Skip to contentSkip to conversation

x86/idle: fix for Intel ISR errata

PATCH In review Roger Pau Monne 3 participants 11 messages May 15, 2020 – May 21, 2020

lore
Share & tools
lore
#0 lore raw

v3 cover x86/idle: fix for Intel ISR errata

Apply a workaround in order to cope with the BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131 erratas.

Roger Pau Monne (2):
x86/idle: rework C6 EOI workaround
x86/idle: prevent entering C6 with in service interrupts on Intel

 docs/misc/xen-command-line.pandoc |  9 ++++
 xen/arch/x86/acpi/cpu_idle.c      | 74 ++++++++++++++++++++++++-------
 xen/arch/x86/cpu/mwait-idle.c     |  3 ++
 xen/include/asm-x86/cpuidle.h     |  1 +
 4 files changed, 70 insertions(+), 17 deletions(-)
Signature · 1 line
2.26.2
#1 lore raw

PATCH v3 1/2 x86/idle: rework C6 EOI workaround

Change the C6 EOI workaround (errata AAJ72) to use x86_match_cpu. Also call the workaround from mwait_idle, previously it was only used by the ACPI idle driver. Finally make sure the routine is called for all states equal or greater than ACPI_STATE_C3, note that the ACPI driver doesn't currently handle them, but the errata condition shouldn't be limited by that.

---
Changes since v2:
- New in this version.

---
 xen/arch/x86/acpi/cpu_idle.c  | 43 +++++++++++++++++++++--------------
 xen/arch/x86/cpu/mwait-idle.c |  3 +++
 xen/include/asm-x86/cpuidle.h |  1 +
 3 files changed, 30 insertions(+), 17 deletions(-)
xen/arch/x86/acpi/cpu_idle.c +26 -17
index b83446e77d..0efdaff21b 100644
@@ -548,26 +548,35 @@ void trace_exit_reason(u32 *irq_traced)
     }
 }
 
-/*
- * "AAJ72. EOI Transaction May Not be Sent if Software Enters Core C6 During 
- * an Interrupt Service Routine"
- * 
- * There was an errata with some Core i7 processors that an EOI transaction 
- * may not be sent if software enters core C6 during an interrupt service 
- * routine. So we don't enter deep Cx state if there is an EOI pending.
- */
-static bool errata_c6_eoi_workaround(void)
+bool errata_c6_eoi_workaround(void)
 {
-    static int8_t fix_needed = -1;
+    static int8_t __read_mostly fix_needed = -1;
 
     if ( unlikely(fix_needed == -1) )
     {
-        int model = boot_cpu_data.x86_model;
-        fix_needed = (cpu_has_apic && !directed_eoi_enabled &&
-                      (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
-                      (boot_cpu_data.x86 == 6) &&
-                      ((model == 0x1a) || (model == 0x1e) || (model == 0x1f) ||
-                       (model == 0x25) || (model == 0x2c) || (model == 0x2f)));
+#define INTEL_FAM6_MODEL(m) { X86_VENDOR_INTEL, 6, m, X86_FEATURE_ALWAYS }
+        /*
+         * Errata AAJ72: EOI Transaction May Not be Sent if Software Enters
+         * Core C6 During an Interrupt Service Routine"
+         *
+         * There was an errata with some Core i7 processors that an EOI
+         * transaction may not be sent if software enters core C6 during an
+         * interrupt service routine. So we don't enter deep Cx state if
+         * there is an EOI pending.
+         */
+        const static struct x86_cpu_id eoi_errata[] = {
+            INTEL_FAM6_MODEL(0x1a),
+            INTEL_FAM6_MODEL(0x1e),
+            INTEL_FAM6_MODEL(0x1f),
+            INTEL_FAM6_MODEL(0x25),
+            INTEL_FAM6_MODEL(0x2c),
+            INTEL_FAM6_MODEL(0x2f),
+            { }
+        };
+#undef INTEL_FAM6_MODEL
+
+        fix_needed = cpu_has_apic && !directed_eoi_enabled &&
+                     x86_match_cpu(eoi_errata);
     }
 
     return (fix_needed && cpu_has_pending_apic_eoi());
@@ -676,7 +685,7 @@ static void acpi_processor_idle(void)
         return;
     }
 
