| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAL_JsqKN4YcCpL9uiOqea4CcqRWgU7Af=V8JNMjypivaVHq4sQ@mail.gmail.com>
Date: Wed, 12 May 2021 09:54:22 -0500
From: Rob Herring <robh@...nel.org>
To: "Liang, Kan" <kan.liang@...ux.intel.com>
Cc: Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Andi Kleen <ak@...ux.intel.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Andy Lutomirski <luto@...capital.net>,
Stephane Eranian <eranian@...gle.com>,
Namhyung Kim <namhyung@...nel.org>
Subject: Re: [PATCH V6] perf: Reset the dirty counter to prevent the leak for
an RDPMC task
On Tue, May 11, 2021 at 4:43 PM Liang, Kan <kan.liang@...ux.intel.com> wrote:
>
>
>
> On 5/11/2021 4:39 PM, Rob Herring wrote:
> > On Tue, May 11, 2021 at 12:59 PM Liang, Kan <kan.liang@...ux.intel.com> wrote:
> >>
> >>
> >>
> >> On 5/10/2021 4:29 PM, Rob Herring wrote:
> >>> On Mon, May 10, 2021 at 2:18 PM Peter Zijlstra <peterz@...radead.org> wrote:
> >>>>
> >>>> On Thu, Apr 22, 2021 at 11:25:52AM -0700, kan.liang@...ux.intel.com wrote:
> >>>>
> >>>>> - Add a new method check_leakage() to check and clear dirty counters
> >>>>> to prevent potential leakage.
> >>>>
> >>>> I really dislike adding spurious callbacks, also because indirect calls
> >>>> are teh suck, but also because it pollutes the interface so.
> >>>>
> >>>> That said, I'm not sure I actually like the below any better :/
> >>>>
> >>
> >> Maybe we can add a atomic variable to track the number of
> >> event_mapped(). Only invoke sched_task() when the number > 0.
> >
> > Except that it is only needed when mapped and user access is allowed/enabled.
> >
> >>
> >> It looks like only X86 implements the event_mapped(). So it should not
> >> impact other ARCHs.
> >
> > Arm will have one if we ever settle on the implementation.
> >
> >>
> >> What do you think?
> >>
> >> diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
> >> index c6fedd2..ae5b0e7 100644
> >> --- a/arch/x86/events/core.c
> >> +++ b/arch/x86/events/core.c
> >> @@ -1636,6 +1636,8 @@ static void x86_pmu_del(struct perf_event *event,
> >> int flags)
> >> if (cpuc->txn_flags & PERF_PMU_TXN_ADD)
> >> goto do_del;
> >>
> >> + __set_bit(event->hw.idx, cpuc->dirty);
> >> +
> >> /*
> >> * Not a TXN, therefore cleanup properly.
> >> */
> >> @@ -2484,6 +2486,31 @@ static int x86_pmu_event_init(struct perf_event
> >> *event)
> >> return err;
> >> }
> >>
> >> +static void x86_pmu_clear_dirty_counters(void)
> >> +{
> >> + struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> >> + int i;
> >> +
> >> + /* Don't need to clear the assigned counter. */
> >> + for (i = 0; i < cpuc->n_events; i++)
> >> + __clear_bit(cpuc->assign[i], cpuc->dirty);
> >> +
> >> + if (bitmap_empty(cpuc->dirty, X86_PMC_IDX_MAX))
> >> + return;
> >> +
> >> + for_each_set_bit(i, cpuc->dirty, X86_PMC_IDX_MAX) {
> >> + /* Metrics and fake events don't have corresponding HW counters. */
> >> + if (is_metric_idx(i) || (i == INTEL_PMC_IDX_FIXED_VLBR))
> >> + continue;
> >> + else if (i >= INTEL_PMC_IDX_FIXED)
> >> + wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + (i - INTEL_PMC_IDX_FIXED), 0);
> >> + else
> >> + wrmsrl(x86_pmu_event_addr(i), 0);
> >> + }
> >> +
> >> + bitmap_zero(cpuc->dirty, X86_PMC_IDX_MAX);
> >> +}
> >> +
> >> static void x86_pmu_event_mapped(struct perf_event *event, struct
> >> mm_struct *mm)
> >> {
> >> if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))
> >> @@ -2507,7 +2534,6 @@ static void x86_pmu_event_mapped(struct perf_event
> >> *event, struct mm_struct *mm)
> >>
> >> static void x86_pmu_event_unmapped(struct perf_event *event, struct
> >> mm_struct *mm)
> >> {
> >> -
> >> if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))
> >> return;
> >>
> >> @@ -2616,6 +2642,14 @@ static const struct attribute_group
> >> *x86_pmu_attr_groups[] = {
> >> static void x86_pmu_sched_task(struct perf_event_context *ctx, bool
> >> sched_in)
> >> {
> >> static_call_cond(x86_pmu_sched_task)(ctx, sched_in);
> >> +
> >> + /*
> >> + * If a new task has the RDPMC enabled, clear the dirty counters
> >> + * to prevent the potential leak.
> >> + */
> >> + if (sched_in && ctx && READ_ONCE(x86_pmu.attr_rdpmc) &&
> >> + current->mm && atomic_read(¤t->mm->context.perf_rdpmc_allowed))
> >> + x86_pmu_clear_dirty_counters();
> >> }
> >>
> >> static void x86_pmu_swap_task_ctx(struct perf_event_context *prev,
> >> diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
> >> index 10c8171..55bd891 100644
> >> --- a/arch/x86/events/perf_event.h
> >> +++ b/arch/x86/events/perf_event.h
> >> @@ -229,6 +229,7 @@ struct cpu_hw_events {
> >> */
> >> struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
> >> unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
> >> + unsigned long dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
> >> int enabled;
> >>
> >> int n_events; /* the # of events in the below arrays */
> >> diff --git a/kernel/events/core.c b/kernel/events/core.c
> >> index 1574b70..ef8f6f4 100644
> >> --- a/kernel/events/core.c
> >> +++ b/kernel/events/core.c
> >> @@ -384,6 +384,7 @@ DEFINE_STATIC_KEY_FALSE(perf_sched_events);
> >> static DECLARE_DELAYED_WORK(perf_sched_work, perf_sched_delayed);
> >> static DEFINE_MUTEX(perf_sched_mutex);
> >> static atomic_t perf_sched_count;
> >> +static atomic_t perf_event_mmap_count;
> >
> > A global count is not going to work. I think it needs to be per PMU at
> > least. In the case of Arm big.LITTLE, user access is constrained to
> > one subset of cores which is 1 PMU instance.
> >
>
> How about this one?
Would you mind splitting this to core and x86 parts.
>
> diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
> index c6fedd2..9052578 100644
> --- a/arch/x86/events/core.c
> +++ b/arch/x86/events/core.c
> @@ -1636,6 +1636,8 @@ static void x86_pmu_del(struct perf_event *event,
> int flags)
> if (cpuc->txn_flags & PERF_PMU_TXN_ADD)
> goto do_del;
>
> + __set_bit(event->hw.idx, cpuc->dirty);
> +
> /*
> * Not a TXN, therefore cleanup properly.
> */
> @@ -2484,12 +2486,43 @@ static int x86_pmu_event_init(struct perf_event
> *event)
> return err;
> }
>
> +static void x86_pmu_clear_dirty_counters(void)
> +{
> + struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> + int i;
> +
> + /* Don't need to clear the assigned counter. */
> + for (i = 0; i < cpuc->n_events; i++)
> + __clear_bit(cpuc->assign[i], cpuc->dirty);
> +
> + if (bitmap_empty(cpuc->dirty, X86_PMC_IDX_MAX))
> + return;
> +
> + for_each_set_bit(i, cpuc->dirty, X86_PMC_IDX_MAX) {
> + /* Metrics and fake events don't have corresponding HW counters. */
> + if (is_metric_idx(i) || (i == INTEL_PMC_IDX_FIXED_VLBR))
> + continue;
> + else if (i >= INTEL_PMC_IDX_FIXED)
> + wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + (i - INTEL_PMC_IDX_FIXED), 0);
> + else
> + wrmsrl(x86_pmu_event_addr(i), 0);
> + }
> +
> + bitmap_zero(cpuc->dirty, X86_PMC_IDX_MAX);
> +}
> +
> static void x86_pmu_event_mapped(struct perf_event *event, struct
> mm_struct *mm)
> {
> if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))
> return;
>
> /*
> + * Enable sched_task() for the RDPMC task.
