[PATCH] drm/aperture: Run fbdev removal before internal helpers

[Javier Martinez Canillas]

[adding Zack and Alex to Cc list]

Hello Thomas,

Thanks a lot for tracking this down and figuring out the root cause!

On 6/17/22 14:10, Thomas Zimmermann wrote:
> Always run fbdev removal first to remove simpledrm via
> sysfb_disable(). This clears the internal state. The later call
> to drm_aperture_detach_drivers() then does nothing. Otherwise,
> with drm_aperture_detach_drivers() running first, the call to
> sysfb_disable() uses inconsistent state.
> 
> Example backtrace show below:
> 
> [   11.663422] ==================================================================
> [   11.663426] BUG: KASAN: use-after-free in device_del+0x79/0x5f0
> [   11.663435] Read of size 8 at addr ffff888108185050 by task systemd-udevd/311
> [   11.663440] CPU: 0 PID: 311 Comm: systemd-udevd Tainted: G            E     5
> 	.19.0-rc2-1-default+ #1689
> [   11.663445] Hardware name: HP ProLiant DL120 G7, BIOS J01 04/21/2011
> [   11.663447] Call Trace:
> [   11.663449]  <TASK>
> [   11.663451]  ? device_del+0x79/0x5f0
> [   11.663456]  dump_stack_lvl+0x5b/0x73
> [   11.663462]  print_address_description.constprop.0+0x1f/0x1b0
> [   11.663468]  ? device_del+0x79/0x5f0
> [   11.663471]  ? device_del+0x79/0x5f0
> [   11.663475]  print_report.cold+0x3c/0x21c
> [   11.663481]  ? lock_acquired+0x87/0x1e0
> [   11.663484]  ? lock_acquired+0x87/0x1e0
> [   11.663489]  ? device_del+0x79/0x5f0
> [   11.663492]  kasan_report+0xbf/0xf0
> [   11.663498]  ? device_del+0x79/0x5f0
> [   11.663503]  device_del+0x79/0x5f0
> [   11.663509]  ? device_remove_attrs+0x170/0x170
> [   11.663514]  ? lock_is_held_type+0xe8/0x140
> [   11.663523]  platform_device_del.part.0+0x19/0xe0
> [   11.663530]  platform_device_unregister+0x1c/0x30
> [   11.663535]  sysfb_disable+0x2d/0x70
> [   11.663540]  remove_conflicting_framebuffers+0x1c/0xf0
> [   11.663546]  remove_conflicting_pci_framebuffers+0x130/0x1a0
> [   11.663554]  drm_aperture_remove_conflicting_pci_framebuffers+0x86/0xb0
> [   11.663561]  ? mgag200_pci_remove+0x30/0x30 [mgag200]
> [   11.663578]  mgag200_pci_probe+0x2d/0x140 [mgag200]
> 

Maybe include a Reported-by: Zack Rusin <zackr at vmware.com> ? since
this seems to be the exact same issue that he reported yesterday.

Patch looks good to me and this seems to be the correct fix IMO.
That way we won't re-introduce the issue that was fixed by the
sysfb_unregister() function, that is to remove a pdev even if wasn't
bound to a driver to prevent a late simpledrm registration to match.

Reviewed-by: Javier Martinez Canillas <javierm at redhat.com>

I wonder what's the best way to coordinate with Alex to merge this
fix and your patch that moves the aperture code out of DRM, since
as far as I can tell both should target the v5.20 release.

> Signed-off-by: Thomas Zimmermann <tzimmermann at suse.de>
> Fixes: 873eb3b11860 ("fbdev: Disable sysfb device registration when removing conflicting FBs")
> Cc: Javier Martinez Canillas <javierm at redhat.com>
> Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
> Cc: Daniel Vetter <daniel at ffwll.ch>
> Cc: Sam Ravnborg <sam at ravnborg.org>
> Cc: Helge Deller <deller at gmx.de>
> Cc: Thomas Zimmermann <tzimmermann at suse.de>
> Cc: Alex Deucher <alexander.deucher at amd.com>
> Cc: Zhen Lei <thunder.leizhen at huawei.com>
> Cc: Changcheng Deng <deng.changcheng at zte.com.cn>
> ---
>  drivers/gpu/drm/drm_aperture.c | 26 +++++++++++++++-----------
>  1 file changed, 15 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_aperture.c b/drivers/gpu/drm/drm_aperture.c
> index 74bd4a76b253..059fd71424f6 100644
> --- a/drivers/gpu/drm/drm_aperture.c
> +++ b/drivers/gpu/drm/drm_aperture.c
> @@ -329,7 +329,20 @@ int drm_aperture_remove_conflicting_pci_framebuffers(struct pci_dev *pdev,
>  						     const struct drm_driver *req_driver)
>  {
>  	resource_size_t base, size;
> -	int bar, ret = 0;
> +	int bar, ret;
> +
> +	/*
> +	 * WARNING: Apparently we must kick fbdev drivers before vgacon,
> +	 * otherwise the vga fbdev driver falls over.
> +	 */
> +#if IS_REACHABLE(CONFIG_FB)
> +	ret = remove_conflicting_pci_framebuffers(pdev, req_driver->name);
> +	if (ret)
> +		return ret;
> +#endif
> +	ret = vga_remove_vgacon(pdev);
> +	if (ret)
> +		return ret;
>  
>  	for (bar = 0; bar < PCI_STD_NUM_BARS; ++bar) {
>  		if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM))
> @@ -339,15 +352,6 @@ int drm_aperture_remove_conflicting_pci_framebuffers(struct pci_dev *pdev,
>  		drm_aperture_detach_drivers(base, size);
>  	}
>  
> -	/*
> -	 * WARNING: Apparently we must kick fbdev drivers before vgacon,
> -	 * otherwise the vga fbdev driver falls over.
> -	 */
> -#if IS_REACHABLE(CONFIG_FB)
> -	ret = remove_conflicting_pci_framebuffers(pdev, req_driver->name);
> -#endif
> -	if (ret == 0)
> -		ret = vga_remove_vgacon(pdev);
> -	return ret;
> +	return 0;
>  }
>  EXPORT_SYMBOL(drm_aperture_remove_conflicting_pci_framebuffers);

-- 
Best regards,

Javier Martinez Canillas
Linux Engineering
Red Hat