Message-ID: <Y8A6rsEXR/rmJY0N@FVFF77S0Q05N>
Date:   Thu, 12 Jan 2023 16:51:58 +0000
From:   Mark Rutland <mark.rutland@....com>
To:     Anshuman Khandual <anshuman.khandual@....com>
Cc:     linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        Catalin Marinas <catalin.marinas@....com>,
        Will Deacon <will@...nel.org>
Subject: Re: [PATCH V7 6/6] arm64/perf: Enable branch stack events via
 FEAT_BRBE

On Thu, Jan 05, 2023 at 08:40:39AM +0530, Anshuman Khandual wrote:
> This enables branch stack sampling events in ARMV8 PMU, via an architecture
> feature FEAT_BRBE aka branch record buffer extension. This defines required
> branch helper functions pmuv8pmu_branch_XXXXX() and the implementation here
> is wrapped with a new config option CONFIG_ARM64_BRBE.
> 
> Cc: Catalin Marinas <catalin.marinas@....com>
> Cc: Will Deacon <will@...nel.org>
> Cc: Mark Rutland <mark.rutland@....com>
> Cc: linux-arm-kernel@...ts.infradead.org
> Cc: linux-kernel@...r.kernel.org
> Signed-off-by: Anshuman Khandual <anshuman.khandual@....com>
> ---
>  arch/arm64/Kconfig                  |  11 +
>  arch/arm64/include/asm/perf_event.h |   9 +
>  arch/arm64/kernel/Makefile          |   1 +
>  arch/arm64/kernel/brbe.c            | 512 ++++++++++++++++++++++++++++
>  arch/arm64/kernel/brbe.h            | 257 ++++++++++++++
>  5 files changed, 790 insertions(+)
>  create mode 100644 arch/arm64/kernel/brbe.c
>  create mode 100644 arch/arm64/kernel/brbe.h
> 
> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
> index 03934808b2ed..915b12709a46 100644
> --- a/arch/arm64/Kconfig
> +++ b/arch/arm64/Kconfig
> @@ -1363,6 +1363,17 @@ config HW_PERF_EVENTS
>  	def_bool y
>  	depends on ARM_PMU
>  
> +config ARM64_BRBE
> +	bool "Enable support for Branch Record Buffer Extension (BRBE)"
> +	depends on PERF_EVENTS && ARM64 && ARM_PMU
> +	default y
> +	help
> +	  Enable perf support for Branch Record Buffer Extension (BRBE) which
> +	  records all branches taken in an execution path. This supports some
> +	  branch types and privilege based filtering. It captured additional
> +	  relevant information such as cycle count, misprediction and branch
> +	  type, branch privilege level etc.
> +
>  # Supported by clang >= 7.0 or GCC >= 12.0.0
>  config CC_HAVE_SHADOW_CALL_STACK
>  	def_bool $(cc-option, -fsanitize=shadow-call-stack -ffixed-x18)
> diff --git a/arch/arm64/include/asm/perf_event.h b/arch/arm64/include/asm/perf_event.h
> index a038902d6874..cf2e88c7b707 100644
> --- a/arch/arm64/include/asm/perf_event.h
> +++ b/arch/arm64/include/asm/perf_event.h
> @@ -277,6 +277,14 @@ struct pmu_hw_events;
>  struct arm_pmu;
>  struct perf_event;
>  
> +#ifdef CONFIG_ARM64_BRBE
> +void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event);
> +bool armv8pmu_branch_valid(struct perf_event *event);
> +void armv8pmu_branch_enable(struct perf_event *event);
> +void armv8pmu_branch_disable(struct perf_event *event);
> +void armv8pmu_branch_probe(struct arm_pmu *arm_pmu);
> +void armv8pmu_branch_reset(void);
> +#else
>  static inline void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event) { }
>  static inline bool armv8pmu_branch_valid(struct perf_event *event) { return false; }
>  static inline void armv8pmu_branch_enable(struct perf_event *event) { }
> @@ -284,3 +292,4 @@ static inline void armv8pmu_branch_disable(struct perf_event *event) { }
>  static inline void armv8pmu_branch_probe(struct arm_pmu *arm_pmu) { }
>  static inline void armv8pmu_branch_reset(void) { }
>  #endif
> +#endif
> diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile
> index ceba6792f5b3..6ee7ccb61621 100644
> --- a/arch/arm64/kernel/Makefile
> +++ b/arch/arm64/kernel/Makefile
> @@ -46,6 +46,7 @@ obj-$(CONFIG_MODULES)			+= module.o
>  obj-$(CONFIG_ARM64_MODULE_PLTS)		+= module-plts.o
>  obj-$(CONFIG_PERF_EVENTS)		+= perf_regs.o perf_callchain.o
>  obj-$(CONFIG_HW_PERF_EVENTS)		+= perf_event.o
> +obj-$(CONFIG_ARM64_BRBE)		+= brbe.o
>  obj-$(CONFIG_HAVE_HW_BREAKPOINT)	+= hw_breakpoint.o
>  obj-$(CONFIG_CPU_PM)			+= sleep.o suspend.o
>  obj-$(CONFIG_CPU_IDLE)			+= cpuidle.o
> diff --git a/arch/arm64/kernel/brbe.c b/arch/arm64/kernel/brbe.c
> new file mode 100644
> index 000000000000..cd03d3531e04
> --- /dev/null
> +++ b/arch/arm64/kernel/brbe.c
> @@ -0,0 +1,512 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Branch Record Buffer Extension Driver.
> + *
> + * Copyright (C) 2022 ARM Limited
> + *
> + * Author: Anshuman Khandual <anshuman.khandual@....com>
> + */
> +#include "brbe.h"
> +
> +static bool valid_brbe_nr(int brbe_nr)
> +{
> +	return brbe_nr == BRBIDR0_EL1_NUMREC_8 ||
> +	       brbe_nr == BRBIDR0_EL1_NUMREC_16 ||
> +	       brbe_nr == BRBIDR0_EL1_NUMREC_32 ||
> +	       brbe_nr == BRBIDR0_EL1_NUMREC_64;
> +}
> +
> +static bool valid_brbe_cc(int brbe_cc)
> +{
> +	return brbe_cc == BRBIDR0_EL1_CC_20_BIT;
> +}
> +
> +static bool valid_brbe_format(int brbe_format)
> +{
> +	return brbe_format == BRBIDR0_EL1_FORMAT_0;
> +}
> +
> +static bool valid_brbe_version(int brbe_version)
> +{
> +	return brbe_version == ID_AA64DFR0_EL1_BRBE_IMP ||
> +	       brbe_version == ID_AA64DFR0_EL1_BRBE_BRBE_V1P1;
> +}
> +
> +static void select_brbe_bank(int bank)
> +{
> +	static int brbe_current_bank = BRBE_BANK_IDX_INVALID;

