kernel.git/arch/powerpc/perf, branch linux-5.17.y

powerpc/perf: Fix the threshold compare group constraint for power9

2022-06-09T08:26:09Z

[ Upstream commit ab0cc6bbf0c812731c703ec757fcc3fc3a457a34 ] Thresh compare bits for a event is used to program thresh compare field in Monitor Mode Control Register A (MMCRA: 9-18 bits for power9). When scheduling events as a group, all events in that group should match value in threshold bits (like thresh compare, thresh control, thresh select). Otherwise event open for the sibling events should fail. But in the current code, incase thresh compare bits are not valid, we are not failing in group_constraint function which can result in invalid group schduling. Fix the issue by returning -1 incase event is threshold and threshold compare value is not valid. Thresh control bits in the event code is used to program thresh_ctl field in Monitor Mode Control Register A (MMCRA: 48-55). In below example, the scheduling of group events PM_MRK_INST_CMPL (873534401e0) and PM_THRESH_MET (8734340101ec) is expected to fail as both event request different thresh control bits and invalid thresh compare value. Result before the patch changes: [command]# perf stat -e "{r8735340401e0,r8734340101ec}" sleep 1 Performance counter stats for 'sleep 1': 11,048 r8735340401e0 1,967 r8734340101ec 1.001354036 seconds time elapsed 0.001421000 seconds user 0.000000000 seconds sys Result after the patch changes: [command]# perf stat -e "{r8735340401e0,r8734340101ec}" sleep 1 Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (r8735340401e0). /bin/dmesg | grep -i perf may provide additional information. Fixes: 78a16d9fc1206 ("powerpc/perf: Avoid FAB_*_MATCH checks for power9") Signed-off-by: Kajol Jain Reviewed-by: Athira Rajeev Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220506061015.43916-2-kjain@linux.ibm.com Signed-off-by: Sasha Levin

powerpc/perf: Fix the threshold compare group constraint for power10

2022-06-09T08:26:09Z

[ Upstream commit 505d31650ba96d6032313480fdb566d289a4698c ] Thresh compare bits for a event is used to program thresh compare field in Monitor Mode Control Register A (MMCRA: 8-18 bits for power10). When scheduling events as a group, all events in that group should match value in threshold bits. Otherwise event open for the sibling events should fail. But in the current code, incase thresh compare bits are not valid, we are not failing in group_constraint function which can result in invalid group schduling. Fix the issue by returning -1 incase event is threshold and threshold compare value is not valid in group_constraint function. Patch also fixes the p10_thresh_cmp_val function to return -1, incase threshold bits are not valid and changes corresponding check in is_thresh_cmp_valid function to return false only when the thresh_cmp value is less then 0. Thresh control bits in the event code is used to program thresh_ctl field in Monitor Mode Control Register A (MMCRA: 48-55). In below example, the scheduling of group events PM_MRK_INST_CMPL (3534401e0) and PM_THRESH_MET (34340101ec) is expected to fail as both event request different thresh control bits. Result before the patch changes: [command]# perf stat -e "{r35340401e0,r34340101ec}" sleep 1 Performance counter stats for 'sleep 1': 8,482 r35340401e0 0 r34340101ec 1.001474838 seconds time elapsed 0.001145000 seconds user 0.000000000 seconds sys Result after the patch changes: [command]# perf stat -e "{r35340401e0,r34340101ec}" sleep 1 Performance counter stats for 'sleep 1': r35340401e0 r34340101ec 1.001499607 seconds time elapsed 0.000204000 seconds user 0.000760000 seconds sys Fixes: 82d2c16b350f7 ("powerpc/perf: Adds support for programming of Thresholding in P10") Signed-off-by: Kajol Jain Reviewed-by: Athira Rajeev Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220506061015.43916-1-kjain@linux.ibm.com Signed-off-by: Sasha Levin

powerpc/perf: Fix 32bit compile

2022-05-09T07:16:28Z

[ Upstream commit bb82c574691daf8f7fa9a160264d15c5804cb769 ] The "read_bhrb" global symbol is only called under CONFIG_PPC64 of arch/powerpc/perf/core-book3s.c but it is compiled for both 32 and 64 bit anyway (and LLVM fails to link this on 32bit). This fixes it by moving bhrb.o to obj64 targets. Signed-off-by: Alexey Kardashevskiy Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220421025756.571995-1-aik@ozlabs.ru Signed-off-by: Sasha Levin

powerpc/perf: Fix power10 event alternatives

2022-04-27T12:41:13Z

[ Upstream commit c6cc9a852f123301d5271f1484df8e961b2b64f1 ] When scheduling a group of events, there are constraint checks done to make sure all events can go in a group. Example, one of the criteria is that events in a group cannot use the same PMC. But platform specific PMU supports alternative event for some of the event codes. During perf_event_open(), if any event group doesn't match constraint check criteria, further lookup is done to find alternative event. By current design, the array of alternatives events in PMU code is expected to be sorted by column 0. This is because in find_alternative() the return criteria is based on event code comparison. ie. "event < ev_alt[i][0])". This optimisation is there since find_alternative() can be called multiple times. In power10 PMU code, the alternative event array is not sorted properly and hence there is breakage in finding alternative event. To work with existing logic, fix the alternative event array to be sorted by column 0 for power10-pmu.c Results: In case where an alternative event is not chosen when we could, events will be multiplexed. ie, time sliced where it could actually run concurrently. Example, in power10 PM_INST_CMPL_ALT(0x00002) has alternative event, PM_INST_CMPL(0x500fa). Without the fix, if a group of events with PMC1 to PMC4 is used along with PM_INST_CMPL_ALT, it will be time sliced since all programmable PMC's are consumed already. But with the fix, when it picks alternative event on PMC5, all events will run concurrently. Before: # perf stat -e r00002,r100fc,r200fa,r300fc,r400fc Performance counter stats for 'system wide': 328668935 r00002 (79.94%) 56501024 r100fc (79.95%) 49564238 r200fa (79.95%) 376 r300fc (80.19%) 660 r400fc (79.97%) 4.039150522 seconds time elapsed With the fix, since alternative event is chosen to run on PMC6, events will be run concurrently. After: # perf stat -e r00002,r100fc,r200fa,r300fc,r400fc Performance counter stats for 'system wide': 23596607 r00002 4907738 r100fc 2283608 r200fa 135 r300fc 248 r400fc 1.664671390 seconds time elapsed Fixes: a64e697cef23 ("powerpc/perf: power10 Performance Monitoring support") Signed-off-by: Athira Rajeev Reviewed-by: Madhavan Srinivasan Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220419114828.89843-2-atrajeev@linux.vnet.ibm.com Signed-off-by: Sasha Levin

powerpc/perf: Fix power9 event alternatives

2022-04-27T12:41:13Z

[ Upstream commit 0dcad700bb2776e3886fe0a645a4bf13b1e747cd ] When scheduling a group of events, there are constraint checks done to make sure all events can go in a group. Example, one of the criteria is that events in a group cannot use the same PMC. But platform specific PMU supports alternative event for some of the event codes. During perf_event_open(), if any event group doesn't match constraint check criteria, further lookup is done to find alternative event. By current design, the array of alternatives events in PMU code is expected to be sorted by column 0. This is because in find_alternative() the return criteria is based on event code comparison. ie. "event < ev_alt[i][0])". This optimisation is there since find_alternative() can be called multiple times. In power9 PMU code, the alternative event array is not sorted properly and hence there is breakage in finding alternative events. To work with existing logic, fix the alternative event array to be sorted by column 0 for power9-pmu.c Results: With alternative events, multiplexing can be avoided. That is, for example, in power9 PM_LD_MISS_L1 (0x3e054) has alternative event, PM_LD_MISS_L1_ALT (0x400f0). This is an identical event which can be programmed in a different PMC. Before: # perf stat -e r3e054,r300fc Performance counter stats for 'system wide': 1057860 r3e054 (50.21%) 379 r300fc (49.79%) 0.944329741 seconds time elapsed Since both the events are using PMC3 in this case, they are multiplexed here. After: # perf stat -e r3e054,r300fc Performance counter stats for 'system wide': 1006948 r3e054 182 r300fc Fixes: 91e0bd1e6251 ("powerpc/perf: Add PM_LD_MISS_L1 and PM_BR_2PATH to power9 event list") Signed-off-by: Athira Rajeev Reviewed-by: Madhavan Srinivasan Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220419114828.89843-1-atrajeev@linux.vnet.ibm.com Signed-off-by: Sasha Levin

