Re: [git pull] habanalabs pull request for kernel 5.15

From: Jeffrey Hugo <quic_jhugo@quicinc.com>
To: Dave Airlie <airlied@gmail.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Jason Gunthorpe <jgg@mellanox.com>,
	Daniel Vetter <daniel.vetter@ffwll.ch>,
	Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oded Gabbay <ogabbay@kernel.org>, LKML <linux-kernel@vger.kernel.org>
Subject: Re: [git pull] habanalabs pull request for kernel 5.15
Date: Tue, 24 Aug 2021 19:16:59 -0600	[thread overview]
Message-ID: <5506552d-5be1-42f6-ead5-51b92a7ced82@quicinc.com> (raw)
In-Reply-To: <CAPM=9tyT_iouVE2v8J0SMJOLV=pr=QJOm88ud=bht4=5Ms2Y2A@mail.gmail.com>

On 8/19/2021 12:48 PM, Dave Airlie wrote:
> On Fri, 20 Aug 2021 at 03:07, Greg KH <gregkh@linuxfoundation.org> wrote:
>>
>> On Thu, Aug 19, 2021 at 02:02:09PM +0300, Oded Gabbay wrote:
>>> Hi Greg,
>>>
>>> This is habanalabs pull request for the merge window of kernel 5.15.
>>> The commits divide roughly 50/50 between adding new features, such
>>> as peer-to-peer support with DMA-BUF or signaling from within a graph,
>>> and fixing various bugs, small improvements, etc.
>>
>> Pulled and pushed out, thanks!
> 
> NAK for adding dma-buf or p2p support to this driver in the upstream
> kernel. There needs to be a hard line between
> "I-can't-believe-its-not-a-drm-driver" drivers which bypass our
> userspace requirements, and I consider this the line.
> 
> This driver was merged into misc on the grounds it wasn't really a
> drm/gpu driver and so didn't have to accept our userspace rules.
> 
> Adding dma-buf/p2p support to this driver is showing it really fits
> the gpu driver model and should be under the drivers/gpu rules since
> what are most GPUs except accelerators.

Care to elaborate?  I'm not trying to be cute, but all I see here is 
that dma-buf/p2p using drivers must be in drivers/gpu, yet many drivers 
outside of the gpu area use those features.  Surely your position can't 
be that only drivers/gpu can use dma-buf or p2p (which is part of the 
PCIe spec).

> We are opening a major can of worms (some would say merging habanalabs
> driver opened it), but this places us in the situation that if a GPU
> vendor just claims their hw is a "vector" accelerator they can use
> Greg to bypass all the work that been done to ensure we have
> maintainability long term. I don't want drivers in the tree using
> dma-buf to interact with other drivers when we don't have access to a
> userspace project to validate the kernel driver assumptions.

Umm, isn't that [1]?  The Habana device has the most open userspace I'm 
aware of.  Seems disingenuous to claim you don't have access to a 
userspace project for this driver.

[1] - https://github.com/HabanaAI/hl-thunk