the link's frequency capability + * @param[in,out] u32* PCBFreqCap = modify to change the link's frequency capability * * --------------------------------------------------------------------------------------- */ @@ -269,7 +275,7 @@ typedef struct { u8 LinkB, u8 *ABLinkWidthLimit, u8 *BALinkWidthLimit, - u16 *PCBFreqCap + u32 *PCBFreqCap ); /**---------------------------------------------------------------------------------------- @@ -291,7 +297,7 @@ typedef struct { * @param[in] u8 Depth = The depth in the I/O chain from the Host * @param[in,out] u8* DownstreamLinkWidthLimit = modify to change the Link Witdh In * @param[in,out] u8* UpstreamLinkWidthLimit = modify to change the Link Witdh Out - * @param[in,out] u16* PCBFreqCap = modify to change the link's frequency capability + * @param[in,out] u32* PCBFreqCap = modify to change the link's frequency capability * * --------------------------------------------------------------------------------------- */ @@ -301,7 +307,7 @@ typedef struct { u8 Depth, u8 *DownstreamLinkWidthLimit, u8 *UpstreamLinkWidthLimit, - u16 *PCBFreqCap + u32 *PCBFreqCap ); /**---------------------------------------------------------------------------------------- diff --git a/src/northbridge/amd/amdht/h3ncmn.c b/src/northbridge/amd/amdht/h3ncmn.c index 8ee929d..29524af 100644 --- a/src/northbridge/amd/amdht/h3ncmn.c +++ b/src/northbridge/amd/amdht/h3ncmn.c @@ -1310,7 +1310,7 @@ static u8 convertWidthToBits(u8 value, cNorthBridge *nb) * @return Frequency mask * ******************************************************************************/ -static u16 ht1NorthBridgeFreqMask(u8 node, cNorthBridge *nb) +static uint32_t ht1NorthBridgeFreqMask(u8 node, cNorthBridge *nb) { /* only up to HT1 speeds */ return (HT_FREQUENCY_LIMIT_HT1_ONLY); @@ -1331,26 +1331,43 @@ static u16 ht1NorthBridgeFreqMask(u8 node, cNorthBridge *nb) * @return = Frequency mask * ******************************************************************************/ -static u16 fam10NorthBridgeFreqMask(u8 node, cNorthBridge *nb) +static uint32_t fam10NorthBridgeFreqMask(u8 node, cNorthBridge *nb) { u8 nbCOF; - u16 supported; + uint32_t supported; nbCOF = getMinNbCOF(); /* * nbCOF is minimum northbridge speed in hundreds of MHz. * HT can not go faster than the minimum speed of the northbridge. */ - if ((nbCOF >= 6) && (nbCOF <= 26)) + if ((nbCOF >= 6) && (nbCOF < 10)) { + /* Generation 1 HT link frequency */ /* Convert frequency to bit and all less significant bits, * by setting next power of 2 and subtracting 1. */ - supported = ((u16)1 << ((nbCOF >> 1) + 2)) - 1; + supported = ((uint32_t)1 << ((nbCOF >> 1) + 2)) - 1; } - else if (nbCOF > 26) + else if ((nbCOF >= 10) && (nbCOF <= 32)) { - supported = HT_FREQUENCY_LIMIT_2600M; + /* Generation 3 HT link frequency + * Assume error retry is enabled on all Gen 3 links + */ + if (is_gt_rev_d()) { + nbCOF *= 2; + if (nbCOF > 32) + nbCOF = 32; + } + + /* Convert frequency to bit and all less significant bits, + * by setting next power of 2 and subtracting 1. + */ + supported = ((uint32_t)1 << ((nbCOF >> 1) + 2)) - 1; + } + else if (nbCOF > 32) + { + supported = HT_FREQUENCY_LIMIT_3200M; } /* unlikely cases, but include as a defensive measure, also avoid trick above */ else if (nbCOF == 4) @@ -1405,8 +1422,13 @@ static void gatherLinkData(sMainData *pDat, cNorthBridge *nb) pDat->PortList[i].PrvWidthInCap = convertBitsToWidth((u8)temp, pDat->nb); AmdPCIReadBits(linkBase + HTHOST_FREQ_REV_REG, 31, 16, &temp); - pDat->PortList[i].PrvFrequencyCap = (u16)temp & 0x7FFF - & nb->northBridgeFreqMask(pDat->PortList[i].NodeID, pDat->nb); /* Mask off bit 15, reserved value */ + pDat->PortList[i].PrvFrequencyCap = temp & 0x7FFF /* Mask off bit 15, reserved value */ + & nb->northBridgeFreqMask(pDat->PortList[i].NodeID, pDat->nb); + if (is_gt_rev_d()) { + AmdPCIReadBits(linkBase + HTHOST_FREQ_REV_REG_2, 15, 1, &temp); + temp &= 0x7; /* Mask off reserved values */ + pDat->PortList[i].PrvFrequencyCap |= (temp << 17); + } } else { @@ -1463,7 +1485,7 @@ static void setLinkData(sMainData *pDat, cNorthBridge *nb) { u8 i; SBDFO linkBase; - u32 temp, widthin, widthout, bits; + u32 temp, temp2, frequency_index, widthin, widthout, bits; for (i = 0; i < pDat->TotalLinks*2; i++) { @@ -1524,10 +1546,19 @@ static void setLinkData(sMainData *pDat, cNorthBridge *nb) temp = pDat->PortList[i].SelFrequency; if (pDat->PortList[i].Type == PORTLIST_TYPE_CPU) { - ASSERT((temp >= HT_FREQUENCY_600M && temp <= HT_FREQUENCY_2600M) + ASSERT((temp >= HT_FREQUENCY_600M && temp <= HT_FREQUENCY_3200M) || (temp == HT_FREQUENCY_200M) || (temp == HT_FREQUENCY_400M)); + frequency_index = temp; + if (temp > 0xf) { + temp2 = (temp >> 4) & 0x1; + temp &= 0xf; + } else { + temp2 = 0x0; + } + if (is_gt_rev_d()) + AmdPCIWriteBits(linkBase + HTHOST_FREQ_REV_REG_2, 0, 0, &temp2); AmdPCIWriteBits(linkBase + HTHOST_FREQ_REV_REG, 11, 8, &temp); - if (temp > HT_FREQUENCY_1000M) /* Gen1 = 200MHz -> 1000MHz, Gen3 = 1200MHz -> 2600MHz */ + if (frequency_index > HT_FREQUENCY_1000M) /* Gen1 = 200MHz -> 1000MHz, Gen3 = 1200MHz -> 3200MHz */ { /* Enable for Gen3 frequencies */ temp = 1; @@ -1537,27 +1568,27 @@ static void setLinkData(sMainData *pDat, cNorthBridge *nb) /* Disable for Gen1 frequencies */ temp = 0; } - /* HT3 retry mode enable / disable */ - AmdPCIWriteBits(MAKE_SBDFO(makePCISegmentFromNode(pDat->PortList[i].NodeID), - makePCIBusFromNode(pDat->PortList[i].NodeID), - makePCIDeviceFromNode(pDat->PortList[i].NodeID), - CPU_HTNB_FUNC_00, - REG_HT_LINK_RETRY0_0X130 + 4*pDat->PortList[i].Link), - 0, 0, &temp); - /* and Scrambling enable / disable */ - AmdPCIWriteBits(MAKE_SBDFO(makePCISegmentFromNode(pDat->PortList[i].NodeID), + /* HT3 retry mode enable / disable */ + AmdPCIWriteBits(MAKE_SBDFO(makePCISegmentFromNode(pDat->PortList[i].NodeID), makePCIBusFromNode(pDat->PortList[i].NodeID), makePCIDeviceFromNode(pDat->PortList[i].NodeID), CPU_HTNB_FUNC_00, - REG_HT_LINK_EXT_CONTROL0_0X170 + 4*pDat->PortList[i].Link), - 3, 3, &temp); + REG_HT_LINK_RETRY0_0X130 + 4*pDat->PortList[i].Link), + 0, 0, &temp); + /* and Scrambling enable / disable */ + AmdPCIWriteBits(MAKE_SBDFO(makePCISegmentFromNode(pDat->PortList[i].NodeID), + makePCIBusFromNode(pDat->PortList[i].NodeID), + makePCIDeviceFromNode(pDat->PortList[i].NodeID), + CPU_HTNB_FUNC_00, + REG_HT_LINK_EXT_CONTROL0_0X170 + 4*pDat->PortList[i].Link), + 3, 3, &temp); } else { SBDFO currentPtr; BOOL isFound; - ASSERT(temp <= HT_FREQUENCY_2600M); + ASSERT(temp <= HT_FREQUENCY_3200M); /* Write the frequency setting */ AmdPCIWriteBits(linkBase + HTSLAVE_FREQ_REV_0_REG, 11, 8, &temp); @@ -1718,6 +1749,9 @@ static void fam0fWriteHTLinkCmdBufferAlloc(u8 node, u8 link, u8 req, u8 preq, u8 /* Probe Command Buffers */ temp = prb; AmdPCIWriteBits(currentPtr, 15, 12, &temp); + /* LockBc */ + temp = 1; + AmdPCIWriteBits(currentPtr, 31, 31, &temp); } #endif /* HT_BUILD_NC_ONLY */ diff --git a/src/northbridge/amd/amdht/h3ncmn.h b/src/northbridge/amd/amdht/h3ncmn.h index 328d909..3c8a346 100644 --- a/src/northbridge/amd/amdht/h3ncmn.h +++ b/src/northbridge/amd/amdht/h3ncmn.h @@ -2,6 +2,7 @@ * This file is part of the coreboot project. * * Copyright (C) 2007 Advanced Micro Devices, Inc. + * Copyright (C) 2015 Timothy Pearson <tpearson(a)raptorengineeringinc.com>, Raptor Engineering * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -103,7 +104,7 @@ struct cNorthBridge /* Public Interfaces for northbridge clients, Optimization */ u8 (*convertBitsToWidth)(u8 value, cNorthBridge *nb); u8 (*convertWidthToBits)(u8 value, cNorthBridge *nb); - u16 (*northBridgeFreqMask)(u8 node, cNorthBridge *nb); + uint32_t (*northBridgeFreqMask)(u8 node, cNorthBridge *nb); void (*gatherLinkData)(sMainData *pDat, cNorthBridge *nb); void (*setLinkData)(sMainData *pDat, cNorthBridge *nb);

1 0

Patch set updated for coreboot: cpu/amd/family_10h-family_15h: Fix incorrect revision detection
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12026 -gerrit commit 739627403f351c7d37e6b40079eb53293ebaa52a Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Sun Aug 2 21:31:48 2015 -0500 cpu/amd/family_10h-family_15h: Fix incorrect revision detection The revision detection code for AMD Family 10h/15h was modified to use a 64-bit value instead of 32-bit in order to accomodate additional processor revisions. The FIDVID code was not updated at that point, leading to incorrect revision use during FIDVID. Change-Id: I7a881a94d62ed455415f9dfc887fd698ac919429 Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/fidvid.c | 32 +++++++++++++++--------------- 1 file changed, 16 insertions(+), 16 deletions(-) diff --git a/src/cpu/amd/family_10h-family_15h/fidvid.c b/src/cpu/amd/family_10h-family_15h/fidvid.c index 8a006cb..3619a9c 100644 --- a/src/cpu/amd/family_10h-family_15h/fidvid.c +++ b/src/cpu/amd/family_10h-family_15h/fidvid.c @@ -371,9 +371,9 @@ static void recalculateVsSlamTimeSettingOnCorePre(device_t dev) pci_write_config32(dev, 0xd8, dtemp); } -static u32 nb_clk_did(int node, u32 cpuRev,u8 procPkg) { - u8 link0isGen3 = 0; - u8 offset; +static u32 nb_clk_did(uint8_t node, uint64_t cpuRev, uint8_t procPkg) { + uint8_t link0isGen3 = 0; + uint8_t offset; if (AMD_CpuFindCapability(node, 0, &offset)) { link0isGen3 = (AMD_checkLinkType(node, 0, offset) & HTPHY_LINKTYPE_HT3 ); } @@ -442,7 +442,7 @@ static u32 power_up_down(int node, u8 procPkg) { return dword; } -static void config_clk_power_ctrl_reg0(int node, u32 cpuRev, u8 procPkg) { +static void config_clk_power_ctrl_reg0(uint8_t node, uint64_t cpuRev, uint8_t procPkg) { device_t dev = NODE_PCI(node, 3); /* Program fields in Clock Power/Control register0 (F3xD4) */ @@ -456,7 +456,7 @@ static void config_clk_power_ctrl_reg0(int node, u32 cpuRev, u8 procPkg) { * ClkRampHystCtl=HW default * ClkRampHystSel=1111b */ - u32 dword= pci_read_config32(dev, 0xd4); + uint32_t dword= pci_read_config32(dev, 0xd4); dword &= CPTC0_MASK; dword |= NB_CLKDID_ALL | LNK_PLL_LOCK | CLK_RAMP_HYST_SEL_VAL; dword |= (nb_clk_did(node,cpuRev,procPkg) << NB_CLKDID_SHIFT); @@ -467,9 +467,9 @@ static void config_clk_power_ctrl_reg0(int node, u32 cpuRev, u8 procPkg) { } -static void config_power_ctrl_misc_reg(device_t dev,u32 cpuRev, u8 procPkg) { +static void config_power_ctrl_misc_reg(device_t dev, uint64_t cpuRev, uint8_t procPkg) { /* check PVI/SVI */ - u32 dword = pci_read_config32(dev, 0xa0); + uint32_t dword = pci_read_config32(dev, 0xa0); /* BKDG r31116 2010-04-22 2.4.1.7 step b F3xA0[VSSlamVidMod] */ /* PllLockTime and PsiVidEn set in ruleset in defaults.h */ @@ -500,14 +500,14 @@ static void config_power_ctrl_misc_reg(device_t dev,u32 cpuRev, u8 procPkg) { pci_write_config32(dev, 0xa0, dword); } -static void config_nb_syn_ptr_adj(device_t dev, u32 cpuRev) { +static void config_nb_syn_ptr_adj(device_t dev, uint64_t cpuRev) { /* Note the following settings are additional from the ported * function setFidVidRegs() */ /* adjust FIFO between nb and core clocks to max allowed values (min latency) */ - u32 nbPstate = pci_read_config32(dev,0x1f0) & NB_PSTATE_MASK; - u8 nbSynPtrAdj; + uint32_t nbPstate = pci_read_config32(dev,0x1f0) & NB_PSTATE_MASK; + uint8_t nbSynPtrAdj; if ((cpuRev & (AMD_DR_Bx | AMD_DA_Cx | AMD_FAM15_ALL) ) || ((cpuRev & AMD_RB_C3) && (nbPstate != 0))) { nbSynPtrAdj = 5; @@ -515,14 +515,14 @@ static void config_nb_syn_ptr_adj(device_t dev, u32 cpuRev) { nbSynPtrAdj = 6; } - u32 dword = pci_read_config32(dev, 0xdc); + uint32_t dword = pci_read_config32(dev, 0xdc); dword &= ~NB_SYN_PTR_ADJ_MASK; dword |= nbSynPtrAdj << NB_SYN_PTR_ADJ_POS; /* NbsynPtrAdj set to 5 or 6 per BKDG (needs reset) */ pci_write_config32(dev, 0xdc, dword); } -static void config_acpi_pwr_state_ctrl_regs(device_t dev, u32 cpuRev, u8 procPkg) { +static void config_acpi_pwr_state_ctrl_regs(device_t dev, uint64_t cpuRev, uint8_t procPkg) { if (is_fam15h()) { /* Family 15h BKDG Rev. 3.14 D18F3x80 recommended settings */ pci_write_config32(dev, 0x80, 0xe20be281); @@ -531,8 +531,8 @@ static void config_acpi_pwr_state_ctrl_regs(device_t dev, u32 cpuRev, u8 procPkg pci_write_config32(dev, 0x84, 0x01e200e2); } else { /* step 1, chapter 2.4.2.6 of AMD Fam 10 BKDG #31116 Rev 3.48 22.4.2010 */ - u32 dword; - u32 c1= 1; + uint32_t dword; + uint32_t c1= 1; if (cpuRev & (AMD_DR_Bx)) { // will coreboot ever enable cache scrubbing ? // if it does, will it be enough to check the current state @@ -565,7 +565,7 @@ static void config_acpi_pwr_state_ctrl_regs(device_t dev, u32 cpuRev, u8 procPkg * which is easier */ - u32 smaf001 = 0xE6; + uint32_t smaf001 = 0xE6; if (cpuRev & AMD_DR_Bx ) { smaf001 = 0xA6; } else { @@ -575,7 +575,7 @@ static void config_acpi_pwr_state_ctrl_regs(device_t dev, u32 cpuRev, u8 procPkg } #endif } - u32 fidvidChange = 0; + uint32_t fidvidChange = 0; if (((cpuRev & AMD_DA_Cx) && (procPkg & AMD_PKGTYPE_S1gX)) || (cpuRev & AMD_RB_C3) ) { fidvidChange=0x0B;

