Re: [PATCH v2 6/7] drm/panfrost: Add support for GPU heap allocations

From: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
To: Steven Price <steven.price@arm.com>
Cc: Tomeu Vizoso <tomeu.vizoso@collabora.com>,
	Maxime Ripard <maxime.ripard@bootlin.com>,
	Sean Paul <sean@poorly.run>,
	"dri-devel@lists.freedesktop.org"
	<dri-devel@lists.freedesktop.org>,
	David Airlie <airlied@linux.ie>,
	Boris Brezillon <boris.brezillon@collabora.com>,
	Robin Murphy <Robin.Murphy@arm.com>
Subject: Re: [PATCH v2 6/7] drm/panfrost: Add support for GPU heap allocations
Date: Fri, 26 Jul 2019 06:57:08 -0700	[thread overview]
Message-ID: <20190726135708.GA8834@kevin> (raw)
In-Reply-To: <1c54bd6c-a594-614d-7de0-badb00269ea1@arm.com>

[-- Attachment #1.1: Type: text/plain, Size: 1680 bytes --]

> It looks like the driver might be allocated a depth or stencil buffer
> without knowing whether it will be used. The buffer is then "grown" if
> it is needed by the GPU. The problem is it is possible for the
> application to access it later.

Hmm. I "think" you should be able to use a dummy (unbacked) Z/S buffer
when it won't be used, and as soon as the *driver* decides it will be
used (e.g. by setting the MALI_MFBD_DEPTH_WRITE bit), *that* is when you
allocate a real memory-backed BO. Neither case needs to be growable;
growable just pushes the logic into kernelspace (instead of handling it
in userspace).

The only wrinkle is if you need to give out addresses a priori, but that
could be solved by a mechanism to mmap a BO to a particular CPU address,
I think. (I recall MEM_ALIAS in kbase might be relevant?)

> * Use HMM: CPU VA==GPU VA. This nicely solves the problem, but falls over
> badly when the GPU VA size is smaller than the user space VA size - which is
> sadly true on many 64 bit integrations.
> 
> * Provide an allocation flag which causes the kernel driver to not pick a
> GPU address until the buffer is mapped on the CPU. The mmap callback would
> then need to look for a region that is free on both the CPU and GPU.
> 
> The second is obviously most similar to the kbase approach. kbase simplifies
> things because the kernel driver has the ultimate say over whether the
> buffer is SAME_VA or not. So on 64 bit user space we upgrade everything to
> be SAME_VA - which means the GPU VA space just follows the CPU VA (similar
> to HMM).

I'll let Rob chime in on this one. Thank you for the detailed write-up!

-Alyssa

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[-- Attachment #2: Type: text/plain, Size: 159 bytes --]

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel