From f436f8bb73138bc74eb1c6527723e00988ad8a8a Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Thu, 1 Oct 2009 16:14:32 +0200 Subject: x86: EDAC: MCE: Fix MCE decoding callback logic Make decoding of MCEs happen only on AMD hardware by registering a non-default callback only on CPU families which support it. While looking at the interaction of decode_mce() with the other MCE code i also noticed a few other things and made the following cleanups/fixes: - Fixed the mce_decode() weak alias - a weak alias is really not good here, it should be a proper callback. A weak alias will be overriden if a piece of code is built into the kernel - not good, obviously. - The patch initializes the callback on AMD family 10h and 11h. - Added the more correct fallback printk of: No support for human readable MCE decoding on this CPU type. Transcribe the message and run it through 'mcelog --ascii' to decode. On CPUs that dont have a decoder. - Made the surrounding code more readable. Note that the callback allows us to have a default fallback - without having to check the CPU versions during the printout itself. When an EDAC module registers itself, it can install the decode-print function. (there's no unregister needed as this is core code.) version -v2 by Borislav Petkov: - add K8 to the set of supported CPUs - always build in edac_mce_amd since we use an early_initcall now - fix checkpatch warnings Signed-off-by: Borislav Petkov Cc: Linus Torvalds Cc: Andi Kleen LKML-Reference: <20091001141432.GA11410@aftab> Signed-off-by: Ingo Molnar --- drivers/edac/Makefile | 2 +- drivers/edac/edac_mce_amd.c | 15 ++++++++++++++- 2 files changed, 15 insertions(+), 2 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile index 7a473bbe8abd..8701cd7ce4e3 100644 --- a/drivers/edac/Makefile +++ b/drivers/edac/Makefile @@ -18,7 +18,7 @@ edac_core-objs += edac_pci.o edac_pci_sysfs.o endif ifdef CONFIG_CPU_SUP_AMD -edac_core-objs += edac_mce_amd.o +obj-$(CONFIG_X86_MCE) += edac_mce_amd.o endif obj-$(CONFIG_EDAC_AMD76X) += amd76x_edac.o diff --git a/drivers/edac/edac_mce_amd.c b/drivers/edac/edac_mce_amd.c index 0c21c370c9dd..83a01a1187d7 100644 --- a/drivers/edac/edac_mce_amd.c +++ b/drivers/edac/edac_mce_amd.c @@ -362,7 +362,7 @@ static inline void amd_decode_err_code(unsigned int ec) pr_warning("Huh? Unknown MCE error 0x%x\n", ec); } -void decode_mce(struct mce *m) +static void amd_decode_mce(struct mce *m) { struct err_regs regs; int node, ecc; @@ -420,3 +420,16 @@ void decode_mce(struct mce *m) amd_decode_err_code(m->status & 0xffff); } + +static int __init mce_amd_init(void) +{ + /* + * We can decode MCEs for Opteron and later CPUs: + */ + if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && + (boot_cpu_data.x86 >= 0xf)) + x86_mce_decode_callback = amd_decode_mce; + + return 0; +} +early_initcall(mce_amd_init); -- cgit v1.2.3 From 0d18b2e34bd1ad8f5bd3f3a17b5e7df132e511a9 Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Fri, 2 Oct 2009 15:31:48 +0200 Subject: x86: EDAC: carve out AMD MCE decoding logic This converts the MCE decoding logic into a standalone config option which can be built-in or a module, the first one being the default for MCEs happening early on in the boot process. This, beyond being separated in a cleaner way, also saves RAM by making the decoding logic modular. Signed-off-by: Borislav Petkov Cc: Linus Torvalds Cc: Andi Kleen LKML-Reference: <20091002133148.GD28682@aftab> Signed-off-by: Ingo Molnar --- drivers/edac/Kconfig | 14 +++++++++++++- drivers/edac/Makefile | 5 +---- drivers/edac/edac_mce_amd.c | 19 ++++++++++++++++++- 3 files changed, 32 insertions(+), 6 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/Kconfig b/drivers/edac/Kconfig index 02127e59fe8e..55c9c59b3f71 100644 --- a/drivers/edac/Kconfig +++ b/drivers/edac/Kconfig @@ -47,6 +47,18 @@ config EDAC_DEBUG_VERBOSE Source file name and line number where debugging message printed will be added to debugging message. + config EDAC_DECODE_MCE + tristate "Decode MCEs in human-readable form (only on AMD for now)" + depends on CPU_SUP_AMD && X86_MCE + default y + ---help--- + Enable this option if you want to decode Machine Check Exceptions + occuring on your machine in human-readable form. + + You should definitely say Y here in case you want to decode MCEs + which occur really early upon boot, before the module infrastructure + has been initialized. + config EDAC_MM_EDAC tristate "Main Memory EDAC (Error Detection And Correction) reporting" help @@ -59,7 +71,7 @@ config EDAC_MM_EDAC config EDAC_AMD64 tristate "AMD64 (Opteron, Athlon64) K8, F10h, F11h" - depends on EDAC_MM_EDAC && K8_NB && X86_64 && PCI && CPU_SUP_AMD + depends on EDAC_MM_EDAC && K8_NB && X86_64 && PCI && EDAC_DECODE_MCE help Support for error detection and correction on the AMD 64 Families of Memory Controllers (K8, F10h and F11h) diff --git a/drivers/edac/Makefile b/drivers/edac/Makefile index 8701cd7ce4e3..bc5dc232a0fb 100644 --- a/drivers/edac/Makefile +++ b/drivers/edac/Makefile @@ -6,7 +6,6 @@ # GNU General Public License. # - obj-$(CONFIG_EDAC) := edac_stub.o obj-$(CONFIG_EDAC_MM_EDAC) += edac_core.o @@ -17,9 +16,7 @@ ifdef CONFIG_PCI edac_core-objs += edac_pci.o edac_pci_sysfs.o endif -ifdef CONFIG_CPU_SUP_AMD -obj-$(CONFIG_X86_MCE) += edac_mce_amd.o -endif +obj-$(CONFIG_EDAC_DECODE_MCE) += edac_mce_amd.o obj-$(CONFIG_EDAC_AMD76X) += amd76x_edac.o obj-$(CONFIG_EDAC_CPC925) += cpc925_edac.o diff --git a/drivers/edac/edac_mce_amd.c b/drivers/edac/edac_mce_amd.c index 83a01a1187d7..713ed7d37247 100644 --- a/drivers/edac/edac_mce_amd.c +++ b/drivers/edac/edac_mce_amd.c @@ -3,6 +3,7 @@ static bool report_gart_errors; static void (*nb_bus_decoder)(int node_id, struct err_regs *regs); +static void (*orig_mce_callback)(struct mce *m); void amd_report_gart_errors(bool v) { @@ -427,9 +428,25 @@ static int __init mce_amd_init(void) * We can decode MCEs for Opteron and later CPUs: */ if ((boot_cpu_data.x86_vendor == X86_VENDOR_AMD) && - (boot_cpu_data.x86 >= 0xf)) + (boot_cpu_data.x86 >= 0xf)) { + /* safe the default decode mce callback */ + orig_mce_callback = x86_mce_decode_callback; + x86_mce_decode_callback = amd_decode_mce; + } return 0; } early_initcall(mce_amd_init); + +#ifdef MODULE +static void __exit mce_amd_exit(void) +{ + x86_mce_decode_callback = orig_mce_callback; +} + +MODULE_DESCRIPTION("AMD MCE decoder"); +MODULE_ALIAS("edac-mce-amd"); +MODULE_LICENSE("GPL"); +module_exit(mce_amd_exit); +#endif -- cgit v1.2.3 From 3011b20da9d60b8168ead403e6aa860d0d8d11e4 Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Mon, 21 Sep 2009 13:23:34 +0200 Subject: amd64_edac: fix driver instance lookup table allocation Allocate memory statically for 8-node machines max for simplicity instead of relying on MAX_NUMNODES which is 0 on !CONFIG_NUMA builds. Spotted by Jan Beulich. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 4 ++-- drivers/edac/amd64_edac.h | 2 ++ 2 files changed, 4 insertions(+), 2 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 4e551e63b6dc..d6186460c846 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -15,8 +15,8 @@ module_param(ecc_enable_override, int, 0644); /* Lookup table for all possible MC control instances */ struct amd64_pvt; -static struct mem_ctl_info *mci_lookup[MAX_NUMNODES]; -static struct amd64_pvt *pvt_lookup[MAX_NUMNODES]; +static struct mem_ctl_info *mci_lookup[EDAC_MAX_NUMNODES]; +static struct amd64_pvt *pvt_lookup[EDAC_MAX_NUMNODES]; /* * See F2x80 for K8 and F2x[1,0]80 for Fam10 and later. The table below is only diff --git a/drivers/edac/amd64_edac.h b/drivers/edac/amd64_edac.h index 8ea07e2715dc..c3f769e017fa 100644 --- a/drivers/edac/amd64_edac.h +++ b/drivers/edac/amd64_edac.h @@ -132,6 +132,8 @@ #define EDAC_AMD64_VERSION " Ver: 3.2.0 " __DATE__ #define EDAC_MOD_STR "amd64_edac" +#define EDAC_MAX_NUMNODES 8 + /* Extended Model from CPUID, for CPU Revision numbers */ #define OPTERON_CPU_LE_REV_C 0 #define OPTERON_CPU_REV_D 1 -- cgit v1.2.3 From 916d11b2b52430576fd4e96138c974cd64f9cfd6 Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Fri, 18 Sep 2009 12:12:46 +0200 Subject: amd64_edac: fix DRAM base and limit address extraction K8 DRAM base and limit addresses from F1x40 +8*i and F1x44 + 8*i, where i in (0..7) are both bits 39-24 and therefore the shifting should be done by 24 and not by 8. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index d6186460c846..6e2ccce225db 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -1130,7 +1130,7 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) debugf0("Reading K8_DRAM_BASE_LOW failed\n"); /* Extract parts into separate data entries */ - pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 8; + pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 24; pvt->dram_IntlvEn[dram] = (low >> 8) & 0x7; pvt->dram_rw_en[dram] = (low & 0x3); @@ -1143,7 +1143,7 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) * Extract parts into separate data entries. Limit is the HIGHEST memory * location of the region, so lower 24 bits need to be all ones */ - pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 8) | 0x00FFFFFF; + pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 24) | 0x00FFFFFF; pvt->dram_IntlvSel[dram] = (low >> 8) & 0x7; pvt->dram_DstNode[dram] = (low & 0x7); } -- cgit v1.2.3 From 72f158fe6fc2033ceb0f3e0e9e5c2aa356fed6da Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Fri, 18 Sep 2009 12:27:27 +0200 Subject: amd64_edac: fix interleave enable tests The pvt->dram_IntlvEn saves the 3 "Interleave Enable" bits already right-shifted by 8 so the check in find_mc_by_sys_addr() by shifting the values to the left 8 bits is wrong. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 6e2ccce225db..854c7c125035 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -289,12 +289,12 @@ static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci, goto found; } - if (unlikely((intlv_en != (0x01 << 8)) && - (intlv_en != (0x03 << 8)) && - (intlv_en != (0x07 << 8)))) { + if (unlikely((intlv_en != 0x01) && + (intlv_en != 0x03) && + (intlv_en != 0x07))) { amd64_printk(KERN_WARNING, "junk value of 0x%x extracted from " "IntlvEn field of DRAM Base Register for node 0: " - "This probably indicates a BIOS bug.\n", intlv_en); + "this probably indicates a BIOS bug.\n", intlv_en); return NULL; } -- cgit v1.2.3 From 8edc5445895dbcf050d6e8ac5488f70a3937753a Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Fri, 18 Sep 2009 12:39:19 +0200 Subject: amd64_edac: fix K8 intlv_sel check The check when DRAM interleaving is enabled should be done against the pvt->dram_IntlvSel field and not against the ->dram_limit. Simplify first loop and fixup printk formatting while at it. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 19 ++++++++----------- 1 file changed, 8 insertions(+), 11 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 854c7c125035..2c869d1fe334 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -279,14 +279,11 @@ static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci, intlv_en = pvt->dram_IntlvEn[0]; if (intlv_en == 0) { - for (node_id = 0; ; ) { + for (node_id = 0; node_id < DRAM_REG_COUNT; node_id++) { if (amd64_base_limit_match(pvt, sys_addr, node_id)) - break; - - if (++node_id >= DRAM_REG_COUNT) - goto err_no_match; + goto found; } - goto found; + goto err_no_match; } if (unlikely((intlv_en != 0x01) && @@ -301,7 +298,7 @@ static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci, bits = (((u32) sys_addr) >> 12) & intlv_en; for (node_id = 0; ; ) { - if ((pvt->dram_limit[node_id] & intlv_en) == bits) + if ((pvt->dram_IntlvSel[node_id] & intlv_en) == bits) break; /* intlv_sel field matches */ if (++node_id >= DRAM_REG_COUNT) @@ -311,10 +308,10 @@ static struct mem_ctl_info *find_mc_by_sys_addr(struct mem_ctl_info *mci, /* sanity test for sys_addr */ if (unlikely(!amd64_base_limit_match(pvt, sys_addr, node_id))) { amd64_printk(KERN_WARNING, - "%s(): sys_addr 0x%lx falls outside base/limit " - "address range for node %d with node interleaving " - "enabled.\n", __func__, (unsigned long)sys_addr, - node_id); + "%s(): sys_addr 0x%llx falls outside base/limit " + "address range for node %d with node interleaving " + "enabled.\n", + __func__, sys_addr, node_id); return NULL; } -- cgit v1.2.3 From 2cff18c22cfaa88216a5d8c62ea64d1fb575c145 Mon Sep 17 00:00:00 2001 From: Keith Mannthey Date: Fri, 18 Sep 2009 14:35:23 +0200 Subject: amd64_edac: simple fix to allow reporting of CECC errors This allows the errors to be further decoded and mapped to csrows. Tested with ECC debug dimms and an Rev F cpu based system. Signed-off-by: Keith Mannthey Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 2c869d1fe334..ecc86c300e2b 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -1190,7 +1190,7 @@ static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci, * different from the node that detected the error. */ src_mci = find_mc_by_sys_addr(mci, SystemAddress); - if (src_mci) { + if (!src_mci) { amd64_mc_printk(mci, KERN_ERR, "failed to map error address 0x%lx to a node\n", (unsigned long)SystemAddress); -- cgit v1.2.3 From 9d858bb10a9907bbbaffbb4a80a31718d548868c Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Mon, 21 Sep 2009 14:35:51 +0200 Subject: amd64_edac: fix chip select handling Different processor families support a different number of chip selects. Handle this in a family-dependent way with the proper values assigned at init time (see amd64_set_dct_base_and_mask). Remove _DCSM_COUNT defines since they're used at one place and originate from public documentation. CC: Keith Mannthey Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 57 +++++++++++++++++++++-------------------------- drivers/edac/amd64_edac.h | 15 +++++-------- 2 files changed, 32 insertions(+), 40 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index ecc86c300e2b..85c308b1fed7 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -189,7 +189,10 @@ static int amd64_get_scrub_rate(struct mem_ctl_info *mci, u32 *bw) /* Map from a CSROW entry to the mask entry that operates on it */ static inline u32 amd64_map_to_dcs_mask(struct amd64_pvt *pvt, int csrow) { - return csrow >> (pvt->num_dcsm >> 3); + if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_F) + return csrow; + else + return csrow >> 1; } /* return the 'base' address the i'th CS entry of the 'dct' DRAM controller */ @@ -374,7 +377,7 @@ static int input_addr_to_csrow(struct mem_ctl_info *mci, u64 input_addr) * base/mask register pair, test the condition shown near the start of * section 3.5.4 (p. 84, BKDG #26094, K8, revA-E). */ - for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) { + for (csrow = 0; csrow < pvt->cs_count; csrow++) { /* This DRAM chip select is disabled on this node */ if ((pvt->dcsb0[csrow] & K8_DCSB_CS_ENABLE) == 0) @@ -731,7 +734,7 @@ static void find_csrow_limits(struct mem_ctl_info *mci, int csrow, u64 base, mask; pvt = mci->pvt_info; - BUG_ON((csrow < 0) || (csrow >= CHIPSELECT_COUNT)); + BUG_ON((csrow < 0) || (csrow >= pvt->cs_count)); base = base_from_dct_base(pvt, csrow); mask = mask_from_dct_mask(pvt, csrow); @@ -959,35 +962,27 @@ err_reg: */ static void amd64_set_dct_base_and_mask(struct amd64_pvt *pvt) { - if (pvt->ext_model >= OPTERON_CPU_REV_F) { + + if (boot_cpu_data.x86 == 0xf && pvt->ext_model < OPTERON_CPU_REV_F) { + pvt->dcsb_base = REV_E_DCSB_BASE_BITS; + pvt->dcsm_mask = REV_E_DCSM_MASK_BITS; + pvt->dcs_mask_notused = REV_E_DCS_NOTUSED_BITS; + pvt->dcs_shift = REV_E_DCS_SHIFT; + pvt->cs_count = 8; + pvt->num_dcsm = 8; + } else { pvt->dcsb_base = REV_F_F1Xh_DCSB_BASE_BITS; pvt->dcsm_mask = REV_F_F1Xh_DCSM_MASK_BITS; pvt->dcs_mask_notused = REV_F_F1Xh_DCS_NOTUSED_BITS; pvt->dcs_shift = REV_F_F1Xh_DCS_SHIFT; - switch (boot_cpu_data.x86) { - case 0xf: - pvt->num_dcsm = REV_F_DCSM_COUNT; - break; - - case 0x10: - pvt->num_dcsm = F10_DCSM_COUNT; - break; - - case 0x11: - pvt->num_dcsm = F11_DCSM_COUNT; - break; - - default: - amd64_printk(KERN_ERR, "Unsupported family!\n"); - break; + if (boot_cpu_data.x86 == 0x11) { + pvt->cs_count = 4; + pvt->num_dcsm = 2; + } else { + pvt->cs_count = 8; + pvt->num_dcsm = 4; } - } else { - pvt->dcsb_base = REV_E_DCSB_BASE_BITS; - pvt->dcsm_mask = REV_E_DCSM_MASK_BITS; - pvt->dcs_mask_notused = REV_E_DCS_NOTUSED_BITS; - pvt->dcs_shift = REV_E_DCS_SHIFT; - pvt->num_dcsm = REV_E_DCSM_COUNT; } } @@ -1000,7 +995,7 @@ static void amd64_read_dct_base_mask(struct amd64_pvt *pvt) amd64_set_dct_base_and_mask(pvt); - for (cs = 0; cs < CHIPSELECT_COUNT; cs++) { + for (cs = 0; cs < pvt->cs_count; cs++) { reg = K8_DCSB0 + (cs * 4); err = pci_read_config_dword(pvt->dram_f2_ctl, reg, &pvt->dcsb0[cs]); @@ -1563,7 +1558,7 @@ static int f10_lookup_addr_in_dct(u32 in_addr, u32 nid, u32 cs) debugf1("InputAddr=0x%x channelselect=%d\n", in_addr, cs); - for (csrow = 0; csrow < CHIPSELECT_COUNT; csrow++) { + for (csrow = 0; csrow < pvt->cs_count; csrow++) { cs_base = amd64_get_dct_base(pvt, cs, csrow); if (!(cs_base & K8_DCSB_CS_ENABLE)) @@ -2494,7 +2489,7 @@ err_reg: * NOTE: CPU Revision Dependent code * * Input: - * @csrow_nr ChipSelect Row Number (0..CHIPSELECT_COUNT-1) + * @csrow_nr ChipSelect Row Number (0..pvt->cs_count-1) * k8 private pointer to --> * DRAM Bank Address mapping register * node_id @@ -2574,7 +2569,7 @@ static int amd64_init_csrows(struct mem_ctl_info *mci) (pvt->nbcfg & K8_NBCFG_ECC_ENABLE) ? "Enabled" : "Disabled" ); - for (i = 0; i < CHIPSELECT_COUNT; i++) { + for (i = 0; i < pvt->cs_count; i++) { csrow = &mci->csrows[i]; if ((pvt->dcsb0[i] & K8_DCSB_CS_ENABLE) == 0) { @@ -2985,7 +2980,7 @@ static int amd64_init_2nd_stage(struct amd64_pvt *pvt) goto err_exit; ret = -ENOMEM; - mci = edac_mc_alloc(0, CHIPSELECT_COUNT, pvt->channel_count, node_id); + mci = edac_mc_alloc(0, pvt->cs_count, pvt->channel_count, node_id); if (!mci) goto err_exit; diff --git a/drivers/edac/amd64_edac.h b/drivers/edac/amd64_edac.h index c3f769e017fa..64193927a05a 100644 --- a/drivers/edac/amd64_edac.h +++ b/drivers/edac/amd64_edac.h @@ -144,7 +144,7 @@ #define OPTERON_CPU_REV_FA 5 /* Hardware limit on ChipSelect rows per MC and processors per system */ -#define CHIPSELECT_COUNT 8 +#define MAX_CS_COUNT 8 #define DRAM_REG_COUNT 8 @@ -195,7 +195,6 @@ */ #define REV_E_DCSB_BASE_BITS (0xFFE0FE00ULL) #define REV_E_DCS_SHIFT 4 -#define REV_E_DCSM_COUNT 8 #define REV_F_F1Xh_DCSB_BASE_BITS (0x1FF83FE0ULL) #define REV_F_F1Xh_DCS_SHIFT 8 @@ -206,9 +205,6 @@ */ #define REV_F_DCSB_BASE_BITS (0x1FF83FE0ULL) #define REV_F_DCS_SHIFT 8 -#define REV_F_DCSM_COUNT 4 -#define F10_DCSM_COUNT 4 -#define F11_DCSM_COUNT 2 /* DRAM CS Mask Registers */ #define K8_DCSM0 0x60 @@ -447,12 +443,12 @@ struct amd64_pvt { u32 dbam1; /* DRAM Base Address Mapping reg for DCT1 */ /* DRAM CS Base Address Registers F2x[1,0][5C:40] */ - u32 dcsb0[CHIPSELECT_COUNT]; - u32 dcsb1[CHIPSELECT_COUNT]; + u32 dcsb0[MAX_CS_COUNT]; + u32 dcsb1[MAX_CS_COUNT]; /* DRAM CS Mask Registers F2x[1,0][6C:60] */ - u32 dcsm0[CHIPSELECT_COUNT]; - u32 dcsm1[CHIPSELECT_COUNT]; + u32 dcsm0[MAX_CS_COUNT]; + u32 dcsm1[MAX_CS_COUNT]; /* * Decoded parts of DRAM BASE and LIMIT Registers @@ -472,6 +468,7 @@ struct amd64_pvt { */ u32 dcsb_base; /* DCSB base bits */ u32 dcsm_mask; /* DCSM mask bits */ + u32 cs_count; /* num chip selects (== num DCSB registers) */ u32 num_dcsm; /* Number of DCSM registers */ u32 dcs_mask_notused; /* DCSM notused mask bits */ u32 dcs_shift; /* DCSB and DCSM shift value */ -- cgit v1.2.3 From 66216a7a15e75d517dddd0ac6514924b15071e4c Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Tue, 22 Sep 2009 16:48:37 +0200 Subject: amd64_edac: fix DRAM base and limit extraction On Fam10h and above, F1x[1, 0][7C:40] are DRAM Base/Limit registers which specify the destination node of a DRAM address. Those address boundaries are being extracted into ->dram_base[] and ->dram_limit[]. Correct the extraction masks to match the respective address bits. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 85c308b1fed7..4f4ac82382f7 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -1368,8 +1368,8 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) pvt->dram_IntlvEn[dram] = (low_base >> 8) & 0x7; - pvt->dram_base[dram] = (((((u64) high_base & 0x000000FF) << 32) | - ((u64) low_base & 0xFFFF0000))) << 8; + pvt->dram_base[dram] = (((u64)high_base & 0x000000FF) << 40) | + (((u64)low_base & 0xFFFF0000) << 24); low_offset = K8_DRAM_LIMIT_LOW + (dram << 3); high_offset = F10_DRAM_LIMIT_HIGH + (dram << 3); @@ -1390,9 +1390,9 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) * Extract address values and form a LIMIT address. Limit is the HIGHEST * memory location of the region, so low 24 bits need to be all ones. */ - low_limit |= 0x0000FFFF; - pvt->dram_limit[dram] = - ((((u64) high_limit << 32) + (u64) low_limit) << 8) | (0xFF); + pvt->dram_limit[dram] = (((u64)high_limit & 0x000000FF) << 40) | + (((u64) low_limit & 0xFFFF0000) << 24) | + 0x00FFFFFF; } static void f10_read_dram_ctl_register(struct amd64_pvt *pvt) -- cgit v1.2.3 From 94baaee4947d84809b289d5ca03677525ffc6da9 Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Thu, 24 Sep 2009 11:05:30 +0200 Subject: amd64_edac: beef up DRAM error injection When injecting DRAM ECC errors (F3xBC_x8), EccVector[15:0] is a bitmask of which bits should be error injected when written to and holds the payload of 16-bit DRAM word when read, respectively. Add /sysfs members to show the DRAM ECC section/word/vector. Fail wrong injection values entered over /sysfs instead of truncating them. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.h | 6 ++---- drivers/edac/amd64_edac_inj.c | 49 +++++++++++++++++++++++++++++++++++++------ 2 files changed, 45 insertions(+), 10 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.h b/drivers/edac/amd64_edac.h index 64193927a05a..c6f359a85207 100644 --- a/drivers/edac/amd64_edac.h +++ b/drivers/edac/amd64_edac.h @@ -372,13 +372,11 @@ enum { #define SET_NB_DRAM_INJECTION_WRITE(word, bits) \ (BIT(((word) & 0xF) + 20) | \ - BIT(17) | \ - ((bits) & 0xF)) + BIT(17) | bits) #define SET_NB_DRAM_INJECTION_READ(word, bits) \ (BIT(((word) & 0xF) + 20) | \ - BIT(16) | \ - ((bits) & 0xF)) + BIT(16) | bits) #define K8_NBCAP 0xE8 #define K8_NBCAP_CORES (BIT(12)|BIT(13)) diff --git a/drivers/edac/amd64_edac_inj.c b/drivers/edac/amd64_edac_inj.c index d3675b76b3a7..29f1f7a612d9 100644 --- a/drivers/edac/amd64_edac_inj.c +++ b/drivers/edac/amd64_edac_inj.c @@ -1,5 +1,11 @@ #include "amd64_edac.h" +static ssize_t amd64_inject_section_show(struct mem_ctl_info *mci, char *buf) +{ + struct amd64_pvt *pvt = mci->pvt_info; + return sprintf(buf, "0x%x\n", pvt->injection.section); +} + /* * store error injection section value which refers to one of 4 16-byte sections * within a 64-byte cacheline @@ -15,12 +21,26 @@ static ssize_t amd64_inject_section_store(struct mem_ctl_info *mci, ret = strict_strtoul(data, 10, &value); if (ret != -EINVAL) { + + if (value > 3) { + amd64_printk(KERN_WARNING, + "%s: invalid section 0x%lx\n", + __func__, value); + return -EINVAL; + } + pvt->injection.section = (u32) value; return count; } return ret; } +static ssize_t amd64_inject_word_show(struct mem_ctl_info *mci, char *buf) +{ + struct amd64_pvt *pvt = mci->pvt_info; + return