aboutsummaryrefslogtreecommitdiff
path: root/sys
diff options
context:
space:
mode:
Diffstat (limited to 'sys')
-rw-r--r--sys/arm64/include/armreg.h5
-rw-r--r--sys/arm64/vmm/arm64.h1
-rw-r--r--sys/arm64/vmm/vmm.c34
-rw-r--r--sys/compat/linuxkpi/common/src/linux_acpi.c30
-rw-r--r--sys/dev/nvme/nvme_private.h6
-rw-r--r--sys/geom/part/g_part.c14
-rw-r--r--sys/net80211/ieee80211.c28
-rw-r--r--sys/net80211/ieee80211_crypto.c87
-rw-r--r--sys/netinet6/in6.c14
-rw-r--r--sys/netlink/netlink_snl.h6
-rw-r--r--sys/vm/uma_core.c22
-rw-r--r--sys/x86/x86/mca.c279
12 files changed, 399 insertions, 127 deletions
diff --git a/sys/arm64/include/armreg.h b/sys/arm64/include/armreg.h
index da051e8f7c8a..393d6d89da0c 100644
--- a/sys/arm64/include/armreg.h
+++ b/sys/arm64/include/armreg.h
@@ -2180,6 +2180,7 @@
#define OSLAR_EL1_CRn 1
#define OSLAR_EL1_CRm 0
#define OSLAR_EL1_op2 4
+#define OSLAR_OSLK (0x1ul << 0)
/* OSLSR_EL1 */
#define OSLSR_EL1_op0 2
@@ -2187,6 +2188,10 @@
#define OSLSR_EL1_CRn 1
#define OSLSR_EL1_CRm 1
#define OSLSR_EL1_op2 4
+#define OSLSR_OSLM_1 (0x1ul << 3)
+#define OSLSR_nTT (0x1ul << 2)
+#define OSLSR_OSLK (0x1ul << 1)
+#define OSLSR_OSLM_0 (0x1ul << 0)
/* PAR_EL1 - Physical Address Register */
#define PAR_F_SHIFT 0
diff --git a/sys/arm64/vmm/arm64.h b/sys/arm64/vmm/arm64.h
index 334b795832a3..f9b74aef7188 100644
--- a/sys/arm64/vmm/arm64.h
+++ b/sys/arm64/vmm/arm64.h
@@ -119,6 +119,7 @@ struct hypctx {
struct vgic_v3_regs vgic_v3_regs;
struct vgic_v3_cpu *vgic_cpu;
bool has_exception;
+ bool dbg_oslock;
};
struct hyp {
diff --git a/sys/arm64/vmm/vmm.c b/sys/arm64/vmm/vmm.c
index 1dcefa1489e9..a551a2807183 100644
--- a/sys/arm64/vmm/vmm.c
+++ b/sys/arm64/vmm/vmm.c
@@ -651,6 +651,33 @@ vmm_reg_wi(struct vcpu *vcpu, uint64_t wval, void *arg)
return (0);
}
+static int
+vmm_write_oslar_el1(struct vcpu *vcpu, uint64_t wval, void *arg)
+{
+ struct hypctx *hypctx;
+
+ hypctx = vcpu_get_cookie(vcpu);
+ /* All other fields are RES0 & we don't do anything with this */
+ /* TODO: Disable access to other debug state when locked */
+ hypctx->dbg_oslock = (wval & OSLAR_OSLK) == OSLAR_OSLK;
+ return (0);
+}
+
+static int
+vmm_read_oslsr_el1(struct vcpu *vcpu, uint64_t *rval, void *arg)
+{
+ struct hypctx *hypctx;
+ uint64_t val;
+
+ hypctx = vcpu_get_cookie(vcpu);
+ val = OSLSR_OSLM_1;
+ if (hypctx->dbg_oslock)
+ val |= OSLSR_OSLK;
+ *rval = val;
+
+ return (0);
+}
+
static const struct vmm_special_reg vmm_special_regs[] = {
#define SPECIAL_REG(_reg, _read, _write) \
{ \
@@ -707,6 +734,13 @@ static const struct vmm_special_reg vmm_special_regs[] = {
SPECIAL_REG(CNTP_TVAL_EL0, vtimer_phys_tval_read,
vtimer_phys_tval_write),
SPECIAL_REG(CNTPCT_EL0, vtimer_phys_cnt_read, vtimer_phys_cnt_write),
+
+ /* Debug registers */
+ SPECIAL_REG(DBGPRCR_EL1, vmm_reg_raz, vmm_reg_wi),
+ SPECIAL_REG(OSDLR_EL1, vmm_reg_raz, vmm_reg_wi),
+ /* TODO: Exceptions on invalid access */
+ SPECIAL_REG(OSLAR_EL1, vmm_reg_raz, vmm_write_oslar_el1),
+ SPECIAL_REG(OSLSR_EL1, vmm_read_oslsr_el1, vmm_reg_wi),
#undef SPECIAL_REG
};
diff --git a/sys/compat/linuxkpi/common/src/linux_acpi.c b/sys/compat/linuxkpi/common/src/linux_acpi.c
index 60e14cfc98c5..c7d62c745c7e 100644
--- a/sys/compat/linuxkpi/common/src/linux_acpi.c
+++ b/sys/compat/linuxkpi/common/src/linux_acpi.c
@@ -119,17 +119,27 @@ acpi_evaluate_dsm(ACPI_HANDLE ObjHandle, const guid_t *guid,
}
static void
-linux_handle_power_suspend_event(void *arg __unused,
- enum power_stype stype __unused)
+linux_handle_power_suspend_event(void *arg __unused, enum power_stype stype)
{
- /*
- * Only support S3 for now.
