user/sven/linux.git/include/linux/perf/arm_pmu.h, branch v6.4.5

drivers/perf: apple_m1: Force 63bit counters for M2 CPUs

2023-07-19T14:35:14Z

[ Upstream commit 8be3593b9efa8903d2ee7bb9cdf57a8e56c66f36 ] Sidharth reports that on M2, the PMU never generates any interrupt when using 'perf record', which is a annoying as you get no sample. I'm temped to say "no sample, no problem", but others may have a different opinion. Upon investigation, it appears that the counters on M2 are significantly different from the ones on M1, as they count on 64 bits instead of 48. Which of course, in the fine M1 tradition, means that we can only use 63 bits, as the top bit is used to signal the interrupt... This results in having to introduce yet another flag to indicate yet another odd counter width. Who knows what the next crazy implementation will do... With this, perf can work out the correct offset, and 'perf record' works as intended. Tested on M2 and M2-Pro CPUs. Cc: Janne Grunau Cc: Hector Martin Cc: Mark Rutland Cc: Will Deacon Fixes: 7d0bfb7c9977 ("drivers/perf: apple_m1: Add Apple M2 support") Reported-by: Sidharth Kshatriya Tested-by: Sidharth Kshatriya Signed-off-by: Marc Zyngier Link: https://lore.kernel.org/r/20230528080205.288446-1-maz@kernel.org Signed-off-by: Will Deacon Signed-off-by: Sasha Levin

arm_pmu: fix event CPU filtering

2023-02-16T21:23:52Z

Janne reports that perf has been broken on Apple M1 as of commit: bd27568117664b8b ("perf: Rewrite core context handling") That commit replaced the pmu::filter_match() callback with pmu::filter(), whose return value has the opposite polarity, with true implying events should be ignored rather than scheduled. While an attempt was made to update the logic in armv8pmu_filter() and armpmu_filter() accordingly, the return value remains inverted in a couple of cases: * If the arm_pmu does not have an arm_pmu::filter() callback, armpmu_filter() will always return whether the CPU is supported rather than whether the CPU is not supported. As a result, the perf core will not schedule events on supported CPUs, resulting in a loss of events. Additionally, the perf core will attempt to schedule events on unsupported CPUs, but this will be rejected by armpmu_add(), which may result in a loss of events from other PMUs on those unsupported CPUs. * If the arm_pmu does have an arm_pmu::filter() callback, and armpmu_filter() is called on a CPU which is not supported by the arm_pmu, armpmu_filter() will return false rather than true. As a result, the perf core will attempt to schedule events on unsupported CPUs, but this will be rejected by armpmu_add(), which may result in a loss of events from other PMUs on those unsupported CPUs. This means a loss of events can be seen with any arm_pmu driver, but with the ARMv8 PMUv3 driver (which is the only arm_pmu driver with an arm_pmu::filter() callback) the event loss will be more limited and may go unnoticed, which is how this issue evaded testing so far. Fix the CPU filtering by performing this consistently in armpmu_filter(), and remove the redundant arm_pmu::filter() callback and armv8pmu_filter() implementation. Commit bd2756811766 also silently removed the CHAIN event filtering from armv8pmu_filter(), which will be addressed by a separate patch without using the filter callback. Fixes: bd2756811766 ("perf: Rewrite core context handling") Reported-by: Janne Grunau Link: https://lore.kernel.org/asahi/20230215-arm_pmu_m1_regression-v1-1-f5a266577c8d@jannau.net/ Signed-off-by: Mark Rutland Cc: Will Deacon Cc: Peter Zijlstra Cc: Ravi Bangoria Cc: Asahi Lina Cc: Eric Curtin Tested-by: Janne Grunau Link: https://lore.kernel.org/r/20230216141240.3833272-2-mark.rutland@arm.com Signed-off-by: Will Deacon

