Skip to contentSkip to conversation

Handle AMD threshold interrupt storms

RFC PATCH In review Smita Koralahalli 4 participants 18 messages Feb 17, 2022 – Mar 15, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover Handle AMD threshold interrupt storms

This series of patches handles interrupt storms in AMD threshold interrupt handler. Patch 1 takes care of handling the storms and Patch 2 does some refactoring and simplification to reuse the existing function.

Smita Koralahalli (2):
x86/mce: Handle AMD threshold interrupt storms
x86/mce: Simplify code in log_and_reset_block()

 arch/x86/kernel/cpu/mce/amd.c      | 131 ++++++++++++++++++++++++++++-
 arch/x86/kernel/cpu/mce/core.c     |  16 +++-
 arch/x86/kernel/cpu/mce/intel.c    |   2 +-
 arch/x86/kernel/cpu/mce/internal.h |   8 +-
 4 files changed, 148 insertions(+), 9 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

RFC PATCH 1/2 x86/mce: Handle AMD threshold interrupt storms

Extend the logic of handling CMCI storms to AMD threshold interrupts.

Similar to CMCI storm handling, keep track of the rate at which each processor sees interrupts. If it exceeds threshold, disable interrupts and switch to polling of machine check banks.

But unlike CMCI, re-enable threshold interrupts per CPU because MCA exceptions and interrupts are directed to a single CPU on AMD systems. As the threshold interrupts are per CPU, a global counter is not required to keep the count of all CPUs in the storm.

---
This patch mostly inherits existing Intel's CMCI storm logic and is not a per bank based approach. Commit 7bee1ef01f38395 ("x86/mce: Add per-bank CMCI storm mitigation") under Tony Luck's Linux Tree makes the existing CMCI storm more fine grained and adds a hook into machine_check_poll() to keep track of per-CPU, per-bank corrected error logs.

---
 arch/x86/kernel/cpu/mce/amd.c      | 126 +++++++++++++++++++++++++++++
 arch/x86/kernel/cpu/mce/core.c     |  16 +++-
 arch/x86/kernel/cpu/mce/intel.c    |   2 +-
 arch/x86/kernel/cpu/mce/internal.h |   8 +-
 4 files changed, 147 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/mce/amd.c +126 -0
index 1940d305db1c..53d9320d1470 100644
@@ -478,6 +478,129 @@ static void mce_threshold_block_init(struct threshold_block *b, int offset)
 	threshold_restart_bank(&tr);
 };
 
+#define MCI_STORM_INTERVAL	(HZ)
+#define MCI_STORM_THRESHOLD	15
+
+enum {
+	MCI_STORM_NONE,
+	MCI_STORM_ACTIVE,
+};
+
+static DEFINE_PER_CPU(unsigned long, mci_time_stamp);
+static DEFINE_PER_CPU(unsigned int, mci_storm_cnt);
+static DEFINE_PER_CPU(unsigned int, mci_storm_state);
+
+static DEFINE_PER_CPU(int, mci_backoff_cnt);
+
+static void _reset_block(struct threshold_block *block)
+{
+	struct thresh_restart tr;
+
+	memset(&tr, 0, sizeof(tr));
+	tr.b = block;
+	threshold_restart_bank(&tr);
+}
+
+static void toggle_interrupt_reset_block(struct threshold_block *block, bool on)
+{
+	if (!block)
+		return;
+
+	block->interrupt_enable = !!on;
+	_reset_block(block);
+}
+
+static void mci_toggle_interrupt_mode(bool on)
+{
+	struct threshold_block *first_block = NULL, *block = NULL, *tmp = NULL;
+	struct threshold_bank **bp = this_cpu_read(threshold_banks);
+	unsigned long flags;
+	unsigned int bank;
+
+	if (!bp)
+		return;
+
+	local_irq_save(flags);
+
+	for (bank = 0; bank < this_cpu_read(mce_num_banks); bank++) {
+		if (!(this_cpu_read(bank_map) & (1 << bank)))
+			continue;
+
+		first_block = bp[bank]->blocks;
+		if (!first_block)
+			continue;
+
+		toggle_interrupt_reset_block(first_block, on);
+
+		list_for_each_entry_safe(block, tmp, &first_block->miscj, miscj)
+			toggle_interrupt_reset_block(block, on);
+	}
+
+	local_irq_restore(flags);
+}
+
+bool mce_amd_mci_poll(bool err_seen)
+{
+	if (__this_cpu_read(mci_storm_state) == MCI_STORM_NONE)
+		return false;
+
+	if (err_seen)
+		this_cpu_write(mci_backoff_cnt, INITIAL_CHECK_INTERVAL);
+	else
+		this_cpu_dec(mci_backoff_cnt);
+
+	return true;
+}
+
+unsigned long mci_amd_adjust_timer(unsigned long interval)
+{
+	if (__this_cpu_read(mci_storm_state) == MCI_STORM_ACTIVE) {
+		if (this_cpu_read(mci_backoff_cnt) > 0) {
+			mce_notify_irq();
+			return MCI_STORM_INTERVAL;
+		}
+
+		__this_cpu_write(mci_storm_state, MCI_STORM_NONE);
+		pr_notice("Storm subsided on CPU %d: switching to interrupt mode\n",
+			  smp_processor_id());
+		mci_toggle_interrupt_mode(true);
+	}
+
+	return interval;
+}
+
+static bool storm_detect(void)
+{
+	unsigned int cnt = this_cpu_read(mci_storm_cnt);
+	unsigned long ts = this_cpu_read(mci_time_stamp);
+	unsigned long now = jiffies;
+
+	if (__this_cpu_read(mci_storm_state) != MCI_STORM_NONE)
+		return true;
+
+	if (time_before_eq(now, ts + MCI_STORM_INTERVAL)) {
+		cnt++;
+	} else {
+		cnt = 1;
+		this_cpu_write(mci_time_stamp, now);
+	}
+
+	this_cpu_write(mci_storm_cnt, cnt);
+
+	if (cnt <= MCI_STORM_THRESHOLD)
+		return false;
+
+	mci_toggle_interrupt_mode(false);
+	__this_cpu_write(mci_storm_state, MCI_STORM_ACTIVE);
+	mce_timer_kick(MCI_STORM_INTERVAL);
+	this_cpu_write(mci_backoff_cnt, INITIAL_CHECK_INTERVAL);
+
+	pr_notice("Storm detected on CPU %d: switching to poll mode\n",
+		  smp_processor_id());
+
+	return true;
+}
+
 static int setup_APIC_mce_threshold(int reserved, int new)
 {
 	if (reserved < 0 && !setup_APIC_eilvt(new, THRESHOLD_APIC_VECTOR,
@@ -868,6 +991,9 @@ static void amd_threshold_interrupt(void)
 	struct threshold_bank **bp = this_cpu_read(threshold_banks);
 	unsigned int bank, cpu = smp_processor_id();
 
+	if (storm_detect())
+		return;
+
 	/*
 	 * Validate that the threshold bank has been initialized already. The
 	 * handler is installed at boot time, but on a hotplug event the
arch/x86/kernel/cpu/mce/core.c +14 -2
index 4c31656503bd..ec89b1115889 100644
@@ -1554,6 +1554,13 @@ static unsigned long mce_adjust_timer_default(unsigned long interval)
 
 static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default;
 
+static bool mce_mci_poll_default(bool err_seen)
+{
+	return false;
+}
+
+static bool (*mce_mci_poll)(bool err_seen) = mce_mci_poll_default;
+
 static void __start_timer(struct timer_list *t, unsigned long interval)
 {
 	unsigned long when = jiffies + interval;
@@ -1577,9 +1584,11 @@ static void mce_timer_fn(struct timer_list *t)
 	iv = __this_cpu_read(mce_next_interval);
 
 	if (mce_available(this_cpu_ptr(&cpu_info))) {
-		machine_check_poll(0, this_cpu_ptr(&mce_poll_banks));
+		bool err_seen;
+
+		err_seen = machine_check_poll(0, this_cpu_ptr(&mce_poll_banks));
 
-		if (mce_intel_cmci_poll()) {
+		if (mce_mci_poll(err_seen)) {
 			iv = mce_adjust_timer(iv);
 			goto done;
 		}
@@ -1938,10 +1947,13 @@ static void __mcheck_cpu_init_vendor(struct cpuinfo_x86 *c)
 	case X86_VENDOR_INTEL:
 		mce_intel_feature_init(c);
 		mce_adjust_timer = cmci_intel_adjust_timer;
+		mce_mci_poll = mce_intel_cmci_poll;
 		break;
 
 	case X86_VENDOR_AMD: {
 		mce_amd_feature_init(c);
+		mce_adjust_timer = mci_amd_adjust_timer;
+		mce_mci_poll = mce_amd_mci_poll;
 		break;
 		}
 
arch/x86/kernel/cpu/mce/intel.c +1 -1
index 95275a5e57e0..6f8006d9620d 100644
@@ -127,7 +127,7 @@ static bool lmce_supported(void)
 	return tmp & FEAT_CTL_LMCE_ENABLED;
 }
 