- * acpi_sleep_event isn't always called so we use power_suspend_early
- * instead which means we don't know what state we're switching to.
- * TODO: Make acpi_sleep_event consistent
- */
- linux_acpi_target_sleep_state = ACPI_STATE_S3;
- pm_suspend_target_state = PM_SUSPEND_MEM;
+ switch (stype) {
+ case POWER_STYPE_SUSPEND_TO_IDLE:
+ /*
+ * XXX: obiwac Not 100% sure this is correct, but
+ * acpi_target_sleep_state does seem to be set to
+ * ACPI_STATE_S3 during s2idle on Linux.
+ */
+ linux_acpi_target_sleep_state = ACPI_STATE_S3;
+ pm_suspend_target_state = PM_SUSPEND_TO_IDLE;
+ break;
+ case POWER_STYPE_SUSPEND_TO_MEM:
+ linux_acpi_target_sleep_state = ACPI_STATE_S3;
+ pm_suspend_target_state = PM_SUSPEND_MEM;
+ break;
+ default:
+ printf("%s: sleep type %d not yet supported\n",
+ __func__, stype);
+ break;
+ }
}
static void
diff --git a/sys/dev/nvme/nvme_private.h b/sys/dev/nvme/nvme_private.h
index 52f9e12f8f9a..52e9fcbbebcd 100644
--- a/sys/dev/nvme/nvme_private.h
+++ b/sys/dev/nvme/nvme_private.h
@@ -463,13 +463,13 @@ static __inline void
nvme_completion_poll(struct nvme_completion_poll_status *status)
{
int timeout = ticks + 10 * hz;
- sbintime_t delta_t = SBT_1US;
+ sbintime_t delta = SBT_1US;
while (!atomic_load_acq_int(&status->done)) {
if (timeout - ticks < 0)
panic("NVME polled command failed to complete within 10s.");
- pause_sbt("nvme", delta_t, 0, C_PREL(1));
- delta_t = min(SBT_1MS, delta_t * 3 / 2);
+ pause_sbt("nvme", delta, 0, C_PREL(1));
+ delta = min(SBT_1MS, delta + delta / 2);
}
}
diff --git a/sys/geom/part/g_part.c b/sys/geom/part/g_part.c
index 4c0d0c3aa902..1e4236507fa4 100644
--- a/sys/geom/part/g_part.c
+++ b/sys/geom/part/g_part.c
@@ -122,13 +122,13 @@ struct g_part_alias_list {
{ "ntfs", G_PART_ALIAS_MS_NTFS },
{ "openbsd-data", G_PART_ALIAS_OPENBSD_DATA },
{ "prep-boot", G_PART_ALIAS_PREP_BOOT },
- { "solaris-boot", G_PART_ALIAS_SOLARIS_BOOT },
- { "solaris-root", G_PART_ALIAS_SOLARIS_ROOT },
- { "solaris-swap", G_PART_ALIAS_SOLARIS_SWAP },
- { "solaris-backup", G_PART_ALIAS_SOLARIS_BACKUP },
- { "solaris-var", G_PART_ALIAS_SOLARIS_VAR },
- { "solaris-home", G_PART_ALIAS_SOLARIS_HOME },
- { "solaris-altsec", G_PART_ALIAS_SOLARIS_ALTSEC },
+ { "solaris-boot", G_PART_ALIAS_SOLARIS_BOOT },
+ { "solaris-root", G_PART_ALIAS_SOLARIS_ROOT },
+ { "solaris-swap", G_PART_ALIAS_SOLARIS_SWAP },
+ { "solaris-backup", G_PART_ALIAS_SOLARIS_BACKUP },
+ { "solaris-var", G_PART_ALIAS_SOLARIS_VAR },
+ { "solaris-home", G_PART_ALIAS_SOLARIS_HOME },
+ { "solaris-altsec", G_PART_ALIAS_SOLARIS_ALTSEC },
{ "solaris-reserved", G_PART_ALIAS_SOLARIS_RESERVED },
{ "u-boot-env", G_PART_ALIAS_U_BOOT_ENV },
{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
diff --git a/sys/net80211/ieee80211.c b/sys/net80211/ieee80211.c
index 2b7cf635b9f5..1299f86ebdc7 100644
--- a/sys/net80211/ieee80211.c
+++ b/sys/net80211/ieee80211.c
@@ -2689,13 +2689,18 @@ ieee80211_channel_type_char(const struct ieee80211_channel *c)
return 'f';
}
-/*
- * Determine whether the given key in the given VAP is a global key.
+/**
+ * @brief Determine whether the given key in the given VAP is a global key.
+ *
* (key index 0..3, shared between all stations on a VAP.)
*
* This is either a WEP key or a GROUP key.
*
* Note this will NOT return true if it is a IGTK key.
+ *
+ * @param vap the current VAP
+ * @param key ieee80211_key to use/check
+ * @returns true if it's a global/WEP key, false otherwise
*/
bool
ieee80211_is_key_global(const struct ieee80211vap *vap,
@@ -2705,8 +2710,23 @@ ieee80211_is_key_global(const struct ieee80211vap *vap,
key < &vap->iv_nw_keys[IEEE80211_WEP_NKID]);
}
-/*
- * Determine whether the given key in the given VAP is a unicast key.
