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 <j@jannau.net>
Link: https://lore.kernel.org/asahi/20230215-arm_pmu_m1_regression-v1-1-f5a266577c8d@jannau.net/
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Asahi Lina <lina@asahilina.net>
Cc: Eric Curtin <ecurtin@redhat.com>
Tested-by: Janne Grunau <j@jannau.net>
Link: https://lore.kernel.org/r/20230216141240.3833272-2-mark.rutland@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
It turns out the optimisation implemented by commit 4f2c3872dde5 is
totally broken, since all the places that consume hw->dtcs_used for
events other than cycle count are still not expecting it to be sparsely
populated, and fail to read all the relevant DTC counters correctly if
so.
If implemented correctly, the optimisation potentially saves up to 3
register reads per event update, which is reasonably significant for
events targeting a single node, but still not worth a massive amount of
additional code complexity overall. Getting it right within the current
design looks a fair bit more involved than it was ever intended to be,
so let's just make a functional revert which restores the old behaviour
while still backporting easily.
Fixes: 4f2c3872dde5 ("perf/arm-cmn: Optimise DTC counter accesses")
Reported-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/b41bb4ed7283c3d8400ce5cf5e6ec94915e6750f.1674498637.git.robin.murphy@arm.com
Signed-off-by: Will Deacon <will@kernel.org>
Pull RISC-V updates from Palmer Dabbelt: - Support for the T-Head PMU via the perf subsystem - ftrace support for rv32 - Support for non-volatile memory devices - Various fixes and cleanups * tag 'riscv-for-linus-6.2-mw1' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux: (52 commits) Documentation: RISC-V: patch-acceptance: s/implementor/implementer Documentation: RISC-V: Mention the UEFI Standards Documentation: RISC-V: Allow patches for non-standard behavior Documentation: RISC-V: Fix a typo in patch-acceptance riscv: Fixup compile error with !MMU riscv: Fix P4D_SHIFT definition for 3-level page table mode riscv: Apply a static assert to riscv_isa_ext_id RISC-V: Add some comments about the shadow and overflow stacks RISC-V: Align the shadow stack RISC-V: Ensure Zicbom has a valid block size RISC-V: Introduce riscv_isa_extension_check RISC-V: Improve use of isa2hwcap[] riscv: Don't duplicate _ALTERNATIVE_CFG* macros riscv: alternatives: Drop the underscores from the assembly macro names riscv: alternatives: Don't name unused macro parameters riscv: Don't duplicate __ALTERNATIVE_CFG in __ALTERNATIVE_CFG_2 riscv: mm: call best_map_size many times during linear-mapping riscv: Move cast inside kernel_mapping_[pv]a_to_[vp]a riscv: Fix crash during early errata patching riscv: boot: add zstd support ...
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()
...
Pull irq updates from Thomas Gleixner:
"Updates for the interrupt core and driver subsystem:
The bulk is the rework of the MSI subsystem to support per device MSI
interrupt domains. This solves conceptual problems of the current
PCI/MSI design which are in the way of providing support for
PCI/MSI[-X] and the upcoming PCI/IMS mechanism on the same device.
IMS (Interrupt Message Store] is a new specification which allows
device manufactures to provide implementation defined storage for MSI
messages (as opposed to PCI/MSI and PCI/MSI-X that has a specified
message store which is uniform accross all devices). The PCI/MSI[-X]
uniformity allowed us to get away with "global" PCI/MSI domains.
IMS not only allows to overcome the size limitations of the MSI-X
table, but also gives the device manufacturer the freedom to store the
message in arbitrary places, even in host memory which is shared with
the device.
There have been several attempts to glue this into the current MSI
code, but after lengthy discussions it turned out that there is a
fundamental design problem in the current PCI/MSI-X implementation.
This needs some historical background.
When PCI/MSI[-X] support was added around 2003, interrupt management
was completely different from what we have today in the actively
developed architectures. Interrupt management was completely
architecture specific and while there were attempts to create common
infrastructure the commonalities were rudimentary and just providing
shared data structures and interfaces so that drivers could be written
in an architecture agnostic way.
The initial PCI/MSI[-X] support obviously plugged into this model
which resulted in some basic shared infrastructure in the PCI core
code for setting up MSI descriptors, which are a pure software
construct for holding data relevant for a particular MSI interrupt,
but the actual association to Linux interrupts was completely
architecture specific. This model is still supported today to keep
museum architectures and notorious stragglers alive.
In 2013 Intel tried to add support for hot-pluggable IO/APICs to the
kernel, which was creating yet another architecture specific mechanism
and resulted in an unholy mess on top of the existing horrors of x86
interrupt handling. The x86 interrupt management code was already an
incomprehensible maze of indirections between the CPU vector
management, interrupt remapping and the actual IO/APIC and PCI/MSI[-X]
implementation.
