drm/nouveau: avoid null deref on bad arguments to nouveau_vma_getmap

Commit Message

Pasi Kärkkäinen Sept. 3, 2013, 5:58 p.m. UTC

On Tue, Sep 03, 2013 at 05:48:48PM +0300, Pasi Kärkkäinen wrote:
> > >>>>> Not it really isn't appropriate..
> > >>>>>
> > >>>>> You'd have to call call nouveau_vm_map_sg_table instead, the only place that doesn't handle that correctly
> > >>>>> is where it's not expected to be called.
> > >>>>>
> > >>>>> Here, have a completely untested patch to fix things...
> > >>>>>
> > >>>> Maarten: I've been testing this stuff with Linux 3.10.x, so I had to modify the patch a bit to make it apply there.. 
> > >>>> I've attached the modified copy that applies to 3.10.9, hopefully I did the backport correctly.
> > >>>>
> > >>>> With Linux 3.10.9 and the patch applied the kernel doesn't crash anymore, and I get this error in dmesg:
> > >>>>
> > >>>> [   76.105643] nouveau W[     DRM] Trying to create a fb in vram with valid_domains=00000004
> > >>>>
> > >>>> Does that help? 
> > >>>>
> > >>> Any comments? 
> > >>>
> > >>> Maarten's patch works for me, I get that warning instead of a kernel crash,
> > >>> but it's also a bigger change that doesn't apply to older kernels as-is. 
> > >>>
> > >>> Ilia's original patch in this thread can be applied to older kernels as-is,
> > >>> and it also prevents the kernel crash from happening.
> > >>>
> > >>> Should we get both applied, so Ilia's patch can be CC'd to stable ? 
> > >>>
> > >> You haven't understood the root cause then. You're trying to move an IMPORTED dma-buf into VRAM.
> > >> Doing so would seem to work, but at that point it's no longer a dma-buf so changes done by the exporter
> > >> would not show up in nouveau because they no longer refer to the same piece of memory.
> > >>
> > > Yes, that's true, I don't understand the root cause. 
> > > That's mostly because I'm not familiar with the nouveau code/internals. 
> > >
> > >
> > >> Failing is the only right option here.
> > >>
> > > Sorry but I'm not sure I understand that correctly.. what exactly do you suggest? What should fail?
> > >
> > >
> > > Because I'm not familiar with the code (yet) understanding and finding the root cause
> > > will take time for me.. that's why I was suggesting to meanwhile apply Ilia's very simple patch
> > > from this thread which actually prevents the hard kernel crash from happening, because it seems 
> > > like an unharmful fix to have to protect against this kind of bugs (the root cause) ?
> > > Or is that unacceptable? 
> > >
> > > (To recap: The kernel crash happens very often when trying to use the nouveau adapter in Optimus mode, with all kernels at least 3.8+, and it's very annoying that your laptop crashes when trying to enable a nouveau output. So Ilia's patch doesn't make the driver working properly, but at least it prevents the hard kernel crash from happening. The crash bug is in many kernel versions, including the long-term v3.10 tree. I've had the crash happening with 3.8.x, 3.9.x and 3.10.x)
> >
> > The fix probably requires commit fdfb8332651db7a280851dfccfc4f0cff4bcd052 to apply cleanly "drm/nouveau: fix some error-path leaks in fbcon handling code"
> > 
> 
> So what you mean is to use fdfb8332651db7a280851dfccfc4f0cff4bcd052 and your patch from this thread? 
> and skip Ilia's patch?
> 

So I tested with Linux 3.10.10. I had to apply these two patches first:

1e2bd5f53b6282e711e9f074765911868f8e7dc1 drm/nouveau: fixup fbcon failure paths
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/patch/?id=1e2bd5f53b6282e711e9f074765911868f8e7dc1

fdfb8332651db7a280851dfccfc4f0cff4bcd052 drm/nouveau: fix some error-path leaks in fbcon handling code
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/patch/?id=fdfb8332651db7a280851dfccfc4f0cff4bcd052

And with those applied I was able to apply Maarten's patch cleanly (also attached to this email).

It's worth noting that I'm able to reproduce the kernel crash bug with Fedora 18 (which has xorg-1.13),
but I'm not able to reproduce it with Fedora 19 (which has xorg-1.14). Both running the same kernel.
Optimus enabled in BIOS on Lenovo T430 laptop. Nvidia GF108 Discreet GPU with Intel integrated GPU.

Maarten's patch fixes the kernel crash problem, and produces a warning message instead in dmesg.
You can add my Tested-By to the patch, assuming Maarten's patch is going to be merged? 

Thanks,

-- Pasi

Patch

diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -138,17 +143,26 @@  nouveau_user_framebuffer_create(struct drm_device *dev,
 {
 	struct nouveau_framebuffer *nouveau_fb;
 	struct drm_gem_object *gem;
+	struct nouveau_bo *nvbo;
 	int ret = -ENOMEM;
 
 	gem = drm_gem_object_lookup(dev, file_priv, mode_cmd->handles[0]);
 	if (!gem)
 		return ERR_PTR(-ENOENT);
 
+	nvbo = nouveau_gem_object(gem);
+	if (!(nvbo->valid_domains & NOUVEAU_GEM_DOMAIN_VRAM)) {
+		nv_warn(nouveau_drm(dev), "Trying to create a fb in vram with"
+			" valid_domains=%08x\n", nvbo->valid_domains);
+		ret = -EINVAL;
+		goto err_unref;
+	}
+
 	nouveau_fb = kzalloc(sizeof(struct nouveau_framebuffer), GFP_KERNEL);
 	if (!nouveau_fb)
 		goto err_unref;
 
-	ret = nouveau_framebuffer_init(dev, nouveau_fb, mode_cmd, nouveau_gem_object(gem));
+	ret = nouveau_framebuffer_init(dev, nouveau_fb, mode_cmd, nvbo);
 	if (ret)
 		goto err;