+/**
+ * @brief Determine whether the given key in the given VAP is a unicast key.
+ *
+ * This only returns true if it's a unicast key.
+ *
+ * Note: For now net80211 only supports a single unicast key, stored in
+ * an ieee80211_node entry.
+ *
+ * Code should use this to know if it's a unicast key and then call
+ * ieee80211_crypto_get_keyid() to get the 802.11 key ID (0..3 for
+ * unicast/global keys, 4..5 for IGTK keys.) Since the unicast
+ * and global key indexes "overlap", callers will need to check
+ * both the type and id.
+ *
+ * @param vap the current VAP
+ * @param key ieee80211_key to use/check
+ * @returns true if the key is a unicast key, false if it is not
*/
bool
ieee80211_is_key_unicast(const struct ieee80211vap *vap,
diff --git a/sys/net80211/ieee80211_crypto.c b/sys/net80211/ieee80211_crypto.c
index 1e63ca46f28f..566f0b2e0c23 100644
--- a/sys/net80211/ieee80211_crypto.c
+++ b/sys/net80211/ieee80211_crypto.c
@@ -611,11 +611,15 @@ ieee80211_crypto_setkey(struct ieee80211vap *vap, struct ieee80211_key *key)
return dev_key_set(vap, key);
}
-/*
- * Return index if the key is a WEP key (0..3); -1 otherwise.
+/**
+ * @brief Return index if the key is a WEP key (0..3); -1 otherwise.
*
* This is different to "get_keyid" which defaults to returning
* 0 for unicast keys; it assumes that it won't be used for WEP.
+ *
+ * @param vap the current VAP
+ * @param k ieee80211_key to check
+ * @returns 0..3 if it's a global/WEP key, -1 otherwise.
*/
int
ieee80211_crypto_get_key_wepidx(const struct ieee80211vap *vap,
@@ -628,8 +632,18 @@ ieee80211_crypto_get_key_wepidx(const struct ieee80211vap *vap,
return (-1);
}
-/*
- * Note: only supports a single unicast key (0).
+/**
+ * @brief Return the index of a unicast, global or IGTK key.
+ *
+ * Return the index of a key. For unicast keys the index is 0..1.
+ * For global/WEP keys it's 0..3. For IGTK keys its 4..5.
+ *
+ * TODO: support >1 unicast key
+ * TODO: support IGTK keys
+ *
+ * @param vap the current VAP
+ * @param k ieee80211_key to check
+ * @returns 0..3 for a WEP/global key, 0..1 for unicast key, 4..5 for IGTK key
*/
uint8_t
ieee80211_crypto_get_keyid(struct ieee80211vap *vap, struct ieee80211_key *k)
@@ -641,6 +655,19 @@ ieee80211_crypto_get_keyid(struct ieee80211vap *vap, struct ieee80211_key *k)
return (0);
}
+/**
+ * @param Return the key to use for encrypting an mbuf frame to a node
+ *
+ * This routine chooses a suitable key used to encrypt the given frame with.
+ * It doesn't do the encryption; it only chooses the key. If a key is not
+ * available then the routine will return NULL.
+ *
+ * It's up to the caller to enforce whether a key is absolutely required or not.
+ *
+ * @param ni The ieee80211_node to send the frame to
+ * @param m the mbuf to encrypt
+ * @returns the ieee80211_key to encrypt with, or NULL if there's no suitable key
+ */
struct ieee80211_key *
ieee80211_crypto_get_txkey(struct ieee80211_node *ni, struct mbuf *m)
{
@@ -676,8 +703,28 @@ ieee80211_crypto_get_txkey(struct ieee80211_node *ni, struct mbuf *m)
return &ni->ni_ucastkey;
}
-/*
- * Add privacy headers appropriate for the specified key.
+/**
+ * @brief Privacy encapsulate and encrypt the given mbuf.
+ *
+ * This routine handles the mechanics of encryption - expanding the
+ * mbuf to add privacy headers, IV, ICV, MIC, MMIC, and then encrypts
+ * the given mbuf if required.
+ *
+ * This should be called by the driver in its TX path as part of
+ * encapsulation before passing frames to the hardware/firmware
+ * queues.
+ *
+ * Drivers/hardware which does its own entirely offload path
+ * should still call this for completeness - it indicates to the
+ * driver that the frame itself should be encrypted.
+ *
+ * The driver should have set capability bits in the attach /
+ * key allocation path to disable various encapsulation/encryption
+ * features.
+ *
+ * @param ni ieee80211_node for this frame
+ * @param mbuf mbuf to modify
+ * @returns the key used if the frame is to be encrypted, NULL otherwise
*/
struct ieee80211_key *
ieee80211_crypto_encap(struct ieee80211_node *ni, struct mbuf *m)
@@ -693,9 +740,31 @@ ieee80211_crypto_encap(struct ieee80211_node *ni, struct mbuf *m)
return NULL;
}
-/*
- * Validate and strip privacy headers (and trailer) for a
- * received frame that has the WEP/Privacy bit set.
+/**
+ * @brief Decapsulate and validate an encrypted frame.
+ *
+ * This handles an encrypted frame (one with the privacy bit set.)
+ * It also obeys the key / config / receive packet flags for how
+ * the driver says its already been processed.
+ *
+ * Unlike ieee80211_crypto_encap(), this isn't called in the driver.
+ * Instead, drivers passed the potentially decrypted frame - fully,
+ * partial, or not at all - and net80211 will call this as appropriate.
+ *
+ * This handles NICs (like ath(4)) which have a variable size between
+ * the 802.11 header and 802.11 payload due to DMA alignment / encryption
+ * engine concerns.
+ *
+ * If the frame was decrypted and validated successfully then 1 is returned
+ * and the mbuf can be treated as an 802.11 frame. If it is not decrypted
+ * successfully or it was decrypted but failed validation/checks, then
+ * 0 is returned.
+ *
+ * @param ni ieee80211_node for received frame
+ * @param m mbuf frame to receive
+ * @param hdrlen length of the 802.11 header, including trailing null bytes
+ * @param key pointer to ieee80211_key that will be set if appropriate
+ * @returns 0 if the frame wasn't decrypted/validated, 1 if decrypted/validated.
*/
int
ieee80211_crypto_decap(struct ieee80211_node *ni, struct mbuf *m, int hdrlen,
diff --git a/sys/netinet6/in6.c b/sys/netinet6/in6.c
index 4f756a75fac7..b98703bdfbfe 100644
--- a/sys/netinet6/in6.c
+++ b/sys/netinet6/in6.c
@@ -1295,8 +1295,8 @@ in6_addifaddr(struct ifnet *ifp, struct in6_aliasreq *ifra, struct in6_ifaddr *i
*/
bzero(&pr0, sizeof(pr0));
pr0.ndpr_ifp = ifp;
- pr0.ndpr_plen = in6_mask2len(&ifra->ifra_prefixmask.sin6_addr,
- NULL);
+ pr0.ndpr_plen = ia->ia_plen =
+ in6_mask2len(&ifra->ifra_prefixmask.sin6_addr, NULL);
if (pr0.ndpr_plen == 128) {
/* we don't need to install a host route. */
goto aifaddr_out;
@@ -1490,16 +1490,16 @@ in6_unlink_ifa(struct in6_ifaddr *ia, struct ifnet *ifp)
* positive reference.
*/
remove_lle = 0;
- if (ia->ia6_ndpr == NULL) {
- nd6log((LOG_NOTICE,
- "in6_unlink_ifa: autoconf'ed address "
- "%s has no prefix\n", ip6_sprintf(ip6buf, IA6_IN6(ia))));
- } else {
+ if (ia->ia6_ndpr != NULL) {
ia->ia6_ndpr->ndpr_addrcnt--;
/* Do not delete lles within prefix if refcont != 0 */
if (ia->ia6_ndpr->ndpr_addrcnt == 0)
remove_lle = 1;
ia->ia6_ndpr = NULL;
+ } else if (ia->ia_plen < 128) {
+ nd6log((LOG_NOTICE,
+ "in6_unlink_ifa: autoconf'ed address "
+ "%s has no prefix\n", ip6_sprintf(ip6buf, IA6_IN6(ia))));
}
nd6_rem_ifa_lle(ia, remove_lle);
diff --git a/sys/netlink/netlink_snl.h b/sys/netlink/netlink_snl.h
index 6dd8a9cbdb35..57f7e1e29d08 100644
--- a/sys/netlink/netlink_snl.h
+++ b/sys/netlink/netlink_snl.h
@@ -1068,14 +1068,14 @@ snl_init_writer(struct snl_state *ss, struct snl_writer *nw)
{
nw->size = SNL_WRITER_BUFFER_SIZE;
nw->base = (char *)snl_allocz(ss, nw->size);
- if (nw->base == NULL) {
+ if (__predict_false(nw->base == NULL)) {
nw->error = true;
nw->size = 0;
- }
+ } else
+ nw->error = false;
nw->offset = 0;
nw->hdr = NULL;
- nw->error = false;
nw->ss = ss;
}
diff --git a/sys/vm/uma_core.c b/sys/vm/uma_core.c
index 679b2e20e88b..b80b5cc781f7 100644
--- a/sys/vm/uma_core.c
+++ b/sys/vm/uma_core.c
@@ -4009,21 +4009,15 @@ restart:
/*
* Use the keg's policy if upper layers haven't already specified a
* domain (as happens with first-touch zones).
- *
- * To avoid races we run the iterator with the keg lock held, but that
- * means that we cannot allow the vm_domainset layer to sleep. Thus,
- * clear M_WAITOK and handle low memory conditions locally.
