linux-kernel - Re: [PATCH v2 0/3] arm64: perf_event: Fix time offset prior to epoch

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20200512091918.GH2978@hirez.programming.kicks-ass.net>
Date:   Tue, 12 May 2020 11:19:18 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Leo Yan <leo.yan@...aro.org>
Cc:     Will Deacon <will@...nel.org>, Marc Zyngier <maz@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        Ingo Molnar <mingo@...hat.com>,
        Arnaldo Carvalho de Melo <acme@...nel.org>,
        Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
        Jiri Olsa <jolsa@...hat.com>,
        Namhyung Kim <namhyung@...nel.org>,
        Catalin Marinas <catalin.marinas@....com>,
        Daniel Lezcano <daniel.lezcano@...aro.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        Allison Randal <allison@...utok.net>,
        Alexios Zavras <alexios.zavras@...el.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Kate Stewart <kstewart@...uxfoundation.org>,
        Enrico Weigelt <info@...ux.net>,
        "Ahmed S. Darwish" <a.darwish@...utronix.de>,
        Paul Cercueil <paul@...pouillou.net>,
        "Ben Dooks (Codethink)" <ben.dooks@...ethink.co.uk>,
        linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH v2 0/3] arm64: perf_event: Fix time offset prior to epoch

On Tue, May 12, 2020 at 02:38:12PM +0800, Leo Yan wrote:
> @@ -1165,28 +1166,26 @@ device_initcall(armv8_pmu_driver_init)
>  void arch_perf_update_userpage(struct perf_event *event,
>  			       struct perf_event_mmap_page *userpg, u64 now)
>  {
> +	struct clock_read_data *rd;
> +	unsigned int seq;
>  
>  	/*
>  	 * Internal timekeeping for enabled/running/stopped times
>  	 * is always computed with the sched_clock.
>  	 */
>  	userpg->cap_user_time = 1;
> +	userpg->cap_user_time_zero = 1;
>  
> +	do {
> +		rd = sched_clock_read_begin(&seq);
> +
> +		userpg->time_mult = rd->mult;
> +		userpg->time_shift = rd->shift;
> +		userpg->time_zero = rd->epoch_ns;
> +
> +		userpg->time_zero -= (rd->epoch_cyc * rd->mult) >> rd->shift;

Damn, I think this is broken vs the counter wrapping.

So what the sched_clock code does is:

	cyc_to_ns((cyc - rd->epoch_cyc) & rd->sched_clock_mask, rd->mult, rd->shift)

But because the perf interface assumes a simple linear relation, we
can't express that properly.

Now, your arm64 counter is 56 bits, so wrapping is rare, but still, we
should probably fix that. And that probably needs an ABI extention
*sigh*.

> +
> +	} while (sched_clock_read_retry(seq));
> +
> +	userpg->time_offset = userpg->time_zero - now;
>  }