/*
- * Copyright (c) 2007 Apple Inc. All rights reserved.
+ * Copyright (c) 2007-2011 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <i386/machine_check.h>
#include <i386/proc_reg.h>
+/*
+ * At the time of the machine-check exception, all hardware-threads panic.
+ * Each thread saves the state of its MCA registers to its per-cpu data area.
+ *
+ * State reporting is serialized so one thread dumps all valid state for all
+ * threads to the panic log. This may entail spinning waiting for other
+ * threads to complete saving state to memory. A timeout applies to this wait
+ * -- in particular, a 3-strikes timeout may prevent a thread from taking
+ * part is the affair.
+ */
+
#define IF(bool,str) ((bool) ? (str) : "")
static boolean_t mca_initialized = FALSE;
static unsigned int mca_error_bank_count = 0;
static boolean_t mca_control_MSR_present = FALSE;
static boolean_t mca_threshold_status_present = FALSE;
+static boolean_t mca_sw_error_recovery_present = FALSE;
static boolean_t mca_extended_MSRs_present = FALSE;
static unsigned int mca_extended_MSRs_count = 0;
+static boolean_t mca_cmci_present = FALSE;
static ia32_mcg_cap_t ia32_mcg_cap;
decl_simple_lock_data(static, mca_lock);
} mca_mci_bank_t;
typedef struct mca_state {
+ boolean_t mca_is_saved;
+ boolean_t mca_is_valid; /* some state is valid */
ia32_mcg_ctl_t mca_mcg_ctl;
ia32_mcg_status_t mca_mcg_status;
mca_mci_bank_t mca_error_bank[0];
mca_get_availability(void)
{
uint64_t features = cpuid_info()->cpuid_features;
- uint32_t family = cpuid_info()->cpuid_family;
+ uint32_t family = cpuid_info()->cpuid_family;
+ uint32_t model = cpuid_info()->cpuid_model;
+ uint32_t stepping = cpuid_info()->cpuid_stepping;
mca_MCE_present = (features & CPUID_FEATURE_MCE) != 0;
mca_MCA_present = (features & CPUID_FEATURE_MCA) != 0;
mca_family = family;
-
+
+ if ((model == CPUID_MODEL_HASWELL && stepping < 3) ||
+ (model == CPUID_MODEL_HASWELL_ULT && stepping < 1) ||
+ (model == CPUID_MODEL_CRYSTALWELL && stepping < 1))
+ panic("Haswell pre-C0 steppings are not supported");
+
/*
* If MCA, the number of banks etc is reported by the IA32_MCG_CAP MSR.
*/
mca_error_bank_count = ia32_mcg_cap.bits.count;
mca_control_MSR_present = ia32_mcg_cap.bits.mcg_ctl_p;
mca_threshold_status_present = ia32_mcg_cap.bits.mcg_tes_p;
+ mca_sw_error_recovery_present = ia32_mcg_cap.bits.mcg_ser_p;
+ mca_cmci_present = ia32_mcg_cap.bits.mcg_ext_corr_err_p;
if (family == 0x0F) {
mca_extended_MSRs_present = ia32_mcg_cap.bits.mcg_ext_p;
mca_extended_MSRs_count = ia32_mcg_cap.bits.mcg_ext_cnt;
}
}
+boolean_t
+mca_is_cmci_present(void)
+{
+ if (!mca_initialized)
+ mca_cpu_init();
+ return mca_cmci_present;
+}
+
void
mca_cpu_alloc(cpu_data_t *cdp)
{
rdmsr64(IA32_MCi_MISC(i)) : 0ULL;
bank->mca_mci_addr = (bank->mca_mci_status.bits.addrv)?