*/
rr = rdomain == UMA_ANYDOMAIN;
+ aflags = flags;
if (rr) {
- aflags = (flags & ~M_WAITOK) | M_NOWAIT;
if (vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain,
&aflags) != 0)
return (NULL);
- } else {
- aflags = flags;
+ } else
domain = rdomain;
- }
for (;;) {
slab = keg_fetch_free_slab(keg, domain, rr, flags);
@@ -4053,13 +4047,8 @@ restart:
if ((flags & M_WAITOK) == 0)
break;
vm_wait_domain(domain);
- } else if (vm_domainset_iter_policy(&di, &domain) != 0) {
- if ((flags & M_WAITOK) != 0) {
- vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask, 0);
- goto restart;
- }
+ } else if (vm_domainset_iter_policy(&di, &domain) != 0)
break;
- }
}
/*
@@ -5245,7 +5234,7 @@ uma_prealloc(uma_zone_t zone, int items)
KEG_GET(zone, keg);
slabs = howmany(items, keg->uk_ipers);
while (slabs-- > 0) {
- aflags = M_NOWAIT;
+ aflags = M_WAITOK;
if (vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain,
&aflags) != 0)
panic("%s: Domainset is empty", __func__);
@@ -5266,7 +5255,8 @@ uma_prealloc(uma_zone_t zone, int items)
break;
}
if (vm_domainset_iter_policy(&di, &domain) != 0)
- vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask, 0);
+ panic("%s: Cannot allocate from any domain",
+ __func__);
}
}
}
diff --git a/sys/x86/x86/mca.c b/sys/x86/x86/mca.c
index e43c88b3a27b..735efe307215 100644
--- a/sys/x86/x86/mca.c
+++ b/sys/x86/x86/mca.c
@@ -46,9 +46,11 @@
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/proc.h>
+#include <sys/sbuf.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
+#include <sys/syslog.h>
#include <sys/systm.h>
#include <sys/taskqueue.h>
#include <machine/intr_machdep.h>
@@ -124,6 +126,22 @@ SYSCTL_INT(_hw_mca, OID_AUTO, erratum383, CTLFLAG_RDTUN,
&workaround_erratum383, 0,
"Is the workaround for Erratum 383 on AMD Family 10h processors enabled?");
+#ifdef DIAGNOSTIC
+static uint64_t fake_status;
+SYSCTL_U64(_hw_mca, OID_AUTO, fake_status, CTLFLAG_RW,
+ &fake_status, 0,
+ "Insert artificial MCA with given status (testing purpose only)");
+static int fake_bank;
+SYSCTL_INT(_hw_mca, OID_AUTO, fake_bank, CTLFLAG_RW,
+ &fake_bank, 0,
+ "Bank to use for artificial MCAs (testing purpose only)");
+#endif
+
+static bool mca_uselog = false;
+SYSCTL_BOOL(_hw_mca, OID_AUTO, uselog, CTLFLAG_RWTUN, &mca_uselog, 0,
+ "Should the system send non-fatal machine check errors to the log "
+ "(instead of the console)?");
+
static STAILQ_HEAD(, mca_internal) mca_freelist;
static int mca_freecount;
static STAILQ_HEAD(, mca_internal) mca_records;
@@ -136,12 +154,40 @@ static struct timeout_task mca_scan_task;
static struct mtx mca_lock;
static bool mca_startup_done = false;
-/* Statistics on number of MCA events by type, updated atomically. */
+/* Static buffer to compose messages while in an interrupt context. */
+static char mca_msg_buf[1024];
+static struct mtx mca_msg_buf_lock;
+
+/* Statistics on number of MCA events by type, updated with the mca_lock. */
static uint64_t mca_stats[MCA_T_COUNT];
SYSCTL_OPAQUE(_hw_mca, OID_AUTO, stats, CTLFLAG_RD | CTLFLAG_SKIP,
mca_stats, MCA_T_COUNT * sizeof(mca_stats[0]),
"S", "Array of MCA events by type");
+/* Variables to track and control message rate limiting. */
+static struct timeval mca_last_log_time;
+static struct timeval mca_log_interval;
+static int mca_log_skipped;
+
+static int
+sysctl_mca_log_interval(SYSCTL_HANDLER_ARGS)
+{
+ int error;
+ u_int val;
+
+ val = mca_log_interval.tv_sec;
+ error = sysctl_handle_int(oidp, &val, 0, req);
+ if (error != 0 || req->newptr == NULL)
+ return (error);
+ mca_log_interval.tv_sec = val;
+ return (0);
+}
+SYSCTL_PROC(_hw_mca, OID_AUTO, log_interval,
+ CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, &mca_log_interval, 0,
+ sysctl_mca_log_interval, "IU",
+ "Minimum number of seconds between logging correctable MCAs"
+ " (0 = no limit)");
+
static unsigned int
mca_ia32_ctl_reg(int bank)
{
@@ -437,98 +483,111 @@ mca_mute(const struct mca_record *rec)
/* Dump details about a single machine check. */
static void
-mca_log(const struct mca_record *rec)
+mca_log(enum scan_mode mode, const struct mca_record *rec, bool fatal)
{
+ int error, numskipped;
uint16_t mca_error;
enum mca_stat_types event_type;
