From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?ISO-8859-1?Q?Christian_K=F6nig?= Subject: Re: [Nouveau] [PATCH 09/17] drm/radeon: use common fence implementation for fences Date: Tue, 22 Jul 2014 17:02:30 +0200 Message-ID: <53CE7D06.2010109@amd.com> References: <20140709093124.11354.3774.stgit@patser> <20140709122953.11354.46381.stgit@patser> <53CE2421.5040906@amd.com> <20140722114607.GL15237@phenom.ffwll.local> <20140722115737.GN15237@phenom.ffwll.local> <53CE56ED.4040109@vodafone.de> <20140722132652.GO15237@phenom.ffwll.local> <53CE6AFA.1060807@vodafone.de> <53CE78C0.9030408@canonical.com> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1"; Format="flowed" Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <53CE78C0.9030408@canonical.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" To: Maarten Lankhorst , =?ISO-8859-1?Q?Christian_K=F6nig?= , Dave Airlie , Thomas Hellstrom , nouveau , LKML , dri-devel , Ben Skeggs , "Deucher, Alexander" List-Id: nouveau.vger.kernel.org Am 22.07.2014 16:44, schrieb Maarten Lankhorst: > op 22-07-14 15:45, Christian K=F6nig schreef: >> Am 22.07.2014 15:26, schrieb Daniel Vetter: >>> On Tue, Jul 22, 2014 at 02:19:57PM +0200, Christian K=F6nig wrote: >>>> Am 22.07.2014 13:57, schrieb Daniel Vetter: >>>>> On Tue, Jul 22, 2014 at 01:46:07PM +0200, Daniel Vetter wrote: >>>>>> On Tue, Jul 22, 2014 at 10:43:13AM +0200, Christian K=F6nig wrote: >>>>>>> Am 22.07.2014 06:05, schrieb Dave Airlie: >>>>>>>> On 9 July 2014 22:29, Maarten Lankhorst wrote: >>>>>>>>> Signed-off-by: Maarten Lankhorst >>>>>>>>> --- >>>>>>>>> drivers/gpu/drm/radeon/radeon.h | 15 +- >>>>>>>>> drivers/gpu/drm/radeon/radeon_device.c | 60 ++++++++- >>>>>>>>> drivers/gpu/drm/radeon/radeon_fence.c | 223 ++++++++++++++++= ++++++++++------ >>>>>>>>> 3 files changed, 248 insertions(+), 50 deletions(-) >>>>>>>>> >>>>>>>> From what I can see this is still suffering from the problem tha= t we >>>>>>>> need to find a proper solution to, >>>>>>>> >>>>>>>> My summary of the issues after talking to Jerome and Ben and >>>>>>>> re-reading things is: >>>>>>>> >>>>>>>> We really need to work out a better interface into the drivers to = be >>>>>>>> able to avoid random atomic entrypoints, >>>>>>> Which is exactly what I criticized from the very first beginning. G= ood to >>>>>>> know that I'm not the only one thinking that this isn't such a good= idea. >>>>>> I guess I've lost context a bit, but which atomic entry point are we >>>>>> talking about? Afaics the only one that's mandatory is the is >>>>>> fence->signaled callback to check whether a fence really has been >>>>>> signalled. It's used internally by the fence code to avoid spurious >>>>>> wakeups. Afaik that should be doable already on any hardware. If tha= t's >>>>>> not the case then we can always track the signalled state in softwar= e and >>>>>> double-check in a worker thread before updating the sw state. And wr= ap >>>>>> this all up into a special fence class if there's more than one driv= er >>>>>> needing this. >>>>> One thing I've forgotten: The i915 scheduler that's floating around r= uns >>>>> its bottom half from irq context. So I really want to be able to check >>>>> fence state from irq context and I also want to make it possible >>>>> (possible! not mandatory) to register callbacks which are run from any >>>>> context asap after the fence is signalled. >>>> NAK, that's just the bad design I've talked about. Checking fence state >>>> inside the same driver from interrupt context is OK, because it's the >>>> drivers interrupt that we are talking about here. >>>> >>>> Checking fence status from another drivers interrupt context is what r= eally >>>> concerns me here, cause your driver doesn't have the slightest idea if= the >>>> called driver is really capable of checking the fence right now. >>> I guess my mail hasn't been clear then. If you don't like it we could a= dd >>> a bit of glue to insulate the madness and bad design i915 might do from >>> radeon. That imo doesn't invalidate the overall fence interfaces. >>> >>> So what about the following: >>> - fence->enabling_signaling is restricted to be called from process >>> context. We don't use any different yet, so would boild down to add= ing a >>> WARN_ON(in_interrupt) or so to fence_enable_sw_signalling. >>> >>> - Make fence->signaled optional (already the case) and don't implement = it >>> in readon (i.e. reduce this patch here). Only downside is that rade= on >>> needs to correctly (i.e. without races or so) call fence_signal. An= d the >>> cross-driver synchronization might be a bit less efficient. Note th= at >>> you can call fence_signal from wherever you want to, so hopefully t= hat >>> doesn't restrict your implementation. >>> >>> End result: No one calls into radeon from interrupt context, and this is >>> guaranteed. >>> >>> Would that be something you can agree to? >> No, the whole enable_signaling stuff should go away. No callback from th= e driver into the fence code, only the other way around. >> >> fence->signaled as well as fence->wait should become mandatory and only = called from process context without holding any locks, neither atomic nor a= ny mutex/semaphore (rcu might be ok). > fence->wait is mandatory, and already requires sleeping. > > If .signaled is not implemented there is no guarantee the fence will be > signaled sometime soon, this is also why enable_signaling exists, to > allow the driver to flush. I get it that it doesn't apply to radeon and n= ouveau, > but for other drivers that could be necessary, like vmwgfx. > > Ironically that is also a part of the ttm fence, except it was called flu= sh there. Then call it flush again and make it optional like in TTM. > I would also like to note that ttm_bo_wait currently is also a function t= hat currently uses is_signaled from atomic_context... I know, but TTM is only called from inside a single driver, no inter = driver needs here. We currently even call the internal fence = implementation from interrupt context as well and at more than one = occasion assume that TTM only uses radeon fences. Christian. > For the more complicated locking worries: Lockdep is your friend, use PRO= VE_LOCKING and find bugs before they trigger. ;-) > >>> Like I've said I think restricting the insanity other people are willing >>> to live with just because you don't like it isn't right. But it is >>> certainly right for you to insist on not being forced into any such >>> design. I think the above would achieve this. >> I don't think so. If it's just me I would say that I'm just to cautious = and the idea is still save to apply to the whole kernel. >> >> But since Dave, Jerome and Ben seems to have similar concerns I think we= need to agree to a minimum and save interface for all drivers. >> >> Christian. >>