[v2] fix for #285

[PATCH v2 0/2] fix for #285

Posted by Emilio Cota 1 year, 2 months ago

Changes since v1:

- Add configure check to only use QTree if Glib still implements gslice.
  If Glib doesn't, then we call Glib directly with inline functions.
- Add TODO's so that in the future (i.e. when the minimum version of
  Glib that we use doesn't implement gslice) we remove QTree.
- Add comment to the top of qtree.h.
- Make qtree-bench results more robust by running longer or more times.
- Drop deprecated API calls (they're unused in QEMU).
- Drop API calls that are too recent (they're unused in QEMU).
- Drop macro benchmark results from the TCG patch since they're too noisy.
- Add test program to the commit log so that we don't lose it in the future
  even if the bug tracker goes away.

Thanks,
		Emilio

Re: [PATCH v2 0/2] fix for #285

Posted by Emilio Cota 1 year, 2 months ago

Ping.

This fixes a bug (admittedly with a big hammer) that affects
users with heavily multi-threaded user-mode workloads.

Thanks,
		Emilio

On Sun, Feb 05, 2023 at 11:37:56 -0500, Emilio Cota wrote:
> Changes since v1:
> 
> - Add configure check to only use QTree if Glib still implements gslice.
>   If Glib doesn't, then we call Glib directly with inline functions.
> - Add TODO's so that in the future (i.e. when the minimum version of
>   Glib that we use doesn't implement gslice) we remove QTree.
> - Add comment to the top of qtree.h.
> - Make qtree-bench results more robust by running longer or more times.
> - Drop deprecated API calls (they're unused in QEMU).
> - Drop API calls that are too recent (they're unused in QEMU).
> - Drop macro benchmark results from the TCG patch since they're too noisy.
> - Add test program to the commit log so that we don't lose it in the future
>   even if the bug tracker goes away.
> 
> Thanks,
> 		Emilio
> 
>

Re: [PATCH v2 0/2] fix for #285

Posted by Emilio Cota 1 year, 1 month ago

Ping. Any feedback on these two patches?

https://patchew.org/QEMU/20230205163758.416992-1-cota@braap.org/
https://lore.kernel.org/qemu-devel/20230205163758.416992-1-cota@braap.org/

Happy to resend if needed.

Thanks,
		Emilio


On Fri, Feb 17, 2023 at 07:44:38 -0500, Emilio Cota wrote:
> Ping.
> 
> This fixes a bug (admittedly with a big hammer) that affects
> users with heavily multi-threaded user-mode workloads.
> 
> Thanks,
> 		Emilio
> 
> On Sun, Feb 05, 2023 at 11:37:56 -0500, Emilio Cota wrote:
> > Changes since v1:
> > 
> > - Add configure check to only use QTree if Glib still implements gslice.
> >   If Glib doesn't, then we call Glib directly with inline functions.
> > - Add TODO's so that in the future (i.e. when the minimum version of
> >   Glib that we use doesn't implement gslice) we remove QTree.
> > - Add comment to the top of qtree.h.
> > - Make qtree-bench results more robust by running longer or more times.
> > - Drop deprecated API calls (they're unused in QEMU).
> > - Drop API calls that are too recent (they're unused in QEMU).
> > - Drop macro benchmark results from the TCG patch since they're too noisy.
> > - Add test program to the commit log so that we don't lose it in the future
> >   even if the bug tracker goes away.
> > 
> > Thanks,
> > 		Emilio
> > 
> >

Re: [PATCH v2 0/2] fix for #285

Posted by Richard Henderson 1 year, 1 month ago

On 3/19/23 07:15, Emilio Cota wrote:
> Ping. Any feedback on these two patches?
> 
> https://patchew.org/QEMU/20230205163758.416992-1-cota@braap.org/
> https://lore.kernel.org/qemu-devel/20230205163758.416992-1-cota@braap.org/

Queued to tcg-next.


