From: Max Gurtovoy maxg@mellanox.com
commit 0ea9ee430e74b16c6b17e70757d1c26d8d140e1f upstream.
Iommu iotlb can be accessed by different cores for performing IO using multiple virt queues. Add a spinlock to synchronize iotlb accesses.
This could be easily reproduced when using more than 1 pktgen threads to inject traffic to vdpa simulator.
Fixes: 2c53d0f64c06f("vdpasim: vDPA device simulator") Cc: stable@vger.kernel.org Signed-off-by: Max Gurtovoy maxg@mellanox.com Signed-off-by: Jason Wang jasowang@redhat.com Link: https://lore.kernel.org/r/20200731073822.13326-1-jasowang@redhat.com Signed-off-by: Michael S. Tsirkin mst@redhat.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org
--- drivers/vdpa/vdpa_sim/vdpa_sim.c | 31 +++++++++++++++++++++++++++---- 1 file changed, 27 insertions(+), 4 deletions(-)
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c @@ -70,6 +70,8 @@ struct vdpasim { u32 status; u32 generation; u64 features; + /* spinlock to synchronize iommu table */ + spinlock_t iommu_lock; };
static struct vdpasim *vdpasim_dev; @@ -118,7 +120,9 @@ static void vdpasim_reset(struct vdpasim for (i = 0; i < VDPASIM_VQ_NUM; i++) vdpasim_vq_reset(&vdpasim->vqs[i]);
+ spin_lock(&vdpasim->iommu_lock); vhost_iotlb_reset(vdpasim->iommu); + spin_unlock(&vdpasim->iommu_lock);
vdpasim->features = 0; vdpasim->status = 0; @@ -235,8 +239,10 @@ static dma_addr_t vdpasim_map_page(struc /* For simplicity, use identical mapping to avoid e.g iova * allocator. */ + spin_lock(&vdpasim->iommu_lock); ret = vhost_iotlb_add_range(iommu, pa, pa + size - 1, pa, dir_to_perm(dir)); + spin_unlock(&vdpasim->iommu_lock); if (ret) return DMA_MAPPING_ERROR;
@@ -250,8 +256,10 @@ static void vdpasim_unmap_page(struct de struct vdpasim *vdpasim = dev_to_sim(dev); struct vhost_iotlb *iommu = vdpasim->iommu;
+ spin_lock(&vdpasim->iommu_lock); vhost_iotlb_del_range(iommu, (u64)dma_addr, (u64)dma_addr + size - 1); + spin_unlock(&vdpasim->iommu_lock); }
static void *vdpasim_alloc_coherent(struct device *dev, size_t size, @@ -263,9 +271,10 @@ static void *vdpasim_alloc_coherent(stru void *addr = kmalloc(size, flag); int ret;
- if (!addr) + spin_lock(&vdpasim->iommu_lock); + if (!addr) { *dma_addr = DMA_MAPPING_ERROR; - else { + } else { u64 pa = virt_to_phys(addr);
ret = vhost_iotlb_add_range(iommu, (u64)pa, @@ -278,6 +287,7 @@ static void *vdpasim_alloc_coherent(stru } else *dma_addr = (dma_addr_t)pa; } + spin_unlock(&vdpasim->iommu_lock);
return addr; } @@ -289,8 +299,11 @@ static void vdpasim_free_coherent(struct struct vdpasim *vdpasim = dev_to_sim(dev); struct vhost_iotlb *iommu = vdpasim->iommu;
+ spin_lock(&vdpasim->iommu_lock); vhost_iotlb_del_range(iommu, (u64)dma_addr, (u64)dma_addr + size - 1); + spin_unlock(&vdpasim->iommu_lock); + kfree(phys_to_virt((uintptr_t)dma_addr)); }
@@ -531,6 +544,7 @@ static int vdpasim_set_map(struct vdpa_d u64 start = 0ULL, last = 0ULL - 1; int ret;
+ spin_lock(&vdpasim->iommu_lock); vhost_iotlb_reset(vdpasim->iommu);
for (map = vhost_iotlb_itree_first(iotlb, start, last); map; @@ -540,10 +554,12 @@ static int vdpasim_set_map(struct vdpa_d if (ret) goto err; } + spin_unlock(&vdpasim->iommu_lock); return 0;
err: vhost_iotlb_reset(vdpasim->iommu); + spin_unlock(&vdpasim->iommu_lock); return ret; }
@@ -551,16 +567,23 @@ static int vdpasim_dma_map(struct vdpa_d u64 pa, u32 perm) { struct vdpasim *vdpasim = vdpa_to_sim(vdpa); + int ret;
- return vhost_iotlb_add_range(vdpasim->iommu, iova, - iova + size - 1, pa, perm); + spin_lock(&vdpasim->iommu_lock); + ret = vhost_iotlb_add_range(vdpasim->iommu, iova, iova + size - 1, pa, + perm); + spin_unlock(&vdpasim->iommu_lock); + + return ret; }
static int vdpasim_dma_unmap(struct vdpa_device *vdpa, u64 iova, u64 size) { struct vdpasim *vdpasim = vdpa_to_sim(vdpa);
+ spin_lock(&vdpasim->iommu_lock); vhost_iotlb_del_range(vdpasim->iommu, iova, iova + size - 1); + spin_unlock(&vdpasim->iommu_lock);
return 0; }