Merge tag 'perf-core-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

2022-12-12T23:19:38Z

Pull perf events updates from Ingo Molnar: - Thoroughly rewrite the data structures that implement perf task context handling, with the goal of fixing various quirks and unfeatures both in already merged, and in upcoming proposed code. The old data structure is the per task and per cpu perf_event_contexts: task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context ^ | ^ | ^ `---------------------------------' | `--> pmu ---' v ^ perf_event ------' In this new design this is replaced with a single task context and a single CPU context, plus intermediate data-structures: task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context ^ | ^ ^ `---------------------------' | | | | perf_cpu_pmu_context <--. | `----. ^ | | | | | | v v | | ,--> perf_event_pmu_context | | | | | | | v v | perf_event ---> pmu ----------------' [ See commit bd2756811766 for more details. ] This rewrite was developed by Peter Zijlstra and Ravi Bangoria. - Optimize perf_tp_event() - Update the Intel uncore PMU driver, extending it with UPI topology discovery on various hardware models. - Misc fixes & cleanups * tag 'perf-core-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (25 commits) perf/x86/intel/uncore: Fix reference count leak in __uncore_imc_init_box() perf/x86/intel/uncore: Fix reference count leak in snr_uncore_mmio_map() perf/x86/intel/uncore: Fix reference count leak in hswep_has_limit_sbox() perf/x86/intel/uncore: Fix reference count leak in sad_cfg_iio_topology() perf/x86/intel/uncore: Make set_mapping() procedure void perf/x86/intel/uncore: Update sysfs-devices-mapping file perf/x86/intel/uncore: Enable UPI topology discovery for Sapphire Rapids perf/x86/intel/uncore: Enable UPI topology discovery for Icelake Server perf/x86/intel/uncore: Get UPI NodeID and GroupID perf/x86/intel/uncore: Enable UPI topology discovery for Skylake Server perf/x86/intel/uncore: Generalize get_topology() for SKX PMUs perf/x86/intel/uncore: Disable I/O stacks to PMU mapping on ICX-D perf/x86/intel/uncore: Clear attr_update properly perf/x86/intel/uncore: Introduce UPI topology type perf/x86/intel/uncore: Generalize IIO topology support perf/core: Don't allow grouping events from different hw pmus perf/amd/ibs: Make IBS a core pmu perf: Fix function pointer case perf/x86/amd: Remove the repeated declaration perf: Fix possible memleak in pmu_dev_alloc() ...

arm_pmu: rework ACPI probing

2022-11-07T16:16:20Z

The current ACPI PMU probing logic tries to associate PMUs with CPUs when the CPU is first brought online, in order to handle late hotplug, though PMUs are only registered during early boot, and so for late hotplugged CPUs this can only associate the CPU with an existing PMU. We tried to be clever and the have the arm_pmu_acpi_cpu_starting() callback allocate a struct arm_pmu when no matching instance is found, in order to avoid duplication of logic. However, as above this doesn't do anything useful for late hotplugged CPUs, and this requires us to allocate memory in an atomic context, which is especially problematic for PREEMPT_RT, as reported by Valentin and Pierre. This patch reworks the probing to detect PMUs for all online CPUs in the arm_pmu_acpi_probe() function, which is more aligned with how DT probing works. The arm_pmu_acpi_cpu_starting() callback only tries to associate CPUs with an existing arm_pmu instance, avoiding the problem of allocating in atomic context. Note that as we didn't previously register PMUs for late-hotplugged CPUs, this change doesn't result in a loss of existing functionality, though we will now warn when we cannot associate a CPU with a PMU. This change allows us to pull the hotplug callback registration into the arm_pmu_acpi_probe() function, as we no longer need the callbacks to be invoked shortly after probing the boot CPUs, and can register it without invoking the calls. For the moment the arm_pmu_acpi_init() initcall remains to register the SPE PMU, though in future this should probably be moved elsewhere (e.g. the arm64 ACPI init code), since this doesn't need to be tied to the regular CPU PMU code. Signed-off-by: Mark Rutland Reported-by: Valentin Schneider Link: https://lore.kernel.org/r/20210810134127.1394269-2-valentin.schneider@arm.com/ Reported-by: Pierre Gondois Link: https://lore.kernel.org/linux-arm-kernel/20220912155105.1443303-1-pierre.gondois@arm.com/ Cc: Pierre Gondois Cc: Valentin Schneider Cc: Will Deacon Reviewed-and-tested-by: Pierre Gondois Link: https://lore.kernel.org/r/20220930111844.1522365-4-mark.rutland@arm.com Signed-off-by: Will Deacon

