* Please revert the addition of the AMD QDMA driver
@ 2024-09-16 7:40 Christoph Hellwig
2024-09-16 23:48 ` Lizhi Hou
0 siblings, 1 reply; 3+ messages in thread
From: Christoph Hellwig @ 2024-09-16 7:40 UTC (permalink / raw)
To: Vinod Koul, Nishad Saraf, Lizhi Hou; +Cc: dmaengine, linux-kernel
Hi Vinod,
I just noticed you added the AMD QDMA driver for this merge window,
which is completely broken in terms of DMA API usage by using the
private get/set_dma_ops APIs. These were never for driver use (
and I've been working for years to fix the few abusers), but with the
DMA changes in 6.12 it actually can't work at all, as the dma-iommu
driver now also sets NULL DMA ops in addition to dma-direct.
As a reminder drivers must never try to inherit dma settings from
one device or another, instead pass the actual DMA device to whatever
layer does the DMA mapping. Without that you break all kinds of
thing.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Please revert the addition of the AMD QDMA driver
2024-09-16 7:40 Please revert the addition of the AMD QDMA driver Christoph Hellwig
@ 2024-09-16 23:48 ` Lizhi Hou
2024-09-17 4:39 ` Christoph Hellwig
0 siblings, 1 reply; 3+ messages in thread
From: Lizhi Hou @ 2024-09-16 23:48 UTC (permalink / raw)
To: Christoph Hellwig, Vinod Koul, Nishad Saraf; +Cc: dmaengine, linux-kernel
Hi Christoph and Vinod,
Sorry, I did not know the limitation of get/set_dma_ops.
Instead of reverting the entire driver, is it ok to put a fix on top to
address this issue?
Thanks,
Lizhi
On 9/16/24 00:40, Christoph Hellwig wrote:
> Hi Vinod,
>
> I just noticed you added the AMD QDMA driver for this merge window,
> which is completely broken in terms of DMA API usage by using the
> private get/set_dma_ops APIs. These were never for driver use (
> and I've been working for years to fix the few abusers), but with the
> DMA changes in 6.12 it actually can't work at all, as the dma-iommu
> driver now also sets NULL DMA ops in addition to dma-direct.
>
> As a reminder drivers must never try to inherit dma settings from
> one device or another, instead pass the actual DMA device to whatever
> layer does the DMA mapping. Without that you break all kinds of
> thing.
>
>
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: Please revert the addition of the AMD QDMA driver
2024-09-16 23:48 ` Lizhi Hou
@ 2024-09-17 4:39 ` Christoph Hellwig
0 siblings, 0 replies; 3+ messages in thread
From: Christoph Hellwig @ 2024-09-17 4:39 UTC (permalink / raw)
To: Lizhi Hou
Cc: Christoph Hellwig, Vinod Koul, Nishad Saraf, dmaengine,
linux-kernel
On Mon, Sep 16, 2024 at 04:48:59PM -0700, Lizhi Hou wrote:
> Hi Christoph and Vinod,
>
>
> Sorry, I did not know the limitation of get/set_dma_ops.
>
> Instead of reverting the entire driver, is it ok to put a fix on top to
> address this issue?
Sure, if there is an easy enough and quick fix.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2024-09-17 4:39 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-09-16 7:40 Please revert the addition of the AMD QDMA driver Christoph Hellwig
2024-09-16 23:48 ` Lizhi Hou
2024-09-17 4:39 ` Christoph Hellwig
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).