Jason Wang
2022-Nov-29 03:04 UTC
[PATCH v3] vhost_vdpa: fix the crash in unmap a large memory
On Fri, Nov 25, 2022 at 3:38 PM Cindy Lu <lulu at redhat.com> wrote:> > / and > > > On Fri, 25 Nov 2022 at 15:17, Jason Wang <jasowang at redhat.com> wrote: > > > > On Fri, Nov 25, 2022 at 10:31 AM Cindy Lu <lulu at redhat.com> wrote: > > > > > > While testing in vIOMMU, sometimes guest will unmap very large memory, > > > which will cause the crash. To fix this,Move the iommu_unmap to > > > vhost_vdpa_pa_unmap/vhost_vdpa_va_unmap and only unmap the memory > > > that saved in iotlb. > > > > > > Call Trace: > > > [ 647.820144] ------------[ cut here ]------------ > > > [ 647.820848] kernel BUG at drivers/iommu/intel/iommu.c:1174! > > > [ 647.821486] invalid opcode: 0000 [#1] PREEMPT SMP PTI > > > [ 647.822082] CPU: 10 PID: 1181 Comm: qemu-system-x86 Not tainted 6.0.0-rc1home_lulu_2452_lulu7_vhost+ #62 > > > [ 647.823139] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.15.0-29-g6a62e0cb0dfe-prebuilt.qem4 > > > [ 647.824365] RIP: 0010:domain_unmap+0x48/0x110 > > > [ 647.825424] Code: 48 89 fb 8d 4c f6 1e 39 c1 0f 4f c8 83 e9 0c 83 f9 3f 7f 18 48 89 e8 48 d3 e8 48 85 c0 75 59 > > > [ 647.828064] RSP: 0018:ffffae5340c0bbf0 EFLAGS: 00010202 > > > [ 647.828973] RAX: 0000000000000001 RBX: ffff921793d10540 RCX: 000000000000001b > > > [ 647.830083] RDX: 00000000080000ff RSI: 0000000000000001 RDI: ffff921793d10540 > > > [ 647.831214] RBP: 0000000007fc0100 R08: ffffae5340c0bcd0 R09: 0000000000000003 > > > [ 647.832388] R10: 0000007fc0100000 R11: 0000000000100000 R12: 00000000080000ff > > > [ 647.833668] R13: ffffae5340c0bcd0 R14: ffff921793d10590 R15: 0000008000100000 > > > [ 647.834782] FS: 00007f772ec90640(0000) GS:ffff921ce7a80000(0000) knlGS:0000000000000000 > > > [ 647.836004] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > > [ 647.836990] CR2: 00007f02c27a3a20 CR3: 0000000101b0c006 CR4: 0000000000372ee0 > > > [ 647.838107] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > > > [ 647.839283] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 > > > [ 647.840666] Call Trace: > > > [ 647.841437] <TASK> > > > [ 647.842107] intel_iommu_unmap_pages+0x93/0x140 > > > [ 647.843112] __iommu_unmap+0x91/0x1b0 > > > [ 647.844003] iommu_unmap+0x6a/0x95 > > > [ 647.844885] vhost_vdpa_unmap+0x1de/0x1f0 [vhost_vdpa] > > > [ 647.845985] vhost_vdpa_process_iotlb_msg+0xf0/0x90b [vhost_vdpa] > > > [ 647.847235] ? _raw_spin_unlock+0x15/0x30 > > > [ 647.848181] ? _copy_from_iter+0x8c/0x580 > > > [ 647.849137] vhost_chr_write_iter+0xb3/0x430 [vhost] > > > [ 647.850126] vfs_write+0x1e4/0x3a0 > > > [ 647.850897] ksys_write+0x53/0xd0 > > > [ 647.851688] do_syscall_64+0x3a/0x90 > > > [ 647.852508] entry_SYSCALL_64_after_hwframe+0x63/0xcd > > > [ 647.853457] RIP: 0033:0x7f7734ef9f4f > > > [ 647.854408] Code: 89 54 24 18 48 89 74 24 10 89 7c 24 08 e8 29 76 f8 ff 48 8b 54 24 18 48 8b 74 24 10 41 89 c8 > > > [ 647.857217] RSP: 002b:00007f772ec8f040 EFLAGS: 00000293 ORIG_RAX: 0000000000000001 > > > [ 647.858486] RAX: ffffffffffffffda RBX: 00000000fef00000 RCX: 00007f7734ef9f4f > > > [ 647.859713] RDX: 0000000000000048 RSI: 00007f772ec8f090 RDI: 0000000000000010 > > > [ 647.860942] RBP: 00007f772ec8f1a0 R08: 0000000000000000 R09: 0000000000000000 > > > [ 647.862206] R10: 0000000000000001 R11: 0000000000000293 R12: 0000000000000010 > > > [ 647.863446] R13: 0000000000000002 R14: 0000000000000000 R15: ffffffff01100000 > > > [ 647.864692] </TASK> > > > [ 647.865458] Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace fscache netfs v] > > > [ 647.874688] ---[ end trace 0000000000000000 ]--- > > > [ 647.876013] RIP: 0010:domain_unmap+0x48/0x110 > > > [ 647.878306] Code: 48 89 fb 8d 4c f6 1e 39 c1 0f 4f c8 83 