-bool mce_intel_cmci_poll(void)
+bool mce_intel_cmci_poll(bool err_seen)
 {
 	if (__this_cpu_read(cmci_storm_state) == CMCI_STORM_NONE)
 		return false;
arch/x86/kernel/cpu/mce/internal.h +6 -2
index a04b61e27827..aa03107a72b5 100644
@@ -42,7 +42,7 @@ extern mce_banks_t mce_banks_ce_disabled;
 
 #ifdef CONFIG_X86_MCE_INTEL
 unsigned long cmci_intel_adjust_timer(unsigned long interval);
-bool mce_intel_cmci_poll(void);
+bool mce_intel_cmci_poll(bool err_seen);
 void mce_intel_hcpu_update(unsigned long cpu);
 void cmci_disable_bank(int bank);
 void intel_init_cmci(void);
@@ -51,7 +51,7 @@ void intel_clear_lmce(void);
 bool intel_filter_mce(struct mce *m);
 #else
 # define cmci_intel_adjust_timer mce_adjust_timer_default
-static inline bool mce_intel_cmci_poll(void) { return false; }
+# define mce_intel_cmci_poll mce_mci_poll_default
 static inline void mce_intel_hcpu_update(unsigned long cpu) { }
 static inline void cmci_disable_bank(int bank) { }
 static inline void intel_init_cmci(void) { }
@@ -186,8 +186,12 @@ enum mca_msr {
 extern bool filter_mce(struct mce *m);
 
 #ifdef CONFIG_X86_MCE_AMD
+unsigned long mci_amd_adjust_timer(unsigned long interval);
 extern bool amd_filter_mce(struct mce *m);
+bool mce_amd_mci_poll(bool err_seen);
 #else
+# define mci_amd_adjust_timer mce_adjust_timer_default
+# define mce_amd_mci_poll mce_mci_poll_default
 static inline bool amd_filter_mce(struct mce *m) { return false; }
 #endif
 
Signature · 1 line
2.17.1
maintainer
#2 lore raw

Re: [RFC PATCH 1/2] x86/mce: Handle AMD threshold interrupt storms

Quoted 5 lines · Smita Koralahalli

Extend the logic of handling CMCI storms to AMD threshold interrupts.

Similar to CMCI storm handling, keep track of the rate at which each processor sees interrupts. If it exceeds threshold, disable interrupts and switch to polling of machine check banks.

I've been sitting on some partially done patches to re-work storm handling for Intel ... which rips out all the existing storm bits and replaces with something all new. I'll post the 2-part series as replies to this.

Two-part motivation:

1) Disabling CMCI globally is an overly big hammer (as you note in your patches which to a more gentle per-CPU disable.

2) Intel signals some UNCORRECTED errors using CMCI (yes, turns out that was a poorly chosen name given the later evolution of the architecture). Since we don't want to miss those, the proposed storm code just bumps the threshold to (almost) maximum to mitigate, but not eliminate the storm. Note that the threshold only applies to corrected errors.

-Tony

maintainer
#3 lore raw

PATCH v2 1/2 x86/mce: Remove old CMCI storm mitigation code

When a "storm" of CMCI is detected this code mitigates by disabling CMCI interrupt signalling from all of the banks owned by the CPU that saw the storm.

There are problems with this approach:

1) It is very coarse grained. In all liklihood only one of the banks was geenrating the interrupts, but CMCI is disabled for all. This means Linux may delay seeing and processing errors logged from other banks.

2) Although CMCI stands for Corrected Machine Check Interrupt, it is also used to signal when an uncorrected error is logged. This is a problem because these errors should be handled in a timely manner.

Delete all this code in preparation for a finer grained solution.

---
 arch/x86/kernel/cpu/mce/core.c     |  20 +---
 arch/x86/kernel/cpu/mce/intel.c    | 145 -----------------------------
 arch/x86/kernel/cpu/mce/internal.h |   6 --
 3 files changed, 1 insertion(+), 170 deletions(-)
arch/x86/kernel/cpu/mce/core.c +1 -19
index 6ed365337a3b..4f9abb66520d 100644
@@ -1510,13 +1510,6 @@ static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
 static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
 static DEFINE_PER_CPU(struct timer_list, mce_timer);
 
-static unsigned long mce_adjust_timer_default(unsigned long interval)
-{
-	return interval;
-}
-
-static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default;
-
 static void __start_timer(struct timer_list *t, unsigned long interval)
 {
 	unsigned long when = jiffies + interval;
@@ -1539,15 +1532,9 @@ static void mce_timer_fn(struct timer_list *t)
 
 	iv = __this_cpu_read(mce_next_interval);
 
-	if (mce_available(this_cpu_ptr(&cpu_info))) {
+	if (mce_available(this_cpu_ptr(&cpu_info)))
 		machine_check_poll(0, this_cpu_ptr(&mce_poll_banks));
 
-		if (mce_intel_cmci_poll()) {
-			iv = mce_adjust_timer(iv);
-			goto done;
-		}
-	}
-
 	/*
 	 * Alert userspace if needed. If we logged an MCE, reduce the polling
 	 * interval, otherwise increase the polling interval.
@@ -1557,7 +1544,6 @@ static void mce_timer_fn(struct timer_list *t)
 	else
 		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
 
-done:
 	__this_cpu_write(mce_next_interval, iv);
 	__start_timer(t, iv);
 }
@@ -1887,7 +1873,6 @@ static void mce_zhaoxin_feature_init(struct cpuinfo_x86 *c)
 
 	intel_init_cmci();
 	intel_init_lmce();
-	mce_adjust_timer = cmci_intel_adjust_timer;
 }
 
 static void mce_zhaoxin_feature_clear(struct cpuinfo_x86 *c)
@@ -1900,7 +1885,6 @@ static void __mcheck_cpu_init_vendor(struct cpuinfo_x86 *c)
 	switch (c->x86_vendor) {
 	case X86_VENDOR_INTEL:
 		mce_intel_feature_init(c);
-		mce_adjust_timer = cmci_intel_adjust_timer;
 		break;
 
 	case X86_VENDOR_AMD: {
@@ -2559,8 +2543,6 @@ static void mce_reenable_cpu(void)
 
 static int mce_cpu_dead(unsigned int cpu)
 {
-	mce_intel_hcpu_update(cpu);
-
 	/* intentionally ignoring frozen here */
 	if (!cpuhp_tasks_frozen)
 		cmci_rediscover();
arch/x86/kernel/cpu/mce/intel.c +0 -145

169 lines not on this page · raw message

arch/x86/kernel/cpu/mce/internal.h +0 -6
index acd61c41846c..f01d6cbeb809 100644
@@ -41,18 +41,12 @@ struct dentry *mce_get_debugfs_dir(void);
 extern mce_banks_t mce_banks_ce_disabled;
 
 #ifdef CONFIG_X86_MCE_INTEL
-unsigned long cmci_intel_adjust_timer(unsigned long interval);
-bool mce_intel_cmci_poll(void);
-void mce_intel_hcpu_update(unsigned long cpu);
 void cmci_disable_bank(int bank);
 void intel_init_cmci(void);
 void intel_init_lmce(void);
 void intel_clear_lmce(void);
 bool intel_filter_mce(struct mce *m);
 #else
-# define cmci_intel_adjust_timer mce_adjust_timer_default
-static inline bool mce_intel_cmci_poll(void) { return false; }
-static inline void mce_intel_hcpu_update(unsigned long cpu) { }
 static inline void cmci_disable_bank(int bank) { }
 static inline void intel_init_cmci(void) { }
 static inline void intel_init_lmce(void) { }
Signature · 1 line
2.35.1
maintainer
#4 lore raw

Re: [PATCH 1/2] x86/mce: Remove old CMCI storm mitigation code

Quoted 7 lines · Tony

When a "storm" of CMCI is detected this code mitigates by disabling CMCI interrupt signalling from all of the banks owned by the CPU that saw the storm.

There are problems with this approach:

1) It is very coarse grained. In all liklihood only one of the

Unknown word [liklihood] in commit message.
Suggestions: ['likelihood', 'livelihood']

Quoted 1 line

banks was geenrating the interrupts, but CMCI is disabled for all.

Unknown word [geenrating] in commit message.
Suggestions: ['generating', 'penetrating', 'germinating', 'entreating', 'ingratiating']

Do I need to give you the whole spiel about using a spellchecker?

:)

Quoted 8 lines

This means Linux may delay seeing and processing errors logged from other banks.

2) Although CMCI stands for Corrected Machine Check Interrupt, it is also used to signal when an uncorrected error is logged. This is a problem because these errors should be handled in a timely manner.

Quoted 8 lines

Delete all this code in preparation for a finer grained solution.

---
 arch/x86/kernel/cpu/mce/core.c     |  20 +---
 arch/x86/kernel/cpu/mce/intel.c    | 145 -----------------------------
 arch/x86/kernel/cpu/mce/internal.h |   6 --
 3 files changed, 1 insertion(+), 170 deletions(-)

Yah, can't complain about diffstats like that.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
maintainer
#5 lore raw

PATCH v2 2/2 x86/mce: Add per-bank CMCI storm mitigation

In reply to Luck, Tony

Add a hook into machine_check_poll() to keep track of per-CPU, per-bank corrected error logs.

Maintain a bitmap history for each bank showing whether the bank logged an corrected error or not each time it is polled.

In normal operation the interval between polls of this banks determines how far to shift the history. The 64 bit width corresponds to about one second.

When a storm is observed the Rate of interrupts is reduced by setting a large threshold value for this bank in IA32_MCi_CTL2. This bank is added to the bitmap of banks for this CPU to poll. The polling rate is increased to once per second.
During a storm each bit in the history indicates the status of the bank each time it is polled. Thus the history covers just over a minute.

Declare a storm for that bank if the number of corrected interrupts seen in that history is above some threshold (5 in this RFC code for ease of testing, likely move to 15 for compatibility with previous storm detection).

A storm on a bank ends if enough consecutive polls of the bank show no corrected errors (currently 30, may also change). That resets the threshold in IA32_MCi_CTL2 back to 1, removes the bank from the bitmap for polling, and changes the polling rate back to the default.

If a CPU with banks in storm mode is taken offline, the new CPU that inherits ownership of those banks takes over management of storm(s) in the inherited bank(s).

---
 arch/x86/kernel/cpu/mce/core.c     |  26 ++++--
 arch/x86/kernel/cpu/mce/intel.c    | 124 ++++++++++++++++++++++++++++-
 arch/x86/kernel/cpu/mce/internal.h |   4 +-
 3 files changed, 143 insertions(+), 11 deletions(-)
arch/x86/kernel/cpu/mce/core.c +18 -8
index 4f9abb66520d..1f3e7c074182 100644
@@ -714,6 +714,8 @@ bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
 		barrier();
 		m.status = mce_rdmsrl(mca_msr_reg(i, MCA_STATUS));
 
+		mce_intel_storm_tracker(i, m.status);
+
 		/* If this entry is not valid, ignore it */
 		if (!(m.status & MCI_STATUS_VAL))
 			continue;
@@ -1509,6 +1511,7 @@ static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
 
 static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
 static DEFINE_PER_CPU(struct timer_list, mce_timer);
+static DEFINE_PER_CPU(bool, storm_poll_mode);
 
 static void __start_timer(struct timer_list *t, unsigned long interval)
 {
@@ -1544,22 +1547,29 @@ static void mce_timer_fn(struct timer_list *t)
 	else
 		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
 
-	__this_cpu_write(mce_next_interval, iv);
-	__start_timer(t, iv);
+	if (__this_cpu_read(storm_poll_mode)) {
+		__start_timer(t, HZ);
+	} else {
+		__this_cpu_write(mce_next_interval, iv);
+		__start_timer(t, iv);
+	}
 }
 
 /*
- * Ensure that the timer is firing in @interval from now.
+ * When a storm starts on any bank on this CPU, switch to polling
+ * once per second. When the storm ends, revert to the default
+ * polling interval.
  */
-void mce_timer_kick(unsigned long interval)
+void mce_timer_kick(bool storm)
 {
 	struct timer_list *t = this_cpu_ptr(&mce_timer);
-	unsigned long iv = __this_cpu_read(mce_next_interval);
 
-	__start_timer(t, interval);
+	__this_cpu_write(storm_poll_mode, storm);
 
-	if (interval < iv)
-		__this_cpu_write(mce_next_interval, interval);
+	if (storm)
+		__start_timer(t, HZ);
+	else
+		__this_cpu_write(mce_next_interval, check_interval * HZ);
 }
 
 /* Must not be called in IRQ context where del_timer_sync() can deadlock */
arch/x86/kernel/cpu/mce/intel.c +122 -2

168 lines not on this page · raw message

arch/x86/kernel/cpu/mce/internal.h +3 -1
index f01d6cbeb809..6c7480bce977 100644
@@ -41,12 +41,14 @@ struct dentry *mce_get_debugfs_dir(void);
 extern mce_banks_t mce_banks_ce_disabled;
 
 #ifdef CONFIG_X86_MCE_INTEL
+void mce_intel_storm_tracker(int bank, u64 status);
 void cmci_disable_bank(int bank);
 void intel_init_cmci(void);
 void intel_init_lmce(void);
 void intel_clear_lmce(void);
 bool intel_filter_mce(struct mce *m);
 #else
+static inline void mce_intel_storm_tracker(int bank, u64 status) { }
 static inline void cmci_disable_bank(int bank) { }
 static inline void intel_init_cmci(void) { }
 static inline void intel_init_lmce(void) { }
@@ -54,7 +56,7 @@ static inline void intel_clear_lmce(void) { }
 static inline bool intel_filter_mce(struct mce *m) { return false; }
 #endif
 
-void mce_timer_kick(unsigned long interval);
+void mce_timer_kick(bool storm);
 
 #ifdef CONFIG_ACPI_APEI
 int apei_write_mce(struct mce *m);
Signature · 1 line
2.35.1
#6 lore raw

Re: [PATCH 2/2] x86/mce: Add per-bank CMCI storm mitigation

Hi,

After reviewing Intel's new implementation to handle storms and
comparing it against the patch here:
https://lore.kernel.org/all/[email protected]/ <https://lore.kernel.org/all/[email protected]/>

AMD's current implementation, already does per CPU disabling of interrupts and the storm handling is done for corrected errors only.

However, we are not doing per bank approach to handle storms. So, its a good idea to incorporate this in our code too.

It looked to me that most of the Intel's current implementation can be shared except in few places where we use different registers (MCx_MISC) for setting threshold limits to fire interrupts.

Other questions/comments below..

Quoted 9 lines · Tony

Add a hook into machine_check_poll() to keep track of per-CPU, per-bank corrected error logs.

Maintain a bitmap history for each bank showing whether the bank logged an corrected error or not each time it is polled.

In normal operation the interval between polls of this banks determines how far to shift the history. The 64 bit width corresponds to about one second.

I did not quite get this paragraph and the associated sentence below ..."Thus the history covers just over a minute".

I'm thinking these timing calculations relate to the statements above?

+ delta = now - this_cpu_read(bank_time_stamp[bank]);
+ shift = this_cpu_read(bank_storm[bank]) ? 1 : (delta + HZBITS) / HZBITS;
+ history = (shift < 64) ? this_cpu_read(bank_history[bank]) << shift : 0;
+ this_cpu_write(bank_time_stamp[bank], now);

Can you please elaborate?

Quoted 51 lines

When a storm is observed the Rate of interrupts is reduced by setting a large threshold value for this bank in IA32_MCi_CTL2. This bank is added to the bitmap of banks for this CPU to poll. The polling rate is increased to once per second.
During a storm each bit in the history indicates the status of the bank each time it is polled. Thus the history covers just over a minute.

Declare a storm for that bank if the number of corrected interrupts seen in that history is above some threshold (5 in this RFC code for ease of testing, likely move to 15 for compatibility with previous storm detection).

A storm on a bank ends if enough consecutive polls of the bank show no corrected errors (currently 30, may also change). That resets the threshold in IA32_MCi_CTL2 back to 1, removes the bank from the bitmap for polling, and changes the polling rate back to the default.

If a CPU with banks in storm mode is taken offline, the new CPU that inherits ownership of those banks takes over management of storm(s) in the inherited bank(s).