powerpc/64s/hash: Make hash faults work in NMI context

2022-04-13T17:27:18Z

[ Upstream commit 8b91cee5eadd2021f55e6775f2d50bd56d00c217 ] Hash faults are not resoved in NMI context, instead causing the access to fail. This is done because perf interrupts can get backtraces including walking the user stack, and taking a hash fault on those could deadlock on the HPTE lock if the perf interrupt hits while the same HPTE lock is being held by the hash fault code. The user-access for the stack walking will notice the access failed and deal with that in the perf code. The reason to allow perf interrupts in is to better profile hash faults. The problem with this is any hash fault on a kernel access that happens in NMI context will crash, because kernel accesses must not fail. Hard lockups, system reset, machine checks that access vmalloc space including modules and including stack backtracing and symbol lookup in modules, per-cpu data, etc could all run into this problem. Fix this by disallowing perf interrupts in the hash fault code (the direct hash fault is covered by MSR[EE]=0 so the PMI disable just needs to extend to the preload case). This simplifies the tricky logic in hash faults and perf, at the cost of reduced profiling of hash faults. perf can still latch addresses when interrupts are disabled, it just won't get the stack trace at that point, so it would still find hot spots, just sometimes with confusing stack chains. An alternative could be to allow perf interrupts here but always do the slowpath stack walk if we are in nmi context, but that slows down all perf interrupt stack walking on hash though and it does not remove as much tricky code. Reported-by: Laurent Dufour Signed-off-by: Nicholas Piggin Tested-by: Laurent Dufour Reviewed-by: Aneesh Kumar K.V Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220204035348.545435-1-npiggin@gmail.com Signed-off-by: Sasha Levin

powerpc/perf: Don't use perf_hw_context for trace IMC PMU

2022-04-08T11:58:01Z

[ Upstream commit 0198322379c25215b2778482bf1221743a76e2b5 ] Trace IMC (In-Memory collection counters) in powerpc is useful for application level profiling. For trace_imc, presently task context (task_ctx_nr) is set to perf_hw_context. But perf_hw_context should only be used for CPU PMU. See commit 26657848502b ("perf/core: Verify we have a single perf_hw_context PMU"). So for trace_imc, even though it is per thread PMU, it is preferred to use sw_context in order to be able to do application level monitoring. Hence change the task_ctx_nr to use perf_sw_context. Fixes: 012ae244845f ("powerpc/perf: Trace imc PMU functions") Signed-off-by: Athira Rajeev Reviewed-by: Madhavan Srinivasan [mpe: Update subject & incorporate notes into change log, reflow comment] Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220202041837.65968-1-atrajeev@linux.vnet.ibm.com Signed-off-by: Sasha Levin

powerpc/perf: Fix power_pmu_disable to call clear_pmi_irq_pending only if PMI is pending