e9 0c 83 f9 3f 7f 18 48 89 e8 48 d3 e8 48 85 c0 75 59 > > > [ 647.884581] RSP: 0018:ffffae5340c0bbf0 EFLAGS: 00010202 > > > [ 647.886308] RAX: 0000000000000001 RBX: ffff921793d10540 RCX: 000000000000001b > > > [ 647.888775] RDX: 00000000080000ff RSI: 0000000000000001 RDI: ffff921793d10540 > > > [ 647.890295] RBP: 0000000007fc0100 R08: ffffae5340c0bcd0 R09: 0000000000000003 > > > [ 647.891660] R10: 0000007fc0100000 R11: 0000000000100000 R12: 00000000080000ff > > > [ 647.893019] R13: ffffae5340c0bcd0 R14: ffff921793d10590 R15: 0000008000100000 > > > [ 647.894506] FS: 00007f772ec90640(0000) GS:ffff921ce7a80000(0000) knlGS:0000000000000000 > > > [ 647.895963] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > > [ 647.897348] CR2: 00007f02c27a3a20 CR3: 0000000101b0c006 CR4: 0000000000372ee0 > > > [ 647.898719] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > > > > > > Cc: stable at vger.kernel.org > > > Fixes: 4c8cf31885f6 ("vhost: introduce vDPA-based backend") > > > Signed-off-by: Cindy Lu <lulu at redhat.com> > > > --- > > > drivers/vhost/vdpa.c | 10 ++++++++-- > > > 1 file changed, 8 insertions(+), 2 deletions(-) > > > > > > diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c > > > index 166044642fd5..e5a07751bf45 100644 > > > --- a/drivers/vhost/vdpa.c > > > +++ b/drivers/vhost/vdpa.c > > > @@ -692,6 +692,8 @@ static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v, > > > struct vhost_iotlb_map *map; > > > struct page *page; > > > unsigned long pfn, pinned; > > > + struct vdpa_device *vdpa = v->vdpa; > > > + const struct vdpa_config_ops *ops = vdpa->config; > > > > > > while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) { > > > pinned = PFN_DOWN(map->size); > > > @@ -703,6 +705,8 @@ static void vhost_vdpa_pa_unmap(struct vhost_vdpa *v, > > > unpin_user_page(page); > > > } > > > atomic64_sub(PFN_DOWN(map->size), &dev->mm->pinned_vm); > > > + if ((ops->dma_map == NULL) && (ops->set_map == NULL)) > > > + iommu_unmap(v->domain, map->start, map->size); > > > > I think we'd better move the ops->dma_unmap() here as well as iommu_unmap()? > > > > > vhost_iotlb_map_free(iotlb, map); > > > } > > > } > > > @@ -713,11 +717,15 @@ static void vhost_vdpa_va_unmap(struct vhost_vdpa *v, > > > { > > > struct vhost_iotlb_map *map; > > > struct vdpa_map_file *map_file; > > > + struct vdpa_device *vdpa = v->vdpa; > > > + const struct vdpa_config_ops *ops = vdpa->config; > > > > > > while ((map = vhost_iotlb_itree_first(iotlb, start, last)) != NULL) { > > > map_file = (struct vdpa_map_file *)map->opaque; > > > fput(map_file->file); > > > kfree(map_file); > > > + if (ops->set_map == NULL) > > > + iommu_unmap(v->domain, map->start, map->size); > > > > Need to check where we have dma_unmap() and call that if it exists? > > > > Thanks > > > Hi Jason? > I think these functions are called in vhost_vdpa_unmap, > Do you want to separate the function in vhost_vdpa_unmap > and move it to vhost_vdpa_va_unmap and vhost_vdpa_pa_unmap? II meant dma_map()/dma_unmap() should be functional equivalent to iommu_map/unmap(). That means we should unmap exactly what is mapped before (vDPA parent may call iommu_unmap in its own dma_unmap() if it needs). If we move the iommu_unmap() from vhost_vdpa_unmap() to vhost_vdpa_{pa|va}_umap, we should move dma_unmap() as well. Thanks> thanks > cindy > > > > vhost_iotlb_map_free(iotlb, map); > > > } > > > } > > > @@ -805,8 +813,6 @@ static void vhost_vdpa_unmap(struct vhost_vdpa *v, > > > } else if (ops->set_map) { > > > if (!v->in_batch) > > > ops->set_map(vdpa, asid, iotlb); > > > - } else { > > > - iommu_unmap(v->domain, iova, size); > > > } > > > > > > /* If we are in the middle of batch processing, delay the free > > > -- > > > 2.34.3 > > > > > >