-    if ( (cx->type == ACPI_STATE_C3) && errata_c6_eoi_workaround() )
+    if ( (cx->type >= ACPI_STATE_C3) && errata_c6_eoi_workaround() )
         cx = power->safe_state;
 
 
xen/arch/x86/cpu/mwait-idle.c +3 -0
index b81937966e..88a3e160c5 100644
@@ -770,6 +770,9 @@ static void mwait_idle(void)
 		return;
 	}
 
+	if ((cx->type >= 3) && errata_c6_eoi_workaround())
+		cx = power->safe_state;
+
 	eax = cx->address;
 	cstate = ((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
 
xen/include/asm-x86/cpuidle.h +1 -0
index 5d7dffd228..13879f58a1 100644
@@ -26,4 +26,5 @@ void update_idle_stats(struct acpi_processor_power *,
 void update_last_cx_stat(struct acpi_processor_power *,
                          struct acpi_processor_cx *, uint64_t);
 
+bool errata_c6_eoi_workaround(void);
 #endif /* __X86_ASM_CPUIDLE_H__ */
Signature · 1 line
2.26.2
Reviewed-by on v3 1/2
#2 lore raw

Re: [PATCH v3 1/2] x86/idle: rework C6 EOI workaround

Quoted 8 lines · Roger Pau Monne

Change the C6 EOI workaround (errata AAJ72) to use x86_match_cpu. Also call the workaround from mwait_idle, previously it was only used by the ACPI idle driver. Finally make sure the routine is called for all states equal or greater than ACPI_STATE_C3, note that the ACPI driver doesn't currently handle them, but the errata condition shouldn't be limited by that.

with two nits:

Quoted 29 lines
xen/arch/x86/acpi/cpu_idle.c +3 -16
@@ -548,26 +548,35 @@ void trace_exit_reason(u32 *irq_traced)
     }
 }
 
-/*
- * "AAJ72. EOI Transaction May Not be Sent if Software Enters Core C6 During 
- * an Interrupt Service Routine"
- * 
- * There was an errata with some Core i7 processors that an EOI transaction 
- * may not be sent if software enters core C6 during an interrupt service 
- * routine. So we don't enter deep Cx state if there is an EOI pending.
- */
-static bool errata_c6_eoi_workaround(void)
+bool errata_c6_eoi_workaround(void)
 {
-    static int8_t fix_needed = -1;
+    static int8_t __read_mostly fix_needed = -1;
 
     if ( unlikely(fix_needed == -1) )
     {
-        int model = boot_cpu_data.x86_model;
-        fix_needed = (cpu_has_apic && !directed_eoi_enabled &&
-                      (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
-                      (boot_cpu_data.x86 == 6) &&
-                      ((model == 0x1a) || (model == 0x1e) || (model == 0x1f) ||
-                       (model == 0x25) || (model == 0x2c) || (model == 0x2f)));
+#define INTEL_FAM6_MODEL(m) { X86_VENDOR_INTEL, 6, m, X86_FEATURE_ALWAYS }

xen/arch/x86/acpi/cpu_idle.c void trace_exit_reason(u32 *irq_traced) · line 567

Quoted 10 lines
xen/arch/x86/acpi/cpu_idle.c +10 -0
+        /*
+         * Errata AAJ72: EOI Transaction May Not be Sent if Software Enters
+         * Core C6 During an Interrupt Service Routine"
+         *
+         * There was an errata with some Core i7 processors that an EOI
+         * transaction may not be sent if software enters core C6 during an
+         * interrupt service routine. So we don't enter deep Cx state if
+         * there is an EOI pending.
+         */
+        const static struct x86_cpu_id eoi_errata[] = {

Commonly we use "static const".

Quoted 17 lines
xen/arch/x86/cpu/mwait-idle.c +3 -0
@@ -770,6 +770,9 @@ static void mwait_idle(void)
 		return;
 	}
 
+	if ((cx->type >= 3) && errata_c6_eoi_workaround())
+		cx = power->safe_state;
+
 	eax = cx->address;
 	cstate = ((eax >> MWAIT_SUBSTATE_SIZE) & MWAIT_CSTATE_MASK) + 1;
 
xen/include/asm-x86/cpuidle.h +0 -0
index 5d7dffd228..13879f58a1 100644

xen/include/asm-x86/cpuidle.h void update_idle_stats(struct acpi_processor_power *, · line 30

Quoted 5 lines
xen/include/asm-x86/cpuidle.h +1 -0
@@ -26,4 +26,5 @@ void update_idle_stats(struct acpi_processor_power *,
 void update_last_cx_stat(struct acpi_processor_power *,
                          struct acpi_processor_cx *, uint64_t);
 
+bool errata_c6_eoi_workaround(void);
 #endif /* __X86_ASM_CPUIDLE_H__ */

I'd prefer if a blank line was left ahead of #ifdef-s of this kind. Both are easy enough to do while committing, I guess.

