Re: [PATCH v2 07/10] powerpc/perf: support BHRB disable bit and new filtering modes

[Athira Rajeev <atrajeev@linux.vnet.ibm.com>]

[-- Attachment #1: Type: text/plain, Size: 10270 bytes --]



> On 08-Jul-2020, at 5:12 PM, Michael Ellerman <mpe@ellerman.id.au> wrote:
> 
> Athira Rajeev <atrajeev@linux.vnet.ibm.com <mailto:atrajeev@linux.vnet.ibm.com>> writes:
> 
>> PowerISA v3.1 has few updates for the Branch History Rolling Buffer(BHRB).
>                   ^
>                   a
>> First is the addition of BHRB disable bit and second new filtering
>                                                      ^
>                                                      is
>> modes for BHRB.
>> 
>> BHRB disable is controlled via Monitor Mode Control Register A (MMCRA)
>> bit 26, namely "BHRB Recording Disable (BHRBRD)". This field controls
> 
> Most people call that bit 37.
> 
>> whether BHRB entries are written when BHRB recording is enabled by other
>> bits. Patch implements support for this BHRB disable bit.
>       ^
>       This
> 
>> Secondly PowerISA v3.1 introduce filtering support for
> 
> .. that should be in a separate patch please.
> 
>> PERF_SAMPLE_BRANCH_IND_CALL/COND. The patch adds BHRB filter support
>                                    ^
>                                    This
>> for "ind_call" and "cond" in power10_bhrb_filter_map().
>> 
>> 'commit bb19af816025 ("powerpc/perf: Prevent kernel address leak to userspace via BHRB buffer")'
> 
> That doesn't need single quotes, and should be wrapped at 72 columns
> like the rest of the text.
> 
>> added a check in bhrb_read() to filter the kernel address from BHRB buffer. Patch here modified
>> it to avoid that check for PowerISA v3.1 based processors, since PowerISA v3.1 allows
>> only MSR[PR]=1 address to be written to BHRB buffer.
> 
> And that should be a separate patch again please.

Sure, I will split these to separate patches

> 
>> Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
>> ---
>> arch/powerpc/perf/core-book3s.c       | 27 +++++++++++++++++++++------
>> arch/powerpc/perf/isa207-common.c     | 13 +++++++++++++
>> arch/powerpc/perf/power10-pmu.c       | 13 +++++++++++--
>> arch/powerpc/platforms/powernv/idle.c | 14 ++++++++++++++
>> 4 files changed, 59 insertions(+), 8 deletions(-)
>> 
>> diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
>> index fad5159..9709606 100644
>> --- a/arch/powerpc/perf/core-book3s.c
>> +++ b/arch/powerpc/perf/core-book3s.c
>> @@ -466,9 +466,13 @@ static void power_pmu_bhrb_read(struct perf_event *event, struct cpu_hw_events *
>> 			 * addresses at this point. Check the privileges before
>> 			 * exporting it to userspace (avoid exposure of regions
>> 			 * where we could have speculative execution)
>> +			 * Incase of ISA 310, BHRB will capture only user-space
>                           ^
>                           In case of ISA v3.1,

Ok, 
> 
>> +			 * address,hence include a check before filtering code
>                           ^                                                  ^
>                           addresses, hence                                   .
>> 			 */
>> -			if (is_kernel_addr(addr) && perf_allow_kernel(&event->attr) != 0)
>> -				continue;
>> +			if (!(ppmu->flags & PPMU_ARCH_310S))
>> +				if (is_kernel_addr(addr) &&
>> +				perf_allow_kernel(&event->attr) != 0)
>> +					continue;
> 
> The indentation is weird. You should just check all three conditions
> with &&.

Ok, will correct this.
> 
>> 
>> 			/* Branches are read most recent first (ie. mfbhrb 0 is
>> 			 * the most recent branch).
>> @@ -1212,7 +1216,7 @@ static void write_mmcr0(struct cpu_hw_events *cpuhw, unsigned long mmcr0)
>> static void power_pmu_disable(struct pmu *pmu)
>> {
>> 	struct cpu_hw_events *cpuhw;
>> -	unsigned long flags, mmcr0, val;
>> +	unsigned long flags, mmcr0, val, mmcra = 0;
> 
> You initialise it below.
> 
>> 	if (!ppmu)
>> 		return;
>> @@ -1245,12 +1249,23 @@ static void power_pmu_disable(struct pmu *pmu)
>> 		mb();
>> 		isync();
>> 
>> +		val = mmcra = cpuhw->mmcr[2];
>> +
> 
> For mmcr0 (above), val is the variable we mutate and mmcr0 is the
> original value. But here you've done the reverse, which is confusing.