This is a per-cpu peroperty, so I don't understand how this can safely be
stored in a static variable. If this is necessary it needs to go in a per-cpu
variable, but I suspect we don't actually need it.

> +	u64 brbfcr;
> +
> +	if (brbe_current_bank == bank)
> +		return;

It looks like this is just for the same of optimizing redundant changes when
armv8pmu_branch_read() iterates over the records?

It'd be simpler to have armv8pmu_branch_read() iterate over each bank, then
within that iterate over each record within that bank.

> +	WARN_ON(bank > BRBE_BANK_IDX_1);
> +	brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
> +	brbfcr &= ~BRBFCR_EL1_BANK_MASK;
> +	brbfcr |= ((bank << BRBFCR_EL1_BANK_SHIFT) & BRBFCR_EL1_BANK_MASK);

You can use SYS_FIELD_PREP() for this:

	brbfcr &= ~BRBFCR_EL1_BANK_MASK;
	brbfcr |= SYS_FIELD_PREP(BRBFCR_EL1, BANK, bank);

Please use FIELD_PREP for this.

> +	write_sysreg_s(brbfcr, SYS_BRBFCR_EL1);
> +	isb();
> +	brbe_current_bank = bank;
> +}
> +
> +static void select_brbe_bank_index(int buffer_idx)
> +{
> +	switch (buffer_idx) {
> +	case BRBE_BANK0_IDX_MIN ... BRBE_BANK0_IDX_MAX:
> +		select_brbe_bank(BRBE_BANK_IDX_0);
> +		break;
> +	case BRBE_BANK1_IDX_MIN ... BRBE_BANK1_IDX_MAX:
> +		select_brbe_bank(BRBE_BANK_IDX_1);
> +		break;
> +	default:
> +		pr_warn("unsupported BRBE index\n");

It would be worth logging the specific index in case we ever have to debug
this. It's probably worth also making this a WARN_ONCE() or WARN_RATELIMITED().