rdmsr64(IA32_MCi_ADDR(i)) : 0ULL;
+ mca_state->mca_is_valid = TRUE;
}
+
+ /*
+ * If we're the first thread with MCA state, point our package to it
+ * and don't care about races
+ */
+ if (x86_package()->mca_state == NULL)
+ x86_package()->mca_state = mca_state;
+
+ mca_state->mca_is_saved = TRUE;
}
void
static void
mca_report_cpu_info(void)
{
- uint64_t microcode;
i386_cpu_info_t *infop = cpuid_info();
- // microcode revision is top 32 bits of MSR_IA32_UCODE_REV
- microcode = rdmsr64(MSR_IA32_UCODE_REV) >> 32;
kdb_printf(" family: %d model: %d stepping: %d microcode: %d\n",
infop->cpuid_family,
infop->cpuid_model,
infop->cpuid_stepping,
- (uint32_t) microcode);
+ infop->cpuid_microcode_version);
kdb_printf(" %s\n", infop->cpuid_brand_string);
}
+static const char *mc8_memory_operation[] = {
+ [MC8_MMM_GENERIC] = "generic",
+ [MC8_MMM_READ] = "read",
+ [MC8_MMM_WRITE] = "write",
+ [MC8_MMM_ADDRESS_COMMAND] = "address/command",
+ [MC8_MMM_RESERVED] = "reserved"
+};
+
+static void
+mca_dump_bank_mc8(mca_state_t *state, int i)
+{
+ mca_mci_bank_t *bank;
+ ia32_mci_status_t status;
+ struct ia32_mc8_specific mc8;
+ int mmm;
+
+ bank = &state->mca_error_bank[i];
+ status = bank->mca_mci_status;
+ mc8 = status.bits_mc8;
+ mmm = MIN(mc8.memory_operation, MC8_MMM_RESERVED);
+
+ kdb_printf(
+ " IA32_MC%d_STATUS(0x%x): 0x%016qx %svalid\n",
+ i, IA32_MCi_STATUS(i), status.u64, IF(!status.bits.val, "in"));
+ if (!status.bits.val)
+ return;
+
+ kdb_printf(
+ " Channel number: %d%s\n"
+ " Memory Operation: %s\n"
+ " Machine-specific error: %s%s%s%s%s%s%s%s%s\n"
+ " COR_ERR_CNT: %d\n",
+ mc8.channel_number,
+ IF(mc8.channel_number == 15, " (unknown)"),
+ mc8_memory_operation[mmm],
+ IF(mc8.read_ecc, "Read ECC "),
+ IF(mc8.ecc_on_a_scrub, "ECC on scrub "),
+ IF(mc8.write_parity, "Write parity "),
+ IF(mc8.redundant_memory, "Redundant memory "),
+ IF(mc8.sparing, "Sparing/Resilvering "),
+ IF(mc8.access_out_of_range, "Access out of Range "),
+ IF(mc8.rtid_out_of_range, "RTID out of Range "),
+ IF(mc8.address_parity, "Address Parity "),
+ IF(mc8.byte_enable_parity, "Byte Enable Parity "),
+ mc8.cor_err_cnt);
+ kdb_printf(
+ " Status bits:\n%s%s%s%s%s%s",
+ IF(status.bits.pcc, " Processor context corrupt\n"),
+ IF(status.bits.addrv, " ADDR register valid\n"),
+ IF(status.bits.miscv, " MISC register valid\n"),
+ IF(status.bits.en, " Error enabled\n"),
+ IF(status.bits.uc, " Uncorrected error\n"),
+ IF(status.bits.over, " Error overflow\n"));
+ if (status.bits.addrv)
+ kdb_printf(
+ " IA32_MC%d_ADDR(0x%x): 0x%016qx\n",
+ i, IA32_MCi_ADDR(i), bank->mca_mci_addr);
+ if (status.bits.miscv) {
+ ia32_mc8_misc_t mc8_misc;
+
+ mc8_misc.u64 = bank->mca_mci_misc;
+ kdb_printf(
+ " IA32_MC%d_MISC(0x%x): 0x%016qx\n"
+ " RTID: %d\n"
+ " DIMM: %d\n"
+ " Channel: %d\n"
+ " Syndrome: 0x%x\n",
+ i, IA32_MCi_MISC(i), mc8_misc.u64,
+ mc8_misc.bits.rtid,
+ mc8_misc.bits.dimm,
+ mc8_misc.bits.channel,
+ (int) mc8_misc.bits.syndrome);
+ }
+}
+
static const char *mca_threshold_status[] = {
- [THRESHOLD_STATUS_NO_TRACKING] "No tracking",
- [THRESHOLD_STATUS_GREEN] "Green",
- [THRESHOLD_STATUS_YELLOW] "Yellow",
- [THRESHOLD_STATUS_RESERVED] "Reserved"
+ [THRESHOLD_STATUS_NO_TRACKING] = "No tracking",
+ [THRESHOLD_STATUS_GREEN] = "Green",
+ [THRESHOLD_STATUS_YELLOW] = "Yellow",
+ [THRESHOLD_STATUS_RESERVED] = "Reserved"
};
static void
mca_threshold_status[threshold] :
"Undefined");
}
+ if (mca_threshold_status_present &&
+ mca_sw_error_recovery_present) {
+ kdb_printf(
+ " Software Error Recovery:\n%s%s",
+ IF(status.bits_tes_p.ar, " Recovery action reqd\n"),
+ IF(status.bits_tes_p.s, " Signaling UCR error\n"));
+ }
kdb_printf(
" Status bits:\n%s%s%s%s%s%s",
IF(status.bits.pcc, " Processor context corrupt\n"),
}
static void
-mca_dump_error_banks(mca_state_t *state)
+mca_cpu_dump_error_banks(mca_state_t *state)
{
unsigned int i;
+ if (!state->mca_is_valid)
+ return;
+
kdb_printf("MCA error-reporting registers:\n");
for (i = 0; i < mca_error_bank_count; i++ ) {
+ if (i == 8 && state == x86_package()->mca_state) {
+ /*
+ * Fatal Memory Error
+ */
+
+ /* Dump MC8 for this package */
+ kdb_printf(" Package %d logged:\n",
+ x86_package()->ppkg_num);
+ mca_dump_bank_mc8(state, 8);
+ continue;
+ }
mca_dump_bank(state, i);
}
}
void
mca_dump(void)
{
- ia32_mcg_status_t status;
- mca_state_t *mca_state = current_cpu_datap()->cpu_mca_state;
+ mca_state_t *mca_state = current_cpu_datap()->cpu_mca_state;
+ uint64_t deadline;
+ unsigned int i = 0;
/*
* Capture local MCA registers to per-cpu data.