2022-01-24T06:31:15Z

Running selftest with CONFIG_PPC_IRQ_SOFT_MASK_DEBUG enabled in kernel triggered below warning: [ 172.851380] ------------[ cut here ]------------ [ 172.851391] WARNING: CPU: 8 PID: 2901 at arch/powerpc/include/asm/hw_irq.h:246 power_pmu_disable+0x270/0x280 [ 172.851402] Modules linked in: dm_mod bonding nft_ct nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 ip_set nf_tables rfkill nfnetlink sunrpc xfs libcrc32c pseries_rng xts vmx_crypto uio_pdrv_genirq uio sch_fq_codel ip_tables ext4 mbcache jbd2 sd_mod t10_pi sg ibmvscsi ibmveth scsi_transport_srp fuse [ 172.851442] CPU: 8 PID: 2901 Comm: lost_exception_ Not tainted 5.16.0-rc5-03218-g798527287598 #2 [ 172.851451] NIP: c00000000013d600 LR: c00000000013d5a4 CTR: c00000000013b180 [ 172.851458] REGS: c000000017687860 TRAP: 0700 Not tainted (5.16.0-rc5-03218-g798527287598) [ 172.851465] MSR: 8000000000029033 CR: 48004884 XER: 20040000 [ 172.851482] CFAR: c00000000013d5b4 IRQMASK: 1 [ 172.851482] GPR00: c00000000013d5a4 c000000017687b00 c000000002a10600 0000000000000004 [ 172.851482] GPR04: 0000000082004000 c0000008ba08f0a8 0000000000000000 00000008b7ed0000 [ 172.851482] GPR08: 00000000446194f6 0000000000008000 c00000000013b118 c000000000d58e68 [ 172.851482] GPR12: c00000000013d390 c00000001ec54a80 0000000000000000 0000000000000000 [ 172.851482] GPR16: 0000000000000000 0000000000000000 c000000015d5c708 c0000000025396d0 [ 172.851482] GPR20: 0000000000000000 0000000000000000 c00000000a3bbf40 0000000000000003 [ 172.851482] GPR24: 0000000000000000 c0000008ba097400 c0000000161e0d00 c00000000a3bb600 [ 172.851482] GPR28: c000000015d5c700 0000000000000001 0000000082384090 c0000008ba0020d8 [ 172.851549] NIP [c00000000013d600] power_pmu_disable+0x270/0x280 [ 172.851557] LR [c00000000013d5a4] power_pmu_disable+0x214/0x280 [ 172.851565] Call Trace: [ 172.851568] [c000000017687b00] [c00000000013d5a4] power_pmu_disable+0x214/0x280 (unreliable) [ 172.851579] [c000000017687b40] [c0000000003403ac] perf_pmu_disable+0x4c/0x60 [ 172.851588] [c000000017687b60] [c0000000003445e4] __perf_event_task_sched_out+0x1d4/0x660 [ 172.851596] [c000000017687c50] [c000000000d1175c] __schedule+0xbcc/0x12a0 [ 172.851602] [c000000017687d60] [c000000000d11ea8] schedule+0x78/0x140 [ 172.851608] [c000000017687d90] [c0000000001a8080] sys_sched_yield+0x20/0x40 [ 172.851615] [c000000017687db0] [c0000000000334dc] system_call_exception+0x18c/0x380 [ 172.851622] [c000000017687e10] [c00000000000c74c] system_call_common+0xec/0x268 The warning indicates that MSR_EE being set(interrupt enabled) when there was an overflown PMC detected. This could happen in power_pmu_disable since it runs under interrupt soft disable condition ( local_irq_save ) and not with interrupts hard disabled. commit 2c9ac51b850d ("powerpc/perf: Fix PMU callbacks to clear pending PMI before resetting an overflown PMC") intended to clear PMI pending bit in Paca when disabling the PMU. It could happen that PMC gets overflown while code is in power_pmu_disable callback function. Hence add a check to see if PMI pending bit is set in Paca before clearing it via clear_pmi_pending. Fixes: 2c9ac51b850d ("powerpc/perf: Fix PMU callbacks to clear pending PMI before resetting an overflown PMC") Reported-by: Sachin Sant Signed-off-by: Athira Rajeev Tested-by: Sachin Sant Reviewed-by: Nicholas Piggin Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220122033429.25395-1-atrajeev@linux.vnet.ibm.com

powerpc/perf: Only define power_pmu_wants_prompt_pmi() for CONFIG_PPC64

2022-01-17T04:04:13Z

power_pmu_wants_prompt_pmi() is used to decide if PMIs should be taken promptly. This is valid only for ppc64 and is used only if CONFIG_PPC_BOOK3S_64=y. Hence include the function under config check for PPC64. Fixes warning for 32-bit compilation: arch/powerpc/perf/core-book3s.c:2455:6: warning: no previous prototype for 'power_pmu_wants_prompt_pmi' 2455 | bool power_pmu_wants_prompt_pmi(void) | ^~~~~~~~~~~~~~~~~~~~~~~~~~ Fixes: 5a7745b96f43 ("powerpc/64s/perf: add power_pmu_wants_prompt_pmi to say whether perf wants PMIs to be soft-NMI") Reported-by: kernel test robot Signed-off-by: Athira Rajeev Reviewed-by: Nicholas Piggin [mpe: Move inside existing CONFIG_PPC64 ifdef block] Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20220114031355.87480-1-atrajeev@linux.vnet.ibm.com

powerpc/perf: Fix spelling of "its"

2021-12-23T11:36:58Z

Use the possessive "its" instead of the contraction of "it is" (it's). Signed-off-by: Randy Dunlap Signed-off-by: Michael Ellerman Link: https://lore.kernel.org/r/20211223003942.22098-1-rdunlap@infradead.org