r~

> 
> Happy to resend if needed.
> 
> Thanks,
> 		Emilio
> 
> 
> On Fri, Feb 17, 2023 at 07:44:38 -0500, Emilio Cota wrote:
>> Ping.
>>
>> This fixes a bug (admittedly with a big hammer) that affects
>> users with heavily multi-threaded user-mode workloads.
>>
>> Thanks,
>> 		Emilio
>>
>> On Sun, Feb 05, 2023 at 11:37:56 -0500, Emilio Cota wrote:
>>> Changes since v1:
>>>
>>> - Add configure check to only use QTree if Glib still implements gslice.
>>>    If Glib doesn't, then we call Glib directly with inline functions.
>>> - Add TODO's so that in the future (i.e. when the minimum version of
>>>    Glib that we use doesn't implement gslice) we remove QTree.
>>> - Add comment to the top of qtree.h.
>>> - Make qtree-bench results more robust by running longer or more times.
>>> - Drop deprecated API calls (they're unused in QEMU).
>>> - Drop API calls that are too recent (they're unused in QEMU).
>>> - Drop macro benchmark results from the TCG patch since they're too noisy.
>>> - Add test program to the commit log so that we don't lose it in the future
>>>    even if the bug tracker goes away.
>>>
>>> Thanks,
>>> 		Emilio
>>>
>>>

Re: [PATCH v2 0/2] fix for #285

Posted by Philippe Mathieu-Daudé 1 year, 1 month ago

Hi Emilio,

On 19/3/23 15:15, Emilio Cota wrote:
> Ping. Any feedback on these two patches?
> 
> https://patchew.org/QEMU/20230205163758.416992-1-cota@braap.org/
> https://lore.kernel.org/qemu-devel/20230205163758.416992-1-cota@braap.org/
> 
> Happy to resend if needed.

Since we are past hard-freeze, this series likely missed the v8.0
release IMO. Note that doesn't mean maintainers can't queue it and
send the pull request later when the next development window opens.

> On Fri, Feb 17, 2023 at 07:44:38 -0500, Emilio Cota wrote:
>> Ping.
>>
>> This fixes a bug (admittedly with a big hammer) that affects
>> users with heavily multi-threaded user-mode workloads.
>>
>> Thanks,
>> 		Emilio
>>
>> On Sun, Feb 05, 2023 at 11:37:56 -0500, Emilio Cota wrote:
>>> Changes since v1:
>>>
>>> - Add configure check to only use QTree if Glib still implements gslice.
>>>    If Glib doesn't, then we call Glib directly with inline functions.
>>> - Add TODO's so that in the future (i.e. when the minimum version of
>>>    Glib that we use doesn't implement gslice) we remove QTree.
>>> - Add comment to the top of qtree.h.
>>> - Make qtree-bench results more robust by running longer or more times.
>>> - Drop deprecated API calls (they're unused in QEMU).
>>> - Drop API calls that are too recent (they're unused in QEMU).
>>> - Drop macro benchmark results from the TCG patch since they're too noisy.
>>> - Add test program to the commit log so that we don't lose it in the future
>>>    even if the bug tracker goes away.
>>>
>>> Thanks,
>>> 		Emilio
>>>
>>>
>

Re: [PATCH v2 0/2] fix for #285

Posted by Daniel P. Berrangé 1 year, 1 month ago

On Mon, Mar 20, 2023 at 04:05:01PM +0100, Philippe Mathieu-Daudé wrote:
> Hi Emilio,
> 
> On 19/3/23 15:15, Emilio Cota wrote:
> > Ping. Any feedback on these two patches?
> > 
> > https://patchew.org/QEMU/20230205163758.416992-1-cota@braap.org/
> > https://lore.kernel.org/qemu-devel/20230205163758.416992-1-cota@braap.org/
> > 
> > Happy to resend if needed.
> 
> Since we are past hard-freeze, this series likely missed the v8.0
> release IMO. Note that doesn't mean maintainers can't queue it and
> send the pull request later when the next development window opens.

This series is a bug fix though, the freeze only applies to
features. For rc0/1 most bug fixes are still permissible.

I'd really recommend we take this series, as it is a clear
bug in linux-user that can be fairly easily reproduced

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|