mca_save_state(mca_state);
/*
- * Serialize in case of multiple simultaneous machine-checks.
- * Only the first caller is allowed to dump MCA registers,
+ * Serialize: the first caller controls dumping MCA registers,
* other threads spin meantime.
*/
simple_lock(&mca_lock);
mca_dump_state = DUMPING;
simple_unlock(&mca_lock);
+ /*
+ * Wait for all other hardware threads to save their state.
+ * Or timeout.
+ */
+ deadline = mach_absolute_time() + LockTimeOut;
+ while (mach_absolute_time() < deadline && i < real_ncpus) {
+ if (!cpu_datap(i)->cpu_mca_state->mca_is_saved) {
+ cpu_pause();
+ continue;
+ }
+ i += 1;
+ }
+
/*
* Report machine-check capabilities:
*/
kdb_printf(
- "Machine-check capabilities (cpu %d) 0x%016qx:\n",
- cpu_number(), ia32_mcg_cap.u64);
+ "Machine-check capabilities 0x%016qx:\n", ia32_mcg_cap.u64);
mca_report_cpu_info();
" control MSR present\n"),
IF(mca_threshold_status_present,
" threshold-based error status present\n"),
- "");
+ IF(mca_cmci_present,
+ " extended corrected memory error handling present\n"));
if (mca_extended_MSRs_present)
kdb_printf(
" %d extended MSRs present\n", mca_extended_MSRs_count);
/*
- * Report machine-check status:
+ * Dump all processor state:
*/
- status.u64 = rdmsr64(IA32_MCG_STATUS);
- kdb_printf(
- "Machine-check status 0x%016qx:\n%s%s%s", status.u64,
- IF(status.bits.ripv, " restart IP valid\n"),
- IF(status.bits.eipv, " error IP valid\n"),
- IF(status.bits.mcip, " machine-check in progress\n"));
+ for (i = 0; i < real_ncpus; i++) {
+ mca_state_t *mcsp = cpu_datap(i)->cpu_mca_state;
+ ia32_mcg_status_t status;
+
+ kdb_printf("Processor %d: ", i);
+ if (mcsp == NULL ||
+ mcsp->mca_is_saved == FALSE ||
+ mcsp->mca_mcg_status.u64 == 0) {
+ kdb_printf("no machine-check status reported\n");
+ continue;
+ }
+ if (!mcsp->mca_is_valid) {
+ kdb_printf("no valid machine-check state\n");
+ continue;
+ }
+ status = mcsp->mca_mcg_status;
+ kdb_printf(
+ "machine-check status 0x%016qx:\n%s%s%s", status.u64,
+ IF(status.bits.ripv, " restart IP valid\n"),
+ IF(status.bits.eipv, " error IP valid\n"),
+ IF(status.bits.mcip, " machine-check in progress\n"));
- /*
- * Dump error-reporting registers:
- */
- mca_dump_error_banks(mca_state);
+ mca_cpu_dump_error_banks(mcsp);
+ }
/*
* Dump any extended machine state:
/* Update state to release any other threads. */
mca_dump_state = DUMPED;
}
+
+
+extern void mca_exception_panic(void);
+extern void mtrr_lapic_cached(void);
+void mca_exception_panic(void)
+{
+#if DEBUG
+ mtrr_lapic_cached();
+#else
+ kprintf("mca_exception_panic() requires DEBUG build\n");
+#endif
+}
projects / apple / xnu.git / blobdiff