1 0

Patch set updated for coreboot: nb/amd/amdfam10: Add HyperTransport probe filter support
by Timothy Pearson 20 Nov '15

20 Nov '15

Timothy Pearson (tpearson(a)raptorengineeringinc.com) just uploaded a new patch set to gerrit, which you can find at http://review.coreboot.org/12021 -gerrit commit de87bb37a68caefd6b408c1818c058ec1fd6929a Author: Timothy Pearson <tpearson(a)raptorengineeringinc.com> Date: Sun Aug 2 21:06:39 2015 -0500 nb/amd/amdfam10: Add HyperTransport probe filter support All modern Opteron processors support the HT probe filter, which helps to increase coherent fabric performance by reducing the number of HT transactions per cache probe. AMD recommends that the probe filter be enabled on all systems with more than two nodes, and it does not hurt to enable it on systems with 2 nodes. Change-Id: I00a27a828260be8685ae622cfa5a4995add95a8e Signed-off-by: Timothy Pearson <tpearson(a)raptorengineeringinc.com> --- src/cpu/amd/family_10h-family_15h/init_cpus.c | 18 ++- .../amd/family_10h-family_15h/model_10xxx_init.c | 22 ++++ src/northbridge/amd/amdfam10/northbridge.c | 144 ++++++++++++++++++++- src/northbridge/amd/amdht/h3ncmn.c | 22 ++++ 4 files changed, 203 insertions(+), 3 deletions(-) diff --git a/src/cpu/amd/family_10h-family_15h/init_cpus.c b/src/cpu/amd/family_10h-family_15h/init_cpus.c index 14d8642..530ffd2 100644 --- a/src/cpu/amd/family_10h-family_15h/init_cpus.c +++ b/src/cpu/amd/family_10h-family_15h/init_cpus.c @@ -347,9 +347,11 @@ static void STOP_CAR_AND_CPU(uint8_t skip_sharedc_config, uint32_t apicid) static u32 init_cpus(u32 cpu_init_detectedx, struct sys_info *sysinfo) { - u32 bsp_apicid = 0; - u32 apicid; + uint32_t bsp_apicid = 0; + uint32_t apicid; + uint32_t dword; uint8_t set_mtrrs; + uint8_t node_count; struct node_core_id id; /* Please refer to the calculations and explaination in cache_as_ram.inc before modifying these values */ @@ -441,6 +443,18 @@ static u32 init_cpus(u32 cpu_init_detectedx, struct sys_info *sysinfo) cpuSetAMDMSR(id.nodeid); + /* Set up HyperTransport probe filter support */ + if (is_gt_rev_d()) { + dword = pci_read_config32(NODE_PCI(id.nodeid, 0), 0x60); + node_count = ((dword >> 4) & 0x7) + 1; + + if (node_count > 1) { + msr_t msr = rdmsr(BU_CFG2_MSR); + msr.hi |= 1 << (42 - 32); + wrmsr(BU_CFG2_MSR, msr); + } + } + #if CONFIG_SET_FIDVID #if CONFIG_LOGICAL_CPUS && CONFIG_SET_FIDVID_CORE0_ONLY // Run on all AP for proper FID/VID setup. diff --git a/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c b/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c index 3bda14c..fe9fb9c 100644 --- a/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c +++ b/src/cpu/amd/family_10h-family_15h/model_10xxx_init.c @@ -50,6 +50,28 @@ static inline uint8_t is_fam15h(void) return fam15h; } +static inline uint8_t