Skip to contentSkip to conversation

x86/resctrl: Fix event counts regression in reused RMIDs

PATCH Superseded Peter Newman 2 participants 5 messages Dec 14, 2022 – Dec 19, 2022

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 x86/resctrl: Fix event counts regression in reused RMIDs

When creating a new monitoring group, the RMID allocated for it may have been used by a group which was previously removed. In this case, the hardware counters will have non-zero values which should be deducted from what is reported in the new group's counts.

resctrl_arch_reset_rmid() initializes the prev_msr value for counters to 0, causing the initial count to be charged to the new group. Resurrect __rmid_read() and use it to initialize prev_msr correctly.

Unlike before, __rmid_read() checks for error bits in the MSR read so that callers don't need to.

Fixes 1d81d15db39c ("x86/resctrl: Move mbm_overflow_count() into resctrl_arch_rmid_read()")
---
 arch/x86/kernel/cpu/resctrl/monitor.c | 49 ++++++++++++++++++---------
 1 file changed, 33 insertions(+), 16 deletions(-)
arch/x86/kernel/cpu/resctrl/monitor.c +33 -16
index efe0c30d3a12..77538abeb72a 100644
@@ -146,6 +146,30 @@ static inline struct rmid_entry *__rmid_entry(u32 rmid)
 	return entry;
 }
 
+static int __rmid_read(u32 rmid, enum resctrl_event_id eventid, u64 *val)
+{
+	u64 msr_val;
+
+	/*
+	 * As per the SDM, when IA32_QM_EVTSEL.EvtID (bits 7:0) is configured
+	 * with a valid event code for supported resource type and the bits
+	 * IA32_QM_EVTSEL.RMID (bits 41:32) are configured with valid RMID,
+	 * IA32_QM_CTR.data (bits 61:0) reports the monitored data.
+	 * IA32_QM_CTR.Error (bit 63) and IA32_QM_CTR.Unavailable (bit 62)
+	 * are error bits.
+	 */
+	wrmsr(MSR_IA32_QM_EVTSEL, eventid, rmid);
+	rdmsrl(MSR_IA32_QM_CTR, msr_val);
+
+	if (msr_val & RMID_VAL_ERROR)
+		return -EIO;
+	if (msr_val & RMID_VAL_UNAVAIL)
+		return -EINVAL;
+
+	*val = msr_val;
+	return 0;
+}
+
 static struct arch_mbm_state *get_arch_mbm_state(struct rdt_hw_domain *hw_dom,
 						 u32 rmid,
 						 enum resctrl_event_id eventid)
@@ -172,8 +196,12 @@ void resctrl_arch_reset_rmid(struct rdt_resource *r, struct rdt_domain *d,
 	struct arch_mbm_state *am;
 
 	am = get_arch_mbm_state(hw_dom, rmid, eventid);
-	if (am)
+	if (am) {
 		memset(am, 0, sizeof(*am));
+
+		/* Record any initial, non-zero count value. */
+		__rmid_read(rmid, eventid, &am->prev_msr);
+	}
 }
 
 static u64 mbm_overflow_count(u64 prev_msr, u64 cur_msr, unsigned int width)
@@ -191,25 +219,14 @@ int resctrl_arch_rmid_read(struct rdt_resource *r, struct rdt_domain *d,
 	struct rdt_hw_domain *hw_dom = resctrl_to_arch_dom(d);
 	struct arch_mbm_state *am;
 	u64 msr_val, chunks;
+	int ret;
 
 	if (!cpumask_test_cpu(smp_processor_id(), &d->cpu_mask))
 		return -EINVAL;
 
-	/*
-	 * As per the SDM, when IA32_QM_EVTSEL.EvtID (bits 7:0) is configured
-	 * with a valid event code for supported resource type and the bits
-	 * IA32_QM_EVTSEL.RMID (bits 41:32) are configured with valid RMID,
-	 * IA32_QM_CTR.data (bits 61:0) reports the monitored data.
-	 * IA32_QM_CTR.Error (bit 63) and IA32_QM_CTR.Unavailable (bit 62)
-	 * are error bits.
-	 */
-	wrmsr(MSR_IA32_QM_EVTSEL, eventid, rmid);
-	rdmsrl(MSR_IA32_QM_CTR, msr_val);
-
-	if (msr_val & RMID_VAL_ERROR)
-		return -EIO;
-	if (msr_val & RMID_VAL_UNAVAIL)
-		return -EINVAL;
+	ret = __rmid_read(rmid, eventid, &msr_val);
+	if (ret)
+		return ret;
 
 	am = get_arch_mbm_state(hw_dom, rmid, eventid);
 	if (am) {

base-commit: 830b3c68c1fb1e9176028d02ef86f3cf76aa2476

Signature · 1 line
2.39.0.rc1.256.g54fd8350bd-goog
#1 lore raw

PATCH v2 2/2 x86/resctrl: Avoid redundant counter read in __mon_event_count()

__mon_event_count() does the per-RMID, per-domain work for user-initiated event count reads and the initialization of new monitor groups.