+ struct sbuf sb;
+ bool uncor, using_shared_buf;
if (mca_mute(rec))
return;
- if (!log_corrected && (rec->mr_status & MC_STATUS_UC) == 0 &&
- (!tes_supported(rec->mr_mcg_cap) ||
+ uncor = (rec->mr_status & MC_STATUS_UC) != 0;
+
+ if (!log_corrected && !uncor && (!tes_supported(rec->mr_mcg_cap) ||
((rec->mr_status & MC_STATUS_TES_STATUS) >> 53) != 0x2))
return;
- printf("MCA: Bank %d, Status 0x%016llx\n", rec->mr_bank,
+ /* Try to use an allocated buffer when not in an interrupt context. */
+ if (mode == POLLED && sbuf_new(&sb, NULL, 512, SBUF_AUTOEXTEND) != NULL)
+ using_shared_buf = false;
+ else {
+ using_shared_buf = true;
+ mtx_lock_spin(&mca_msg_buf_lock);
+ sbuf_new(&sb, mca_msg_buf, sizeof(mca_msg_buf), SBUF_FIXEDLEN);
+ }
+
+ sbuf_printf(&sb, "MCA: Bank %d, Status 0x%016llx\n", rec->mr_bank,
(long long)rec->mr_status);
- printf("MCA: Global Cap 0x%016llx, Status 0x%016llx\n",
+ sbuf_printf(&sb, "MCA: Global Cap 0x%016llx, Status 0x%016llx\n",
(long long)rec->mr_mcg_cap, (long long)rec->mr_mcg_status);
- printf("MCA: Vendor \"%s\", ID 0x%x, APIC ID %d\n", cpu_vendor,
- rec->mr_cpu_id, rec->mr_apic_id);
- printf("MCA: CPU %d ", rec->mr_cpu);
+ sbuf_printf(&sb, "MCA: Vendor \"%s\", ID 0x%x, APIC ID %d\n",
+ cpu_vendor, rec->mr_cpu_id, rec->mr_apic_id);
+ sbuf_printf(&sb, "MCA: CPU %d ", rec->mr_cpu);
if (rec->mr_status & MC_STATUS_UC)
- printf("UNCOR ");
+ sbuf_printf(&sb, "UNCOR ");
else {
- printf("COR ");
+ sbuf_printf(&sb, "COR ");
if (cmci_supported(rec->mr_mcg_cap))
- printf("(%lld) ", ((long long)rec->mr_status &
+ sbuf_printf(&sb, "(%lld) ", ((long long)rec->mr_status &
MC_STATUS_COR_COUNT) >> 38);
if (tes_supported(rec->mr_mcg_cap)) {
switch ((rec->mr_status & MC_STATUS_TES_STATUS) >> 53) {
case 0x1:
- printf("(Green) ");
+ sbuf_printf(&sb, "(Green) ");
break;
case 0x2:
- printf("(Yellow) ");
+ sbuf_printf(&sb, "(Yellow) ");
break;
}
}
}
if (rec->mr_status & MC_STATUS_EN)
- printf("EN ");
+ sbuf_printf(&sb, "EN ");
if (rec->mr_status & MC_STATUS_PCC)
- printf("PCC ");
+ sbuf_printf(&sb, "PCC ");
if (ser_supported(rec->mr_mcg_cap)) {
if (rec->mr_status & MC_STATUS_S)
- printf("S ");
+ sbuf_printf(&sb, "S ");
if (rec->mr_status & MC_STATUS_AR)
- printf("AR ");
+ sbuf_printf(&sb, "AR ");
}
if (rec->mr_status & MC_STATUS_OVER)
- printf("OVER ");
+ sbuf_printf(&sb, "OVER ");
mca_error = rec->mr_status & MC_STATUS_MCA_ERROR;
event_type = MCA_T_COUNT;
switch (mca_error) {
/* Simple error codes. */
case 0x0000:
- printf("no error");
+ sbuf_printf(&sb, "no error");
event_type = MCA_T_NONE;
break;
case 0x0001:
- printf("unclassified error");
+ sbuf_printf(&sb, "unclassified error");
event_type = MCA_T_UNCLASSIFIED;
break;
case 0x0002:
- printf("ucode ROM parity error");
+ sbuf_printf(&sb, "ucode ROM parity error");
event_type = MCA_T_UCODE_ROM_PARITY;
break;
case 0x0003:
- printf("external error");
+ sbuf_printf(&sb, "external error");
event_type = MCA_T_EXTERNAL;
break;
case 0x0004:
- printf("FRC error");
+ sbuf_printf(&sb, "FRC error");
event_type = MCA_T_FRC;
break;
case 0x0005:
- printf("internal parity error");
+ sbuf_printf(&sb, "internal parity error");
event_type = MCA_T_INTERNAL_PARITY;
break;
case 0x0006:
- printf("SMM handler code access violation");
+ sbuf_printf(&sb, "SMM handler code access violation");
event_type = MCA_T_SMM_HANDLER;
break;
case 0x0400:
- printf("internal timer error");
+ sbuf_printf(&sb, "internal timer error");
event_type = MCA_T_INTERNAL_TIMER;
break;
case 0x0e0b:
- printf("generic I/O error");
+ sbuf_printf(&sb, "generic I/O error");
event_type = MCA_T_GENERIC_IO;
if (rec->mr_cpu_vendor_id == CPU_VENDOR_INTEL &&
(rec->mr_status & MC_STATUS_MISCV)) {
- printf(" (pci%d:%d:%d:%d)",
+ sbuf_printf(&sb, " (pci%d:%d:%d:%d)",
(int)((rec->mr_misc & MC_MISC_PCIE_SEG) >> 32),
(int)((rec->mr_misc & MC_MISC_PCIE_BUS) >> 24),
(int)((rec->mr_misc & MC_MISC_PCIE_SLOT) >> 19),
@@ -537,7 +596,8 @@ mca_log(const struct mca_record *rec)
break;
default:
if ((mca_error & 0xfc00) == 0x0400) {
- printf("internal error %x", mca_error & 0x03ff);
+ sbuf_printf(&sb, "internal error %x",
+ mca_error & 0x03ff);
event_type = MCA_T_INTERNAL;
break;
}
