[patch 32/86] drm/nouveau/gem: fix fence_sync race / oops

[Greg KH]

3.2-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Ben Skeggs <bskeggs@xxxxxxxxxx>

commit 525895ba388c949aa906f26e3ec5cb1ab041f56b upstream.

Due to a race it was possible for a fence to be destroyed while another
thread was trying to synchronise with it.  If this happened in the fallback
non-semaphore path, it lead to the following oops due to fence->channel
being NULL.

BUG: unable to handle kernel NULL pointer dereference at   (null)
IP: [<fa9632ce>] nouveau_fence_update+0xe/0xe0 [nouveau]
*pde = a649c067
SMP
Modules linked in: fuse nouveau(O) ttm(O) drm_kms_helper(O) drm(O) mxm_wmi video wmi netconsole configfs lockd bnep bluetooth rfkill ip6t_REJECT nf_conntrack_ipv6 nf_defrag_ipv6 nf_conntrack_ipv4 nf_defrag_ipv4 xt_state nf_conntrack ip6table_filter ip6_tables snd_hda_codec_realtek snd_hda_intel snd_hda_cobinfmt_misc uinput ata_generic pata_acpi pata_aet2c_algo_bit i2c_core [last unloaded: wmi]

Pid: 2255, comm: gnome-shell Tainted: G           O 3.2.0-0.rc5.git0.1.fc17.i686 #1 System manufacturer System Product Name/M2A-VM
EIP: 0060:[<fa9632ce>] EFLAGS: 00010296 CPU: 1
EIP is at nouveau_fence_update+0xe/0xe0 [nouveau]
EAX: 00000000 EBX: ddfc6dd0 ECX: dd111580 EDX: 00000000
ESI: 00003e80 EDI: dd111580 EBP: dd121d00 ESP: dd121ce8
 DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
Process gnome-shell (pid: 2255, ti=dd120000 task=dd111580 task.ti=dd120000)
Stack:
 7dc86c76 00000000 00003e80 ddfc6dd0 00003e80 dd111580 dd121d0c fa96371f
 00000000 dd121d3c fa963773 dd111580 01000246 000ec53d 00000000 ddfc6dd0
 00001f40 00000000 ddfc6dd0 00000010 dc7df840 dd121d6c fa9639a0 00000000
Call Trace:
 [<fa96371f>] __nouveau_fence_signalled+0x1f/0x30 [nouveau]
 [<fa963773>] __nouveau_fence_wait+0x43/0xd0 [nouveau]
 [<fa9639a0>] nouveau_fence_sync+0x1a0/0x1c0 [nouveau]
 [<fa964046>] validate_list+0x176/0x300 [nouveau]
 [<f7d9c9c0>] ? ttm_bo_mem_put+0x30/0x30 [ttm]
 [<fa964b8a>] nouveau_gem_ioctl_pushbuf+0x48a/0xfd0 [nouveau]
 [<c0406481>] ? die+0x31/0x80
 [<f7c93d98>] drm_ioctl+0x388/0x490 [drm]
 [<c0406481>] ? die+0x31/0x80
 [<fa964700>] ? nouveau_gem_ioctl_new+0x150/0x150 [nouveau]
 [<c0635c7b>] ? file_has_perm+0xcb/0xe0
 [<f7c93a10>] ? drm_copy_field+0x80/0x80 [drm]
 [<c0564f56>] do_vfs_ioctl+0x86/0x5b0
 [<c0406481>] ? die+0x31/0x80
 [<c0635f22>] ? selinux_file_ioctl+0x62/0x130
 [<c0554f30>] ? fget_light+0x30/0x340
 [<c05654ef>] sys_ioctl+0x6f/0x80
 [<c099e3a4>] syscall_call+0x7/0xb
 [<c0406481>] ? die+0x31/0x80
 [<c0406481>] ? die+0x31/0x80

Signed-off-by: Ben Skeggs <bskeggs@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
 drivers/gpu/drm/nouveau/nouveau_gem.c |   23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)

--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -380,6 +380,25 @@ retry:
 }
 
 static int
+validate_sync(struct nouveau_channel *chan, struct nouveau_bo *nvbo)
+{
+	struct nouveau_fence *fence = NULL;
+	int ret = 0;
+
+	spin_lock(&nvbo->bo.bdev->fence_lock);
+	if (nvbo->bo.sync_obj)
+		fence = nouveau_fence_ref(nvbo->bo.sync_obj);
+	spin_unlock(&nvbo->bo.bdev->fence_lock);
+
+	if (fence) {
+		ret = nouveau_fence_sync(fence, chan);
+		nouveau_fence_unref(&fence);
+	}
+
+	return ret;
+}
+
+static int
 validate_list(struct nouveau_channel *chan, struct list_head *list,
 	      struct drm_nouveau_gem_pushbuf_bo *pbbo, uint64_t user_pbbo_ptr)
 {
@@ -393,7 +412,7 @@ validate_list(struct nouveau_channel *ch
 	list_for_each_entry(nvbo, list, entry) {
 		struct drm_nouveau_gem_pushbuf_bo *b = &pbbo[nvbo->pbbo_index];
 
-		ret = nouveau_fence_sync(nvbo->bo.sync_obj, chan);
+		ret = validate_sync(chan, nvbo);
 		if (unlikely(ret)) {
 			NV_ERROR(dev, "fail pre-validate sync\n");
 			return ret;
@@ -416,7 +435,7 @@ validate_list(struct nouveau_channel *ch
 			return ret;
 		}
 
-		ret = nouveau_fence_sync(nvbo->bo.sync_obj, chan);
+		ret = validate_sync(chan, nvbo);
 		if (unlikely(ret)) {
 			NV_ERROR(dev, "fail post-validate sync\n");
 			return ret;


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/