> +	}
> +}
> +
> +static const char branch_filter_error_msg[] = "branch filter not supported";
> +
> +bool armv8pmu_branch_valid(struct perf_event *event)
> +{
> +	u64 branch_type = event->attr.branch_sample_type;
> +
> +	/*
> +	 * If the event does not have at least one of the privilege
> +	 * branch filters as in PERF_SAMPLE_BRANCH_PLM_ALL, the core
> +	 * perf will adjust its value based on perf event's existing
> +	 * privilege level via attr.exclude_[user|kernel|hv].
> +	 *
> +	 * As event->attr.branch_sample_type might have been changed
> +	 * when the event reaches here, it is not possible to figure
> +	 * out whether the event originally had HV privilege request
> +	 * or got added via the core perf. Just report this situation
> +	 * once and continue ignoring if there are other instances.
> +	 */
> +	if ((branch_type & PERF_SAMPLE_BRANCH_HV) && !is_kernel_in_hyp_mode())
> +		pr_warn_once("%s - hypervisor privilege\n", branch_filter_error_msg);
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ABORT_TX) {
> +		pr_warn_once("%s - aborted transaction\n", branch_filter_error_msg);
> +		return false;
> +	}
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_NO_TX) {
> +		pr_warn_once("%s - no transaction\n", branch_filter_error_msg);
> +		return false;
> +	}
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_IN_TX) {
> +		pr_warn_once("%s - in transaction\n", branch_filter_error_msg);
> +		return false;
> +	}
> +	return true;
> +}

Is this called when validating user input? If so, NAK to printing anything to a
higher leval than debug. If there are constraints the user needs to be aware of
we should expose the relevant information under sysfs, but it seems that these
are just generic perf options that BRBE doesn't support.

It would be better to whitelist what we do support rather than blacklisting
what we don't.

> +
> +static void branch_records_alloc(struct arm_pmu *armpmu)
> +{
> +	struct pmu_hw_events *events;
> +	int cpu;
> +
> +	for_each_possible_cpu(cpu) {
> +		events = per_cpu_ptr(armpmu->hw_events, cpu);
> +
> +		events->branches = kzalloc(sizeof(struct branch_records), GFP_KERNEL);
> +		WARN_ON(!events->branches);
> +	}
> +}

It would be simpler for this to be a percpu allocation.

If the allocation fails, we should propogate that error rather than just
WARNing, and fail probing the PMU.

Also, if the generic allocator fails it will print a warning (unless
__GFP_NOWARN was used), so we don't need the warning here.

> +
> +static int brbe_attributes_probe(struct arm_pmu *armpmu, u32 brbe)
> +{
> +	struct brbe_hw_attr *brbe_attr = kzalloc(sizeof(struct brbe_hw_attr), GFP_KERNEL);

Same comments as for the failure path in branch_records_alloc().

> +	u64 brbidr = read_sysreg_s(SYS_BRBIDR0_EL1);

Which context is this run in? Unless this is affine to a relevant CPU we can't
read the sysreg safely, and if we're in a cross-call we cannot allocate memory,
so this doesn't look right to me.

I suspect CONFIG_DEBUG_ATOMIC_SLEEP=y and/or CONFIG_PROVE_LOCKING=y will complain here.

Please follow the approach of armv8pmu_probe_pmu(), where we use a probe_info
structure that the callee can fill with information. Then we can do the
allocation in the main thread from a non-atomic context.

> +
> +	WARN_ON(!brbe_attr);
> +	armpmu->private = brbe_attr;
> +
> +	brbe_attr->brbe_version = brbe;
> +	brbe_attr->brbe_format = brbe_fetch_format(brbidr);
> +	brbe_attr->brbe_cc = brbe_fetch_cc_bits(brbidr);
> +	brbe_attr->brbe_nr = brbe_fetch_numrec(brbidr);

As a minor thing, could we please s/fetch/get/ ? To me, 'fetch' sounds like a
memory operation, and elsewhere we use 'get' for this sort of getter function.