@@ -546,14 +606,16 @@ mca_log(const struct mca_record *rec)
/* Memory hierarchy error. */
if ((mca_error & 0xeffc) == 0x000c) {
- printf("%s memory error", mca_error_level(mca_error));
+ sbuf_printf(&sb, "%s memory error",
+ mca_error_level(mca_error));
event_type = MCA_T_MEMORY;
break;
}
/* TLB error. */
if ((mca_error & 0xeff0) == 0x0010) {
- printf("%sTLB %s error", mca_error_ttype(mca_error),
+ sbuf_printf(&sb, "%sTLB %s error",
+ mca_error_ttype(mca_error),
mca_error_level(mca_error));
event_type = MCA_T_TLB;
break;
@@ -561,19 +623,19 @@ mca_log(const struct mca_record *rec)
/* Memory controller error. */
if ((mca_error & 0xef80) == 0x0080) {
- printf("%s channel ", mca_error_mmtype(mca_error,
- &event_type));
+ sbuf_printf(&sb, "%s channel ",
+ mca_error_mmtype(mca_error, &event_type));
if ((mca_error & 0x000f) != 0x000f)
- printf("%d", mca_error & 0x000f);
+ sbuf_printf(&sb, "%d", mca_error & 0x000f);
else
- printf("??");
- printf(" memory error");
+ sbuf_printf(&sb, "??");
+ sbuf_printf(&sb, " memory error");
break;
}
/* Cache error. */
if ((mca_error & 0xef00) == 0x0100) {
- printf("%sCACHE %s %s error",
+ sbuf_printf(&sb, "%sCACHE %s %s error",
mca_error_ttype(mca_error),
mca_error_level(mca_error),
mca_error_request(mca_error));
@@ -583,77 +645,129 @@ mca_log(const struct mca_record *rec)
/* Extended memory error. */
if ((mca_error & 0xef80) == 0x0280) {
- printf("%s channel ", mca_error_mmtype(mca_error,
- &event_type));
+ sbuf_printf(&sb, "%s channel ",
+ mca_error_mmtype(mca_error, &event_type));
if ((mca_error & 0x000f) != 0x000f)
- printf("%d", mca_error & 0x000f);
+ sbuf_printf(&sb, "%d", mca_error & 0x000f);
else
- printf("??");
- printf(" extended memory error");
+ sbuf_printf(&sb, "??");
+ sbuf_printf(&sb, " extended memory error");
break;
}
/* Bus and/or Interconnect error. */
if ((mca_error & 0xe800) == 0x0800) {
- printf("BUS%s ", mca_error_level(mca_error));
+ sbuf_printf(&sb, "BUS%s ", mca_error_level(mca_error));
event_type = MCA_T_BUS;
switch ((mca_error & 0x0600) >> 9) {
case 0:
- printf("Source");
+ sbuf_printf(&sb, "Source");
break;
case 1:
- printf("Responder");
+ sbuf_printf(&sb, "Responder");
break;
case 2:
- printf("Observer");
+ sbuf_printf(&sb, "Observer");
break;
default:
- printf("???");
+ sbuf_printf(&sb, "???");
break;
}
- printf(" %s ", mca_error_request(mca_error));
+ sbuf_printf(&sb, " %s ", mca_error_request(mca_error));
switch ((mca_error & 0x000c) >> 2) {
case 0:
- printf("Memory");
+ sbuf_printf(&sb, "Memory");
break;
case 2:
- printf("I/O");
+ sbuf_printf(&sb, "I/O");
break;
case 3:
- printf("Other");
+ sbuf_printf(&sb, "Other");
break;
default:
- printf("???");
+ sbuf_printf(&sb, "???");
break;
}
if (mca_error & 0x0100)
- printf(" timed out");
+ sbuf_printf(&sb, " timed out");
break;
}
- printf("unknown error %x", mca_error);
+ sbuf_printf(&sb, "unknown error %x", mca_error);
event_type = MCA_T_UNKNOWN;
break;
}
- printf("\n");
+ sbuf_printf(&sb, "\n");
if (rec->mr_status & MC_STATUS_ADDRV) {
- printf("MCA: Address 0x%llx", (long long)rec->mr_addr);
+ sbuf_printf(&sb, "MCA: Address 0x%llx",
+ (long long)rec->mr_addr);
if (ser_supported(rec->mr_mcg_cap) &&
(rec->mr_status & MC_STATUS_MISCV)) {
- printf(" (Mode: %s, LSB: %d)",
+ sbuf_printf(&sb, " (Mode: %s, LSB: %d)",
mca_addres_mode(rec->mr_misc),
(int)(rec->mr_misc & MC_MISC_RA_LSB));
}
- printf("\n");
+ sbuf_printf(&sb, "\n");
}
if (rec->mr_status & MC_STATUS_MISCV)
- printf("MCA: Misc 0x%llx\n", (long long)rec->mr_misc);
+ sbuf_printf(&sb, "MCA: Misc 0x%llx\n", (long long)rec->mr_misc);
+
if (event_type < 0 || event_type >= MCA_T_COUNT) {
KASSERT(0, ("%s: invalid event type (%d)", __func__,
event_type));
event_type = MCA_T_UNKNOWN;
}
- atomic_add_64(&mca_stats[event_type], 1);
+ numskipped = 0;
+ if (!fatal && !uncor) {
+ /*
+ * Update statistics and check the rate limit for
+ * correctable errors. The rate limit is only applied
+ * after the system records a reasonable number of errors
+ * of the same type. The goal is to reduce the impact of
+ * the system seeing and attempting to log a burst of
+ * similar errors, which (especially when printed to the
+ * console) can be expensive.