In the initialization case, after resctrl_arch_reset_rmid() calls __rmid_read() to record an initial count for a new monitor group, it immediately calls resctrl_arch_rmid_read(). This re-read of the hardware counter is unnecessary.

Following return from resctrl_arch_reset_rmid(), just clear the mbm_state and return.

Also, move the mbm_state lookup into the rr->first case, as it's not needed for regular event count reads: the QOS_L3_OCCUP_EVENT_ID case was redundant with the accumulating logic at the end of the function.

---
 arch/x86/kernel/cpu/resctrl/monitor.c | 43 ++++++++++++---------------
 1 file changed, 19 insertions(+), 24 deletions(-)
arch/x86/kernel/cpu/resctrl/monitor.c +19 -24
index 77538abeb72a..e708df478077 100644
@@ -366,41 +366,36 @@ void free_rmid(u32 rmid)
 		list_add_tail(&entry->list, &rmid_free_lru);
 }
 
+static struct mbm_state *get_mbm_state(struct rdt_domain *d, u32 rmid,
+				       enum resctrl_event_id evtid)
+{
+	switch (evtid) {
+	case QOS_L3_MBM_TOTAL_EVENT_ID:
+		return &d->mbm_total[rmid];
+	case QOS_L3_MBM_LOCAL_EVENT_ID:
+		return &d->mbm_local[rmid];
+	default:
+		return NULL;
+	}
+}
+
 static int __mon_event_count(u32 rmid, struct rmid_read *rr)
 {
 	struct mbm_state *m;
 	u64 tval = 0;
 
-	if (rr->first)
+	if (rr->first) {
 		resctrl_arch_reset_rmid(rr->r, rr->d, rmid, rr->evtid);
+		m = get_mbm_state(rr->d, rmid, rr->evtid);
+		if (m)
+			memset(m, 0, sizeof(struct mbm_state));
+		return 0;
+	}
 
 	rr->err = resctrl_arch_rmid_read(rr->r, rr->d, rmid, rr->evtid, &tval);
 	if (rr->err)
 		return rr->err;
 
-	switch (rr->evtid) {
-	case QOS_L3_OCCUP_EVENT_ID:
-		rr->val += tval;
-		return 0;
-	case QOS_L3_MBM_TOTAL_EVENT_ID:
-		m = &rr->d->mbm_total[rmid];
-		break;
-	case QOS_L3_MBM_LOCAL_EVENT_ID:
-		m = &rr->d->mbm_local[rmid];
-		break;
-	default:
-		/*
-		 * Code would never reach here because an invalid
-		 * event id would fail in resctrl_arch_rmid_read().
-		 */
-		return -EINVAL;
-	}
-
-	if (rr->first) {
-		memset(m, 0, sizeof(struct mbm_state));
-		return 0;
-	}
-
 	rr->val += tval;
 
 	return 0;
Signature · 1 line
2.39.0.rc1.256.g54fd8350bd-goog
maintainer Reviewed-by on v2 2/2
#2 lore raw

Re: [PATCH v2 2/2] x86/resctrl: Avoid redundant counter read in __mon_event_count()

Hi Peter,

Quoted 8 lines · Peter Newman

__mon_event_count() does the per-RMID, per-domain work for user-initiated event count reads and the initialization of new monitor groups.

In the initialization case, after resctrl_arch_reset_rmid() calls __rmid_read() to record an initial count for a new monitor group, it immediately calls resctrl_arch_rmid_read(). This re-read of the hardware counter is unnecessary.

"... and the following computations are ignored by the caller during initialization."

Quoted 7 lines

Following return from resctrl_arch_reset_rmid(), just clear the mbm_state and return.

Also, move the mbm_state lookup into the rr->first case, as it's not needed for regular event count reads: the QOS_L3_OCCUP_EVENT_ID case was redundant with the accumulating logic at the end of the function.