> + */
> + if (x86_pmu.sched_task && event->hw.target)
> + atomic_inc(&event->pmu->sched_cb_usages);
> +
> + /*
> * This function relies on not being called concurrently in two
> * tasks in the same mm. Otherwise one task could observe
> * perf_rdpmc_allowed > 1 and return all the way back to
> @@ -2507,10 +2540,12 @@ static void x86_pmu_event_mapped(struct
> perf_event *event, struct mm_struct *mm)
>
> static void x86_pmu_event_unmapped(struct perf_event *event, struct
> mm_struct *mm)
> {
> -
> if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))
> return;
>
> + if (x86_pmu.sched_task && event->hw.target)
> + atomic_dec(&event->pmu->sched_cb_usages);
> +
> if (atomic_dec_and_test(&mm->context.perf_rdpmc_allowed))
> on_each_cpu_mask(mm_cpumask(mm), cr4_update_pce, NULL, 1);
> }
> @@ -2616,6 +2651,14 @@ static const struct attribute_group
> *x86_pmu_attr_groups[] = {
> static void x86_pmu_sched_task(struct perf_event_context *ctx, bool
> sched_in)
> {
> static_call_cond(x86_pmu_sched_task)(ctx, sched_in);
> +
> + /*
> + * If a new task has the RDPMC enabled, clear the dirty counters
> + * to prevent the potential leak.
> + */
> + if (sched_in && ctx && READ_ONCE(x86_pmu.attr_rdpmc) &&
> + current->mm && atomic_read(¤t->mm->context.perf_rdpmc_allowed))
> + x86_pmu_clear_dirty_counters();
> }
>
> static void x86_pmu_swap_task_ctx(struct perf_event_context *prev,
> diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
> index 10c8171..55bd891 100644
> --- a/arch/x86/events/perf_event.h
> +++ b/arch/x86/events/perf_event.h
> @@ -229,6 +229,7 @@ struct cpu_hw_events {
> */
> struct perf_event *events[X86_PMC_IDX_MAX]; /* in counter order */
> unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
> + unsigned long dirty[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
> int enabled;
>
> int n_events; /* the # of events in the below arrays */
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index c8a3388..3a85dbe 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -301,6 +301,9 @@ struct pmu {
> /* number of address filters this PMU can do */
> unsigned int nr_addr_filters;
>
> + /* Track the per PMU sched_task() callback users */
> + atomic_t sched_cb_usages;
To align with the per cpu one: s/usages/usage/
I think we should be able to use refcount_t here instead?
> +
> /*
> * Fully disable/enable this PMU, can be used to protect from the PMI
> * as well as for lazy/batch writing of the MSRs.
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 1574b70..8216acc 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -3851,7 +3851,7 @@ static void perf_event_context_sched_in(struct
> perf_event_context *ctx,
> cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
> perf_event_sched_in(cpuctx, ctx, task);
>
> - if (cpuctx->sched_cb_usage && pmu->sched_task)
> + if (pmu->sched_task && (cpuctx->sched_cb_usage ||
> atomic_read(&pmu->sched_cb_usages)))
For completeness, shouldn't this condition be added everywhere
->sched_task() can be called perhaps with the exception of
__perf_pmu_sched_task() which is only called when the task context
doesn't change.
> pmu->sched_task(cpuctx->task_ctx, true);
>
> perf_pmu_enable(pmu);
>
> Thanks,
> Kan
Powered by blists - more mailing lists