Re: [Xen-devel] Ping: [PATCH 0/2] x86/Xen/32: xen_iret_crit_fixup adjustments

From: Jan Beulich
Date: Wed Nov 20 2019 - 02:19:00 EST


On 20.11.2019 03:39, Boris Ostrovsky wrote:
> On 11/19/19 9:17 PM, Boris Ostrovsky wrote:
>> On 11/19/19 12:50 PM, Boris Ostrovsky wrote:
>>> On 11/19/19 7:58 AM, Jan Beulich wrote:
>>>> On 11.11.2019 15:30, Jan Beulich wrote:
>>>>> The first patch here fixes another regression from 3c88c692c287
>>>>> ("x86/stackframe/32: Provide consistent pt_regs"), besides the
>>>>> one already addressed by
>>>>> https://lists.xenproject.org/archives/html/xen-devel/2019-10/msg01988.html.
>>>>> The second patch is a minimal bit of cleanup on top.
>>>>>
>>>>> 1: make xen_iret_crit_fixup independent of frame layout
>>>>> 2: simplify xen_iret_crit_fixup's ring check
>>>> Seeing that the other regression fix has been taken into -tip,
>>>> what is the situation here? Should 5.4 really ship with this
>>>> still unfixed?
>>> I am still unable to boot a 32-bit guest with those patches, crashing in
>>> int3_exception_notify with regs->sp zero.
>>>
>>> When I revert to 3c88c692c287 the guest actually boots so my (?) problem
>>> was introduced somewhere in-between.
>> Nevermind this. I didn't read your patches correctly.
>
> BTW, I'd rather this not go into 5.4 this late. 3c88c692c287 has been
> there since 5.2 and noone complained.

Afaict the issues were introduced in 5.3, and my first patch (including
a note [complaint if you will] of the second issue) was sent around
5.4-rc2. This has been blocking osstest's linux-linus forever since, so
even without my mail everyone could have been aware by paying attention
to the flight reports (the bisection ones, unfortunately, are pretty
useless here, as in cases like this one they seem to tend to point at
huge merge commits).

Jan