At roughly the same time ARM struggled with the ever growing SoC
specific extensions which were glued on top of the architected GIC
interrupt controller.
This resulted in a fundamental redesign of interrupt management and
provided the today prevailing concept of hierarchical interrupt
domains. This allowed to disentangle the interactions between x86
vector domain and interrupt remapping and also allowed ARM to handle
the zoo of SoC specific interrupt components in a sane way.
The concept of hierarchical interrupt domains aims to encapsulate the
functionality of particular IP blocks which are involved in interrupt
delivery so that they become extensible and pluggable. The X86
encapsulation looks like this:
|--- device 1
[Vector]---[Remapping]---[PCI/MSI]--|...
|--- device N
where the remapping domain is an optional component and in case that
it is not available the PCI/MSI[-X] domains have the vector domain as
their parent. This reduced the required interaction between the
domains pretty much to the initialization phase where it is obviously
required to establish the proper parent relation ship in the
components of the hierarchy.
While in most cases the model is strictly representing the chain of IP
blocks and abstracting them so they can be plugged together to form a
hierarchy, the design stopped short on PCI/MSI[-X]. Looking at the
hardware it's clear that the actual PCI/MSI[-X] interrupt controller
is not a global entity, but strict a per PCI device entity.
Here we took a short cut on the hierarchical model and went for the
easy solution of providing "global" PCI/MSI domains which was possible
because the PCI/MSI[-X] handling is uniform across the devices. This
also allowed to keep the existing PCI/MSI[-X] infrastructure mostly
unchanged which in turn made it simple to keep the existing
architecture specific management alive.
A similar problem was created in the ARM world with support for IP
block specific message storage. Instead of going all the way to stack
a IP block specific domain on top of the generic MSI domain this ended
in a construct which provides a "global" platform MSI domain which
allows overriding the irq_write_msi_msg() callback per allocation.
In course of the lengthy discussions we identified other abuse of the
MSI infrastructure in wireless drivers, NTB etc. where support for
implementation specific message storage was just mindlessly glued into
the existing infrastructure. Some of this just works by chance on
particular platforms but will fail in hard to diagnose ways when the
driver is used on platforms where the underlying MSI interrupt
management code does not expect the creative abuse.
Another shortcoming of today's PCI/MSI-X support is the inability to
allocate or free individual vectors after the initial enablement of
MSI-X. This results in an works by chance implementation of VFIO (PCI
pass-through) where interrupts on the host side are not set up upfront
to avoid resource exhaustion. They are expanded at run-time when the
guest actually tries to use them. The way how this is implemented is
that the host disables MSI-X and then re-enables it with a larger
number of vectors again. That works by chance because most device
drivers set up all interrupts before the device actually will utilize
them. But that's not universally true because some drivers allocate a
large enough number of vectors but do not utilize them until it's
actually required, e.g. for acceleration support. But at that point
other interrupts of the device might be in active use and the MSI-X
disable/enable dance can just result in losing interrupts and
therefore hard to diagnose subtle problems.
Last but not least the "global" PCI/MSI-X domain approach prevents to
utilize PCI/MSI[-X] and PCI/IMS on the same device due to the fact
that IMS is not longer providing a uniform storage and configuration
model.
The solution to this is to implement the missing step and switch from
global PCI/MSI domains to per device PCI/MSI domains. The resulting
hierarchy then looks like this:
|--- [PCI/MSI] device 1
[Vector]---[Remapping]---|...
|--- [PCI/MSI] device N
which in turn allows to provide support for multiple domains per
device:
|--- [PCI/MSI] device 1
|--- [PCI/IMS] device 1
[Vector]---[Remapping]---|...
|--- [PCI/MSI] device N
|--- [PCI/IMS] device N
This work converts the MSI and PCI/MSI core and the x86 interrupt
domains to the new model, provides new interfaces for post-enable
allocation/free of MSI-X interrupts and the base framework for
PCI/IMS. PCI/IMS has been verified with the work in progress IDXD
driver.
There is work in progress to convert ARM over which will replace the
platform MSI train-wreck. The cleanup of VFIO, NTB and other creative
"solutions" are in the works as well.
