From a63d7408afbd108944a6b05bdf0b0d75f32755b9 Mon Sep 17 00:00:00 2001 From: Mark Rutland Date: Mon, 18 Nov 2024 12:02:04 +0000 Subject: [PATCH 1/3] arm64: disable ARCH_CORRECT_STACKTRACE_ON_KRETPROBE tests The kprobes_test suite's test_stacktrace_on_nested_kretprobe() test currently fails on arm64, e.g. | KTAP version 1 | 1..1 | KTAP version 1 | # Subtest: kprobes_test | # module: test_kprobes | 1..7 | ok 1 test_kprobe | ok 2 test_kprobes | ok 3 test_kprobe_missed | ok 4 test_kretprobe | ok 5 test_kretprobes | ok 6 test_stacktrace_on_kretprobe | # test_stacktrace_on_nested_kretprobe: EXPECTATION FAILED at lib/test_kprobes.c:327 | Expected stack_buf[i + 1] == target_return_address[1], but | stack_buf[i + 1] == -96519936577004 (0xffffa83733777214) | target_return_address[1] == -96519936577136 (0xffffa83733777190) | # test_stacktrace_on_nested_kretprobe: EXPECTATION FAILED at lib/test_kprobes.c:338 | Expected stack_buf[1] == target_return_address[1], but | stack_buf[1] == -96519936577004 (0xffffa83733777214) | target_return_address[1] == -96519936577136 (0xffffa83733777190) | not ok 7 test_stacktrace_on_nested_kretprobe | # kprobes_test: pass:6 fail:1 skip:0 total:7 | # Totals: pass:6 fail:1 skip:0 total:7 | not ok 1 kprobes_test The test assumes that when a stacktrace straddles an exception boundary, no necessary entries will be omitted and no extraneous entries will be reported, and when unwinding from a kretprobed callee, the next entry in the trace will be its immediate caller (whether kretprobed or not). Recently the arm64 stacktrace code was changed to always report the LR at an exception boundary, where we don't know whether the LR is live. In the case of the kretprobe trampoline the LR is not live at the time the stacktrace is performed, and so the entry in the trace for the LR is extraneous. This can be seen if a call to show_stack() is added to stacktrace_internal_return_handler(): | Call trace: | show_stack+0x18/0x30 (C) | stacktrace_internal_return_handler+0x130/0x43c | __kretprobe_trampoline_handler+0xa0/0x130 | kretprobe_breakpoint_handler+0x50/0x70 | call_break_hook+0x74/0x8c | brk_handler+0x1c/0x60 | do_debug_exception+0x68/0x114 | el1_dbg+0x70/0x94 | el1h_64_sync_handler+0xc4/0xe4 | el1h_64_sync+0x6c/0x70 | kprobe_stacktrace_target+0x34/0x48 (P) | kprobe_stacktrace_target+0x34/0x48 (LK) <-------- extra entry here | kprobe_stacktrace_driver+0x24/0x40 (K) | test_stacktrace_on_nested_kretprobe+0x84/0x160 | kunit_try_run_case+0x6c/0x160 | kunit_generic_run_threadfn_adapter+0x28/0x4c | kthread+0x110/0x114 | ret_from_fork+0x10/0x20 This breaks test_stacktrace_on_nested_kretprobe() because while the caller (kprobe_stacktrace_driver()) appears in the trace, it doesn't occur *immediately* after the first instance of callee (kprobe_stacktrace_target()). While this behaviour is unfortunate for the kretprobes tests, the behaviour is desirable elsewhere (e.g. anywhere a human will read the trace), and is otherwise not harmful. For the moment, deselect ARCH_CORRECT_STACKTRACE_ON_KRETPROBE on arm64 to disable the tests which depend on this behaviour. With ARCH_CORRECT_STACKTRACE_ON_KRETPROBE deselected, the remaining tests work as expected, e.g. | KTAP version 1 | 1..1 | KTAP version 1 | # Subtest: kprobes_test | # module: test_kprobes | 1..5 | ok 1 test_kprobe | ok 2 test_kprobes | ok 3 test_kprobe_missed | ok 4 test_kretprobe | ok 5 test_kretprobes | # kprobes_test: pass:5 fail:0 skip:0 total:5 | # Totals: pass:5 fail:0 skip:0 total:5 | ok 1 kprobes_test In future we have several options to improve matters, e.g. * Add metadata and update arm64's unwinder to skip the LR in this case. This is likely to happen as part of work for RELIABLE_STACKTRACE for other reasons, and might solve this case by coincidence. * Modify the kretprobes tests to only require that the caller appears in the trace after the callee, rather than requiring that it is *immediately* after the callee. We might want separate strict/not-strict options for this. * Use reliable stacktrace for these tests, so that architectures which cannot unwind across