sprintf(buf, "0x%x\n", pvt->injection.word); +} + /* * store error injection word value which refers to one of 9 16-bit word of the * 16-byte (128-bit + ECC bits) section @@ -37,14 +57,25 @@ static ssize_t amd64_inject_word_store(struct mem_ctl_info *mci, ret = strict_strtoul(data, 10, &value); if (ret != -EINVAL) { - value = (value <= 8) ? value : 0; - pvt->injection.word = (u32) value; + if (value > 8) { + amd64_printk(KERN_WARNING, + "%s: invalid word 0x%lx\n", + __func__, value); + return -EINVAL; + } + pvt->injection.word = (u32) value; return count; } return ret; } +static ssize_t amd64_inject_ecc_vector_show(struct mem_ctl_info *mci, char *buf) +{ + struct amd64_pvt *pvt = mci->pvt_info; + return sprintf(buf, "0x%x\n", pvt->injection.bit_map); +} + /* * store 16 bit error injection vector which enables injecting errors to the * corresponding bit within the error injection word above. When used during a @@ -60,8 +91,14 @@ static ssize_t amd64_inject_ecc_vector_store(struct mem_ctl_info *mci, ret = strict_strtoul(data, 16, &value); if (ret != -EINVAL) { - pvt->injection.bit_map = (u32) value & 0xFFFF; + if (value & 0xFFFF0000) { + amd64_printk(KERN_WARNING, + "%s: invalid EccVector: 0x%lx\n", + __func__, value); + return -EINVAL; + } + pvt->injection.bit_map = (u32) value; return count; } return ret; @@ -147,7 +184,7 @@ struct mcidev_sysfs_attribute amd64_inj_attrs[] = { .name = "inject_section", .mode = (S_IRUGO | S_IWUSR) }, - .show = NULL, + .show = amd64_inject_section_show, .store = amd64_inject_section_store, }, { @@ -155,7 +192,7 @@ struct mcidev_sysfs_attribute amd64_inj_attrs[] = { .name = "inject_word", .mode = (S_IRUGO | S_IWUSR) }, - .show = NULL, + .show = amd64_inject_word_show, .store = amd64_inject_word_store, }, { @@ -163,7 +200,7 @@ struct mcidev_sysfs_attribute amd64_inj_attrs[] = { .name = "inject_ecc_vector", .mode = (S_IRUGO | S_IWUSR) }, - .show = NULL, + .show = amd64_inject_ecc_vector_show, .store = amd64_inject_ecc_vector_store, }, { -- cgit v1.2.3 From 4997811e3b9e4d6f37380701894f063c62f14929 Mon Sep 17 00:00:00 2001 From: Borislav Petkov Date: Mon, 12 Oct 2009 17:23:03 +0200 Subject: amd64_edac: fix DRAM base and limit extraction masks, v2 This is a proper fix as a follow-up to 66216a7 and 916d11b. Signed-off-by: Borislav Petkov --- drivers/edac/amd64_edac.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c index 4f4ac82382f7..d4560d9d5a83 100644 --- a/drivers/edac/amd64_edac.c +++ b/drivers/edac/amd64_edac.c @@ -1122,7 +1122,7 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) debugf0("Reading K8_DRAM_BASE_LOW failed\n"); /* Extract parts into separate data entries */ - pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 24; + pvt->dram_base[dram] = ((u64) low & 0xFFFF0000) << 8; pvt->dram_IntlvEn[dram] = (low >> 8) & 0x7; pvt->dram_rw_en[dram] = (low & 0x3); @@ -1135,7 +1135,7 @@ static void k8_read_dram_base_limit(struct amd64_pvt *pvt, int dram) * Extract parts into separate data entries. Limit is the HIGHEST memory * location of the region, so lower 24 bits need to be all ones */ - pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 24) | 0x00FFFFFF; + pvt->dram_limit[dram] = (((u64) low & 0xFFFF0000) << 8) | 0x00FFFFFF; pvt->dram_IntlvSel[dram] = (low >> 8) & 0x7; pvt->dram_DstNode[dram] = (low & 0x7); } @@ -1369,7 +1369,7 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) pvt->dram_IntlvEn[dram] = (low_base >> 8) & 0x7; pvt->dram_base[dram] = (((u64)high_base & 0x000000FF) << 40) | - (((u64)low_base & 0xFFFF0000) << 24); + (((u64)low_base & 0xFFFF0000) << 8); low_offset = K8_DRAM_LIMIT_LOW + (dram << 3); high_offset = F10_DRAM_LIMIT_HIGH + (dram << 3); @@ -1391,7 +1391,7 @@ static void f10_read_dram_base_limit(struct amd64_pvt *pvt, int dram) * memory location of the region, so low 24 bits need to be all ones. */ pvt->dram_limit[dram] = (((u64)high_limit & 0x000000FF) << 40) | - (((u64) low_limit & 0xFFFF0000) << 24) | + (((u64) low_limit & 0xFFFF0000) << 8) | 0x00FFFFFF; } -- cgit v1.2.3 From 156edd4aaa819ec5867ced83c7b8dba9193789ea Mon Sep 17 00:00:00 2001 From: Jeff Roberson Date: Mon, 26 Oct 2009 16:50:09 -0700 Subject: edac: i5400 fix csrow mapping The i5400 EDAC driver has several bugs with chip-select row computation which most likely lead to bugs in detailed error reporting. Attempts to contact the authors have gone mostly unanswered so I am presenting my diff here. I do not subscribe to lkml and would appreciate being kept in the cc. The most egregious problem was miscalculating the addresses of MTR registers after register 0 by assuming they are 32bit rather than 16. This caused the