Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
In-Reply-To: <20180126140033.ix2n7j62k6iccmdh@oak.lan>
References: <b3c1e952cfff95e0289772d6af5ec2ebd6cc3141.1516935110.git.baolin.wang@linaro.org>
 <CAK8P3a2eHfepGXrWAVSwYQZX-vFS84SiKXCRuGdw+QTbproZhg@mail.gmail.com> <20180126140033.ix2n7j62k6iccmdh@oak.lan>
From:   Baolin Wang <baolin.wang@linaro.org>
Date:   Fri, 26 Jan 2018 22:20:51 +0800
Message-ID: <CAMz4kuK_mQViv4XuuLmyuDjKMaVmO_1Sx4o9NTYHhrGBB3B3SQ@mail.gmail.com>
Subject: Re: [PATCH] kdb: use ktime_get_seconds() instead of ktime_get_ts()
To:     Daniel Thompson <daniel.thompson@linaro.org>
Cc:     Arnd Bergmann <arnd@arndb.de>,
        Jason Wessel <jason.wessel@windriver.com>,
        Ingo Molnar <mingo@kernel.org>,
        Mark Brown <broonie@kernel.org>,
        kgdb-bugreport@lists.sourceforge.net,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 26 January 2018 at 22:00, Daniel Thompson <daniel.thompson@linaro.org> wrote:
> On Fri, Jan 26, 2018 at 10:21:58AM +0100, Arnd Bergmann wrote:
>> On Fri, Jan 26, 2018 at 4:03 AM, Baolin Wang <baolin.wang@linaro.org> wrote:
>> > The kdb code will print the monotonic time by ktime_get_ts(), but
>> > the ktime_get_ts() will be protected by a sequence lock, that will
>> > introduce one deadlock risk if the lock was already held in the
>> > context from which we entered the debugger.
>> >
>> > Since kdb is only interested in the second field, we can use the
>> > ktime_get_seconds() to get the monotonic time without a lock,
>> > moreover we can remove the 'struct timespec', which is not y2038
>> > safe.
>> >
>> > Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
>> > ---
>> >  kernel/debug/kdb/kdb_main.c |    4 +---
>> >  1 file changed, 1 insertion(+), 3 deletions(-)
>> >
>> > diff --git a/kernel/debug/kdb/kdb_main.c b/kernel/debug/kdb/kdb_main.c
>> > index 69e70f4..f0fc6f7 100644
>> > --- a/kernel/debug/kdb/kdb_main.c
>> > +++ b/kernel/debug/kdb/kdb_main.c
>> > @@ -2486,10 +2486,8 @@ static int kdb_kill(int argc, const char **argv)
>> >   */
>> >  static void kdb_sysinfo(struct sysinfo *val)
>> >  {
>> > -       struct timespec uptime;
>> > -       ktime_get_ts(&uptime);
>> >         memset(val, 0, sizeof(*val));
>> > -       val->uptime = uptime.tv_sec;
>> > +       val->uptime = ktime_get_seconds();
>> >         val->loads[0] = avenrun[0];
>> >         val->loads[1] = avenrun[1];
>> >         val->loads[2] = avenrun[2];
>>
>> Using ktime_get_seconds() avoids locking problems, but I wonder what
>> would happen if we trigger the 'WARN_ON(timekeeping_suspended)'
>> from kdb. Is that a problem? If it is, we have to use ktime_get_mono_fast_ns()
>> and div_u64() instead.
>
> Normally a WARN_ON() doesn't triggered a kgdb_breakpoint() so (apart
> from bugs) we can start executing the warning. Unfortunately
> kdb_trap_printk isn't set when we call ktime_get_seconds() so printing
> the warning isn't safe.
>
> If we had no choice of time function we could work around by
> enabling printk() trapping for the call but since ktime_get_mono_fast_ns()
> already exists its probably best just to use that.
>

If timekeeping_suspended is set, which means the system had been in
suspend state. So now we still need debugger the system? But cores
were already powered down.

The ktime_get_mono_fast_ns() will access the the clocksource driver,
if the timekeeping is suspended following system suspend and the
clocksource is not SUSPEND_NONSTOP, we may meet some unexpected issue
to access the timer's register without clock. So I am not sure if
ktime_get_mono_fast_ns() can work well for this case.

-- 
Baolin.wang
Best Regards