[PATCH] drm: qxl: Don't alloc fbdev if emulation is not supported

Tue Feb 28 09:00:39 UTC 2017

On Mon, Feb 27, 2017 at 05:33:30PM -0300, Gabriel Krisman Bertazi wrote:
> If fbdev emulation is disabled, the QXL shutdown path will try to clean
> a framebuffer that wasn't initialized, hitting the Oops below.  The
> problem is that even when FBDEV_EMULATION is disabled we allocate the
> qfbdev strutucture, but we don't initialize it.  The fix is to stop
> allocating the memory, since it won't be used.  This allows the existing
> verification in the cleanup hook to do it's job preventing the oops.
> 
> Now that we don't allocate the unused fbdev structure, we need to be
> careful when dereferencing it in the PM suspend hook.
> 
> [   24.284684] BUG: unable to handle kernel NULL pointer dereference at 00000000000002e0
> [   24.285627] IP: mutex_lock+0x18/0x30
> [   24.286049] PGD 78cdf067
> [   24.286050] PUD 7940f067
> [   24.286344] PMD 0
> [   24.286649]
> [   24.287072] Oops: 0002 [#1] SMP
> [   24.287422] Modules linked in: qxl
> [   24.287806] CPU: 0 PID: 2328 Comm: bash Not tainted 4.10.0-rc5+ #97
> [   24.288515] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.9.3-20161025_171302-gandalf 04/01/2014
> [   24.289681] task: ffff88007c4c0000 task.stack: ffffc90001b58000
> [   24.290354] RIP: 0010:mutex_lock+0x18/0x30
> [   24.290812] RSP: 0018:ffffc90001b5bcb0 EFLAGS: 00010246
> [   24.291401] RAX: 0000000000000000 RBX: 00000000000002e0 RCX: 0000000000000000
> [   24.292209] RDX: ffff88007c4c0000 RSI: 0000000000000001 RDI: 00000000000002e0
> [   24.292987] RBP: ffffc90001b5bcb8 R08: fffffffffffffffe R09: 0000000000000001
> [   24.293797] R10: ffff880078d80b80 R11: 0000000000011400 R12: 0000000000000000
> [   24.294601] R13: 00000000000002e0 R14: ffffffffa0009c28 R15: 0000000000000060
> [   24.295439] FS:  00007f30e3acbb40(0000) GS:ffff88007fc00000(0000) knlGS:0000000000000000
> [   24.296364] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [   24.296997] CR2: 00000000000002e0 CR3: 0000000078c7b000 CR4: 00000000000006f0
> [   24.297813] Call Trace:
> [   24.298097]  drm_framebuffer_cleanup+0x1f/0x70
> [   24.298612]  qxl_fbdev_fini+0x68/0x90 [qxl]
> [   24.299074]  qxl_modeset_fini+0xd/0x30 [qxl]
> [   24.299562]  qxl_pci_remove+0x22/0x50 [qxl]
> [   24.300025]  pci_device_remove+0x34/0xb0
> [   24.300507]  device_release_driver_internal+0x150/0x200
> [   24.301082]  device_release_driver+0xd/0x10
> [   24.301587]  unbind_store+0x108/0x150
> [   24.301993]  drv_attr_store+0x20/0x30
> [   24.302402]  sysfs_kf_write+0x32/0x40
> [   24.302827]  kernfs_fop_write+0x108/0x190
> [   24.303269]  __vfs_write+0x23/0x120
> [   24.303678]  ? security_file_permission+0x36/0xb0
> [   24.304193]  ? rw_verify_area+0x49/0xb0
> [   24.304636]  vfs_write+0xb0/0x190
> [   24.305004]  SyS_write+0x41/0xa0
> [   24.305362]  entry_SYSCALL_64_fastpath+0x1a/0xa9
> [   24.305887] RIP: 0033:0x7f30e31d9620
> [   24.306285] RSP: 002b:00007ffc54b47e68 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
> [   24.307128] RAX: ffffffffffffffda RBX: 00007f30e3497600 RCX: 00007f30e31d9620
> [   24.307928] RDX: 000000000000000d RSI: 0000000000da2008 RDI: 0000000000000001
> [   24.308727] RBP: 000000000070bc60 R08: 00007f30e3498760 R09: 00007f30e3acbb40
> [   24.309504] R10: 0000000000000073 R11: 0000000000000246 R12: 0000000000000001
> [   24.310295] R13: 0000000000000000 R14: 0000000000000000 R15: 00007ffc54b47f34
> [   24.311095] Code: 0e 01 e9 7b fe ff ff 66 90 66 2e 0f 1f 84 00 00 00 00 00
> 55 48 89 e5 53 48 89 fb e8 83 e8 ff ff 65 48 8b 14 25 40 c4 00 00 31 c0 <3e>
> 48 0f b1 13 48 85 c0 74 08 48 89 df e8 66 fd ff ff 5b 5d c3
> [   24.313182] RIP: mutex_lock+0x18/0x30 RSP: ffffc90001b5bcb0
> [   24.313811] CR2: 00000000000002e0
> [   24.314208] ---[ end trace 29669c1593cae14b ]---
> 
> Signed-off-by: Gabriel Krisman Bertazi <krisman at collabora.co.uk>

Hm, I bit silly that we need #ifdef here, all this stuff is meant to Just
Work. Can't we just fix the oops with suitable checks in the fini paths
instead?

Also 0day has hit some oops in bochs, it might suffer from similar bugs.
-Daniel

> ---
>  drivers/gpu/drm/qxl/qxl_fb.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/qxl/qxl_fb.c b/drivers/gpu/drm/qxl/qxl_fb.c
> index 35124737666e..14e2a49a4dcf 100644
> --- a/drivers/gpu/drm/qxl/qxl_fb.c
> +++ b/drivers/gpu/drm/qxl/qxl_fb.c
> @@ -368,9 +368,11 @@ static const struct drm_fb_helper_funcs qxl_fb_helper_funcs = {
>  
>  int qxl_fbdev_init(struct qxl_device *qdev)
>  {
> +	int ret = 0;
> +
> +#ifdef CONFIG_DRM_FBDEV_EMULATION
>  	struct qxl_fbdev *qfbdev;
>  	int bpp_sel = 32; /* TODO: parameter from somewhere? */
> -	int ret;
>  
>  	qfbdev = kzalloc(sizeof(struct qxl_fbdev), GFP_KERNEL);
>  	if (!qfbdev)
> @@ -403,6 +405,8 @@ int qxl_fbdev_init(struct qxl_device *qdev)
>  	drm_fb_helper_fini(&qfbdev->helper);
>  free:
>  	kfree(qfbdev);
> +#endif
> +
>  	return ret;
>  }
>  
> @@ -418,6 +422,9 @@ void qxl_fbdev_fini(struct qxl_device *qdev)
>  
>  void qxl_fbdev_set_suspend(struct qxl_device *qdev, int state)
>  {
> +	if (!qdev->mode_info.qfbdev)
> +		return;
> +
>  	drm_fb_helper_set_suspend(&qdev->mode_info.qfbdev->helper, state);
>  }
>  
> -- 
> 2.11.0
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel

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