> +
> +	if (!valid_brbe_version(brbe_attr->brbe_version) ||
> +	   !valid_brbe_format(brbe_attr->brbe_format) ||
> +	   !valid_brbe_cc(brbe_attr->brbe_cc) ||
> +	   !valid_brbe_nr(brbe_attr->brbe_nr))
> +		return -EOPNOTSUPP;
> +
> +	return 0;
> +}
> +
> +void armv8pmu_branch_probe(struct arm_pmu *armpmu)
> +{
> +	u64 aa64dfr0 = read_sysreg_s(SYS_ID_AA64DFR0_EL1);
> +	u32 brbe;
> +
> +	brbe = cpuid_feature_extract_unsigned_field(aa64dfr0, ID_AA64DFR0_EL1_BRBE_SHIFT);
> +	if (!brbe)
> +		return;
> +
> +	if (brbe_attributes_probe(armpmu, brbe))
> +		return;
> +
> +	branch_records_alloc(armpmu);
> +	armpmu->features |= ARM_PMU_BRANCH_STACK;
> +}
> +
> +static u64 branch_type_to_brbfcr(int branch_type)
> +{
> +	u64 brbfcr = 0;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY) {
> +		brbfcr |= BRBFCR_EL1_BRANCH_FILTERS;
> +		return brbfcr;
> +	}
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY_CALL) {
> +		brbfcr |= BRBFCR_EL1_INDCALL;
> +		brbfcr |= BRBFCR_EL1_DIRCALL;
> +	}
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY_RETURN)
> +		brbfcr |= BRBFCR_EL1_RTN;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_IND_CALL)
> +		brbfcr |= BRBFCR_EL1_INDCALL;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_COND)
> +		brbfcr |= BRBFCR_EL1_CONDDIR;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_IND_JUMP)
> +		brbfcr |= BRBFCR_EL1_INDIRECT;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_CALL)
> +		brbfcr |= BRBFCR_EL1_DIRCALL;
> +
> +	return brbfcr;
> +}
> +
> +static u64 branch_type_to_brbcr(int branch_type)
> +{
> +	u64 brbcr = (BRBCR_EL1_FZP | BRBCR_EL1_DEFAULT_TS);
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_USER)
> +		brbcr |= BRBCR_EL1_E0BRE;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_KERNEL)
> +		brbcr |= BRBCR_EL1_E1BRE;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_HV) {
> +		if (is_kernel_in_hyp_mode())
> +			brbcr |= BRBCR_EL1_E1BRE;
> +	}

I assume that in that case we're actually writing to BRBCR_EL2, and this is
actually the E2BRE bit, which is at the same position? If so, I think that's
worth a comment above the USER/KERNEL/HV bits here.

How do the BRB* control registers work with E2H? Is BRBCR_EL1 rewritten to
BRBCR_EL2 by the hardware?