Drivers:
- Updates for the LoongArch interrupt chip drivers
- Support for MTK CIRQv2
- The usual small fixes and updates all over the place"
* tag 'irq-core-2022-12-10' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (134 commits)
irqchip/ti-sci-inta: Fix kernel doc
irqchip/gic-v2m: Mark a few functions __init
irqchip/gic-v2m: Include arm-gic-common.h
irqchip/irq-mvebu-icu: Fix works by chance pointer assignment
iommu/amd: Enable PCI/IMS
iommu/vt-d: Enable PCI/IMS
x86/apic/msi: Enable PCI/IMS
PCI/MSI: Provide pci_ims_alloc/free_irq()
PCI/MSI: Provide IMS (Interrupt Message Store) support
genirq/msi: Provide constants for PCI/IMS support
x86/apic/msi: Enable MSI_FLAG_PCI_MSIX_ALLOC_DYN
PCI/MSI: Provide post-enable dynamic allocation interfaces for MSI-X
PCI/MSI: Provide prepare_desc() MSI domain op
PCI/MSI: Split MSI-X descriptor setup
genirq/msi: Provide MSI_FLAG_MSIX_ALLOC_DYN
genirq/msi: Provide msi_domain_alloc_irq_at()
genirq/msi: Provide msi_domain_ops:: Prepare_desc()
genirq/msi: Provide msi_desc:: Msi_data
genirq/msi: Provide struct msi_map
x86/apic/msi: Remove arch_create_remap_msi_irq_domain()
...
* for-next/perf: (21 commits) arm_pmu: Drop redundant armpmu->map_event() in armpmu_event_init() drivers/perf: hisi: Add TLP filter support Documentation: perf: Indent filter options list of hisi-pcie-pmu docs: perf: Fix PMU instance name of hisi-pcie-pmu drivers/perf: hisi: Fix some event id for hisi-pcie-pmu arm64/perf: Replace PMU version number '0' with ID_AA64DFR0_EL1_PMUVer_NI perf/amlogic: Remove unused header inclusions of <linux/version.h> perf/amlogic: Fix build error for x86_64 allmodconfig dt-binding: perf: Add Amlogic DDR PMU docs/perf: Add documentation for the Amlogic G12 DDR PMU perf/amlogic: Add support for Amlogic meson G12 SoC DDR PMU driver MAINTAINERS: Update HiSilicon PMU maintainers perf: arm_cspmu: Fix module cyclic dependency perf: arm_cspmu: Fix build failure on x86_64 perf: arm_cspmu: Fix modular builds due to missing MODULE_LICENSE()s perf: arm_cspmu: Add support for NVIDIA SCF and MCF attribute perf: arm_cspmu: Add support for ARM CoreSight PMU driver perf/smmuv3: Fix hotplug callback leak in arm_smmu_pmu_init() perf/arm_dmc620: Fix hotplug callback leak in dmc620_pmu_init() drivers: perf: marvell_cn10k: Fix hotplug callback leak in tad_pmu_init() ...
__hw_perf_event_init() already calls armpmu->map_event() callback, and also returns its error code including -ENOENT, along with a debug callout. Hence an additional armpmu->map_event() check for -ENOENT is redundant. Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Will Deacon <will@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: linux-arm-kernel@lists.infradead.org Cc: linux-kernel@vger.kernel.org Acked-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com> Link: https://lore.kernel.org/r/20221202015611.338499-1-anshuman.khandual@arm.com Signed-off-by: Will Deacon <will@kernel.org>
The PMU support to filter the TLP when counting the bandwidth with below options: - only count the TLP headers - only count the TLP payloads - count both TLP headers and payloads In the current driver it's default to count the TLP payloads only, which will have an implicity side effects that on the traffic only have header only TLPs, we'll get no data. Make this user configuration through "len_mode" parameter and make it default to count both TLP headers and payloads when user not specified. Also update the documentation for it. Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> Signed-off-by: Yicong Yang <yangyicong@hisilicon.com> Link: https://lore.kernel.org/r/20221117084136.53572-5-yangyicong@huawei.com Signed-off-by: Will Deacon <will@kernel.org>
Some event id of hisi-pcie-pmu is incorrect, fix them.
Fixes: 8404b0fbc7fb ("drivers/perf: hisi: Add driver for HiSilicon PCIe PMU")
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Link: https://lore.kernel.org/r/20221117084136.53572-2-yangyicong@huawei.com
Signed-off-by: Will Deacon <will@kernel.org>
According to the "Abaci Robot": | ./drivers/perf/amlogic/meson_g12_ddr_pmu.c:15 linux/version.h not needed. | ./drivers/perf/amlogic/meson_ddr_pmu_core.c: 19 linux/version.h not needed. So drop the unnecessary #include directives. Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3280 Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3282 Reported-by: Abaci Robot <abaci@linux.alibaba.com> Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com> Link: https://lore.kernel.org/r/20221129032108.119661-1-jiapeng.chong@linux.alibaba.com Link: https://lore.kernel.org/r/20221129032108.119661-2-jiapeng.chong@linux.alibaba.com [will: Squashed patches together, filled out commit message a bit more] Signed-off-by: Will Deacon <will@kernel.org>