Starting the above with "Also" creates impression that two logical changes are merged into a single patch. Since this is not the case you can avoid the impression with something like:

Following return from resctrl_arch_reset_rmid(), just clear the mbm_state and return. This involves moving the mbm_state lookup into the rr->first case, as it's not needed for regular event count reads:
the QOS_L3_OCCUP_EVENT_ID case was redundant with the accumulating logic at the end of the function.

Quoted 61 lines
---
 arch/x86/kernel/cpu/resctrl/monitor.c | 43 ++++++++++++---------------
 1 file changed, 19 insertions(+), 24 deletions(-)
arch/x86/kernel/cpu/resctrl/monitor.c +19 -17
index 77538abeb72a..e708df478077 100644
@@ -366,41 +366,36 @@ void free_rmid(u32 rmid)
 		list_add_tail(&entry->list, &rmid_free_lru);
 }
 
+static struct mbm_state *get_mbm_state(struct rdt_domain *d, u32 rmid,
+				       enum resctrl_event_id evtid)
+{
+	switch (evtid) {
+	case QOS_L3_MBM_TOTAL_EVENT_ID:
+		return &d->mbm_total[rmid];
+	case QOS_L3_MBM_LOCAL_EVENT_ID:
+		return &d->mbm_local[rmid];
+	default:
+		return NULL;
+	}
+}
+
 static int __mon_event_count(u32 rmid, struct rmid_read *rr)
 {
 	struct mbm_state *m;
 	u64 tval = 0;
 
-	if (rr->first)
+	if (rr->first) {
 		resctrl_arch_reset_rmid(rr->r, rr->d, rmid, rr->evtid);
+		m = get_mbm_state(rr->d, rmid, rr->evtid);
+		if (m)
+			memset(m, 0, sizeof(struct mbm_state));
+		return 0;
+	}
 
 	rr->err = resctrl_arch_rmid_read(rr->r, rr->d, rmid, rr->evtid, &tval);
 	if (rr->err)
 		return rr->err;
 
-	switch (rr->evtid) {
-	case QOS_L3_OCCUP_EVENT_ID:
-		rr->val += tval;
-		return 0;
-	case QOS_L3_MBM_TOTAL_EVENT_ID:
-		m = &rr->d->mbm_total[rmid];
-		break;
-	case QOS_L3_MBM_LOCAL_EVENT_ID:
-		m = &rr->d->mbm_local[rmid];
-		break;
-	default:
-		/*
-		 * Code would never reach here because an invalid
-		 * event id would fail in resctrl_arch_rmid_read().
-		 */
-		return -EINVAL;

arch/x86/kernel/cpu/resctrl/monitor.c void free_rmid(u32 rmid) · line 401

Quoted 10 lines
arch/x86/kernel/cpu/resctrl/monitor.c +0 -7
-	}
-
-	if (rr->first) {
-		memset(m, 0, sizeof(struct mbm_state));
-		return 0;
-	}
-
 	rr->val += tval;
 
 	return 0;

Thank you very much.

Reinette

maintainer Reviewed-by on v2 1/2
#3 lore raw
In reply to Peter Newman

Hi Peter,

Quoted 9 lines · Peter Newman

When creating a new monitoring group, the RMID allocated for it may have been used by a group which was previously removed. In this case, the hardware counters will have non-zero values which should be deducted from what is reported in the new group's counts.

resctrl_arch_reset_rmid() initializes the prev_msr value for counters to 0, causing the initial count to be charged to the new group. Resurrect __rmid_read() and use it to initialize prev_msr correctly.

Quoted 5 lines

Unlike before, __rmid_read() checks for error bits in the MSR read so that callers don't need to.

Fixes 1d81d15db39c ("x86/resctrl: Move mbm_overflow_count() into resctrl_arch_rmid_read()")

This does look like a candidate for stable?

Quoted 1 line

---

It is helpful to have a summary here of what changed since previous version.