perf: Rewrite core context handling

2022-10-27T18:12:16Z

There have been various issues and limitations with the way perf uses (task) contexts to track events. Most notable is the single hardware PMU task context, which has resulted in a number of yucky things (both proposed and merged). Notably: - HW breakpoint PMU - ARM big.little PMU / Intel ADL PMU - Intel Branch Monitoring PMU - AMD IBS PMU - S390 cpum_cf PMU - PowerPC trace_imc PMU *Current design:* Currently we have a per task and per cpu perf_event_contexts: task_struct::perf_events_ctxp[] <-> perf_event_context <-> perf_cpu_context ^ | ^ | ^ `---------------------------------' | `--> pmu ---' v ^ perf_event ------' Each task has an array of pointers to a perf_event_context. Each perf_event_context has a direct relation to a PMU and a group of events for that PMU. The task related perf_event_context's have a pointer back to that task. Each PMU has a per-cpu pointer to a per-cpu perf_cpu_context, which includes a perf_event_context, which again has a direct relation to that PMU, and a group of events for that PMU. The perf_cpu_context also tracks which task context is currently associated with that CPU and includes a few other things like the hrtimer for rotation etc. Each perf_event is then associated with its PMU and one perf_event_context. *Proposed design:* New design proposed by this patch reduce to a single task context and a single CPU context but adds some intermediate data-structures: task_struct::perf_event_ctxp -> perf_event_context <- perf_cpu_context ^ | ^ ^ `---------------------------' | | | | perf_cpu_pmu_context <--. | `----. ^ | | | | | | v v | | ,--> perf_event_pmu_context | | | | | | | v v | perf_event ---> pmu ----------------' With the new design, perf_event_context will hold all events for all pmus in the (respective pinned/flexible) rbtrees. This can be achieved by adding pmu to rbtree key: {cpu, pmu, cgroup, group_index} Each perf_event_context carries a list of perf_event_pmu_context which is used to hold per-pmu-per-context state. For example, it keeps track of currently active events for that pmu, a pmu specific task_ctx_data, a flag to tell whether rotation is required or not etc. Additionally, perf_cpu_pmu_context is used to hold per-pmu-per-cpu state like hrtimer details to drive the event rotation, a pointer to perf_event_pmu_context of currently running task and some other ancillary information. Each perf_event is associated to it's pmu, perf_event_context and perf_event_pmu_context. Further optimizations to current implementation are possible. For example, ctx_resched() can be optimized to reschedule only single pmu events. Much thanks to Ravi for picking this up and pushing it towards completion. Signed-off-by: Peter Zijlstra (Intel) Co-developed-by: Ravi Bangoria Signed-off-by: Ravi Bangoria Signed-off-by: Peter Zijlstra (Intel) Link: https://lkml.kernel.org/r/20221008062424.313-1-ravi.bangoria@amd.com

arm64/perf: Assert all platform event flags are within PERF_EVENT_FLAG_ARCH

2022-09-07T19:54:01Z

Ensure all platform specific event flags are within PERF_EVENT_FLAG_ARCH. Signed-off-by: Anshuman Khandual Signed-off-by: Peter Zijlstra (Intel) Reviewed-by: James Clark Link: https://lkml.kernel.org/r/20220907091924.439193-4-anshuman.khandual@arm.com

drivers/perf: arm_pmu: Handle 47 bit counters

2022-03-08T13:32:48Z

The current ARM PMU framework can only deal with 32 or 64bit counters. Teach it about a 47bit flavour. Yes, this is odd. Reviewed-by: Hector Martin Signed-off-by: Marc Zyngier Signed-off-by: Will Deacon