---
  arch/x86/kernel/cpu/mce/core.c     |  26 ++++--
  arch/x86/kernel/cpu/mce/intel.c    | 124 ++++++++++++++++++++++++++++-
  arch/x86/kernel/cpu/mce/internal.h |   4 +-
  3 files changed, 143 insertions(+), 11 deletions(-)
arch/x86/kernel/cpu/mce/core.c +3 -0
index 4f9abb66520d..1f3e7c074182 100644
@@ -714,6 +714,8 @@ bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
  		barrier();
  		m.status = mce_rdmsrl(mca_msr_reg(i, MCA_STATUS));
 
+		mce_intel_storm_tracker(i, m.status);
+
  		/* If this entry is not valid, ignore it */
  		if (!(m.status & MCI_STATUS_VAL))
  			continue;
@@ -1509,6 +1511,7 @@ static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
 
  static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
  static DEFINE_PER_CPU(struct timer_list, mce_timer);
+static DEFINE_PER_CPU(bool, storm_poll_mode);
 
  static void __start_timer(struct timer_list *t, unsigned long interval)
  {

arch/x86/kernel/cpu/mce/core.c static void mce_timer_fn(struct timer_list *t) · line 1551

Quoted 7 lines
arch/x86/kernel/cpu/mce/core.c +2 -2
@@ -1544,22 +1547,29 @@ static void mce_timer_fn(struct timer_list *t)
  	else
  		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
 
-	__this_cpu_write(mce_next_interval, iv);
-	__start_timer(t, iv);
+	if (__this_cpu_read(storm_poll_mode)) {
+		__start_timer(t, HZ);

So, this is where the timer is fired to poll once per second on a storm?

Quoted 15 lines
arch/x86/kernel/cpu/mce/core.c +8 -2
+	} else {
+		__this_cpu_write(mce_next_interval, iv);
+		__start_timer(t, iv);
+	}
  }
 
  /*
- * Ensure that the timer is firing in @interval from now.
+ * When a storm starts on any bank on this CPU, switch to polling
+ * once per second. When the storm ends, revert to the default
+ * polling interval.
   */
-void mce_timer_kick(unsigned long interval)
+void mce_timer_kick(bool storm)
  {

arch/x86/kernel/cpu/mce/core.c static void mce_timer_fn(struct timer_list *t) · line 1570

Quoted 10 lines
arch/x86/kernel/cpu/mce/core.c +3 -4
  	struct timer_list *t = this_cpu_ptr(&mce_timer);
-	unsigned long iv = __this_cpu_read(mce_next_interval);
 
-	__start_timer(t, interval);
+	__this_cpu_write(storm_poll_mode, storm);
 
-	if (interval < iv)
-		__this_cpu_write(mce_next_interval, interval);
+	if (storm)
+		__start_timer(t, HZ);

.. and here?

Quoted 10 lines
arch/x86/kernel/cpu/mce/core.c +2 -0
+	else
+		__this_cpu_write(mce_next_interval, check_interval * HZ);
  }
 
  /* Must not be called in IRQ context where del_timer_sync() can deadlock */
arch/x86/kernel/cpu/mce/intel.c +0 -0
index cee9d989f791..2ed5634ec277 100644

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 55

Quoted 9 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
@@ -47,8 +47,48 @@ static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
   */
  static DEFINE_RAW_SPINLOCK(cmci_discover_lock);
 
+/*
+ * CMCI storm tracking state
+ */
+static DEFINE_PER_CPU(int, stormy_bank_count);
+static DEFINE_PER_CPU(u64 [MAX_NR_BANKS], bank_history);
+static DEFINE_PER_CPU(bool [MAX_NR_BANKS], bank_storm);

So, we maintain two bitmaps.
bank_history: per bank bitmap which contains history of whether the bank logged corrected error or not each time it is polled.

bank_storm: bitmap of banks in a storm.
Am I right?

Quoted 18 lines
arch/x86/kernel/cpu/mce/intel.c +16 -0
+static DEFINE_PER_CPU(unsigned long [MAX_NR_BANKS], bank_time_stamp);
+static int cmci_threshold[MAX_NR_BANKS];
+
  #define CMCI_THRESHOLD		1
 
+/*
+ * High threshold to limit CMCI rate during storms. Max supported is
+ * 0x7FFF. Use this slightly smaller value so it has a distinctive
+ * signature when some asks "Why am I not seeing all corrected errors?"
+ */
+#define CMCI_STORM_THRESHOLD	0x7FED
+
+/*
+ * How many errors within the history buffer mark the start of a storm
+ */
+#define STORM_BEGIN	5
+
+/*

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 83

Quoted 10 lines
arch/x86/kernel/cpu/mce/intel.c +10 -0
+ * How many polls of machine check bank without an error before declaring
+ * the storm is over
+ */
+#define STORM_END	30
+
+/*
+ * If there is no poll data for a bank for this amount of time, just
+ * discard the history.
+ */
+#define STORM_INTERVAL (1 * HZ)

Looks like STORM_INTERVAL isn't been used anywhere..

Quoted 16 lines
arch/x86/kernel/cpu/mce/intel.c +9 -0
+
+/*
+ * When there is no storm each "bit" in the history represents
+ * this many jiffies. When there is a storm every poll() takes
+ * one history bit.
+ */
+#define HZBITS (HZ / 64)
+
  static int cmci_supported(int *banks)
  {
  	u64 cap;
@@ -103,6 +143,70 @@ static bool lmce_supported(void)
  	return tmp & FEAT_CTL_LMCE_ENABLED;
  }
 
+/*

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 156

Quoted 10 lines
arch/x86/kernel/cpu/mce/intel.c +10 -0
+ * Set a new CMCI threshold value. Preserve the state of the
+ * MCI_CTL2_CMCI_EN bit in case this happens during a
+ * cmci_rediscover() operation.
+ */
+static void cmci_set_threshold(int bank, int thresh)
+{
+	unsigned long flags;
+	u64 val;
+
+	raw_spin_lock_irqsave(&cmci_discover_lock, flags);

Why not local_irq_save(flags) instead?

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 167

Quoted 11 lines
arch/x86/kernel/cpu/mce/intel.c +11 -0
+	rdmsrl(MSR_IA32_MCx_CTL2(bank), val);
+	val &= ~MCI_CTL2_CMCI_THRESHOLD_MASK;
+	wrmsrl(MSR_IA32_MCx_CTL2(bank), val | thresh);
+	raw_spin_unlock_irqrestore(&cmci_discover_lock, flags);
+}
+
+static void cmci_storm_begin(int bank)
+{
+	__set_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_storm[bank], true);
+	if (this_cpu_inc_return(stormy_bank_count) == 1)

What is the significance of stormy_bank_count?

Quoted 65 lines
arch/x86/kernel/cpu/mce/intel.c +46 -1
+		mce_timer_kick(true);
+}
+
+static void cmci_storm_end(int bank)
+{
+	__clear_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_history[bank], 0ull);
+	this_cpu_write(bank_storm[bank], false);
+	if (this_cpu_dec_return(stormy_bank_count) == 0)
+		mce_timer_kick(false);
+}
+
+void mce_intel_storm_tracker(int bank, u64 status)
+{
+	unsigned long now = jiffies, delta;
+	unsigned int shift;
+	u64 history;
+
+	delta = now - this_cpu_read(bank_time_stamp[bank]);
+	shift = this_cpu_read(bank_storm[bank]) ? 1 : (delta + HZBITS) / HZBITS;
+	history = (shift < 64) ? this_cpu_read(bank_history[bank]) << shift : 0;
+	this_cpu_write(bank_time_stamp[bank], now);
+
+	if ((status & (MCI_STATUS_VAL | MCI_STATUS_UC)) == MCI_STATUS_VAL)
+		history |= 1;
+	this_cpu_write(bank_history[bank], history);
+
+	if (this_cpu_read(bank_storm[bank])) {
+		if (history & GENMASK_ULL(STORM_END - 1, 0))
+			return;
+		pr_notice("CPU%d BANK%d CMCI storm subsided\n", smp_processor_id(), bank);
+		cmci_set_threshold(bank, cmci_threshold[bank]);
+		cmci_storm_end(bank);
+	} else {
+		if (hweight64(history) < STORM_BEGIN)
+			return;
+		pr_notice("CPU%d BANK%d CMCI storm detected\n", smp_processor_id(), bank);
+		cmci_set_threshold(bank, CMCI_STORM_THRESHOLD);
+		cmci_storm_begin(bank);
+	}
+}
+
  /*
   * The interrupt handler. This is called on every event.
   * Just call the poller directly to log any events.
@@ -147,6 +251,9 @@ static void cmci_discover(int banks)
  			continue;
  		}
 
+		if ((val & MCI_CTL2_CMCI_THRESHOLD_MASK) == CMCI_STORM_THRESHOLD)
+			goto storm;
+
  		if (!mca_cfg.bios_cmci_threshold) {
  			val &= ~MCI_CTL2_CMCI_THRESHOLD_MASK;
  			val |= CMCI_THRESHOLD;
@@ -159,7 +266,7 @@ static void cmci_discover(int banks)
  			bios_zero_thresh = 1;
  			val |= CMCI_THRESHOLD;
  		}
-
+storm:
  		val |= MCI_CTL2_CMCI_EN;
  		wrmsrl(MSR_IA32_MCx_CTL2(i), val);
  		rdmsrl(MSR_IA32_MCx_CTL2(i), val);

arch/x86/kernel/cpu/mce/intel.c static void cmci_discover(int banks) · line 277

Quoted 5 lines
arch/x86/kernel/cpu/mce/intel.c +1 -1
@@ -167,7 +274,14 @@ static void cmci_discover(int banks)
  		/* Did the enable bit stick? -- the bank supports CMCI */
  		if (val & MCI_CTL2_CMCI_EN) {
  			set_bit(i, owned);
-			__clear_bit(i, this_cpu_ptr(mce_poll_banks));
+			if ((val & MCI_CTL2_CMCI_THRESHOLD_MASK) == CMCI_STORM_THRESHOLD) {

Why val is checked twice? Before goto storm and after?

Quoted 11 lines
arch/x86/kernel/cpu/mce/intel.c +7 -0
+				pr_notice("CPU%d BANK%d CMCI inherited storm\n", smp_processor_id(), i);
+				this_cpu_write(bank_history[i], ~0ull);
+				this_cpu_write(bank_time_stamp[i], jiffies);
+				cmci_storm_begin(i);
+			} else {
+				__clear_bit(i, this_cpu_ptr(mce_poll_banks));
+			}
  			/*
  			 * We are able to set thresholds for some banks that
  			 * had a threshold of 0. This means the BIOS has not

arch/x86/kernel/cpu/mce/intel.c static void cmci_discover(int banks) · line 297

Quoted 7 lines
arch/x86/kernel/cpu/mce/intel.c +4 -0
@@ -177,6 +291,10 @@ static void cmci_discover(int banks)
  			if (mca_cfg.bios_cmci_threshold && bios_zero_thresh &&
  					(val & MCI_CTL2_CMCI_THRESHOLD_MASK))
  				bios_wrong_thresh = 1;
+
+			/* Save default threshold for each bank */
+			if (cmci_threshold[i] == 0)
+				cmci_threshold[i] = val & MCI_CTL2_CMCI_THRESHOLD_MASK;

So, this is the default threshold value for interrupts to fire after storm subsides?

Is this defaulted to 1? As the commit message reads.. .."That resets the threshold in IA32_MCi_CTL2 back to 1"..

Quoted 32 lines
arch/x86/kernel/cpu/mce/intel.c +2 -0
  		} else {
  			WARN_ON(!test_bit(i, this_cpu_ptr(mce_poll_banks)));
  		}
@@ -218,6 +336,8 @@ static void __cmci_disable_bank(int bank)
  	val &= ~MCI_CTL2_CMCI_EN;
  	wrmsrl(MSR_IA32_MCx_CTL2(bank), val);
  	__clear_bit(bank, this_cpu_ptr(mce_banks_owned));
+	if ((val & MCI_CTL2_CMCI_THRESHOLD_MASK) == CMCI_STORM_THRESHOLD)
+		cmci_storm_end(bank);
  }
 
  /*
arch/x86/kernel/cpu/mce/internal.h +2 -0
index f01d6cbeb809..6c7480bce977 100644
@@ -41,12 +41,14 @@ struct dentry *mce_get_debugfs_dir(void);
  extern mce_banks_t mce_banks_ce_disabled;
 
  #ifdef CONFIG_X86_MCE_INTEL
+void mce_intel_storm_tracker(int bank, u64 status);
  void cmci_disable_bank(int bank);
  void intel_init_cmci(void);
  void intel_init_lmce(void);
  void intel_clear_lmce(void);
  bool intel_filter_mce(struct mce *m);
  #else
+static inline void mce_intel_storm_tracker(int bank, u64 status) { }
  static inline void cmci_disable_bank(int bank) { }
  static inline void intel_init_cmci(void) { }
  static inline void intel_init_lmce(void) { }

arch/x86/kernel/cpu/mce/internal.h static inline void intel_clear_lmce(void) { } · line 62

Quoted 8 lines
arch/x86/kernel/cpu/mce/internal.h +1 -1
@@ -54,7 +56,7 @@ static inline void intel_clear_lmce(void) { }
  static inline bool intel_filter_mce(struct mce *m) { return false; }
  #endif
 
-void mce_timer_kick(unsigned long interval);
+void mce_timer_kick(bool storm);
 
  #ifdef CONFIG_ACPI_APEI
  int apei_write_mce(struct mce *m);

Seems to me that most of the code can be shared except in few places.

Should I come up with a shared code by keeping Tony's patch as reference and incorporating AMD's changes in them?

Thanks,
Smita

maintainer
#7 lore raw

Re: [PATCH 2/2] x86/mce: Add per-bank CMCI storm mitigation

In reply to Luck, Tony

arch/x86/kernel/cpu/mce/core.c bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b) · line 717

Quoted 9 lines · Tony
arch/x86/kernel/cpu/mce/core.c +1 -0
index 4f9abb66520d..1f3e7c074182 100644
@@ -714,6 +714,8 @@ bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
 		barrier();
 		m.status = mce_rdmsrl(mca_msr_reg(i, MCA_STATUS));
 
+		mce_intel_storm_tracker(i, m.status);

Why is this called before the VALID bit check?

Because you want to still run the tracker on each polling - not only when it sees a valid error?

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 52

Quoted 11 lines
arch/x86/kernel/cpu/mce/intel.c +3 -0
index cee9d989f791..2ed5634ec277 100644
@@ -47,8 +47,48 @@ static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
  */
 static DEFINE_RAW_SPINLOCK(cmci_discover_lock);
 
+/*
+ * CMCI storm tracking state
+ */

Those could use some comments explaining what is tracking what:

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 55

Quoted 3 lines
arch/x86/kernel/cpu/mce/intel.c +3 -0
+static DEFINE_PER_CPU(int, stormy_bank_count);
+static DEFINE_PER_CPU(u64 [MAX_NR_BANKS], bank_history);
+static DEFINE_PER_CPU(bool [MAX_NR_BANKS], bank_storm);

AFAICT, this says whether a bank is in storm mode?

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 56

Quoted 1 line
arch/x86/kernel/cpu/mce/intel.c +1 -0
+static DEFINE_PER_CPU(unsigned long [MAX_NR_BANKS], bank_time_stamp);

This looks like it collects the jiffies when the bank was looked at in the storm tracker.

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 66

Quoted 10 lines
arch/x86/kernel/cpu/mce/intel.c +8 -0
+static int cmci_threshold[MAX_NR_BANKS];
+
 #define CMCI_THRESHOLD		1
 
+/*
+ * High threshold to limit CMCI rate during storms. Max supported is
+ * 0x7FFF. Use this slightly smaller value so it has a distinctive
+ * signature when some asks "Why am I not seeing all corrected errors?"
+ */
+#define CMCI_STORM_THRESHOLD	0x7FED

Why is a "threshold" in hex?

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 71

Quoted 5 lines
arch/x86/kernel/cpu/mce/intel.c +5 -0
+
+/*
+ * How many errors within the history buffer mark the start of a storm
+ */
+#define STORM_BEGIN	5

That looks like a STORM_BEGIN_THRESHOLD to me.

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 77

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+
+/*
+ * How many polls of machine check bank without an error before declaring
+ * the storm is over
+ */
+#define STORM_END	30

Similarly:

STORM_END_POLL_THRESHOLD

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 83

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+
+/*
+ * If there is no poll data for a bank for this amount of time, just
+ * discard the history.
+ */
+#define STORM_INTERVAL (1 * HZ)

That looks unused.

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 167

Quoted 5 lines
arch/x86/kernel/cpu/mce/intel.c +5 -0
+static void cmci_storm_begin(int bank)
+{
+	__set_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_storm[bank], true);
+	if (this_cpu_inc_return(stormy_bank_count) == 1)

s/ == 1//

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 176

Quoted 9 lines
arch/x86/kernel/cpu/mce/intel.c +9 -0
+		mce_timer_kick(true);
+}
+
+static void cmci_storm_end(int bank)
+{
+	__clear_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_history[bank], 0ull);
+	this_cpu_write(bank_storm[bank], false);
+	if (this_cpu_dec_return(stormy_bank_count) == 0)
	if (!...

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 180

Quoted 4 lines
arch/x86/kernel/cpu/mce/intel.c +4 -0
+		mce_timer_kick(false);
+}
+
+void mce_intel_storm_tracker(int bank, u64 status)

Function name needs a verb.

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 187

Quoted 7 lines
arch/x86/kernel/cpu/mce/intel.c +7 -0
+{
+	unsigned long now = jiffies, delta;
+	unsigned int shift;
+	u64 history;
+
+	delta = now - this_cpu_read(bank_time_stamp[bank]);
+	shift = this_cpu_read(bank_storm[bank]) ? 1 : (delta + HZBITS) / HZBITS;

Do

	shift = 1;

on function entry to simplify this assignment.

Also, I'm having trouble with this shift calculation. The laptop here has HZ=250. Let's say delta is 2000 jiffies.

So if this bank wasn't in storm mode, I'd have

shift = (2000 + (250 / 64)) / (250 / 64) = 513

...

Aaaha, so only when the diff is < 250 in my case, i.e., it polls the same bank within a second, only then it would shift the history. I.e., what you mean with that "The 64 bit width corresponds to about one second."

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 197

Quoted 10 lines
arch/x86/kernel/cpu/mce/intel.c +10 -0
+	history = (shift < 64) ? this_cpu_read(bank_history[bank]) << shift : 0;
+	this_cpu_write(bank_time_stamp[bank], now);
+
+	if ((status & (MCI_STATUS_VAL | MCI_STATUS_UC)) == MCI_STATUS_VAL)
+		history |= 1;
+	this_cpu_write(bank_history[bank], history);
+
+	if (this_cpu_read(bank_storm[bank])) {
+		if (history & GENMASK_ULL(STORM_END - 1, 0))
+			return;

Aha, under STORM_END polls you don't declare the storm as being over.

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 203

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+		pr_notice("CPU%d BANK%d CMCI storm subsided\n", smp_processor_id(), bank);
+		cmci_set_threshold(bank, cmci_threshold[bank]);
+		cmci_storm_end(bank);
+	} else {
+		if (hweight64(history) < STORM_BEGIN)
+			return;

Aha, so you need STORM_BEGIN errors within the last second to cause the storm polling. Ok I guess.

So all in all I can't find anything eeewy in this - it would need to have a lot more documentation, though, as this is not the most trivial thing to stare at.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
maintainer
#8 lore raw

Re: [PATCH 2/2] x86/mce: Add per-bank CMCI storm mitigation

Quoted 3 lines · Borislav Petkov

So all in all I can't find anything eeewy in this - it would need to have a lot more documentation, though, as this is not the most trivial thing to stare at.

Thanks for the review. I'll see about re-naming things and adding comments to make this easier to read.

Quoted 8 lines

arch/x86/kernel/cpu/mce/core.c bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b) · line 717

Quoted 3 lines
arch/x86/kernel/cpu/mce/core.c +1 -0
 		m.status = mce_rdmsrl(mca_msr_reg(i, MCA_STATUS));
 
+		mce_intel_storm_tracker(i, m.status);

Why is this called before the VALID bit check?

Because you want to still run the tracker on each polling - not only when it sees a valid error?

Yes. The tracker cares both about polls that find errors, and polls that don't. Storm detection is triggered with some threshold of positive scans, even if they are mixed in with some negative ones. End of storm is determined by a number of consecutive negative polls.

Quoted 8 lines
Quoted 8 lines
arch/x86/kernel/cpu/mce/intel.c +0 -0
index cee9d989f791..2ed5634ec277 100644
@@ -47,8 +47,48 @@ static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned);
  */
 static DEFINE_RAW_SPINLOCK(cmci_discover_lock);
 
Quoted 7 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 52

Quoted 3 lines
arch/x86/kernel/cpu/mce/intel.c +3 -0
+/*
+ * CMCI storm tracking state
+ */

Those could use some comments explaining what is tracking what:

Quoted 1 line
arch/x86/kernel/cpu/mce/intel.c +1 -0
+static DEFINE_PER_CPU(int, stormy_bank_count);

This one is a count of how many banks on this CPU are in storm mode.

Quoted 1 line

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 54

Quoted 1 line
arch/x86/kernel/cpu/mce/intel.c +1 -0
+static DEFINE_PER_CPU(u64 [MAX_NR_BANKS], bank_history);

Bitmask history of the most recent 64 polls for each bank.

Quoted 3 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 55

Quoted 1 line
arch/x86/kernel/cpu/mce/intel.c +1 -0
+static DEFINE_PER_CPU(bool [MAX_NR_BANKS], bank_storm);

AFAICT, this says whether a bank is in storm mode?

Yes.

Quoted 4 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 56

Quoted 1 line
arch/x86/kernel/cpu/mce/intel.c +1 -0
+static DEFINE_PER_CPU(unsigned long [MAX_NR_BANKS], bank_time_stamp);

This looks like it collects the jiffies when the bank was looked at in the storm tracker.

Yes.

Quoted 4 lines
Quoted 4 lines
arch/x86/kernel/cpu/mce/intel.c +2 -0
+static int cmci_threshold[MAX_NR_BANKS];
+
 #define CMCI_THRESHOLD		1
 
Quoted 8 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 66

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+/*
+ * High threshold to limit CMCI rate during storms. Max supported is
+ * 0x7FFF. Use this slightly smaller value so it has a distinctive
+ * signature when some asks "Why am I not seeing all corrected errors?"
+ */
+#define CMCI_STORM_THRESHOLD	0x7FED

Why is a "threshold" in hex?

For debugging ... I was using rdmsr tool to read the MCi_CTL2 machine check bank MSRs. It could be defined in decimal.

Quoted 7 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 71

Quoted 5 lines
arch/x86/kernel/cpu/mce/intel.c +5 -0
+
+/*
+ * How many errors within the history buffer mark the start of a storm
+ */
+#define STORM_BEGIN	5

That looks like a STORM_BEGIN_THRESHOLD to me.

Yes.

Quoted 10 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 77

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+
+/*
+ * How many polls of machine check bank without an error before declaring
+ * the storm is over
+ */
+#define STORM_END	30

Similarly:

STORM_END_POLL_THRESHOLD

Ditto yes.

Quoted 8 lines

arch/x86/kernel/cpu/mce/intel.c static DEFINE_PER_CPU(mce_banks_t, mce_banks_owned); · line 83

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+
+/*
+ * If there is no poll data for a bank for this amount of time, just
+ * discard the history.
+ */
+#define STORM_INTERVAL (1 * HZ)

That looks unused.

It was at one point a replacement for CMCI_STORM_INTERVAL deleted in patch 0001. But I must have dropped the place it was used. Will delete.

Quoted 7 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 167

Quoted 5 lines
arch/x86/kernel/cpu/mce/intel.c +5 -0
+static void cmci_storm_begin(int bank)
+{
+	__set_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_storm[bank], true);
+	if (this_cpu_inc_return(stormy_bank_count) == 1)

s/ == 1//

Not sure about this. The stormy_bank_count variable keeps track of how many banks on this CPU are in storm mode. So the transition from zero to one is significant. Need to start polling. But for 1->2, 2->3 etc. nothing needs to happen. The CPU is already polling.

Quoted 11 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 176

Quoted 9 lines
arch/x86/kernel/cpu/mce/intel.c +9 -0
+		mce_timer_kick(true);
+}
+
+static void cmci_storm_end(int bank)
+{
+	__clear_bit(bank, this_cpu_ptr(mce_poll_banks));
+	this_cpu_write(bank_history[bank], 0ull);
+	this_cpu_write(bank_storm[bank], false);
+	if (this_cpu_dec_return(stormy_bank_count) == 0)
	if (!...

OK.

Quoted 6 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 180

Quoted 4 lines
arch/x86/kernel/cpu/mce/intel.c +4 -0
+		mce_timer_kick(false);
+}
+
+void mce_intel_storm_tracker(int bank, u64 status)

Function name needs a verb.

How about: track_cmci_storm(int bank, u64 status) ?

Quoted 10 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 187

Quoted 7 lines
arch/x86/kernel/cpu/mce/intel.c +7 -0
+{
+	unsigned long now = jiffies, delta;
+	unsigned int shift;
+	u64 history;
+
+	delta = now - this_cpu_read(bank_time_stamp[bank]);
+	shift = this_cpu_read(bank_storm[bank]) ? 1 : (delta + HZBITS) / HZBITS;

Do

Quoted 3 lines
	shift = 1;

on function entry to simplify this assignment.

Yes. That will be easier to read.

Quoted 5 lines

Also, I'm having trouble with this shift calculation. The laptop here has HZ=250. Let's say delta is 2000 jiffies.

So if this bank wasn't in storm mode, I'd have

Quoted 8 lines

shift = (2000 + (250 / 64)) / (250 / 64) = 513

...

Aaaha, so only when the diff is < 250 in my case, i.e., it polls the same bank within a second, only then it would shift the history. I.e., what you mean with that "The 64 bit width corresponds to about one second."

Yes, you got it ... but I'll add a comment so code readers don't have to redo that deduction every time.

