Re: [PATCH 5.15 000/245] 5.15.150-rc1 review

[Guenter Roeck]

On 2/29/24 10:42, Greg Kroah-Hartman wrote:
> On Thu, Feb 29, 2024 at 10:15:49AM -0800, Guenter Roeck wrote:
> > On 2/27/24 05:23, Greg Kroah-Hartman wrote:
> > > This is the start of the stable review cycle for the 5.15.150 release.
> > > There are 245 patches in this series, all will be posted as a response
> > > to this one.  If anyone has any issues with these being applied, please
> > > let me know.
> > >
> > > Responses should be made by Thu, 29 Feb 2024 13:15:36 +0000.
> > > Anything received after that time might be too late.
> >
> > $ git grep dma_fence_allocate_private_stub
> > drivers/dma-buf/dma-fence-unwrap.c:             return dma_fence_allocate_private_stub(timestamp);
> >                                                                                         ^^^^^^^^^
> > drivers/dma-buf/dma-fence-unwrap.c:             tmp = dma_fence_allocate_private_stub(ktime_get());
> >                                                                                        ^^^^^^^^^^^
> > drivers/dma-buf/dma-fence.c: * dma_fence_allocate_private_stub - return a private, signaled fence
> > drivers/dma-buf/dma-fence.c:struct dma_fence *dma_fence_allocate_private_stub(void)
> >                                                                                ^^^^
> > drivers/dma-buf/dma-fence.c:EXPORT_SYMBOL(dma_fence_allocate_private_stub);
> > drivers/gpu/drm/drm_syncobj.c:  struct dma_fence *fence = dma_fence_allocate_private_stub();
> > include/linux/dma-fence.h:struct dma_fence *dma_fence_allocate_private_stub(void);
> >                                                                              ^^^^
>
> How is any of this building then?  Does no one actually use
> dma-fence-unwrap.c?

It was found as part of the merge of v5.15.149 into ChromeOS. I did not check the
detailed circumstances and/or why ChromeOS stumbled over the problem. The merge introduces
other problems for us. We are busy tracking those down, so I don't have much time to look
into this further (not blaming v5.15.y, more likely it is the result of conflicts between
deviating code bases).

> > This was introduced with commit 4e82b9c11d3cd ("dma-buf: add dma_fence_timestamp helper") in
> > v5.15.149. The additional parameter to dma_fence_allocate_private_stub() was introduced in the
> > upstream kernel with commit f781f661e8c99 ("dma-buf: keep the signaling time of merged fences
> > v3") which is missing in v5.15.y.
>
> f781f661e8c99 still uses an option to dma_fence_allocate_private_stub():
> -               return dma_fence_get_stub();
> +               return dma_fence_allocate_private_stub(timestamp);
>
> So backporting that will continue the breakage, right?
>
> confused,

Most likely I missed some other commit when running "git blame". Never mind, though.
We'll deal with the problem internally. It was just a note to let people know
that there is some inconsistency in v5.15.y.

Thanks,
Guenter