> +
> +	if (!(branch_type & PERF_SAMPLE_BRANCH_NO_CYCLES))
> +		brbcr |= BRBCR_EL1_CC;
> +
> +	if (!(branch_type & PERF_SAMPLE_BRANCH_NO_FLAGS))
> +		brbcr |= BRBCR_EL1_MPRED;
> +
> +	/*
> +	 * The exception and exception return branches could be
> +	 * captured, irrespective of the perf event's privilege.
> +	 * If the perf event does not have enough privilege for
> +	 * a given exception level, then addresses which falls
> +	 * under that exception level will be reported as zero
> +	 * for the captured branch record, creating source only
> +	 * or target only records.
> +	 */
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY) {
> +		brbcr |= BRBCR_EL1_EXCEPTION;
> +		brbcr |= BRBCR_EL1_ERTN;
> +	}
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY_CALL)
> +		brbcr |= BRBCR_EL1_EXCEPTION;
> +
> +	if (branch_type & PERF_SAMPLE_BRANCH_ANY_RETURN)
> +		brbcr |= BRBCR_EL1_ERTN;
> +
> +	return brbcr & BRBCR_EL1_DEFAULT_CONFIG;
> +}
> +
> +void armv8pmu_branch_enable(struct perf_event *event)
> +{
> +	u64 branch_type = event->attr.branch_sample_type;
> +	u64 brbfcr, brbcr;
> +
> +	brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
> +	brbfcr &= ~BRBFCR_EL1_DEFAULT_CONFIG;
> +	brbfcr |= branch_type_to_brbfcr(branch_type);
> +	write_sysreg_s(brbfcr, SYS_BRBFCR_EL1);
> +	isb();
> +
> +	brbcr = read_sysreg_s(SYS_BRBCR_EL1);
> +	brbcr &= ~BRBCR_EL1_DEFAULT_CONFIG;
> +	brbcr |= branch_type_to_brbcr(branch_type);
> +	write_sysreg_s(brbcr, SYS_BRBCR_EL1);
> +	isb();
> +	armv8pmu_branch_reset();
> +}
> +
> +void armv8pmu_branch_disable(struct perf_event *event)
> +{
> +	u64 brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
> +	u64 brbcr = read_sysreg_s(SYS_BRBCR_EL1);
> +
> +	brbcr &= ~(BRBCR_EL1_E0BRE | BRBCR_EL1_E1BRE);
> +	brbfcr |= BRBFCR_EL1_PAUSED;
> +	write_sysreg_s(brbcr, SYS_BRBCR_EL1);
> +	write_sysreg_s(brbfcr, SYS_BRBFCR_EL1);
> +	isb();
> +}
> +
> +static int brbe_fetch_perf_type(u64 brbinf, bool *new_branch_type)

It's a bit confusing to return the type and new_type fields in this way.

I think this would be clearer as a setter function, even if that results in it
being a bit longer, since it keeps all the type and new_type relationships in
one place and has a single path for returning the value:

static void brbe_set_perf_entry_type(struct perf_branch_entry *entry,
				     u64 brbinf)
{
	int brbe_type = brbe_fetch_type(brbinf);

	switch (brbe_type) {
	case BRBINF_EL1_TYPE_UNCOND_DIR;
		entry->type = PERF_BR_UNCOND;
		break;
	...
	case BRBINF_EL1_TYPE_DEBUG_HALT;
		entry->type = PERF_BR_EXTEND_ABI;
		entry->new_type = PERF_BR_ARM64_DEBUG_HALT;
		break;
	...
	default:
		...
	}
}

... and in theory that makes it easier to propogate an error in future if we
want to.

> +{
> +	int brbe_type = brbe_fetch_type(brbinf);
> +	*new_branch_type = false;
> +
> +	switch (brbe_type) {
> +	case BRBINF_EL1_TYPE_UNCOND_DIR:
> +		return PERF_BR_UNCOND;
> +	case BRBINF_EL1_TYPE_INDIR:
> +		return PERF_BR_IND;
> +	case BRBINF_EL1_TYPE_DIR_LINK:
> +		return PERF_BR_CALL;
> +	case BRBINF_EL1_TYPE_INDIR_LINK:
> +		return PERF_BR_IND_CALL;
> +	case BRBINF_EL1_TYPE_RET_SUB:
> +		return PERF_BR_RET;
> +	case BRBINF_EL1_TYPE_COND_DIR:
> +		return PERF_BR_COND;
> +	case BRBINF_EL1_TYPE_CALL:
> +		return PERF_BR_CALL;
> +	case BRBINF_EL1_TYPE_TRAP:
> +		return PERF_BR_SYSCALL;
> +	case BRBINF_EL1_TYPE_RET_EXCPT:
> +		return PERF_BR_ERET;
> +	case BRBINF_EL1_TYPE_IRQ:
> +		return PERF_BR_IRQ;
> +	case BRBINF_EL1_TYPE_DEBUG_HALT:
> +		*new_branch_type = true;
> +		return PERF_BR_ARM64_DEBUG_HALT;
> +	case BRBINF_EL1_TYPE_SERROR:
> +		return PERF_BR_SERROR;
> +	case BRBINF_EL1_TYPE_INST_DEBUG:
> +		*new_branch_type = true;
> +		return PERF_BR_ARM64_DEBUG_INST;
> +	case BRBINF_EL1_TYPE_DATA_DEBUG:
> +		*new_branch_type = true;
> +		return PERF_BR_ARM64_DEBUG_DATA;
> +	case BRBINF_EL1_TYPE_ALGN_FAULT:
> +		*new_branch_type = true;
> +		return PERF_BR_NEW_FAULT_ALGN;
> +	case BRBINF_EL1_TYPE_INST_FAULT:
> +		*new_branch_type = true;
> +		return PERF_BR_NEW_FAULT_INST;
> +	case BRBINF_EL1_TYPE_DATA_FAULT:
> +		*new_branch_type = true;
> +		return PERF_BR_NEW_FAULT_DATA;
> +	case BRBINF_EL1_TYPE_FIQ:
> +		*new_branch_type = true;
> +		return PERF_BR_ARM64_FIQ;
> +	case BRBINF_EL1_TYPE_DEBUG_EXIT:
> +		*new_branch_type = true;
> +		return PERF_BR_ARM64_DEBUG_EXIT;
> +	default:
> +		pr_warn("unknown branch type captured\n");
> +		return PERF_BR_UNKNOWN;

It would be worth logging the specific value in case we ever have to debug
this. This should also be marked as _ratelimited or _once.

> +	}
> +}
> +
> +static int brbe_fetch_perf_priv(u64 brbinf)
> +{
> +	int brbe_el = brbe_fetch_el(brbinf);
> +
> +	switch (brbe_el) {
> +	case BRBINF_EL1_EL_EL0:
> +		return PERF_BR_PRIV_USER;
> +	case BRBINF_EL1_EL_EL1:
> +		return PERF_BR_PRIV_KERNEL;
> +	case BRBINF_EL1_EL_EL2:
> +		if (is_kernel_in_hyp_mode())
> +			return PERF_BR_PRIV_KERNEL;
> +		return PERF_BR_PRIV_HV;
> +	default:
> +		pr_warn("unknown branch privilege captured\n");
> +		return PERF_BR_PRIV_UNKNOWN;

It would be worth logging the specific value in case we ever have to debug
this. This should also be marked as _ratelimited or _once.

> +	}
> +}
> +
> +static void capture_brbe_flags(struct pmu_hw_events *cpuc, struct perf_event *event,
> +			       u64 brbinf, u64 brbcr, int idx)
> +{
> +	struct perf_branch_entry *entry = &cpuc->branches->branch_entries[idx];
> +	bool new_branch_type;
> +	int branch_type;
> +
> +	if (branch_sample_type(event)) {
> +		branch_type = brbe_fetch_perf_type(brbinf, &new_branch_type);
> +		if (new_branch_type) {
> +			entry->type = PERF_BR_EXTEND_ABI;
> +			entry->new_type = branch_type;
> +		} else {
> +			entry->type = branch_type;
> +		}
> +	}

With the suggestions bove, this would become:

	if (branch_sample_type(event))
		brbe_set_perf_entry_type(entry, brbinf);

> +	if (!branch_sample_no_cycles(event)) {
> +		WARN_ON_ONCE(!(brbcr & BRBCR_EL1_CC));
> +		entry->cycles = brbe_fetch_cycles(brbinf);
> +	}
> +
> +	if (!branch_sample_no_flags(event)) {
> +		/*
> +		 * BRBINF_LASTFAILED does not indicate whether last transaction
> +		 * got failed or aborted during the current branch record itself.
> +		 * Rather, this indicates that all the branch records which were
> +		 * in transaction until the curret branch record have failed. So
> +		 * the entire BRBE buffer needs to be processed later on to find
> +		 * all branch records which might have failed.
> +		 */

This is quite difficult to follow.

I took in the ARM ARM, and it looks like this is all about TME transactions
(which Linux doesn't currently support). Per ARM DDI 0487I.a, page D15-5506:

| R_GVCJH
|   When an entire transaction is executed in a BRBE Non-prohibited region and
|   the transaction fails or is canceled then BRBFCR_EL1.LASTFAILED is set to
|   1.

| R_KBSZM
|   When a Branch record is generated, other than through the injection
|   mechanism, the value of BRBFCR_EL1.LASTFAILED is copied to the LASTFAILED
|   field in the Branch record and BRBFCR_EL1.LASTFAILED is set to 0.

| I_JBPHS
|   When a transaction fails or is canceled, Branch records generated in the
|   transaction are not removed from the Branch record buffer.

