Message-ID: <20201030092457.GI401619@phenom.ffwll.local>
Date:   Fri, 30 Oct 2020 10:24:57 +0100
From:   Daniel Vetter <daniel@...ll.ch>
To:     Greg KH <gregkh@...uxfoundation.org>,
        Christian König <christian.koenig@....com>,
        Deepak R Varma <mh12gx2825@...il.com>,
        outreachy-kernel@...glegroups.com,
        Alex Deucher <alexander.deucher@....com>,
        David Airlie <airlied@...ux.ie>, amd-gfx@...ts.freedesktop.org,
        dri-devel@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        melissa.srw@...il.com
Subject: Re: [Outreachy kernel] [PATCH] drm/amdgpu: use
 DEFINE_DEBUGFS_ATTRIBUTE with debugfs_create_file_unsafe()

On Fri, Oct 30, 2020 at 10:15:21AM +0100, Daniel Vetter wrote:
> On Fri, Oct 30, 2020 at 09:25:18AM +0100, Greg KH wrote:
> > On Fri, Oct 30, 2020 at 09:00:04AM +0100, Christian König wrote:
> > > Am 30.10.20 um 08:57 schrieb Deepak R Varma:
> > > > On Fri, Oct 30, 2020 at 08:11:20AM +0100, Greg KH wrote:
> > > > > On Fri, Oct 30, 2020 at 08:52:45AM +0530, Deepak R Varma wrote:
> > > > > > Using DEFINE_DEBUGFS_ATTRIBUTE macro with debugfs_create_file_unsafe()
> > > > > > function in place of the debugfs_create_file() function will make the
> > > > > > file operation struct "reset" aware of the file's lifetime. Additional
> > > > > > details here: https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flists.archive.carbon60.com%2Flinux%2Fkernel%2F2369498&amp;data=04%7C01%7Cchristian.koenig%40amd.com%7Cddd7a6ac8164415a639708d87ca97004%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637396414464384011%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&amp;sdata=o6GOHvMxNMuOPlC4nhDyURCHBLqfQZhYQq%2BiIMt3D3s%3D&amp;reserved=0
> > > > > > 
> > > > > > Issue reported by Coccinelle script:
> > > > > > scripts/coccinelle/api/debugfs/debugfs_simple_attr.cocci
> > > > > > 
> > > > > > Signed-off-by: Deepak R Varma <mh12gx2825@...il.com>
> > > > > > ---
> > > > > > Please Note: This is a Outreachy project task patch.
> > > > > > 
> > > > > >   drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 20 ++++++++++----------
> > > > > >   1 file changed, 10 insertions(+), 10 deletions(-)
> > > > > > 
> > > > > > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> > > > > > index 2d125b8b15ee..f076b1ba7319 100644
> > > > > > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> > > > > > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
> > > > > > @@ -1551,29 +1551,29 @@ static int amdgpu_debugfs_sclk_set(void *data, u64 val)
> > > > > >   	return 0;
> > > > > >   }
> > > > > > -DEFINE_SIMPLE_ATTRIBUTE(fops_ib_preempt, NULL,
> > > > > > -			amdgpu_debugfs_ib_preempt, "%llu\n");
> > > > > > +DEFINE_DEBUGFS_ATTRIBUTE(fops_ib_preempt, NULL,
> > > > > > +			 amdgpu_debugfs_ib_preempt, "%llu\n");
> > > > > Are you sure this is ok?  Do these devices need this additional
> > > > > "protection"?  Do they have the problem that these macros were written
> > > > > for?
> > > > > 
> > > > > Same for the other patches you just submitted here, I think you need to
> > > > > somehow "prove" that these changes are necessary, checkpatch isn't able
> > > > > to determine this all the time.
> > > > Hi Greg,
> > > > Based on my understanding, the current function debugfs_create_file()
> > > > adds an overhead of lifetime managing proxy for such fop structs. This
> > > > should be applicable to these set of drivers as well. Hence I think this
> > > > change will be useful.
> > > 
> > > Well since this is only created once per device instance I don't really care
> > > about this little overhead.
> > > 
> > > But what exactly is debugfs doing or not doing here?
> > 
> > It is trying to save drivers from having debugfs files open that point
> > to memory that can go away at any time.  For graphics devices, I doubt
> > that is the case.
> 
> So for anything we add/remove we have two-stage cleanup
> 
> 1. drm_dev_unregister (or drm_connector_unregisters, those are the only
> two hotunpluggable things we have) unregister all the uapi interfaces.
> This deletes all the sysfs and debugfs files.
> 
> 2. Once all the references to the underlying object disappear from the
> kernel, we free up the data structure.
> 
> Now for chardev and similar uapi, we hold full references for open files.
> But for sysfs and debugfs we assume that those uapi layers will make sure
> that after we deleted the files in step 1 all access through these
> functions are guaranteed to have finished. If that's not the case, then we
> need to rework our refcounting and also refcount the underlying drm
> structure (drm_device or drm_connector) from sysfs/debugfs files.
> 
> Now I tried to look at the patch Deepak references, and I'm not really
> clear what changes. Or whether we made a wrong assumption previously about
> what debugfs/sysfs guarantee when we delete the files.

I read some more code and kerneldoc, and I still have no idea what this
new _unsafe variant is used for. Only ones I've found seem to use
debugfs_file_get/put like the normal variant, to protect against
concurrently removed files due to hotunplug. Which is kinda what we've
been expecting debugfs to do for us.

What's a use-case for _unsafe _without_ debugfs_file_get/put?

Decently confused me over here doesn't get this.
-Daniel

> -Daniel
> 
> > 
> > thanks,
> > 
> > greg k-h
> 
> -- 
> Daniel Vetter
> Software Engineer, Intel Corporation
> http://blog.ffwll.ch

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch

Powered by blists - more mailing lists