Commit bfe33492 authored by Peter Zijlstra's avatar Peter Zijlstra Committed by Ingo Molnar
Browse files

perf/x86: Fix RDPMC vs. mm_struct tracking 



Vince reported the following rdpmc() testcase failure:

 > Failing test case:
 >
 >	fd=perf_event_open();
 >	addr=mmap(fd);
 >	exec()  // without closing or unmapping the event
 >	fd=perf_event_open();
 >	addr=mmap(fd);
 >	rdpmc()	// GPFs due to rdpmc being disabled

The problem is of course that exec() plays tricks with what is
current->mm, only destroying the old mappings after having
installed the new mm.

Fix this confusion by passing along vma->vm_mm instead of relying on
current->mm.

Reported-by: default avatarVince Weaver <vincent.weaver@maine.edu>
Tested-by: default avatarVince Weaver <vincent.weaver@maine.edu>
Signed-off-by: default avatarPeter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: default avatarAndy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Fixes: 1e0fb9ec ("perf: Add pmu callbacks to track event mapping and unmapping")
Link: http://lkml.kernel.org/r/20170802173930.cstykcqefmqt7jau@hirez.programming.kicks-ass.net


[ Minor cleanups. ]
Signed-off-by: default avatarIngo Molnar <mingo@kernel.org>
parent 8d31f80e

arch/x86/events/core.c

+7 −9
Original line number Diff line number Diff line
@@ -2114,7 +2114,7 @@ static void refresh_pce(void *ignored) 
	load_mm_cr4(this_cpu_read(cpu_tlbstate.loaded_mm));

}



static void x86_pmu_event_mapped(struct perf_event *event)

static void x86_pmu_event_mapped(struct perf_event *event, struct mm_struct *mm)

{

	if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))

		return;
@@ -2129,22 +2129,20 @@ static void x86_pmu_event_mapped(struct perf_event *event) 
	 * For now, this can't happen because all callers hold mmap_sem

	 * for write.  If this changes, we'll need a different solution.

	 */

	lockdep_assert_held_exclusive(&current->mm->mmap_sem);

	lockdep_assert_held_exclusive(&mm->mmap_sem);



	if (atomic_inc_return(&current->mm->context.perf_rdpmc_allowed) == 1)

		on_each_cpu_mask(mm_cpumask(current->mm), refresh_pce, NULL, 1);

	if (atomic_inc_return(&mm->context.perf_rdpmc_allowed) == 1)

		on_each_cpu_mask(mm_cpumask(mm), refresh_pce, NULL, 1);

}



static void x86_pmu_event_unmapped(struct perf_event *event)

static void x86_pmu_event_unmapped(struct perf_event *event, struct mm_struct *mm)

{

	if (!current->mm)

		return;



	if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))

		return;



	if (atomic_dec_and_test(&current->mm->context.perf_rdpmc_allowed))

		on_each_cpu_mask(mm_cpumask(current->mm), refresh_pce, NULL, 1);

	if (atomic_dec_and_test(&mm->context.perf_rdpmc_allowed))

		on_each_cpu_mask(mm_cpumask(mm), refresh_pce, NULL, 1);

}



static int x86_pmu_event_idx(struct perf_event *event)

include/linux/perf_event.h

+2 −2
Original line number Diff line number Diff line
@@ -310,8 +310,8 @@ struct pmu { 
	 * Notification that the event was mapped or unmapped.  Called

	 * in the context of the mapping task.

	 */

	void (*event_mapped)		(struct perf_event *event); /*optional*/

	void (*event_unmapped)		(struct perf_event *event); /*optional*/

	void (*event_mapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */

	void (*event_unmapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */



	/*

	 * Flags for ->add()/->del()/ ->start()/->stop(). There are

kernel/events/core.c

+3 −3
Original line number Diff line number Diff line
@@ -5090,7 +5090,7 @@ static void perf_mmap_open(struct vm_area_struct *vma) 
		atomic_inc(&event->rb->aux_mmap_count);



	if (event->pmu->event_mapped)

		event->pmu->event_mapped(event);

		event->pmu->event_mapped(event, vma->vm_mm);

}



static void perf_pmu_output_stop(struct perf_event *event);
@@ -5113,7 +5113,7 @@ static void perf_mmap_close(struct vm_area_struct *vma) 
	unsigned long size = perf_data_size(rb);



	if (event->pmu->event_unmapped)

		event->pmu->event_unmapped(event);

		event->pmu->event_unmapped(event, vma->vm_mm);



	/*

	 * rb->aux_mmap_count will always drop before rb->mmap_count and
@@ -5411,7 +5411,7 @@ aux_unlock: 
	vma->vm_ops = &perf_mmap_vmops;



	if (event->pmu->event_mapped)

		event->pmu->event_mapped(event);

		event->pmu->event_mapped(event, vma->vm_mm);



	return ret;

}

CRA Git | Maintained and supported by SUSTech CRA and CCSE