I think what this is saying is:

	/*
	 * BRBINFx_EL1.LASTFAILED indicates that a TME transaction failed (or
	 * was cancelled) prior to this record, and some number of records
	 * prior to this one may have been generated during an attempt to
	 * execute the transaction.
	 *
	 * We will remove such entries later in process_branch_aborts().
	 */

Is that right?

> +
> +		/*
> +		 * All these information (i.e transaction state and mispredicts)
> +		 * are not available for target only branch records.
> +		 */
> +		if (!brbe_target(brbinf)) {

Could we rename these heleprs for clarity, e.g.
brbe_record_is_{target_only,source_only,complete}()

With that, it would also be clearer to have:

	/*
	 * These fields only exist for complete and source-only records.
	 */
	if (brbe_record_is_complete(brbinf) ||
	    brbe_record_is_source_only()) {

... and explicilty match the cases we care about[


> +			WARN_ON_ONCE(!(brbcr & BRBCR_EL1_MPRED));

Huh? Why does the value of BRBCR matter here?

> +			entry->mispred = brbe_fetch_mispredict(brbinf);
> +			entry->predicted = !entry->mispred;
> +			entry->in_tx = brbe_fetch_in_tx(brbinf);
> +		}
> +	}
> +
> +	if (branch_sample_priv(event)) {
> +		/*
> +		 * All these information (i.e branch privilege level) are not
> +		 * available for source only branch records.
> +		 */
> +		if (!brbe_source(brbinf))
> +			entry->priv = brbe_fetch_perf_priv(brbinf);

Same style comment as above.

> +	}
> +}
> +
> +/*
> + * A branch record with BRBINF_EL1.LASTFAILED set, implies that all
> + * preceding consecutive branch records, that were in a transaction
> + * (i.e their BRBINF_EL1.TX set) have been aborted.
> + *
> + * Similarly BRBFCR_EL1.LASTFAILED set, indicate that all preceding
> + * consecutive branch records upto the last record, which were in a
> + * transaction (i.e their BRBINF_EL1.TX set) have been aborted.
> + *
> + * --------------------------------- -------------------
> + * | 00 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX success]
> + * --------------------------------- -------------------
> + * | 01 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX success]
> + * --------------------------------- -------------------
> + * | 02 | BRBSRC | BRBTGT | BRBINF | | TX = 0 | LF = 0 |
> + * --------------------------------- -------------------
> + * | 03 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX failed]
> + * --------------------------------- -------------------
> + * | 04 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX failed]
> + * --------------------------------- -------------------
> + * | 05 | BRBSRC | BRBTGT | BRBINF | | TX = 0 | LF = 1 |
> + * --------------------------------- -------------------
> + * | .. | BRBSRC | BRBTGT | BRBINF | | TX = 0 | LF = 0 |
> + * --------------------------------- -------------------
> + * | 61 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX failed]
> + * --------------------------------- -------------------
> + * | 62 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX failed]
> + * --------------------------------- -------------------
> + * | 63 | BRBSRC | BRBTGT | BRBINF | | TX = 1 | LF = 0 | [TX failed]
> + * --------------------------------- -------------------

Are we guaranteed to have a record between two transactions with TX = 0?

AFAICT you could have a sequence where a TCOMMIT is immediately followed by a
TSTART, and IIUC in that case you could have back-to-back records for distinct
transactions all with TX = 1, where the first transaction could be commited,
and the second might fail/cancel.

... or do TCOMMIT/TCANCEL/TSTART get handled specially?

