Re: [PATCH net v3] net/mlx5: fix calling mlx5_cmd_init() before DMA mask is set

[Niklas Schnelle]

On Thu, 2023-10-12 at 12:53 +0200, Niklas Schnelle wrote:
> On Wed, 2023-10-11 at 11:56 -0700, Saeed Mahameed wrote:
> > On 11 Oct 11:20, Saeed Mahameed wrote:
> > > On 11 Oct 09:57, Niklas Schnelle wrote:
> > > > Since commit 06cd555f73ca ("net/mlx5: split mlx5_cmd_init() to probe and
> > > > reload routines") mlx5_cmd_init() is called in mlx5_mdev_init() which is
> > > > called in probe_one() before mlx5_pci_init(). This is a problem because
> > > > mlx5_pci_init() is where the DMA and coherent mask is set but
> > > > mlx5_cmd_init() already does a dma_alloc_coherent(). Thus a DMA
> > > > allocation is done during probe before the correct mask is set. This
> > > > causes probe to fail initialization of the cmdif SW structs on s390x
> > > > after that is converted to the common dma-iommu code. This is because on
> > > > s390x DMA addresses below 4 GiB are reserved on current machines and
> > > > unlike the old s390x specific DMA API implementation common code
> > > > enforces DMA masks.
> > > > 
> > > > Fix this by moving set_dma_caps() out of mlx5_pci_init() and into
> > > > probe_one() before mlx5_mdev_init(). To match the overall naming scheme
> > > > rename it to mlx5_dma_init().
> > > 
> > > How about we just call mlx5_pci_init() before mlx5_mdev_init(), instead of
> > > breaking it apart ?
> > 
> > I just posted this RFC patch [1]:
> 
> This patch works to solve the problem as well.
> 
> > 
> > I am working in very limited conditions these days, and I don't have strong
> > opinion on which approach to take, Leon, Niklas, please advise.
> > 
> > The three possible solutions:
> > 
> > 1) mlx5_pci_init() before mlx5_mdev_init(), I don't think enabling pci
> > before initializing cmd dma would be a problem.
> > 
> > 2) This patch.
> > 
> > 3) Shay's patch from the link below:
> > [1] https://patchwork.kernel.org/project/netdevbpf/patch/20231011184511.19818-1-saeed@xxxxxxxxxx/
> > 
> > Thanks,
> > Saeed.
> 
> My first gut feeling was option 1) but I'm just as happy with 2) or 3).
> For me option 2 is the least invasive but not by much.
> 
> For me the important thing is what Jason also said yesterday. We need
> to merge something now to unbreak linux-next on s390x and to make sure
> we don't end up with a broken v6.7-rc1. This is already hampering our
> CI tests with linux-next. So let's do whatever can be merged the
> quickest and then feel free to do any refactoring ideas that this
> discussion might have spawned on top of that. My guess for this
> criteria would be 2).
> 
> Thanks,
> Niklas
> 

Looking closer at the patch from Shay I do like that it changes the
order in the disable/tear down path too. So since that also fixes a PPC
issue I guess that may indeed be the best solution if we can get it
merged quickly. I'll comment with my Tested-by there too.

Thanks,
Niklas