vdpasim: protect concurrent access to iommu iotlb

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: 2c53d0f("vdpasim: vDPA device simulator")
Cc: [email protected]
Signed-off-by: Max Gurtovoy <[email protected]>
Signed-off-by: Jason Wang <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Michael S. Tsirkin <[email protected]>

Author: Max Gurtovoy

drivers/vdpa/vdpa_sim/vdpa_sim.c

@@ -71,6 +71,8 @@ struct vdpasim {
 	u32 status;
 	u32 generation;
 	u64 features;
+	/* spinlock to synchronize iommu table */
+	spinlock_t iommu_lock;
 };
 
 /* TODO: cross-endian support */
@@ -136,7 +138,9 @@ static void vdpasim_reset(struct vdpasim *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;
@@ -254,8 +258,10 @@ static dma_addr_t vdpasim_map_page(struct device *dev, struct page *page,
 	/* 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;
 
@@ -269,8 +275,10 @@ static void vdpasim_unmap_page(struct device *dev, dma_addr_t dma_addr,
 	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,
@@ -282,9 +290,10 @@ static void *vdpasim_alloc_coherent(struct device *dev, size_t size,
 	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,
@@ -297,6 +306,7 @@ static void *vdpasim_alloc_coherent(struct device *dev, size_t size,
 		} else
 			*dma_addr = (dma_addr_t)pa;
 	}
+	spin_unlock(&vdpasim->iommu_lock);
 
 	return addr;
 }
@@ -308,8 +318,11 @@ static void vdpasim_free_coherent(struct device *dev, size_t size,
 	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));
 }
 
@@ -555,6 +568,7 @@ static int vdpasim_set_map(struct vdpa_device *vdpa,
 	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;
@@ -564,27 +578,36 @@ static int vdpasim_set_map(struct vdpa_device *vdpa,
 		if (ret)
 			goto err;
 	}
+	spin_unlock(&vdpasim->iommu_lock);
 	return 0;
 
 err:
 	vhost_iotlb_reset(vdpasim->iommu);
+	spin_unlock(&vdpasim->iommu_lock);
 	return ret;
 }
 
 static int vdpasim_dma_map(struct vdpa_device *vdpa, u64 iova, u64 size,
 			   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;
 }