> + *
> + * BRBFCR_EL1.LASTFAILED == 1
> + *
> + * Here BRBFCR_EL1.LASTFAILED failes all those consecutive and also
> + * in transaction branches near the end of the BRBE buffer.
> + */
> +static void process_branch_aborts(struct pmu_hw_events *cpuc)
> +{
> +	struct brbe_hw_attr *brbe_attr = (struct brbe_hw_attr *)cpuc->percpu_pmu->private;
> +	u64 brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
> +	bool lastfailed = !!(brbfcr & BRBFCR_EL1_LASTFAILED);
> +	int idx = brbe_attr->brbe_nr - 1;
> +	struct perf_branch_entry *entry;
> +
> +	do {
> +		entry = &cpuc->branches->branch_entries[idx];
> +		if (entry->in_tx) {
> +			entry->abort = lastfailed;
> +		} else {
> +			lastfailed = entry->abort;
> +			entry->abort = false;
> +		}
> +	} while (idx--, idx >= 0);
> +}
> +
> +void armv8pmu_branch_reset(void)
> +{
> +	asm volatile(BRB_IALL);
> +	isb();
> +}
> +
> +void armv8pmu_branch_read(struct pmu_hw_events *cpuc, struct perf_event *event)
> +{
> +	struct brbe_hw_attr *brbe_attr = (struct brbe_hw_attr *)cpuc->percpu_pmu->private;
> +	u64 brbinf, brbfcr, brbcr;
> +	int idx;
> +
> +	brbcr = read_sysreg_s(SYS_BRBCR_EL1);
> +	brbfcr = read_sysreg_s(SYS_BRBFCR_EL1);
> +
> +	/* Ensure pause on PMU interrupt is enabled */
> +	WARN_ON_ONCE(!(brbcr & BRBCR_EL1_FZP));

As above, I think this needs commentary in the interrupt handler, since this
presumably needs us to keep the IRQ asserted until we're done
reading/manipulating records in the IRQ handler.

Do we ever read this outside of the IRQ handler? AFAICT we don't, and that
makes it seem like some of this is redundant.

> +
> +	/* Save and clear the privilege */
> +	write_sysreg_s(brbcr & ~(BRBCR_EL1_E0BRE | BRBCR_EL1_E1BRE), SYS_BRBCR_EL1);

Why? Later on we restore this, and AFAICT we don't modify it.

If it's paused, why do we care about the privilege?

> +
> +	/* Pause the buffer */
> +	write_sysreg_s(brbfcr | BRBFCR_EL1_PAUSED, SYS_BRBFCR_EL1);
> +	isb();

Why? If we're in the IRQ handler it's already paused, and if we're not in the
IRQ handler what prevents us racing with an IRQ?

> +
> +	for (idx = 0; idx < brbe_attr->brbe_nr; idx++) {
> +		struct perf_branch_entry *entry = &cpuc->branches->branch_entries[idx];
> +
> +		select_brbe_bank_index(idx);
> +		brbinf = get_brbinf_reg(idx);
> +		/*
> +		 * There are no valid entries anymore on the buffer.
> +		 * Abort the branch record processing to save some
> +		 * cycles and also reduce the capture/process load
> +		 * for the user space as well.
> +		 */
> +		if (brbe_invalid(brbinf))
> +			break;
> +
> +		perf_clear_branch_entry_bitfields(entry);
> +		if (brbe_valid(brbinf)) {
> +			entry->from = get_brbsrc_reg(idx);
> +			entry->to = get_brbtgt_reg(idx);
> +		} else if (brbe_source(brbinf)) {
> +			entry->from = get_brbsrc_reg(idx);
> +			entry->to = 0;
> +		} else if (brbe_target(brbinf)) {
> +			entry->from = 0;
> +			entry->to = get_brbtgt_reg(idx);
> +		}
> +		capture_brbe_flags(cpuc, event, brbinf, brbcr, idx);
> +	}
> +	cpuc->branches->branch_stack.nr = idx;
> +	cpuc->branches->branch_stack.hw_idx = -1ULL;
> +	process_branch_aborts(cpuc);
> +
> +	/* Restore privilege, enable pause on PMU interrupt */
> +	write_sysreg_s(brbcr | BRBCR_EL1_FZP, SYS_BRBCR_EL1);

Why do we have to save/restore this?

> +
> +	/* Unpause the buffer */
> +	write_sysreg_s(brbfcr & ~BRBFCR_EL1_PAUSED, SYS_BRBFCR_EL1);
> +	isb();
> +	armv8pmu_branch_reset();
> +}

Why do we enable it before we reset it?

Surely it would make sense to reset it first, and ammortize the cost of the ISB?

That said, as above, do we actually need to pause/unpause it? Or is it already
paused by virtue of the IRQ?

Thanks,
Mark.

Powered by blists - more mailing lists