Re: [PATCH drm-misc-next v4 6/8] drm/gpuvm: add drm_gpuvm_flags to drm_gpuvm

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, 27 Sep 2023 18:52:55 +0200
Danilo Krummrich <dakr@xxxxxxxxxx> wrote:

> On 9/22/23 13:58, Boris Brezillon wrote:
> > On Wed, 20 Sep 2023 16:42:39 +0200
> > Danilo Krummrich <dakr@xxxxxxxxxx> wrote:
> >   
> >> +/**
> >> + * enum drm_gpuvm_flags - flags for struct drm_gpuvm
> >> + */
> >> +enum drm_gpuvm_flags {
> >> +	/**
> >> +	 * @DRM_GPUVM_USERBITS: user defined bits
> >> +	 */
> >> +	DRM_GPUVM_USERBITS = (1 << 0),  
> > 
> > Nit: I tried declaring driver-specific flags, and I find this
> > counter-intuitive. You basically end up with something like:
> > 
> > enum my_gpuvm_flags {
> > 	MY_FLAG_X = DRM_GPUVM_USERBITS,
> > 	MY_FLAG_Y = DRM_GPUVM_USERBITS << 1,
> > };
> > 
> > instead of the usual
> > 
> > enum my_gpuvm_flags {
> > 	MY_FLAG_X = BIT(0),
> > 	MY_FLAG_Y = BIT(1),
> > };
> > 
> > pattern.  
> 
> Right, same as with dma_fence flags.
> 
> > 
> > Another issue I see coming is if we end up adding more core flags and
> > drivers start falling short of bits for their own flags. This makes me
> > wonder if we shouldn't kill this notion of USER flags and let drivers
> > store their flags in some dedicated field, given they're likely to
> > derive drm_gpuvm and drm_gpuva with their own object anyway.  
> 
> The only reason I have this in the code is that Xe asked for this with
> drm_gpuva_flags. Hence, for consistency reasons I added it for drm_gpuvm_flags
> too.

Yeah, my comment stands for both drm_gpuva_flags and drm_gpuvm_flags
actually.

> 
> Drivers can still have their own flag fields if needed, otherwise I guess it
> doesn't really hurt to keep DRM_GPUVM_USERBITS in case someone wants to use it.

Sure, it doesn't hurt, but given drivers are inheriting from this
object anyway, I thought it'd be simpler/more future proof to let them
have their flags in a separate field. It's not like we care about
saving 4 bytes in such a big object. Might be a bit different for
drm_gpuva given the amount of live mappings one VM might have, but even
there, I suspect the current drm_gpuva size is going to hurt if we have
millions of 4k mappings, so, four more bytes won't make a huge
difference...

Anyway, I don't think that's a blocker, I just thought I'd mention it,
that's all.



[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux