From: Dmitry Vyukov <dvyukov@google.com>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>,
Arnaldo Carvalho de Melo <acme@kernel.org>,
Mark Rutland <mark.rutland@arm.com>,
Alexander Shishkin <alexander.shishkin@linux.intel.com>,
Jiri Olsa <jolsa@redhat.com>, Namhyung Kim <namhyung@kernel.org>,
Will Deacon <will@kernel.org>,
LKML <linux-kernel@vger.kernel.org>,
Matt Morehouse <mascasa@google.com>
Subject: Re: Process-wide watchpoints
Date: Mon, 1 Feb 2021 09:50:20 +0100 [thread overview]
Message-ID: <CACT4Y+YeRtOTsMQ8xxZg-=nbv+yuJvYYhBErT46M8jtSHmiw6g@mail.gmail.com> (raw)
In-Reply-To: <CACT4Y+ZsKXfAxrzJGQc5mJ+QiP5sAw7zKWtciS+07qZzSf33mw@mail.gmail.com>
On Sun, Jan 31, 2021 at 11:28 AM Dmitry Vyukov <dvyukov@google.com> wrote:
>
> On Sun, Jan 31, 2021 at 11:04 AM Dmitry Vyukov <dvyukov@google.com> wrote:
> >
> > On Thu, Nov 12, 2020 at 11:43 AM Dmitry Vyukov <dvyukov@google.com> wrote:
> > > > > for sampling race detection),
> > > > > number of threads in the process can be up to, say, ~~10K and the
> > > > > watchpoint is intended to be set for a very brief period of time
> > > > > (~~few ms).
> > > >
> > > > Performance is a consideration here, doing lots of IPIs in such a short
> > > > window, on potentially large machines is a DoS risk.
> > > >
> > > > > This can be done today with both perf_event_open and ptrace.
> > > > > However, the problem is that both APIs work on a single thread level
> > > > > (? perf_event_open can be inherited by children, but not for existing
> > > > > siblings). So doing this would require iterating over, say, 10K
> > > >
> > > > One way would be to create the event before the process starts spawning
> > > > threads and keeping it disabled. Then every thread will inherit it, but
> > > > it'll be inactive.
> > > >
> > > > > I see at least one potential problem: what do we do if some sibling
> > > > > thread already has all 4 watchpoints consumed?
> > > >
> > > > That would be immediately avoided by this, since it will have the
> > > > watchpoint reserved per inheriting the event.
> > > >
> > > > Then you can do ioctl(PERF_EVENT_IOC_{MODIFY_ATTRIBUTES,ENABLE,DISABLE})
> > > > to update the watch location and enable/disable it. This _will_ indeed
> > > > result in a shitload of IPIs if the threads are active, but it should
> > > > work.
> > >
> > > Aha! That's the possibility I missed.
> > > We will try to prototype this and get back with more questions if/when
> > > we have them.
> > > Thanks!
> >
> > Hi Peter,
> >
> > I've tested this approach and it works, but only in half.
> > PERF_EVENT_IOC_{ENABLE,DISABLE} work as advertised.
> > However, PERF_EVENT_IOC_MODIFY_ATTRIBUTES does not work for inherited
> > child events.
> > Does something like this make any sense to you? Are you willing to
> > accept such change?
> >
> > diff --git a/kernel/events/core.c b/kernel/events/core.c
> > index 55d18791a72d..f6974807a32c 100644
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -3174,7 +3174,7 @@ int perf_event_refresh(struct perf_event *event,
> > int refresh)
> > }
> > EXPORT_SYMBOL_GPL(perf_event_refresh);
> >
> > -static int perf_event_modify_breakpoint(struct perf_event *bp,
> > +static int _perf_event_modify_breakpoint(struct perf_event *bp,
> > struct perf_event_attr *attr)
> > {
> > int err;
> > @@ -3189,6 +3189,28 @@ static int perf_event_modify_breakpoint(struct
> > perf_event *bp,
> > return err;
> > }
> >
> > +static int perf_event_modify_breakpoint(struct perf_event *bp,
> > + struct perf_event_attr *attr)
> > +{
> > + struct perf_event *child;
> > + int err;
> > +
> > + WARN_ON_ONCE(bp->ctx->parent_ctx);
> > +
> > + mutex_lock(&bp->child_mutex);
> > + err = _perf_event_modify_breakpoint(bp, attr);
> > + if (err)
> > + goto unlock;
> > + list_for_each_entry(child, &bp->child_list, child_list) {
> > + err = _perf_event_modify_breakpoint(child, attr);
> > + if (err)
> > + goto unlock;
> > + }
> > +unlock:
> > + mutex_unlock(&bp->child_mutex);
> > + return err;
> > +}
> > +
> > static int perf_event_modify_attr(struct perf_event *event,
> > struct perf_event_attr *attr)
>
>
> Not directly related to the above question, but related to my use case.