Quoted 13 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 197

Quoted 10 lines
arch/x86/kernel/cpu/mce/intel.c +10 -0
+	history = (shift < 64) ? this_cpu_read(bank_history[bank]) << shift : 0;
+	this_cpu_write(bank_time_stamp[bank], now);
+
+	if ((status & (MCI_STATUS_VAL | MCI_STATUS_UC)) == MCI_STATUS_VAL)
+		history |= 1;
+	this_cpu_write(bank_history[bank], history);
+
+	if (this_cpu_read(bank_storm[bank])) {
+		if (history & GENMASK_ULL(STORM_END - 1, 0))
+			return;

Aha, under STORM_END polls you don't declare the storm as being over.

Quoted 9 lines

arch/x86/kernel/cpu/mce/intel.c static bool lmce_supported(void) · line 203

Quoted 6 lines
arch/x86/kernel/cpu/mce/intel.c +6 -0
+		pr_notice("CPU%d BANK%d CMCI storm subsided\n", smp_processor_id(), bank);
+		cmci_set_threshold(bank, cmci_threshold[bank]);
+		cmci_storm_end(bank);
+	} else {
+		if (hweight64(history) < STORM_BEGIN)
+			return;

Aha, so you need STORM_BEGIN errors within the last second to cause the storm polling. Ok I guess.

Agreed. This needs comments to explain what is going on.

Quoted 3 lines

So all in all I can't find anything eeewy in this - it would need to have a lot more documentation, though, as this is not the most trivial thing to stare at.

Thanks again for the review.

-Tony

maintainer
#9 lore raw

Re: [RFC PATCH 1/2] x86/mce: Handle AMD threshold interrupt storms

In reply to Luck, Tony
Quoted 4 lines · Tony

I've been sitting on some partially done patches to re-work storm handling for Intel ... which rips out all the existing storm bits and replaces with something all new. I'll post the 2-part series as replies to this.

Which begs the obvious question: how much of that code can be shared between the two?

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#10 lore raw

Re: [RFC PATCH 1/2] x86/mce: Handle AMD threshold interrupt storms

Quoted 8 lines · Borislav Petkov
Quoted 4 lines · Tony

I've been sitting on some partially done patches to re-work storm handling for Intel ... which rips out all the existing storm bits and replaces with something all new. I'll post the 2-part series as replies to this.

Which begs the obvious question: how much of that code can be shared between the two?

It looks to me most of the code can be shared except in few places where AMD and Intel use different registers to set error thresholds. And the fact that AMD's threshold interrupts just handles corrected errors unlike CMCI.

I'm thinking of coming up with a shared code between both by keeping the Intel's new storm handling code as base and incorporating AMD changes on them and send for review.

Let me know if thats okay?

Thanks,
Smita

maintainer
#11 lore raw

RE: [RFC PATCH 1/2] x86/mce: Handle AMD threshold interrupt storms

Quoted 2 lines

It looks to me most of the code can be shared except in few places where AMD and Intel use different registers to set error thresholds.

Hopefully easy to abstract.

Quoted 2 lines

And the fact that AMD's threshold interrupts just handles corrected errors unlike CMCI.

That makes your life much simpler than mine :-)

Quoted 5 lines

I'm thinking of coming up with a shared code between both by keeping the Intel's new storm handling code as base and incorporating AMD changes on them and send for review.

Let me know if thats okay?

My new Intel code hasn't had Boris look through it yet to point out all the bits where I have bugs, or just make things more complex than they need to be.

So it would be helpful if Boris could do at least a quick scan to say my code is a worthy base. I'd hate to see you waste time building a merged version and then have Boris say "Nack".

-Tony

maintainer
#12 lore raw

v2 cover New CMCI storm mitigation for Intel CPUs

In reply to Luck, Tony

Two-part motivation:

1) Disabling CMCI globally is an overly big hammer

2) Intel signals some UNCORRECTED errors using CMCI (yes, turns out that was a poorly chosen name given the later evolution of the architecture). Since we don't want to miss those, the proposed storm code just bumps the threshold to (almost) maximum to mitigate, but not eliminate the storm. Note that the threshold only applies to corrected errors.

Patch 1 deletes the parts of the old storm code that are no longer needed.

Patch 2 adds the new per-bank mitigation.

Smita: Unless Boris finds a some more stuff for me to fix, this version will be a better starting point to merge with your changes.

Changes since v1 (based on feedback from Boris)

- Spelling fixes in commit message
- Many more comments explaining what is going on
- Change name of function that does tracking
- Change names for #defines for storm BEGIN/END
- #define for high threshold in decimal, not hex

Tony Luck (2):
x86/mce: Remove old CMCI storm mitigation code
x86/mce: Add per-bank CMCI storm mitigation

 arch/x86/kernel/cpu/mce/core.c     |  46 +++---
 arch/x86/kernel/cpu/mce/intel.c    | 241 ++++++++++++++---------------
 arch/x86/kernel/cpu/mce/internal.h |  10 +-
 3 files changed, 141 insertions(+), 156 deletions(-)

base-commit: ffb217a13a2eaf6d5bd974fc83036a53ca69f1e2

Signature · 1 line
2.35.1
maintainer
#13 lore raw

PATCH v2 1/2 x86/mce: Remove old CMCI storm mitigation code

When a "storm" of CMCI is detected this code mitigates by disabling CMCI interrupt signalling from all of the banks owned by the CPU that saw the storm.

There are problems with this approach:

1) It is very coarse grained. In all likelihood only one of the banks was generating the interrupts, but CMCI is disabled for all. This means Linux may delay seeing and processing errors logged from other banks.

2) Although CMCI stands for Corrected Machine Check Interrupt, it is also used to signal when an uncorrected error is logged. This is a problem because these errors should be handled in a timely manner.

Delete all this code in preparation for a finer grained solution.

---
 arch/x86/kernel/cpu/mce/core.c     |  20 +---
 arch/x86/kernel/cpu/mce/intel.c    | 145 -----------------------------
 arch/x86/kernel/cpu/mce/internal.h |   6 --
 3 files changed, 1 insertion(+), 170 deletions(-)
arch/x86/kernel/cpu/mce/core.c +1 -19
index 5818b837fd4d..396484141ee1 100644
@@ -1572,13 +1572,6 @@ static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
 static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
 static DEFINE_PER_CPU(struct timer_list, mce_timer);
 
-static unsigned long mce_adjust_timer_default(unsigned long interval)
-{
-	return interval;
-}
-
-static unsigned long (*mce_adjust_timer)(unsigned long interval) = mce_adjust_timer_default;
-
 static void __start_timer(struct timer_list *t, unsigned long interval)
 {
 	unsigned long when = jiffies + interval;
@@ -1601,15 +1594,9 @@ static void mce_timer_fn(struct timer_list *t)
 
 	iv = __this_cpu_read(mce_next_interval);
 
-	if (mce_available(this_cpu_ptr(&cpu_info))) {
+	if (mce_available(this_cpu_ptr(&cpu_info)))
 		machine_check_poll(0, this_cpu_ptr(&mce_poll_banks));
 
-		if (mce_intel_cmci_poll()) {
-			iv = mce_adjust_timer(iv);
-			goto done;
-		}
-	}
-
 	/*
 	 * Alert userspace if needed. If we logged an MCE, reduce the polling
 	 * interval, otherwise increase the polling interval.
@@ -1619,7 +1606,6 @@ static void mce_timer_fn(struct timer_list *t)
 	else
 		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
 
-done:
 	__this_cpu_write(mce_next_interval, iv);
 	__start_timer(t, iv);
 }
@@ -1949,7 +1935,6 @@ static void mce_zhaoxin_feature_init(struct cpuinfo_x86 *c)
 
 	intel_init_cmci();
 	intel_init_lmce();
-	mce_adjust_timer = cmci_intel_adjust_timer;
 }
 
 static void mce_zhaoxin_feature_clear(struct cpuinfo_x86 *c)
@@ -1962,7 +1947,6 @@ static void __mcheck_cpu_init_vendor(struct cpuinfo_x86 *c)
 	switch (c->x86_vendor) {
 	case X86_VENDOR_INTEL:
 		mce_intel_feature_init(c);
-		mce_adjust_timer = cmci_intel_adjust_timer;
 		break;
 
 	case X86_VENDOR_AMD: {
@@ -2621,8 +2605,6 @@ static void mce_reenable_cpu(void)
 
 static int mce_cpu_dead(unsigned int cpu)
 {
-	mce_intel_hcpu_update(cpu);
-
 	/* intentionally ignoring frozen here */
 	if (!cpuhp_tasks_frozen)
 		cmci_rediscover();
arch/x86/kernel/cpu/mce/intel.c +0 -145