is_gt_rev_d(void) +{ + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + family = ((family & 0xf00000) >> 16) | ((family & 0xf00) >> 8); + + if (family >= 0x6f) + /* Family 15h or later */ + fam15h = 1; + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + return rev_gte_d; +} + static volatile uint8_t fam15h_startup_flags[MAX_NODES_SUPPORTED][MAX_CORES_SUPPORTED] = {{ 0 }}; static void model_10xxx_init(device_t dev) diff --git a/src/northbridge/amd/amdfam10/northbridge.c b/src/northbridge/amd/amdfam10/northbridge.c index a922496..2684d89 100644 --- a/src/northbridge/amd/amdfam10/northbridge.c +++ b/src/northbridge/amd/amdfam10/northbridge.c @@ -26,10 +26,13 @@ #include <lib.h> #include <smbios.h> #include <cpu/cpu.h> +#include <delay.h> #include <cpu/x86/lapic.h> +#include <cpu/x86/cache.h> #include <cpu/amd/mtrr.h> #include <cpu/amd/amdfam10_sysconf.h> +#include <cpu/amd/model_10xxx_msr.h> #include <cpu/amd/family_10h-family_15h/ram_calc.h> #if CONFIG_LOGICAL_CPUS @@ -1523,7 +1526,7 @@ static void cpu_bus_scan(device_t dev) if(i>=32) { busn--; devn-=32; - pbus = pci_domain->link_list->next); + pbus = pci_domain->link_list->next; } #endif @@ -1643,8 +1646,147 @@ static void cpu_bus_scan(device_t dev) } } +static void detect_and_enable_probe_filter(device_t dev) +{ + uint32_t dword; + + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint8_t dual_node = 0; + unsigned nb_cfg_54; + uint32_t f3xe8; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + + if (is_fam15h()) { + /* Family 15h or later */ + fam15h = 1; + nb_cfg_54 = 1; + } + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + if (rev_gte_d) + /* Check for dual node capability */ + if (f3xe8 & 0x20000000) + dual_node = 1; + + if (rev_gte_d && (sysconf.nodes > 1)) { + /* Enable the probe filter */ + uint8_t i; + uint8_t pfmode = 0x0; + + uint32_t f3x58[MAX_NODES_SUPPORTED]; + uint32_t f3x5c[MAX_NODES_SUPPORTED]; + + printk(BIOS_DEBUG, "Enabling probe filter\n"); + + /* Disable L3 and DRAM scrubbers and configure system for probe filter support */ + for (i = 0; i < sysconf.nodes; i++) { + device_t f2x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 2)); + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + f3x58[i] = pci_read_config32(f3x_dev, 0x58); + f3x5c[i] = pci_read_config32(f3x_dev, 0x5c); + pci_write_config32(f3x_dev, 0x58, f3x58[i] & ~((0x1f << 24) | 0x1f)); + pci_write_config32(f3x_dev, 0x5c, f3x5c[i] & ~0x1); + + dword = pci_read_config32(f2x_dev, 0x1b0); + dword &= ~(0x7 << 8); /* CohPrefPrbLmt = 0x0 */ + pci_write_config32(f2x_dev, 0x1b0, dword); + + msr_t msr = rdmsr_amd(BU_CFG2_MSR); + msr.hi |= 1 << (42 - 32); + wrmsr_amd(BU_CFG2_MSR, msr); + + if (is_fam15h()) { + uint8_t subcache_size = 0x0; + uint8_t pref_so_repl = 0x0; + uint32_t f3x1c4 = pci_read_config32(f3x_dev, 0x1c4); + if ((f3x1c4 & 0xffff) == 0xcccc) { + subcache_size = 0x1; + pref_so_repl = 0x2; + pfmode = 0x3; + } else { + pfmode = 0x2; + } + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword |= 0x1 << 29; /* PFLoIndexHashEn = 0x1 */ + dword &= ~(0x3 << 20); /* PFPreferredSORepl = pref_so_repl */ + dword |= (pref_so_repl & 0x3) << 