*/
#include <kern/kalloc.h>
+#include <mach/mach_time.h>
#include <i386/cpu_data.h>
#include <i386/cpuid.h>
+#include <i386/cpu_topology.h>
+#include <i386/cpu_threads.h>
+#include <i386/machine_cpu.h>
#include <i386/machine_check.h>
#include <i386/proc_reg.h>
static boolean_t mca_threshold_status_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;
-static boolean_t mca_exception_taken = FALSE;
-
decl_simple_lock_data(static, mca_lock);
typedef struct {
mca_mci_bank_t mca_error_bank[0];
} mca_state_t;
+typedef enum {
+ CLEAR,
+ DUMPING,
+ DUMPED
+} mca_dump_state_t;
+static volatile mca_dump_state_t mca_dump_state = CLEAR;
+
static void
mca_get_availability(void)
{
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_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)
{
}
static void
-mca_save_state(void)
+mca_save_state(mca_state_t *mca_state)
{
- mca_state_t *mca_state;
mca_mci_bank_t *bank;
unsigned int i;
assert(!ml_get_interrupts_enabled() || get_preemption_level() > 0);
- mca_state = (mca_state_t *) current_cpu_datap()->cpu_mca_state;
if (mca_state == NULL)
return;
bank->mca_mci_addr = (bank->mca_mci_status.bits.addrv)?
rdmsr64(IA32_MCi_ADDR(i)) : 0ULL;
}
+
+ /*
+ * 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;
}
void
mca_check_save(void)
{
- if (mca_exception_taken)
- mca_save_state();
+ if (mca_dump_state > CLEAR)
+ mca_save_state(current_cpu_datap()->cpu_mca_state);
}
static void mca_dump_64bit_state(void)
// 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 revision %d\n",
+ kdb_printf(" family: %d model: %d stepping: %d microcode: %d\n",
infop->cpuid_family,
infop->cpuid_model,
infop->cpuid_stepping,
(uint32_t) microcode);
- kdb_printf("%s\n", infop->cpuid_brand_string);
+ 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\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.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"
+ " DIMM: %d\n"
+ " Channel: %d\n"
+ " Syndrome: 0x%x\n",
+ i, IA32_MCi_MISC(i), mc8_misc.u64,
+ 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",
};
static void
-mca_dump_error_banks(void)
+mca_dump_bank(mca_state_t *state, int i)
{
- unsigned int i;
+ mca_mci_bank_t *bank;
ia32_mci_status_t status;
- kdb_printf("MCA error-reporting registers:\n");
- for (i = 0; i < mca_error_bank_count; i++ ) {
- status.u64 = rdmsr64(IA32_MCi_STATUS(i));
+ bank = &state->mca_error_bank[i];
+ status = bank->mca_mci_status;
+ 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(
+ " MCA error code: 0x%04x\n",
+ status.bits.mca_error);
+ kdb_printf(
+ " Model specific error code: 0x%04x\n",
+ status.bits.model_specific_error);
+ if (!mca_threshold_status_present) {
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)
- continue;
+ " Other information: 0x%08x\n",
+ status.bits.other_information);
+ } else {
+ int threshold = status.bits_tes_p.threshold;
kdb_printf(
- " MCA error code : 0x%04x\n",
- status.bits.mca_error);
+ " Other information: 0x%08x\n"
+ " Threshold-based status: %s\n",
+ status.bits_tes_p.other_information,
+ (status.bits_tes_p.uc == 0) ?
+ mca_threshold_status[threshold] :
+ "Undefined");
+ }
+ 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(
- " Model specific error code: 0x%04x\n",
- status.bits.model_specific_error);
- if (!mca_threshold_status_present) {
- kdb_printf(
- " Other information : 0x%08x\n",
- status.bits.other_information);
- } else {
- int threshold = status.bits_tes_p.threshold;
- kdb_printf(
- " Other information : 0x%08x\n"
- " Threshold-based status : %s\n",
- status.bits_tes_p.other_information,
- (status.bits_tes_p.uc == 0) ?
- mca_threshold_status[threshold] :
- "Undefined");
- }
+ " IA32_MC%d_ADDR(0x%x): 0x%016qx\n",
+ i, IA32_MCi_ADDR(i), bank->mca_mci_addr);
+ if (status.bits.miscv)
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), rdmsr64(IA32_MCi_ADDR(i)));
- if (status.bits.miscv)
- kdb_printf(
- " IA32_MC%d_MISC(0x%x): 0x%016qx\n",
- i, IA32_MCi_MISC(i), rdmsr64(IA32_MCi_MISC(i)));
+ " IA32_MC%d_MISC(0x%x): 0x%016qx\n",
+ i, IA32_MCi_MISC(i), bank->mca_mci_misc);
+}
+
+static void
+mca_dump_error_banks(mca_state_t *state)
+{
+ unsigned int i;
+
+ kdb_printf("MCA error-reporting registers:\n");
+ for (i = 0; i < mca_error_bank_count; i++ ) {
+ if (i == 8) {
+ /*
+ * Fatal Memory Error
+ */
+
+ /* Dump MC8 for local package */
+ kdb_printf(" Package %d logged:\n",
+ x86_package()->ppkg_num);
+ mca_dump_bank_mc8(state, 8);
+
+ /* If there's other packages, report their MC8s */
+ x86_pkg_t *pkg;
+ uint64_t deadline;
+ for (pkg = x86_pkgs; pkg != NULL; pkg = pkg->next) {
+ if (pkg == x86_package())
+ continue;
+ deadline = mach_absolute_time() + LockTimeOut;
+ while (pkg->mca_state == NULL &&
+ mach_absolute_time() < deadline)
+ cpu_pause();
+ if (pkg->mca_state) {
+ kdb_printf(" Package %d logged:\n",
+ pkg->ppkg_num);
+ mca_dump_bank_mc8(pkg->mca_state, 8);
+ } else {
+ kdb_printf(" Package %d timed out!\n",
+ pkg->ppkg_num);
+ }
+ }
+ continue;
+ }
+ mca_dump_bank(state, i);
}
}
mca_dump(void)
{
ia32_mcg_status_t status;
+ mca_state_t *mca_state = current_cpu_datap()->cpu_mca_state;
- mca_save_state();
+ /*
+ * 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 print MCA registers.
+ * Only the first caller is allowed to dump MCA registers,
+ * other threads spin meantime.
*/
simple_lock(&mca_lock);
- if (mca_exception_taken) {
+ if (mca_dump_state > CLEAR) {
simple_unlock(&mca_lock);
+ while (mca_dump_state == DUMPING)
+ cpu_pause();
return;
}
- mca_exception_taken = TRUE;
+ mca_dump_state = DUMPING;
+ simple_unlock(&mca_lock);
/*
* Report machine-check capabilities:
mca_report_cpu_info();
kdb_printf(
- " %d error-reporting banks\n%s%s", mca_error_bank_count,
+ " %d error-reporting banks\n%s%s%s", mca_error_bank_count,
IF(mca_control_MSR_present,
" control MSR present\n"),
IF(mca_threshold_status_present,
- " threshold-based error status present\n"));
+ " 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);
*/
status.u64 = rdmsr64(IA32_MCG_STATUS);
kdb_printf(
- "Machine-check status 0x%016qx\n%s%s%s", status.u64,
+ "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_dump_error_banks(mca_state);
/*
* Dump any extended machine state:
mca_dump_32bit_state();
}
- simple_unlock(&mca_lock);
+ /* Update state to release any other threads. */
+ mca_dump_state = DUMPED;
}
projects / apple / xnu.git / blobdiff