169 lines not on this page · raw message

arch/x86/kernel/cpu/mce/internal.h +0 -6
index 52c633950b38..dfbd0bca67a0 100644
@@ -41,18 +41,12 @@ struct dentry *mce_get_debugfs_dir(void);
 extern mce_banks_t mce_banks_ce_disabled;
 
 #ifdef CONFIG_X86_MCE_INTEL
-unsigned long cmci_intel_adjust_timer(unsigned long interval);
-bool mce_intel_cmci_poll(void);
-void mce_intel_hcpu_update(unsigned long cpu);
 void cmci_disable_bank(int bank);
 void intel_init_cmci(void);
 void intel_init_lmce(void);
 void intel_clear_lmce(void);
 bool intel_filter_mce(struct mce *m);
 #else
-# define cmci_intel_adjust_timer mce_adjust_timer_default
-static inline bool mce_intel_cmci_poll(void) { return false; }
-static inline void mce_intel_hcpu_update(unsigned long cpu) { }
 static inline void cmci_disable_bank(int bank) { }
 static inline void intel_init_cmci(void) { }
 static inline void intel_init_lmce(void) { }
Signature · 1 line
2.35.1
maintainer
#14 lore raw

PATCH v2 2/2 x86/mce: Add per-bank CMCI storm mitigation

In reply to Tony Luck

Add a hook into machine_check_poll() to keep track of per-CPU, per-bank corrected error logs.

Maintain a bitmap history for each bank showing whether the bank logged an corrected error or not each time it is polled.

In normal operation the interval between polls of this banks determines how far to shift the history. The 64 bit width corresponds to about one second.

When a storm is observed the Rate of interrupts is reduced by setting a large threshold value for this bank in IA32_MCi_CTL2. This bank is added to the bitmap of banks for this CPU to poll. The polling rate is increased to once per second.
During a storm each bit in the history indicates the status of the bank each time it is polled. Thus the history covers just over a minute.

Declare a storm for that bank if the number of corrected interrupts seen in that history is above some threshold (5 in this RFC code for ease of testing, likely move to 15 for compatibility with previous storm detection).

A storm on a bank ends if enough consecutive polls of the bank show no corrected errors (currently 30, may also change). That resets the threshold in IA32_MCi_CTL2 back to 1, removes the bank from the bitmap for polling, and changes the polling rate back to the default.

If a CPU with banks in storm mode is taken offline, the new CPU that inherits ownership of those banks takes over management of storm(s) in the inherited bank(s).

---
 arch/x86/kernel/cpu/mce/core.c     |  26 +++--
 arch/x86/kernel/cpu/mce/intel.c    | 146 ++++++++++++++++++++++++++++-
 arch/x86/kernel/cpu/mce/internal.h |   4 +-
 3 files changed, 165 insertions(+), 11 deletions(-)
arch/x86/kernel/cpu/mce/core.c +18 -8
index 396484141ee1..6e62140eed97 100644
@@ -726,6 +726,8 @@ bool machine_check_poll(enum mcp_flags flags, mce_banks_t *b)
 		barrier();
 		m.status = mce_rdmsrl(mca_msr_reg(i, MCA_STATUS));
 
+		track_cmci_storm(i, m.status);
+
 		/* If this entry is not valid, ignore it */
 		if (!(m.status & MCI_STATUS_VAL))
 			continue;
@@ -1571,6 +1573,7 @@ static unsigned long check_interval = INITIAL_CHECK_INTERVAL;
 
 static DEFINE_PER_CPU(unsigned long, mce_next_interval); /* in jiffies */
 static DEFINE_PER_CPU(struct timer_list, mce_timer);
+static DEFINE_PER_CPU(bool, storm_poll_mode);
 
 static void __start_timer(struct timer_list *t, unsigned long interval)
 {
@@ -1606,22 +1609,29 @@ static void mce_timer_fn(struct timer_list *t)
 	else
 		iv = min(iv * 2, round_jiffies_relative(check_interval * HZ));
 
-	__this_cpu_write(mce_next_interval, iv);
-	__start_timer(t, iv);
+	if (__this_cpu_read(storm_poll_mode)) {
+		__start_timer(t, HZ);
+	} else {
+		__this_cpu_write(mce_next_interval, iv);
+		__start_timer(t, iv);
+	}
 }
 
 /*
- * Ensure that the timer is firing in @interval from now.
+ * When a storm starts on any bank on this CPU, switch to polling
+ * once per second. When the storm ends, revert to the default
+ * polling interval.
  */
-void mce_timer_kick(unsigned long interval)
+void mce_timer_kick(bool storm)
 {
 	struct timer_list *t = this_cpu_ptr(&mce_timer);
-	unsigned long iv = __this_cpu_read(mce_next_interval);
 
-	__start_timer(t, interval);
+	__this_cpu_write(storm_poll_mode, storm);
 
-	if (interval < iv)
-		__this_cpu_write(mce_next_interval, interval);
+	if (storm)
+		__start_timer(t, HZ);
+	else
+		__this_cpu_write(mce_next_interval, check_interval * HZ);
 }
 
 /* Must not be called in IRQ context where del_timer_sync() can deadlock */
arch/x86/kernel/cpu/mce/intel.c +144 -2

190 lines not on this page · raw message

arch/x86/kernel/cpu/mce/internal.h +3 -1
index dfbd0bca67a0..4822fd0ab477 100644
@@ -41,12 +41,14 @@ struct dentry *mce_get_debugfs_dir(void);
 extern mce_banks_t mce_banks_ce_disabled;
 
 #ifdef CONFIG_X86_MCE_INTEL
+void track_cmci_storm(int bank, u64 status);
 void cmci_disable_bank(int bank);
 void intel_init_cmci(void);
 void intel_init_lmce(void);
 void intel_clear_lmce(void);
 bool intel_filter_mce(struct mce *m);
 #else
+static inline void track_cmci_storm(int bank, u64 status) { }
 static inline void cmci_disable_bank(int bank) { }
 static inline void intel_init_cmci(void) { }
 static inline void intel_init_lmce(void) { }
@@ -54,7 +56,7 @@ static inline void intel_clear_lmce(void) { }
 static inline bool intel_filter_mce(struct mce *m) { return false; }
 #endif
 
-void mce_timer_kick(unsigned long interval);
+void mce_timer_kick(bool storm);
 
 #ifdef CONFIG_ACPI_APEI
 int apei_write_mce(struct mce *m);
Signature · 1 line
2.35.1
maintainer
#15 lore raw

Re: [PATCH v2 0/2] New CMCI storm mitigation for Intel CPUs

In reply to Tony Luck
Quoted 2 lines · Tony Luck

Smita: Unless Boris finds a some more stuff for me to fix, this version will be a better starting point to merge with your changes.

Right, I'm wondering if AMD can use the same scheme so that abstracting out the hw-specific accesses (MSR writes, etc) would be enough...

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#16 lore raw

Re: [PATCH v2 0/2] New CMCI storm mitigation for Intel CPUs

Quoted 5 lines · Borislav Petkov
Quoted 2 lines · Tony Luck

Smita: Unless Boris finds a some more stuff for me to fix, this version will be a better starting point to merge with your changes.

Right, I'm wondering if AMD can use the same scheme so that abstracting out the hw-specific accesses (MSR writes, etc) would be enough...

Thanks Tony.

Agreed. Most of this would apply for AMD's threshold interrupts too.

Will come up with a merged patch and move the storm handling to mce/core.c and just keep the hw-specific accesses separate for Intel and AMD in their respective files.

Thanks
Smita.

#17 lore raw

RFC PATCH 2/2 x86/mce: Simplify code in log_and_reset_block()

In reply to Smita Koralahalli

Reuse the existing _reset_block() to reset the threshold block after logging error in log_and_reset_block().

---
 arch/x86/kernel/cpu/mce/amd.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
arch/x86/kernel/cpu/mce/amd.c +1 -4
index 53d9320d1470..823733468973 100644
@@ -960,7 +960,6 @@ static void log_error_thresholding(unsigned int bank, u64 misc)
 
 static void log_and_reset_block(struct threshold_block *block)
 {
-	struct thresh_restart tr;
 	u32 low = 0, high = 0;
 
 	if (!block)
@@ -976,9 +975,7 @@ static void log_and_reset_block(struct threshold_block *block)
 	log_error_thresholding(block->bank, ((u64)high << 32) | low);
 
 	/* Reset threshold block after logging error. */
-	memset(&tr, 0, sizeof(tr));
-	tr.b = block;
-	threshold_restart_bank(&tr);
+	_reset_block(block);
 }
 
 /*
Signature · 1 line
2.17.1