driver to miss half of the memories. Most motherboards tend to have only 8 dimm slots and not 16, so this may not have been noticed before. Further, the row calculations multiplied the number of dimms several times, ultimately ending up with a maximum row of 32. The chipset only supports 4 dimms in each of 4 channels, so csrow could not be higher than 4 unless you use a row per-rank with dual-rank dimms. I opted to eliminate this behavior as it is confusing to the user and the error reporting works by slot and not rank. This gives a much clearer view of memory by slot and channel in /sys. Signed-off-by: Jeff Roberson Signed-off-by: Doug Thompson Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/edac/i5400_edac.c | 89 +++++++++++++++-------------------------------- 1 file changed, 28 insertions(+), 61 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/i5400_edac.c b/drivers/edac/i5400_edac.c index b08b6d8e2dc7..f99d10655ed4 100644 --- a/drivers/edac/i5400_edac.c +++ b/drivers/edac/i5400_edac.c @@ -46,9 +46,10 @@ /* Limits for i5400 */ #define NUM_MTRS_PER_BRANCH 4 #define CHANNELS_PER_BRANCH 2 +#define MAX_DIMMS_PER_CHANNEL NUM_MTRS_PER_BRANCH #define MAX_CHANNELS 4 -#define MAX_DIMMS (MAX_CHANNELS * 4) /* Up to 4 DIMM's per channel */ -#define MAX_CSROWS (MAX_DIMMS * 2) /* max possible csrows per channel */ +/* max possible csrows per channel */ +#define MAX_CSROWS (MAX_DIMMS_PER_CHANNEL) /* Device 16, * Function 0: System Address @@ -331,7 +332,6 @@ static const struct i5400_dev_info i5400_devs[] = { struct i5400_dimm_info { int megabytes; /* size, 0 means not present */ - int dual_rank; }; /* driver private data structure */ @@ -849,11 +849,9 @@ static int determine_mtr(struct i5400_pvt *pvt, int csrow, int channel) int n; /* There is one MTR for each slot pair of FB-DIMMs, - Each slot may have one or two ranks (2 csrows), Each slot pair may be at branch 0 or branch 1. - So, csrow should be divided by eight */ - n = csrow >> 3; + n = csrow; if (n >= NUM_MTRS_PER_BRANCH) { debugf0("ERROR: trying to access an invalid csrow: %d\n", @@ -905,25 +903,22 @@ static void handle_channel(struct i5400_pvt *pvt, int csrow, int channel, amb_present_reg = determine_amb_present_reg(pvt, channel); /* Determine if there is a DIMM present in this DIMM slot */ - if (amb_present_reg & (1 << (csrow >> 1))) { - dinfo->dual_rank = MTR_DIMM_RANK(mtr); - - if (!((dinfo->dual_rank == 0) && - ((csrow & 0x1) == 0x1))) { - /* Start with the number of bits for a Bank - * on the DRAM */ - addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr); - /* Add thenumber of ROW bits */ - addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr); - /* add the number of COLUMN bits */ - addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr); - - addrBits += 6; /* add 64 bits per DIMM */ - addrBits -= 20; /* divide by 2^^20 */ - addrBits -= 3; /* 8 bits per bytes */ - - dinfo->megabytes = 1 << addrBits; - } + if (amb_present_reg & (1 << csrow)) { + /* Start with the number of bits for a Bank + * on the DRAM */ + addrBits = MTR_DRAM_BANKS_ADDR_BITS(mtr); + /* Add thenumber of ROW bits */ + addrBits += MTR_DIMM_ROWS_ADDR_BITS(mtr); + /* add the number of COLUMN bits */ + addrBits += MTR_DIMM_COLS_ADDR_BITS(mtr); + /* add the number of RANK bits */ + addrBits += MTR_DIMM_RANK(mtr); + + addrBits += 6; /* add 64 bits per DIMM */ + addrBits -= 20; /* divide by 2^^20 */ + addrBits -= 3; /* 8 bits per bytes */ + + dinfo->megabytes = 1 << addrBits; } } } @@ -951,12 +946,12 @@ static void calculate_dimm_size(struct i5400_pvt *pvt) return; } - /* Scan all the actual CSROWS (which is # of DIMMS * 2) + /* Scan all the actual CSROWS * and calculate the information for each DIMM * Start with the highest csrow first, to display it first * and work toward the 0th csrow */ - max_csrows = pvt->maxdimmperch * 2; + max_csrows = pvt->maxdimmperch; for (csrow = max_csrows - 1; csrow >= 0; csrow--) { /* on an odd csrow, first output a 'boundary' marker, @@ -1064,7 +1059,7 @@ static void i5400_get_mc_regs(struct mem_ctl_info *mci) /* Get the set of MTR[0-3] regs by each branch */ for (slot_row = 0; slot_row < NUM_MTRS_PER_BRANCH; slot_row++) { - int where = MTR0 + (slot_row * sizeof(u32)); + int where = MTR0 + (slot_row * sizeof(u16)); /* Branch 0 set of MTR registers */ pci_read_config_word(pvt->branch_0, where, @@ -1146,7 +1141,7 @@ static int i5400_init_csrows(struct mem_ctl_info *mci) pvt = mci->pvt_info; channel_count = pvt->maxch; - max_csrows = pvt->maxdimmperch * 2; + max_csrows = pvt->maxdimmperch; empty = 1; /* Assume NO memory */ @@ -1214,28 +1209,6 @@ static void i5400_enable_error_reporting(struct