exception boundaries can explicitly handle this by returning an error. Fixes: c2c6b27b5aa1 ("arm64: stacktrace: unwind exception boundaries") Signed-off-by: Mark Rutland Reported-by: Kristina Martsenko Cc: Ard Biesheuvel Cc: Josh Poimboeuf Cc: Kalesh Singh Cc: Madhavan T. Venkataraman Cc: Marc Zyngier Cc: Mark Brown Cc: Masami Hiramatsu Cc: Miroslav Benes Cc: Puranjay Mohan Cc: Steven Rostedt Cc: Will Deacon Reviewed-by: Mark Brown Link: https://lore.kernel.org/r/20241118120204.3961548-1-mark.rutland@arm.com Signed-off-by: Catalin Marinas --- arch/arm64/Kconfig | 1 - 1 file changed, 1 deletion(-) diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index cd900e640b10d..9d8e97291c8c8 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -14,7 +14,6 @@ config ARM64 select ARCH_HAS_DEBUG_WX select ARCH_BINFMT_ELF_EXTRA_PHDRS select ARCH_BINFMT_ELF_STATE - select ARCH_CORRECT_STACKTRACE_ON_KRETPROBE select ARCH_ENABLE_HUGEPAGE_MIGRATION if HUGETLB_PAGE && MIGRATION select ARCH_ENABLE_MEMORY_HOTPLUG select ARCH_ENABLE_MEMORY_HOTREMOVE From 02a55f2743012a8089f09f6867220c3d57f16564 Mon Sep 17 00:00:00 2001 From: Chun-Tse Shao Date: Fri, 8 Nov 2024 05:08:05 +0000 Subject: [PATCH 2/3] perf/arm-smmuv3: Fix lockdep assert in ->event_init() Same as https://lore.kernel.org/all/20240514180050.182454-1-namhyung@kernel.org/, we should skip `for_each_sibling_event()` for group leader since it doesn't have the ctx yet. Fixes: f3c0eba28704 ("perf: Add a few assertions") Reported-by: Greg Thelen Cc: Namhyung Kim Cc: Robin Murphy Cc: Tuan Phan Signed-off-by: Chun-Tse Shao Acked-by: Will Deacon Link: https://lore.kernel.org/r/20241108050806.3730811-1-ctshao@google.com Signed-off-by: Catalin Marinas --- drivers/perf/arm_smmuv3_pmu.c | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) diff --git a/drivers/perf/arm_smmuv3_pmu.c b/drivers/perf/arm_smmuv3_pmu.c index b1510f660c7a6..621f02a7f43be 100644 --- a/drivers/perf/arm_smmuv3_pmu.c +++ b/drivers/perf/arm_smmuv3_pmu.c @@ -431,6 +431,17 @@ static int smmu_pmu_event_init(struct perf_event *event) return -EINVAL; } + /* + * Ensure all events are on the same cpu so all events are in the + * same cpu context, to avoid races on pmu_enable etc. + */ + event->cpu = smmu_pmu->on_cpu; + + hwc->idx = -1; + + if (event->group_leader == event) + return 0; + for_each_sibling_event(sibling, event->group_leader) { if (is_software_event(sibling)) continue; @@ -442,14 +453,6 @@ static int smmu_pmu_event_init(struct perf_event *event) return -EINVAL; } - hwc->idx = -1; - - /* - * Ensure all events are on the same cpu so all events are in the - * same cpu context, to avoid races on pmu_enable etc. - */ - event->cpu = smmu_pmu->on_cpu; - return 0; } From dfdf714fed559c09021df1d2a4bb64c0ad5f53bc Mon Sep 17 00:00:00 2001 From: Namhyung Kim Date: Wed, 20 Nov 2024 16:13:34 -0800 Subject: [PATCH 3/3] perf/arm-cmn: Ensure port and device id bits are set properly The portid_bits and deviceid_bits were set only for XP type nodes in the arm_cmn_discover() and it confused other nodes to find XP nodes. Copy the both bits from the XP nodes directly when it sets up a new node. Fixes: e79634b53e39 ("perf/arm-cmn: Refactor node ID handling. Again.") Signed-off-by: Namhyung Kim Acked-by: Will Deacon Reviewed-by: Robin Murphy Link: https://lore.kernel.org/r/20241121001334.331334-1-namhyung@kernel.org Signed-off-by: Catalin Marinas --- drivers/perf/arm-cmn.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c index 49bd811c6fd6e..b20fa600e510c 100644 --- a/drivers/perf/arm-cmn.c +++ b/drivers/perf/arm-cmn.c @@ -2178,8 +2178,6 @@ static int arm_cmn_init_dtcs(struct arm_cmn *cmn) continue; xp = arm_cmn_node_to_xp(cmn, dn); - dn->portid_bits = xp->portid_bits; - dn->deviceid_bits = xp->deviceid_bits; dn->dtc = xp->dtc; dn->dtm = xp->dtm; if (cmn->multi_dtm) @@ -2420,6 +2418,8 @@ static int arm_cmn_discover(struct arm_cmn *cmn, unsigned int rgn_offset) } arm_cmn_init_node_info(cmn, reg & CMN_CHILD_NODE_ADDR, dn); + dn->portid_bits = xp->portid_bits; + dn->deviceid_bits = xp->deviceid_bits; switch (dn->type) { case CMN_TYPE_DTC: