Hi Am 23.09.22 um 12:58 schrieb Javier Martinez Canillas:
On 9/23/22 12:43, Thomas Zimmermann wrote:Hi Am 23.09.22 um 10:06 schrieb Javier Martinez Canillas:On 9/22/22 15:09, Thomas Zimmermann wrote:Synchronize CPU access to GEM BOs with other drivers when updating the screen buffer. Imported buffers might otherwise contain stale data.Can you please elaborate what the problem is? The framebuffers memory is setup by the firmware and would never come from an imported dma-buf, so could the GEM BOs even be shared with other drivers? Or is this done just for the sake of correctness ?This isn't about the scanout buffer that we get from the firmware. This is about synchronizing access to the GEM BO memory buffers. Our BOs' memory buffer are allocated from SHMEM or could be imported via dma-buf. In the latter case, another driver or hardware might modify their content concurrently. We need to synchronize before reading the memory from our CPU. The pattern is 1) sync BO memory content via begin_cpu_access() 2) blit from BO mem to scanout buffer 3) release BO memory via end_cpu_access()Yeah, I got that part. What I was asking is if importing from dma-buf is a real use case for simpledrm or if this patch was more about making the driver correct and aligned with what other DRM drivers do?
In a multi-GPU scenario, it could happen that one display uses simpledrm. If the other GPU has a native driver that exports the BO for screen mirroring, the synchronization might be needed. Admittedly it's not a likely scenario. But still...
Best regards Thomas
In any case I agree with the change, it's just the rationale for it that wasn't clear to me.
-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Ivo Totev
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