Jan

#3 lore raw

PATCH v3 2/2 x86/idle: prevent entering C6 with in service interrupts on Intel

In reply to Roger Pau Monne

Apply a workaround for Intel errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131: "A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of The Same Priority Completes".

Apply the errata to all server and client models (big cores) from Broadwell to Cascade Lake. The workaround is grouped together with the existing fix for errata AAJ72, and the eoi from the function name is removed.

---
Changes since v2:
- Use x86_match_cpu and apply the workaround to all models from Broadwell to Cascade Lake.
- Rename command line option to disable-c6-errata.

Changes since v1:
- Unify workaround with errata_c6_eoi_workaround.
- Properly check state in both acpi and mwait drivers.

---
 docs/misc/xen-command-line.pandoc |  9 +++++++
 xen/arch/x86/acpi/cpu_idle.c      | 39 +++++++++++++++++++++++++++----
 xen/arch/x86/cpu/mwait-idle.c     |  2 +-
 xen/include/asm-x86/cpuidle.h     |  2 +-
 4 files changed, 46 insertions(+), 6 deletions(-)
docs/misc/xen-command-line.pandoc +9 -0
index ee12b0f53f..8dd944b357 100644
@@ -652,6 +652,15 @@ Specify the size of the console debug trace buffer. By specifying `cpu:`
 additionally a trace buffer of the specified size is allocated per cpu.
 The debug trace feature is only enabled in debugging builds of Xen.
 
+### disable-c6-errata
+> `= <boolean>`
+
+> Default: `true for affected Intel CPUs`
+
+Workaround for Intel errata AAJ72 and BDX99, CLX30, SKX100, CFW125, BDF104,
+BDH85, BDM135, KWB131. Prevent entering C6 idle states when certain conditions
+are meet in order to avoid triggering the listed erratas.
+
 ### dma_bits
 > `= <integer>`
 
xen/arch/x86/acpi/cpu_idle.c +35 -4
index 0efdaff21b..2fa1ccc031 100644
@@ -548,9 +548,10 @@ void trace_exit_reason(u32 *irq_traced)
     }
 }
 
-bool errata_c6_eoi_workaround(void)
+bool errata_c6_workaround(void)
 {
     static int8_t __read_mostly fix_needed = -1;
+    boolean_param("disable-c6-errata", fix_needed);
 
     if ( unlikely(fix_needed == -1) )
     {
@@ -573,10 +574,40 @@ bool errata_c6_eoi_workaround(void)
             INTEL_FAM6_MODEL(0x2f),
             { }
         };
+        /*
+         * Errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131:
+         * A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of
+         * The Same Priority Completes.
+         *
+         * Resuming from C6 Sleep-State, with Fixed Interrupts of the same
+         * priority queued (in the corresponding bits of the IRR and ISR APIC
+         * registers), the processor may dispatch the second interrupt (from
+         * the IRR bit) before the first interrupt has completed and written to
+         * the EOI register, causing the first interrupt to never complete.
+         */
+        const static struct x86_cpu_id isr_errata[] = {
+            /* Broadwell */
+            INTEL_FAM6_MODEL(0x47),
+            INTEL_FAM6_MODEL(0x3d),
+            INTEL_FAM6_MODEL(0x4f),
+            INTEL_FAM6_MODEL(0x56),
+            /* Skylake (client) */
+            INTEL_FAM6_MODEL(0x5e),
+            INTEL_FAM6_MODEL(0x4e),
+            /* {Sky/Cascade}lake (server) */
+            INTEL_FAM6_MODEL(0x55),
+            /* {Kaby/Coffee/Whiskey/Amber} Lake */
+            INTEL_FAM6_MODEL(0x9e),
+            INTEL_FAM6_MODEL(0x8e),
+            /* Cannon Lake */
+            INTEL_FAM6_MODEL(0x66),
+            { }
+        };
 #undef INTEL_FAM6_MODEL
 