Yes, I am altering mmcra here and using val as original value. I should have done it reverse.

> 
>> 		/*
>> 		 * Disable instruction sampling if it was enabled
>> 		 */
>> -		if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) {
>> -			mtspr(SPRN_MMCRA,
>> -			      cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
>> +		if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE)
>> +			mmcra = cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE;
> 
> You just loaded cpuhw->mmcr[2] into mmcra, use it rather than referring
> back to cpuhw->mmcr[2] over and over.
> 

Ok,
>> +
>> +		/* Disable BHRB via mmcra [:26] for p10 if needed */
>> +		if (!(cpuhw->mmcr[2] & MMCRA_BHRB_DISABLE))
> 
> You don't need to check that it's clear AFAICS. Just always set disable
> and the check against val below will catch the nop case.

My thought here was to avoid writing to MMCRA ( also avoid mb() and isync() ) if its not needed.
But as you suggested, since I am comparing against original value before writing, I may not need this check.
And I missed feature check here. Will correct it.
 
> 
>> +			mmcra |= MMCRA_BHRB_DISABLE;
>> +
>> +		/* Write SPRN_MMCRA if mmcra has either disabled
> 
> Comment format is wrong.
> 
>> +		 * instruction sampling or BHRB
> 
> Full stop please.

Sure
> 
>> +		 */
>> +		if (val != mmcra) {
>> +			mtspr(SPRN_MMCRA, mmcra);
>> 			mb();
>> 			isync();
>> 		}
>> diff --git a/arch/powerpc/perf/isa207-common.c b/arch/powerpc/perf/isa207-common.c
>> index 7d4839e..463d925 100644
>> --- a/arch/powerpc/perf/isa207-common.c
>> +++ b/arch/powerpc/perf/isa207-common.c
>> @@ -404,6 +404,12 @@ int isa207_compute_mmcr(u64 event[], int n_ev,
>> 
>> 	mmcra = mmcr1 = mmcr2 = mmcr3 = 0;
>> 
>> +	/* Disable bhrb unless explicitly requested
>> +	 * by setting MMCRA [:26] bit.
>> +	 */
> 
> Comment format again.
> 
>> +	if (cpu_has_feature(CPU_FTR_ARCH_31))
>> +		mmcra |= MMCRA_BHRB_DISABLE;
> 
> Here we do a feature check before setting MMCRA_BHRB_DISABLE, but you
> didn't above?
> 
>> +
>> 	/* Second pass: assign PMCs, set all MMCR1 fields */
>> 	for (i = 0; i < n_ev; ++i) {
>> 		pmc     = (event[i] >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
>> @@ -475,10 +481,17 @@ int isa207_compute_mmcr(u64 event[], int n_ev,
>> 		}
>> 
>> 		if (event[i] & EVENT_WANTS_BHRB) {
>> +			/* set MMCRA[:26] to 0 for Power10 to enable BHRB */
> 
> "set MMCRA[:26] to 0" == "clear MMCRA[:26]”
> 
Ok

>> +			if (cpu_has_feature(CPU_FTR_ARCH_31))
>> +				mmcra &= ~MMCRA_BHRB_DISABLE;
> 
> Newline please.
> 
>> 			val = (event[i] >> EVENT_IFM_SHIFT) & EVENT_IFM_MASK;
>> 			mmcra |= val << MMCRA_IFM_SHIFT;
>> 		}
>> 
>> +		/* set MMCRA[:26] to 0 if there is user request for BHRB */
>> +		if (cpu_has_feature(CPU_FTR_ARCH_31) && has_branch_stack(pevents[i]))
>> +			mmcra &= ~MMCRA_BHRB_DISABLE;
>> +
> 
> I think it would be cleaner if you did a single test, eg:
> 
> 		if (cpu_has_feature(CPU_FTR_ARCH_31) &&
>                   (has_branch_stack(pevents[i]) || (event[i] & EVENT_WANTS_BHRB)))
> 			mmcra &= ~MMCRA_BHRB_DISABLE;