KVM: arm64: Fix PMU probe ordering

2021-09-20T11:43:34Z

Russell reported that since 5.13, KVM's probing of the PMU has started to fail on his HW. As it turns out, there is an implicit ordering dependency between the architectural PMU probing code and and KVM's own probing. If, due to probe ordering reasons, KVM probes before the PMU driver, it will fail to detect the PMU and prevent it from being advertised to guests as well as the VMM. Obviously, this is one probing too many, and we should be able to deal with any ordering. Add a callback from the PMU code into KVM to advertise the registration of a host CPU PMU, allowing for any probing order. Fixes: 5421db1be3b1 ("KVM: arm64: Divorce the perf code from oprofile helpers") Reported-by: "Russell King (Oracle)" Tested-by: Russell King (Oracle) Signed-off-by: Marc Zyngier Link: https://lore.kernel.org/r/YUYRKVflRtUytzy5@shell.armlinux.org.uk Cc: stable@vger.kernel.org

Revert "arm64: Enable perf events based hard lockup detector"

2021-01-13T15:08:41Z

This reverts commit 367c820ef08082e68df8a3bc12e62393af21e4b5. lockup_detector_init() makes heavy use of per-cpu variables and must be called with preemption disabled. Usually, it's handled early during boot in kernel_init_freeable(), before SMP has been initialised. Since we do not know whether or not our PMU interrupt can be signalled as an NMI until considerably later in the boot process, the Arm PMU driver attempts to re-initialise the lockup detector off the back of a device_initcall(). Unfortunately, this is called from preemptible context and results in the following splat: | BUG: using smp_processor_id() in preemptible [00000000] code: swapper/0/1 | caller is debug_smp_processor_id+0x20/0x2c | CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.10.0+ #276 | Hardware name: linux,dummy-virt (DT) | Call trace: | dump_backtrace+0x0/0x3c0 | show_stack+0x20/0x6c | dump_stack+0x2f0/0x42c | check_preemption_disabled+0x1cc/0x1dc | debug_smp_processor_id+0x20/0x2c | hardlockup_detector_event_create+0x34/0x18c | hardlockup_detector_perf_init+0x2c/0x134 | watchdog_nmi_probe+0x18/0x24 | lockup_detector_init+0x44/0xa8 | armv8_pmu_driver_init+0x54/0x78 | do_one_initcall+0x184/0x43c | kernel_init_freeable+0x368/0x380 | kernel_init+0x1c/0x1cc | ret_from_fork+0x10/0x30 Rather than bodge this with raw_smp_processor_id() or randomly disabling preemption, simply revert the culprit for now until we figure out how to do this properly. Reported-by: Lecopzer Chen Signed-off-by: Will Deacon Acked-by: Mark Rutland Cc: Sumit Garg Cc: Alexandru Elisei Link: https://lore.kernel.org/r/20201221162249.3119-1-lecopzer.chen@mediatek.com Link: https://lore.kernel.org/r/20210112221855.10666-1-will@kernel.org Signed-off-by: Catalin Marinas

arm64: Enable perf events based hard lockup detector

2020-11-25T15:18:39Z

With the recent feature added to enable perf events to use pseudo NMIs as interrupts on platforms which support GICv3 or later, its now been possible to enable hard lockup detector (or NMI watchdog) on arm64 platforms. So enable corresponding support. One thing to note here is that normally lockup detector is initialized just after the early initcalls but PMU on arm64 comes up much later as device_initcall(). So we need to re-initialize lockup detection once PMU has been initialized. Signed-off-by: Sumit Garg Acked-by: Alexandru Elisei Link: https://lore.kernel.org/r/1602060704-10921-1-git-send-email-sumit.garg@linaro.org Signed-off-by: Will Deacon