+ */
+ mtx_lock_spin(&mca_lock);
+ mca_stats[event_type]++;
+ if (mca_log_interval.tv_sec > 0 && mca_stats[event_type] > 50 &&
+ ratecheck(&mca_last_log_time, &mca_log_interval) == 0) {
+ mca_log_skipped++;
+ mtx_unlock_spin(&mca_lock);
+ goto done;
+ }
+ numskipped = mca_log_skipped;
+ mca_log_skipped = 0;
+ mtx_unlock_spin(&mca_lock);
+ }
+
+ error = sbuf_finish(&sb);
+ if (fatal || !mca_uselog) {
+ if (numskipped > 0)
+ printf("MCA: %d events skipped due to rate limit\n",
+ numskipped);
+ if (error)
+ printf("MCA: error logging message (sbuf error %d)\n",
+ error);
+ else
+ sbuf_putbuf(&sb);
+ } else {
+ if (numskipped > 0)
+ log(LOG_ERR,
+ "MCA: %d events skipped due to rate limit\n",
+ numskipped);
+ if (error)
+ log(LOG_ERR,
+ "MCA: error logging message (sbuf error %d)\n",
+ error);
+ else
+ log(uncor ? LOG_CRIT : LOG_ERR, "%s", sbuf_data(&sb));
+ }
+
+done:
+ sbuf_delete(&sb);
+ if (using_shared_buf)
+ mtx_unlock_spin(&mca_msg_buf_lock);
}
static bool
@@ -701,8 +815,24 @@ mca_check_status(enum scan_mode mode, uint64_t mcg_cap, int bank,
bool mce, recover;
status = rdmsr(mca_msr_ops.status(bank));
- if (!(status & MC_STATUS_VAL))
+ if (!(status & MC_STATUS_VAL)) {
+#ifdef DIAGNOSTIC
+ /*
+ * Check if we have a pending artificial event to generate.
+ * Note that this is potentially racy with the sysctl. The
+ * tradeoff is deemed acceptable given the test nature
+ * of the code.
+ */
+ if (fake_status && bank == fake_bank) {
+ status = fake_status;
+ fake_status = 0;
+ }
+ if (!(status & MC_STATUS_VAL))
+ return (0);
+#else
return (0);
+#endif
+ }
recover = *recoverablep;
mce = mca_is_mce(mcg_cap, status, &recover);
@@ -796,9 +926,9 @@ mca_record_entry(enum scan_mode mode, const struct mca_record *record)
mtx_lock_spin(&mca_lock);
rec = STAILQ_FIRST(&mca_freelist);
if (rec == NULL) {
- printf("MCA: Unable to allocate space for an event.\n");
- mca_log(record);
mtx_unlock_spin(&mca_lock);
+ printf("MCA: Unable to allocate space for an event.\n");
+ mca_log(mode, record, false);
return;
}
STAILQ_REMOVE_HEAD(&mca_freelist, link);
@@ -955,7 +1085,7 @@ mca_scan(enum scan_mode mode, bool *recoverablep)
if (*recoverablep)
mca_record_entry(mode, &rec);
else
- mca_log(&rec);
+ mca_log(mode, &rec, true);
}
#ifdef DEV_APIC
@@ -1017,6 +1147,7 @@ static void
mca_process_records(enum scan_mode mode)
{
struct mca_internal *mca;
+ STAILQ_HEAD(, mca_internal) tmplist;
/*
* If in an interrupt context, defer the post-scan activities to a
@@ -1028,10 +1159,21 @@ mca_process_records(enum scan_mode mode)
return;
}
+ /*
+ * Copy the pending list to the stack so we can drop the spin lock
+ * while we are emitting logs.
+ */
+ STAILQ_INIT(&tmplist);
+ mtx_lock_spin(&mca_lock);
+ STAILQ_SWAP(&mca_pending, &tmplist, mca_internal);
+ mtx_unlock_spin(&mca_lock);
+
+ STAILQ_FOREACH(mca, &tmplist, link)
+ mca_log(mode, &mca->rec, false);
+
mtx_lock_spin(&mca_lock);
- while ((mca = STAILQ_FIRST(&mca_pending)) != NULL) {
- STAILQ_REMOVE_HEAD(&mca_pending, link);
- mca_log(&mca->rec);
+ while ((mca = STAILQ_FIRST(&tmplist)) != NULL) {
+ STAILQ_REMOVE_HEAD(&tmplist, link);
mca_store_record(mca);
}
mtx_unlock_spin(&mca_lock);
@@ -1192,6 +1334,7 @@ mca_setup(uint64_t mcg_cap)
mca_banks = mcg_cap & MCG_CAP_COUNT;
mtx_init(&mca_lock, "mca", NULL, MTX_SPIN);
+ mtx_init(&mca_msg_buf_lock, "mca_msg_buf", NULL, MTX_SPIN);
STAILQ_INIT(&mca_records);
STAILQ_INIT(&mca_pending);
mca_tq = taskqueue_create_fast("mca", M_WAITOK,
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-09 06:06:12 +0000