Sure Michael

Thanks for the review. I will address all these changes in the next version

Thanks
Athira 
> 
>> 		if (pevents[i]->attr.exclude_user)
>> 			mmcr2 |= MMCR2_FCP(pmc);
>> 
>> diff --git a/arch/powerpc/perf/power10-pmu.c b/arch/powerpc/perf/power10-pmu.c
>> index d64d69d..07fb919 100644
>> --- a/arch/powerpc/perf/power10-pmu.c
>> +++ b/arch/powerpc/perf/power10-pmu.c
>> @@ -82,6 +82,8 @@
>> 
>> /* MMCRA IFM bits - POWER10 */
>> #define POWER10_MMCRA_IFM1		0x0000000040000000UL
>> +#define POWER10_MMCRA_IFM2		0x0000000080000000UL
>> +#define POWER10_MMCRA_IFM3		0x00000000C0000000UL
>> #define POWER10_MMCRA_BHRB_MASK	0x00000000C0000000UL
>> 
>> /* Table of alternatives, sorted by column 0 */
>> @@ -233,8 +235,15 @@ static u64 power10_bhrb_filter_map(u64 branch_sample_type)
>> 	if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_RETURN)
>> 		return -1;
>> 
>> -	if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL)
>> -		return -1;
>> +	if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL) {
>> +		pmu_bhrb_filter |= POWER10_MMCRA_IFM2;
>> +		return pmu_bhrb_filter;
>> +	}
>> +
>> +	if (branch_sample_type & PERF_SAMPLE_BRANCH_COND) {
>> +		pmu_bhrb_filter |= POWER10_MMCRA_IFM3;
>> +		return pmu_bhrb_filter;
>> +	}
>> 
>> 	if (branch_sample_type & PERF_SAMPLE_BRANCH_CALL)
>> 		return -1;
>> diff --git a/arch/powerpc/platforms/powernv/idle.c b/arch/powerpc/platforms/powernv/idle.c
>> index 2dd4673..7db99c7 100644
>> --- a/arch/powerpc/platforms/powernv/idle.c
>> +++ b/arch/powerpc/platforms/powernv/idle.c
>> @@ -611,6 +611,7 @@ static unsigned long power9_idle_stop(unsigned long psscr, bool mmu_on)
>> 	unsigned long srr1;
>> 	unsigned long pls;
>> 	unsigned long mmcr0 = 0;
>> +	unsigned long mmcra_bhrb = 0;
>> 	struct p9_sprs sprs = {}; /* avoid false used-uninitialised */
>> 	bool sprs_saved = false;
>> 
>> @@ -657,6 +658,15 @@ static unsigned long power9_idle_stop(unsigned long psscr, bool mmu_on)
>> 		  */
>> 		mmcr0		= mfspr(SPRN_MMCR0);
>> 	}
>> +
>> +	if (cpu_has_feature(CPU_FTR_ARCH_31)) {
>> +		/* POWER10 uses MMCRA[:26] as BHRB disable bit
> 
> Comment format.
> 
>> +		 * to disable BHRB logic when not used. Hence Save and
>> +		 * restore MMCRA after a state-loss idle.
>> +		 */
>> +		mmcra_bhrb		= mfspr(SPRN_MMCRA);
>> +	}
> 
> It's the whole mmcra it should be called mmcra?

Yes, we are saving the whole mmcra. 
> 
>> +
>> 	if ((psscr & PSSCR_RL_MASK) >= pnv_first_spr_loss_level) {
>> 		sprs.lpcr	= mfspr(SPRN_LPCR);
>> 		sprs.hfscr	= mfspr(SPRN_HFSCR);
>> @@ -721,6 +731,10 @@ static unsigned long power9_idle_stop(unsigned long psscr, bool mmu_on)
>> 			mtspr(SPRN_MMCR0, mmcr0);
>> 		}
>> 
>> +		/* Reload MMCRA to restore BHRB disable bit for POWER10 */
>> +		if (cpu_has_feature(CPU_FTR_ARCH_31))
>> +			mtspr(SPRN_MMCRA, mmcra_bhrb);
>> +
>> 		/*
>> 		 * DD2.2 and earlier need to set then clear bit 60 in MMCRA
>> 		 * to ensure the PMU starts running.
> 
> 
> cheers


[-- Attachment #2: Type: text/html, Size: 85847 bytes --]