> Could we extend bpf_perf_event_data with some more data re breakpoint events?
>
> struct bpf_perf_event_data {
> bpf_user_pt_regs_t regs;
> __u64 sample_period;
> __u64 addr;
> };
>
> Ideally, I would like to have an actual access address, size and
> read/write type (may not match bp addr/size). Is that info easily
> available at the point of bpf hook call?
> Or, if that's not available at least breakpoint bp_type/bp_size.
>
> Is it correct that we can materialize in bpf_perf_event_data anything
> that's available in bpf_perf_event_data_kern (if it makes sense in the
> public interface of course)?
>
> struct bpf_perf_event_data_kern {
> bpf_user_pt_regs_t *regs;
> struct perf_sample_data *data;
> struct perf_event *event;
> };
>
> Unfortunately I don't see perf_event_attr.bp_type/bp_size
> stored/accessible anywhere in bpf_perf_event_data_kern. What would be
> the right way to expose them in bpf_perf_event_data?
Or, alternatively would it be reasonable for perf to generate SIGTRAP
directly on watchpoint hit (like ptrace does)? That's what I am
ultimately trying to do by attaching a bpf program.
next prev parent reply other threads:[~2021-02-01 8:51 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-12 7:46 Process-wide watchpoints Dmitry Vyukov
2020-11-12 10:31 ` Peter Zijlstra
2020-11-12 10:43 ` Dmitry Vyukov
[not found] ` <CACT4Y+bW1gpv8bz0vswaVUt-OB07oJ3NBeTi+vchAe8TTWK+mg@mail.gmail.com>
[not found] ` <CACT4Y+ZsKXfAxrzJGQc5mJ+QiP5sAw7zKWtciS+07qZzSf33mw@mail.gmail.com>
2021-02-01 8:50 ` Dmitry Vyukov [this message]
2021-02-03 12:29 ` Peter Zijlstra
2021-02-03 12:49 ` Dmitry Vyukov
2021-02-03 12:50 ` Dmitry Vyukov
2021-02-03 13:37 ` Peter Zijlstra
2021-02-04 8:10 ` Dmitry Vyukov
2021-02-04 9:38 ` Peter Zijlstra
2021-02-04 9:54 ` Dmitry Vyukov
2021-02-04 12:09 ` Peter Zijlstra
2021-02-04 12:53 ` Dmitry Vyukov
2021-02-04 13:10 ` Peter Zijlstra
2021-02-04 13:35 ` Dmitry Vyukov
2021-02-04 13:45 ` Peter Zijlstra
2021-02-04 14:59 ` Peter Zijlstra
2021-02-04 13:33 ` Peter Zijlstra
2021-02-04 13:37 ` Dmitry Vyukov
2021-02-04 13:06 ` Peter Zijlstra
2021-02-03 5:38 ` Namhyung Kim
2021-02-04 8:10 ` Dmitry Vyukov
2021-02-03 12:22 ` Peter Zijlstra
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CACT4Y+YeRtOTsMQ8xxZg-=nbv+yuJvYYhBErT46M8jtSHmiw6g@mail.gmail.com' \
--to=dvyukov@google.com \
--cc=acme@kernel.org \
--cc=alexander.shishkin@linux.intel.com \
--cc=jolsa@redhat.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=mascasa@google.com \
--cc=mingo@redhat.com \
--cc=namhyung@kernel.org \
--cc=peterz@infradead.org \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).