Quoted 83 lines
 arch/x86/kernel/cpu/resctrl/monitor.c | 49 ++++++++++++++++++---------
 1 file changed, 33 insertions(+), 16 deletions(-)
arch/x86/kernel/cpu/resctrl/monitor.c +33 -16
index efe0c30d3a12..77538abeb72a 100644
@@ -146,6 +146,30 @@ static inline struct rmid_entry *__rmid_entry(u32 rmid)
 	return entry;
 }
 
+static int __rmid_read(u32 rmid, enum resctrl_event_id eventid, u64 *val)
+{
+	u64 msr_val;
+
+	/*
+	 * As per the SDM, when IA32_QM_EVTSEL.EvtID (bits 7:0) is configured
+	 * with a valid event code for supported resource type and the bits
+	 * IA32_QM_EVTSEL.RMID (bits 41:32) are configured with valid RMID,
+	 * IA32_QM_CTR.data (bits 61:0) reports the monitored data.
+	 * IA32_QM_CTR.Error (bit 63) and IA32_QM_CTR.Unavailable (bit 62)
+	 * are error bits.
+	 */
+	wrmsr(MSR_IA32_QM_EVTSEL, eventid, rmid);
+	rdmsrl(MSR_IA32_QM_CTR, msr_val);
+
+	if (msr_val & RMID_VAL_ERROR)
+		return -EIO;
+	if (msr_val & RMID_VAL_UNAVAIL)
+		return -EINVAL;
+
+	*val = msr_val;
+	return 0;
+}
+
 static struct arch_mbm_state *get_arch_mbm_state(struct rdt_hw_domain *hw_dom,
 						 u32 rmid,
 						 enum resctrl_event_id eventid)
@@ -172,8 +196,12 @@ void resctrl_arch_reset_rmid(struct rdt_resource *r, struct rdt_domain *d,
 	struct arch_mbm_state *am;
 
 	am = get_arch_mbm_state(hw_dom, rmid, eventid);
-	if (am)
+	if (am) {
 		memset(am, 0, sizeof(*am));
+
+		/* Record any initial, non-zero count value. */
+		__rmid_read(rmid, eventid, &am->prev_msr);
+	}
 }
 
 static u64 mbm_overflow_count(u64 prev_msr, u64 cur_msr, unsigned int width)
@@ -191,25 +219,14 @@ int resctrl_arch_rmid_read(struct rdt_resource *r, struct rdt_domain *d,
 	struct rdt_hw_domain *hw_dom = resctrl_to_arch_dom(d);
 	struct arch_mbm_state *am;
 	u64 msr_val, chunks;
+	int ret;
 
 	if (!cpumask_test_cpu(smp_processor_id(), &d->cpu_mask))
 		return -EINVAL;
 
-	/*
-	 * As per the SDM, when IA32_QM_EVTSEL.EvtID (bits 7:0) is configured
-	 * with a valid event code for supported resource type and the bits
-	 * IA32_QM_EVTSEL.RMID (bits 41:32) are configured with valid RMID,
-	 * IA32_QM_CTR.data (bits 61:0) reports the monitored data.
-	 * IA32_QM_CTR.Error (bit 63) and IA32_QM_CTR.Unavailable (bit 62)
-	 * are error bits.
-	 */
-	wrmsr(MSR_IA32_QM_EVTSEL, eventid, rmid);
-	rdmsrl(MSR_IA32_QM_CTR, msr_val);
-
-	if (msr_val & RMID_VAL_ERROR)
-		return -EIO;
-	if (msr_val & RMID_VAL_UNAVAIL)
-		return -EINVAL;
+	ret = __rmid_read(rmid, eventid, &msr_val);
+	if (ret)
+		return ret;
 
 	am = get_arch_mbm_state(hw_dom, rmid, eventid);
 	if (am) {
Quoted 1 line

base-commit: 830b3c68c1fb1e9176028d02ef86f3cf76aa2476

Thank you very much for catching and fixing this.

Reinette

#4 lore raw

Hi Reinette,

Quoted 10 lines · Reinette Chatre
Quoted 9 lines · Peter Newman

When creating a new monitoring group, the RMID allocated for it may have been used by a group which was previously removed. In this case, the hardware counters will have non-zero values which should be deducted from what is reported in the new group's counts.

resctrl_arch_reset_rmid() initializes the prev_msr value for counters to 0, causing the initial count to be charged to the new group. Resurrect __rmid_read() and use it to initialize prev_msr correctly.

Quoted 7 lines
Quoted 5 lines

Unlike before, __rmid_read() checks for error bits in the MSR read so that callers don't need to.

Fixes 1d81d15db39c ("x86/resctrl: Move mbm_overflow_count() into resctrl_arch_rmid_read()")

This does look like a candidate for stable?

Yes, this bug is serious and reproducible. Every RMID reuse would have up to one overflow's-worth of measurement error.

Should I elaborate on the impact more in the changelog?

Quoted 4 lines
Quoted 2 lines

---

It is helpful to have a summary here of what changed since previous version.

ok, I'll add this

Quoted 3 lines

Thank you very much for catching and fixing this.

Thanks, Reinette!

-Peter