20; + dword |= 0x1 << 17; /* PFWayHashEn = 0x1 */ + dword |= 0xf << 12; /* PFSubCacheEn = 0xf */ + dword &= ~(0x3 << 10); /* PFSubCacheSize3 = subcache_size */ + dword |= (subcache_size & 0x3) << 10; + dword &= ~(0x3 << 8); /* PFSubCacheSize2 = subcache_size */ + dword |= (subcache_size & 0x3) << 8; + dword &= ~(0x3 << 6); /* PFSubCacheSize1 = subcache_size */ + dword |= (subcache_size & 0x3) << 6; + dword &= ~(0x3 << 4); /* PFSubCacheSize0 = subcache_size */ + dword |= (subcache_size & 0x3) << 4; + dword &= ~(0x3 << 2); /* PFWayNum = 0x2 */ + dword |= 0x2 << 2; + pci_write_config32(f3x_dev, 0x1d4, dword); + } else { + pfmode = 0x2; + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword |= 0x1 << 29; /* PFLoIndexHashEn = 0x1 */ + dword &= ~(0x3 << 20); /* PFPreferredSORepl = 0x2 */ + dword |= 0x2 << 20; + dword |= 0xf << 12; /* PFSubCacheEn = 0xf */ + dword &= ~(0x3 << 10); /* PFSubCacheSize3 = 0x0 */ + dword &= ~(0x3 << 8); /* PFSubCacheSize2 = 0x0 */ + dword &= ~(0x3 << 6); /* PFSubCacheSize1 = 0x0 */ + dword &= ~(0x3 << 4); /* PFSubCacheSize0 = 0x0 */ + dword &= ~(0x3 << 2); /* PFWayNum = 0x2 */ + dword |= 0x2 << 2; + pci_write_config32(f3x_dev, 0x1d4, dword); + } + } + + udelay(40); + + disable_cache(); + wbinvd(); + for (i = 0; i < sysconf.nodes; i++) { + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + dword = pci_read_config32(f3x_dev, 0x1c4); + dword |= (0x1 << 31); /* L3TagInit = 1 */ + pci_write_config32(f3x_dev, 0x1c4, dword); + do { + } while (pci_read_config32(f3x_dev, 0x1c4) & (0x1 << 31)); + + dword = pci_read_config32(f3x_dev, 0x1d4); + dword &= ~0x3; /* PFMode = pfmode */ + dword |= pfmode & 0x3; + pci_write_config32(f3x_dev, 0x1d4, dword); + do { + } while (!(pci_read_config32(f3x_dev, 0x1d4) & (0x1 << 19))); + } + enable_cache(); + + /* Reenable L3 and DRAM scrubbers */ + for (i = 0; i < sysconf.nodes; i++) { + device_t f3x_dev = dev_find_slot(0, PCI_DEVFN(0x18 + i, 3)); + + pci_write_config32(f3x_dev, 0x58, f3x58[i]); + pci_write_config32(f3x_dev, 0x5c, f3x5c[i]); + } + + } +} + static void cpu_bus_init(device_t dev) { + detect_and_enable_probe_filter(dev); initialize_cpus(dev->link_list); #if CONFIG_AMD_SB_CIMX sb_After_Pci_Init(); diff --git a/src/northbridge/amd/amdht/h3ncmn.c b/src/northbridge/amd/amdht/h3ncmn.c index 01c0e48..8ee929d 100644 --- a/src/northbridge/amd/amdht/h3ncmn.c +++ b/src/northbridge/amd/amdht/h3ncmn.c @@ -103,6 +103,28 @@ static inline uint8_t is_fam15h(void) return fam15h; } +static inline uint8_t is_gt_rev_d(void) +{ + uint8_t fam15h = 0; + uint8_t rev_gte_d = 0; + uint32_t family; + uint32_t model; + + family = model = cpuid_eax(0x80000001); + model = ((model & 0xf0000) >> 12) | ((model & 0xf0) >> 4); + family = ((family & 0xf00000) >> 16) | ((family & 0xf00) >> 8); + + if (family >= 0x6f) + /* Family 15h or later */ + fam15h = 1; + + if ((model >= 0x8) || fam15h) + /* Revision D or later */ + rev_gte_d = 1; + + return rev_gte_d; +} + /***************************************************************************//** * * SBDFO

1 0

Jump to page:

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

coreboot-gerrit November 2015