-        fix_needed = cpu_has_apic && !directed_eoi_enabled &&
-                     x86_match_cpu(eoi_errata);
+        fix_needed = cpu_has_apic &&
+                     ((!directed_eoi_enabled && x86_match_cpu(eoi_errata)) ||
+                      x86_match_cpu(isr_errata));
     }
 
     return (fix_needed && cpu_has_pending_apic_eoi());
@@ -685,7 +716,7 @@ static void acpi_processor_idle(void)
         return;
     }
 
-    if ( (cx->type >= ACPI_STATE_C3) && errata_c6_eoi_workaround() )
+    if ( (cx->type >= ACPI_STATE_C3) && errata_c6_workaround() )
         cx = power->safe_state;
 
 
xen/arch/x86/cpu/mwait-idle.c +1 -1
index 88a3e160c5..52eab81bf8 100644
@@ -770,7 +770,7 @@ static void mwait_idle(void)
 		return;
 	}
 
-	if ((cx->type >= 3) && errata_c6_eoi_workaround())
+	if ((cx->type >= 3) && errata_c6_workaround())
 		cx = power->safe_state;
 
 	eax = cx->address;
xen/include/asm-x86/cpuidle.h +1 -1
index 13879f58a1..dc7298a538 100644
@@ -26,5 +26,5 @@ void update_idle_stats(struct acpi_processor_power *,
 void update_last_cx_stat(struct acpi_processor_power *,
                          struct acpi_processor_cx *, uint64_t);
 
-bool errata_c6_eoi_workaround(void);
+bool errata_c6_workaround(void);
 #endif /* __X86_ASM_CPUIDLE_H__ */
Signature · 1 line
2.26.2
#4 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

docs/misc/xen-command-line.pandoc Specify the size of the console debug trace buffer. By specifying `cpu:` · line 655

Quoted 7 lines · Roger Pau Monne
docs/misc/xen-command-line.pandoc +1 -0
@@ -652,6 +652,15 @@ Specify the size of the console debug trace buffer. By specifying `cpu:`
 additionally a trace buffer of the specified size is allocated per cpu.
 The debug trace feature is only enabled in debugging builds of Xen.
 
+### disable-c6-errata

Hmm, yes please - a disable for errata! ;-)

How about "avoid-c6-errata", and then perhaps as a sub-option to "cpuidle="? (If we really want a control for this in the first place.)

Quoted 6 lines
docs/misc/xen-command-line.pandoc +2 -0
@@ -573,10 +574,40 @@ bool errata_c6_eoi_workaround(void)
             INTEL_FAM6_MODEL(0x2f),
             { }
         };
+        /*
+         * Errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131:

docs/misc/xen-command-line.pandoc bool errata_c6_eoi_workaround(void) · line 588

Quoted 10 lines
docs/misc/xen-command-line.pandoc +10 -0
+         * A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of
+         * The Same Priority Completes.
+         *
+         * Resuming from C6 Sleep-State, with Fixed Interrupts of the same
+         * priority queued (in the corresponding bits of the IRR and ISR APIC
+         * registers), the processor may dispatch the second interrupt (from
+         * the IRR bit) before the first interrupt has completed and written to
+         * the EOI register, causing the first interrupt to never complete.
+         */
+        const static struct x86_cpu_id isr_errata[] = {

Same nit as for patch 1 here.

Jan

#5 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

Quoted 9 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

Quoted 6 lines · Roger Pau Monne
docs/misc/xen-command-line.pandoc +0 -0
@@ -652,6 +652,15 @@ Specify the size of the console debug trace buffer. By specifying `cpu:`
 additionally a trace buffer of the specified size is allocated per cpu.
 The debug trace feature is only enabled in debugging builds of Xen.
 
Quoted 7 lines

docs/misc/xen-command-line.pandoc Specify the size of the console debug trace buffer. By specifying `cpu:` · line 655

Quoted 1 line
docs/misc/xen-command-line.pandoc +1 -0
+### disable-c6-errata

Hmm, yes please - a disable for errata! ;-)

How about "avoid-c6-errata", and then perhaps as a sub-option to "cpuidle="? (If we really want a control for this in the first place.)

Right, I see I'm very bad at naming. Not sure it's even worth it maybe?

I can remove it completely from the patch if that is OK.

Thanks, Roger.

#6 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

Quoted 14 lines · Roger Pau Monné
Quoted 13 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

docs/misc/xen-command-line.pandoc Specify the size of the console debug trace buffer. By specifying `cpu:` · line 655

Quoted 7 lines · Roger Pau Monne
docs/misc/xen-command-line.pandoc +1 -0
@@ -652,6 +652,15 @@ Specify the size of the console debug trace buffer. By specifying `cpu:`
 additionally a trace buffer of the specified size is allocated per cpu.
 The debug trace feature is only enabled in debugging builds of Xen.
 
+### disable-c6-errata

Hmm, yes please - a disable for errata! ;-)

Quoted 8 lines
Quoted 3 lines

How about "avoid-c6-errata", and then perhaps as a sub-option to "cpuidle="? (If we really want a control for this in the first place.)

Right, I see I'm very bad at naming. Not sure it's even worth it maybe?

I can remove it completely from the patch if that is OK.

I'd be fine without. Andrew?

Jan

#7 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

Quoted 16 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

Quoted 13 lines · Roger Pau Monné
Quoted 12 lines · Jan Beulich

[CAUTION - EXTERNAL EMAIL] DO NOT reply, click links, or open attachments unless you have verified the sender and know the content is safe.

docs/misc/xen-command-line.pandoc Specify the size of the console debug trace buffer. By specifying `cpu:` · line 655

Quoted 7 lines · Roger Pau Monne
docs/misc/xen-command-line.pandoc +1 -0
@@ -652,6 +652,15 @@ Specify the size of the console debug trace buffer. By specifying `cpu:`
 additionally a trace buffer of the specified size is allocated per cpu.
 The debug trace feature is only enabled in debugging builds of Xen.
 
+### disable-c6-errata

Hmm, yes please - a disable for errata! ;-)

Quoted 8 lines
Quoted 7 lines
Quoted 3 lines

How about "avoid-c6-errata", and then perhaps as a sub-option to "cpuidle="? (If we really want a control for this in the first place.)

Right, I see I'm very bad at naming. Not sure it's even worth it maybe?

I can remove it completely from the patch if that is OK.

I'd be fine without. Andrew?

Yeah - the only thing people can do with this is shoot themselves in the foot.

There's frankly no need to give them the option in the first place.

~Andrew

#8 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

In reply to Roger Pau Monne
Quoted 3 lines · Roger Pau Monne

Apply a workaround for Intel errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131: "A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of The Same Priority Completes".

HSM175 et al, so presumably a HSD, and HSE as well.

On the broadwell side at least, BDD BDW in addition

~Andrew

#9 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

Quoted 8 lines · Andrew Cooper
Quoted 3 lines · Roger Pau Monne

Apply a workaround for Intel errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131: "A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of The Same Priority Completes".

HSM175 et al, so presumably a HSD, and HSE as well.

On the broadwell side at least, BDD BDW in addition

But those are a different errata AFAICT ('An APIC Timer Interrupt During Core C6 Entry May be Lost') and the workaround should also be different I think. We should mark the lapic timer as not reliable on C6 or higher states in lapic_timer_reliable_states, so that it's disabled before entering sleep?

Thanks, Roger.

#10 lore raw

Re: [PATCH v3 2/2] x86/idle: prevent entering C6 with in service interrupts on Intel

Quoted 11 lines · Roger Pau Monné
Quoted 7 lines · Andrew Cooper
Quoted 3 lines · Roger Pau Monne

Apply a workaround for Intel errata BDX99, CLX30, SKX100, CFW125, BDF104, BDH85, BDM135, KWB131: "A Pending Fixed Interrupt May Be Dispatched Before an Interrupt of The Same Priority Completes".

HSM175 et al, so presumably a HSD, and HSE as well.

On the broadwell side at least, BDD BDW in addition

But those are a different errata AFAICT ('An APIC Timer Interrupt During Core C6 Entry May be Lost') and the workaround should also be different I think.

Hmm, so it is.

The issue in question here definitely does affect Haswell, because that is where we first observed it.  There was also a report on xen-devel against Haswell.

If the errata are missing, then I think Intel needs some more chasing to work out the real extent of the problems.

Quoted 3 lines

We should mark the lapic timer as not reliable on
C6 or higher states in lapic_timer_reliable_states, so that it's disabled before entering sleep?

Probably should.

~Andrew