mem_ctl_info *mci) fbd_error_mask); } -/* - * i5400_get_dimm_and_channel_counts(pdev, &num_csrows, &num_channels) - * - * ask the device how many channels are present and how many CSROWS - * as well - */ -static void i5400_get_dimm_and_channel_counts(struct pci_dev *pdev, - int *num_dimms_per_channel, - int *num_channels) -{ - u8 value; - - /* Need to retrieve just how many channels and dimms per channel are - * supported on this memory controller - */ - pci_read_config_byte(pdev, MAXDIMMPERCH, &value); - *num_dimms_per_channel = (int)value * 2; - - pci_read_config_byte(pdev, MAXCH, &value); - *num_channels = (int)value; -} - /* * i5400_probe1 Probe for ONE instance of device to see if it is * present. @@ -1263,22 +1236,16 @@ static int i5400_probe1(struct pci_dev *pdev, int dev_idx) if (PCI_FUNC(pdev->devfn) != 0) return -ENODEV; - /* Ask the devices for the number of CSROWS and CHANNELS so - * that we can calculate the memory resources, etc - * - * The Chipset will report what it can handle which will be greater - * or equal to what the motherboard manufacturer will implement. - * - * As we don't have a motherboard identification routine to determine + /* As we don't have a motherboard identification routine to determine * actual number of slots/dimms per channel, we thus utilize the * resource as specified by the chipset. Thus, we might have * have more DIMMs per channel than actually on the mobo, but this * allows the driver to support upto the chipset max, without * some fancy mobo determination. */ - i5400_get_dimm_and_channel_counts(pdev, &num_dimms_per_channel, - &num_channels); - num_csrows = num_dimms_per_channel * 2; + num_dimms_per_channel = MAX_DIMMS_PER_CHANNEL; + num_channels = MAX_CHANNELS; + num_csrows = num_dimms_per_channel; debugf0("MC: %s(): Number of - Channels= %d DIMMS= %d CSROWS= %d\n", __func__, num_channels, num_dimms_per_channel, num_csrows); -- cgit v1.2.3 From 0616fb003d4f799c4be62275242fc7ff9a968f84 Mon Sep 17 00:00:00 2001 From: "Ira W. Snyder" Date: Mon, 26 Oct 2009 16:50:10 -0700 Subject: edac: i5400 fix missing CONFIG_PCI define When building without CONFIG_PCI the edac_pci_idx variable is unused, causing a build-time warning. Wrap the variable in #ifdef CONFIG_PCI, just like the rest of the PCI support. Signed-off-by: Ira W. Snyder Signed-off-by: Doug Thompson Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/edac/mpc85xx_edac.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/edac') diff --git a/drivers/edac/mpc85xx_edac.c b/drivers/edac/mpc85xx_edac.c index 157f6504f25e..cf27402af97b 100644 --- a/drivers/edac/mpc85xx_edac.c +++ b/drivers/edac/mpc85xx_edac.c @@ -26,7 +26,9 @@ #include "mpc85xx_edac.h" static int edac_dev_idx; +#ifdef CONFIG_PCI static int edac_pci_idx; +#endif static int edac_mc_idx; static u32 orig_ddr_err_disable; -- cgit v1.2.3 From c2494ace990c5d37cfe66911b85d28e6945eadfc Mon Sep 17 00:00:00 2001 From: Keith Mannthey Date: Mon, 26 Oct 2009 16:50:11 -0700 Subject: edac: i5100 fix initialization code Allow csrows to properly initialize when the topology only has active channels on 2 and 3. This new check allows proper detection and initialization in this topology. Only checking the first mrt that represented channels 0 and 1 is not sufficient. I also fixed up the related debug information path. I can submit as a 2nd patch if needed. Signed-off-by: Keith Mannthey Acked-by: Aristeu Rozanski Signed-off-by: Doug Thompson Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/edac/i5000_edac.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) (limited to 'drivers/edac') diff --git a/drivers/edac/i5000_edac.c b/drivers/edac/i5000_edac.c index d335086f4a26..77a9579d7167 100644 --- a/drivers/edac/i5000_edac.c +++ b/drivers/edac/i5000_edac.c @@ -1173,7 +1173,7 @@ static void i5000_get_mc_regs(struct mem_ctl_info *mci) pci_read_config_word(pvt->branch_1, where, &pvt->b1_mtr[slot_row]); debugf2("MTR%d where=0x%x B1 value=0x%x\n", slot_row, - where, pvt->b0_mtr[slot_row]); + where, pvt->b1_mtr[slot_row]); } else { pvt->b1_mtr[slot_row] = 0; } @@ -1232,7 +1232,7 @@ static int i5000_init_csrows(struct mem_ctl_info *mci) struct csrow_info *p_csrow; int empty, channel_count; int max_csrows; - int mtr; + int mtr, mtr1; int csrow_megs; int channel; int csrow; @@ -1251,9 +1251,10 @@ static int i5000_init_csrows(struct mem_ctl_info *mci) /* use branch 0 for the basis */ mtr = pvt->b0_mtr[csrow >> 1]; + mtr1 = pvt->b1_mtr[csrow >> 1]; /* if no DIMMS on this row, continue */ - if (!MTR_DIMMS_PRESENT(mtr)) + if (!MTR_DIMMS_PRESENT(mtr) && !MTR_DIMMS_PRESENT(mtr1)) continue; /* FAKE OUT VALUES, FIXME */ -- cgit v1.2.3