Re: Linux 4.15-rc2: Regression in resume from ACPI S3

From: Andy Lutomirski <luto@kernel.org>
To: Zhang Rui <rui.zhang@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Pavel Machek <pavel@ucw.cz>, Andrew Lutomirski <luto@kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Jarkko Nikula <jarkko.nikula@linux.intel.com>,
	"Rafael J. Wysocki" <rjw@rjwysocki.net>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	"the arch/x86 maintainers" <x86@kernel.org>
Subject: Re: Linux 4.15-rc2: Regression in resume from ACPI S3
Date: Mon, 11 Dec 2017 08:28:28 -0800	[thread overview]
Message-ID: <CALCETrVPDNc1FHzY5AR1r5mk0pRU_vr5_Mmgi0Y6Y-0H2LLAvw@mail.gmail.com> (raw)
In-Reply-To: <1513001367.2981.11.camel@intel.com>

On Mon, Dec 11, 2017 at 6:09 AM, Zhang Rui <rui.zhang@intel.com> wrote:
> On Sun, 2017-12-10 at 12:30 -0800, Linus Torvalds wrote:
>> On Sun, Dec 10, 2017 at 10:56 AM, Pavel Machek <pavel@ucw.cz> wrote:
>> >
>> >
>> > Confirmed, revert fixes it. You see how it moves
>> > fix_processor_context
>> > around #ifdef CONFIG_X86_32 block? And how people forget 32-bit
>> > machines exist? Aha.
>> Yeah, people do.
>>
>> Andy?
>>
>> >
>> > Which brings me to .. various people do automated testing of
>> > kernel. Testing 32-bit kernel for boot, and both 32-bit and 64-bit
>> > for
>> > boot and suspend would be very nice. The last item is not hard,
>> > either:
>> >
>> > sudo rtcwake -l -m mem -s 5
>> >
>> > ...should take 10 seconds or so.
>> I'm told 0day does *some* suspend/resume testing, but I think it's
>> pretty limited, partly because the kinds of machines it primarily
>> works on don't really support suspend/resume at all.
>
> currently, we're running suspend test on 1 platform only, with 64 bit
> kernel. suspend test will be enabled on more platforms (laptops) in
> next two weeks.
>
> I will check why it does not find the first regression introduced by
> ca37e57bbe0c ("x86/entry/64: Add missing irqflags tracing to
> native_load_gs_index()").
>
>>  I'm also not sure
>> just how many of those machines are 32-bit at all..
>
> for this, I suppose it can be reproduced if we use 32-bit kernel and
> rootfs, right? Then it's easier to enable this in 0Day.
>

Yes.

The 64-bit problem should also be reproducible with rtcwake even in a vm.

Also, on this topic, could make run_tests in
tools/testing/selftests/x86 be added to the rotation as well?  The
testing dir should match the kernel being tested IMO.

> thanks,
> rui
>>
>> But I'm adding Zhang Rui to the cc, to see if my recollection is
>> right.
>>
>> Because you're right, more suspend/resume automated testing would be
>> good to have. And yes, people test mainly 64-bit these days.
>>
>> Also, I'm not even sure what the 0day rules are for just plain
>> mainline. I don't tend to see a lot of breakage reports, even though
>> I'd expect to. This came in from the x86 trees (and those do their
>> own
>> tests too, but probably not suspend/resume either), but it hit my
>> tree
>> fairly soon after going into the x86 -tip trees.
>>
>>             Linus