| Message ID | 20210223115048.435-7-xieyongji@bytedance.com (mailing list archive) |
|---|---|
| State | RFC |
| Headers | show |
| Series | Introduce VDUSE - vDPA Device in Userspace | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On 2021/2/23 7:50 下午, Xie Yongji wrote: > This implements a MMU-based IOMMU driver to support mapping > kernel dma buffer into userspace. The basic idea behind it is > treating MMU (VA->PA) as IOMMU (IOVA->PA). The driver will set > up MMU mapping instead of IOMMU mapping for the DMA transfer so > that the userspace process is able to use its virtual address to > access the dma buffer in kernel. > > And to avoid security issue, a bounce-buffering mechanism is > introduced to prevent userspace accessing the original buffer > directly. > > Signed-off-by: Xie Yongji <xieyongji@bytedance.com> > --- > drivers/vdpa/vdpa_user/iova_domain.c | 486 +++++++++++++++++++++++++++++++++++ > drivers/vdpa/vdpa_user/iova_domain.h | 61 +++++ > 2 files changed, 547 insertions(+) > create mode 100644 drivers/vdpa/vdpa_user/iova_domain.c > create mode 100644 drivers/vdpa/vdpa_user/iova_domain.h > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c > new file mode 100644 > index 000000000000..9285d430d486 > --- /dev/null > +++ b/drivers/vdpa/vdpa_user/iova_domain.c > @@ -0,0 +1,486 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* > + * MMU-based IOMMU implementation > + * > + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > + * > + * Author: Xie Yongji <xieyongji@bytedance.com> > + * > + */ > + > +#include <linux/slab.h> > +#include <linux/file.h> > +#include <linux/anon_inodes.h> > +#include <linux/highmem.h> > + > +#include "iova_domain.h" > + > +#define IOVA_START_PFN 1 > +#define IOVA_ALLOC_ORDER 12 > +#define IOVA_ALLOC_SIZE (1 << IOVA_ALLOC_ORDER) > + > +static inline struct page * > +vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova) > +{ > + u64 index = iova >> PAGE_SHIFT; > + > + return domain->bounce_pages[index]; > +} > + > +static inline void > +vduse_domain_set_bounce_page(struct vduse_iova_domain *domain, > + u64 iova, struct page *page) > +{ > + u64 index = iova >> PAGE_SHIFT; > + > + domain->bounce_pages[index] = page; > +} > + > +static enum dma_data_direction perm_to_dir(int perm) > +{ > + enum dma_data_direction dir; > + > + switch (perm) { > + case VHOST_MAP_WO: > + dir = DMA_FROM_DEVICE; > + break; > + case VHOST_MAP_RO: > + dir = DMA_TO_DEVICE; > + break; > + case VHOST_MAP_RW: > + dir = DMA_BIDIRECTIONAL; > + break; > + default: > + break; > + } > + > + return dir; > +} > + > +static int dir_to_perm(enum dma_data_direction dir) > +{ > + int perm = -EFAULT; > + > + switch (dir) { > + case DMA_FROM_DEVICE: > + perm = VHOST_MAP_WO; > + break; > + case DMA_TO_DEVICE: > + perm = VHOST_MAP_RO; > + break; > + case DMA_BIDIRECTIONAL: > + perm = VHOST_MAP_RW; > + break; > + default: > + break; > + } > + > + return perm; > +} Let's move the above two helpers to vhost_iotlb.h so they could be used by other driver e.g (vpda_sim) > + > +static void do_bounce(phys_addr_t orig, void *addr, size_t size, > + enum dma_data_direction dir) > +{ > + unsigned long pfn = PFN_DOWN(orig); > + > + if (PageHighMem(pfn_to_page(pfn))) { > + unsigned int offset = offset_in_page(orig); > + char *buffer; > + unsigned int sz = 0; > + unsigned long flags; > + > + while (size) { > + sz = min_t(size_t, PAGE_SIZE - offset, size); > + > + local_irq_save(flags); > + buffer = kmap_atomic(pfn_to_page(pfn)); > + if (dir == DMA_TO_DEVICE) > + memcpy(addr, buffer + offset, sz); > + else > + memcpy(buffer + offset, addr, sz); > + kunmap_atomic(buffer); > + local_irq_restore(flags); I wonder why we need to deal with highmem and irq flags explicitly like this. Doesn't kmap_atomic() will take care all of those? > + > + size -= sz; > + pfn++; > + addr += sz; > + offset = 0; > + } > + } else if (dir == DMA_TO_DEVICE) { > + memcpy(addr, phys_to_virt(orig), size); > + } else { > + memcpy(phys_to_virt(orig), addr, size); > + } > +} > + > +static struct page * > +vduse_domain_get_mapping_page(struct vduse_iova_domain *domain, u64 iova) > +{ > + u64 start = iova & PAGE_MASK; > + u64 last = start + PAGE_SIZE - 1; > + struct vhost_iotlb_map *map; > + struct page *page = NULL; > + > + spin_lock(&domain->iotlb_lock); > + map = vhost_iotlb_itree_first(domain->iotlb, start, last); > + if (!map) > + goto out; > + > + page = pfn_to_page((map->addr + iova - map->start) >> PAGE_SHIFT); > + get_page(page); > +out: > + spin_unlock(&domain->iotlb_lock); > + > + return page; > +} > + > +static struct page * > +vduse_domain_alloc_bounce_page(struct vduse_iova_domain *domain, u64 iova) > +{ > + u64 start = iova & PAGE_MASK; > + u64 last = start + PAGE_SIZE - 1; > + struct vhost_iotlb_map *map; > + struct page *page = NULL, *new_page = alloc_page(GFP_KERNEL); > + > + if (!new_page) > + return NULL; > + > + spin_lock(&domain->iotlb_lock); > + if (!vhost_iotlb_itree_first(domain->iotlb, start, last)) { > + __free_page(new_page); > + goto out; > + } > + page = vduse_domain_get_bounce_page(domain, iova); > + if (page) { > + get_page(page); > + __free_page(new_page); > + goto out; > + } > + vduse_domain_set_bounce_page(domain, iova, new_page); > + get_page(new_page); > + page = new_page; > + > + for (map = vhost_iotlb_itree_first(domain->iotlb, start, last); map; > + map = vhost_iotlb_itree_next(map, start, last)) { > + unsigned int src_offset = 0, dst_offset = 0; > + phys_addr_t src; > + void *dst; > + size_t sz; > + > + if (perm_to_dir(map->perm) == DMA_FROM_DEVICE) > + continue; > + > + if (start > map->start) > + src_offset = start - map->start; > + else > + dst_offset = map->start - start; > + > + src = map->addr + src_offset; > + dst = page_address(page) + dst_offset; > + sz = min_t(size_t, map->size - src_offset, > + PAGE_SIZE - dst_offset); > + do_bounce(src, dst, sz, DMA_TO_DEVICE); > + } > +out: > + spin_unlock(&domain->iotlb_lock); > + > + return page; > +} > + > +static void > +vduse_domain_free_bounce_pages(struct vduse_iova_domain *domain, > + u64 iova, size_t size) > +{ > + struct page *page; > + > + spin_lock(&domain->iotlb_lock); > + if (WARN_ON(vhost_iotlb_itree_first(domain->iotlb, iova, > + iova + size - 1))) > + goto out; > + > + while (size > 0) { > + page = vduse_domain_get_bounce_page(domain, iova); > + if (page) { > + vduse_domain_set_bounce_page(domain, iova, NULL); > + __free_page(page); > + } > + size -= PAGE_SIZE; > + iova += PAGE_SIZE; > + } > +out: > + spin_unlock(&domain->iotlb_lock); > +} > + > +static void vduse_domain_bounce(struct vduse_iova_domain *domain, > + dma_addr_t iova, phys_addr_t orig, > + size_t size, enum dma_data_direction dir) > +{ > + unsigned int offset = offset_in_page(iova); > + > + while (size) { > + struct page *p = vduse_domain_get_bounce_page(domain, iova); > + size_t sz = min_t(size_t, PAGE_SIZE - offset, size); > + > + WARN_ON(!p && dir == DMA_FROM_DEVICE); > + > + if (p) > + do_bounce(orig, page_address(p) + offset, sz, dir); > + > + size -= sz; > + orig += sz; > + iova += sz; > + offset = 0; > + } > +} > + > +static dma_addr_t vduse_domain_alloc_iova(struct iova_domain *iovad, > + unsigned long size, unsigned long limit) > +{ > + unsigned long shift = iova_shift(iovad); > + unsigned long iova_len = iova_align(iovad, size) >> shift; > + unsigned long iova_pfn; > + > + if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1))) > + iova_len = roundup_pow_of_two(iova_len); > + iova_pfn = alloc_iova_fast(iovad, iova_len, limit >> shift, true); > + > + return iova_pfn << shift; > +} > + > +static void vduse_domain_free_iova(struct iova_domain *iovad, > + dma_addr_t iova, size_t size) > +{ > + unsigned long shift = iova_shift(iovad); > + unsigned long iova_len = iova_align(iovad, size) >> shift; > + > + free_iova_fast(iovad, iova >> shift, iova_len); > +} > + > +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, > + struct page *page, unsigned long offset, > + size_t size, enum dma_data_direction dir, > + unsigned long attrs) > +{ > + struct iova_domain *iovad = &domain->stream_iovad; > + unsigned long limit = domain->bounce_size - 1; > + phys_addr_t pa = page_to_phys(page) + offset; > + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > + int ret; > + > + if (!iova) > + return DMA_MAPPING_ERROR; > + > + spin_lock(&domain->iotlb_lock); > + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > + (u64)iova + size - 1, > + pa, dir_to_perm(dir)); > + spin_unlock(&domain->iotlb_lock); > + if (ret) { > + vduse_domain_free_iova(iovad, iova, size); > + return DMA_MAPPING_ERROR; > + } > + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) > + vduse_domain_bounce(domain, iova, pa, size, DMA_TO_DEVICE); > + > + return iova; > +} > + > +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, > + dma_addr_t dma_addr, size_t size, > + enum dma_data_direction dir, unsigned long attrs) > +{ > + struct iova_domain *iovad = &domain->stream_iovad; > + struct vhost_iotlb_map *map; > + phys_addr_t pa; > + > + spin_lock(&domain->iotlb_lock); > + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > + (u64)dma_addr + size - 1); > + if (WARN_ON(!map)) { > + spin_unlock(&domain->iotlb_lock); > + return; > + } > + pa = map->addr; > + vhost_iotlb_map_free(domain->iotlb, map); > + spin_unlock(&domain->iotlb_lock); > + > + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) > + vduse_domain_bounce(domain, dma_addr, pa, > + size, DMA_FROM_DEVICE); > + > + vduse_domain_free_iova(iovad, dma_addr, size); > +} > + > +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, > + size_t size, dma_addr_t *dma_addr, > + gfp_t flag, unsigned long attrs) > +{ > + struct iova_domain *iovad = &domain->consistent_iovad; > + unsigned long limit = domain->iova_limit; > + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > + void *orig = alloc_pages_exact(size, flag); > + int ret; > + > + if (!iova || !orig) > + goto err; > + > + spin_lock(&domain->iotlb_lock); > + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > + (u64)iova + size - 1, > + virt_to_phys(orig), VHOST_MAP_RW); > + spin_unlock(&domain->iotlb_lock); > + if (ret) > + goto err; > + > + *dma_addr = iova; > + > + return orig; > +err: > + *dma_addr = DMA_MAPPING_ERROR; > + if (orig) > + free_pages_exact(orig, size); > + if (iova) > + vduse_domain_free_iova(iovad, iova, size); > + > + return NULL; > +} > + > +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, > + void *vaddr, dma_addr_t dma_addr, > + unsigned long attrs) > +{ > + struct iova_domain *iovad = &domain->consistent_iovad; > + struct vhost_iotlb_map *map; > + phys_addr_t pa; > + > + spin_lock(&domain->iotlb_lock); > + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > + (u64)dma_addr + size - 1); > + if (WARN_ON(!map)) { > + spin_unlock(&domain->iotlb_lock); > + return; > + } > + pa = map->addr; > + vhost_iotlb_map_free(domain->iotlb, map); > + spin_unlock(&domain->iotlb_lock); > + > + vduse_domain_free_iova(iovad, dma_addr, size); > + free_pages_exact(phys_to_virt(pa), size); > +} > + > +static vm_fault_t vduse_domain_mmap_fault(struct vm_fault *vmf) > +{ > + struct vduse_iova_domain *domain = vmf->vma->vm_private_data; > + unsigned long iova = vmf->pgoff << PAGE_SHIFT; > + struct page *page; > + > + if (!domain) > + return VM_FAULT_SIGBUS; > + > + if (iova < domain->bounce_size) > + page = vduse_domain_alloc_bounce_page(domain, iova); > + else > + page = vduse_domain_get_mapping_page(domain, iova); > + > + if (!page) > + return VM_FAULT_SIGBUS; > + > + vmf->page = page; > + > + return 0; > +} > + > +static const struct vm_operations_struct vduse_domain_mmap_ops = { > + .fault = vduse_domain_mmap_fault, > +}; > + > +static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma) > +{ > + struct vduse_iova_domain *domain = file->private_data; > + > + vma->vm_flags |= VM_DONTDUMP | VM_DONTEXPAND; > + vma->vm_private_data = domain; > + vma->vm_ops = &vduse_domain_mmap_ops; > + > + return 0; > +} > + > +static int vduse_domain_release(struct inode *inode, struct file *file) > +{ > + struct vduse_iova_domain *domain = file->private_data; > + > + vduse_domain_free_bounce_pages(domain, 0, domain->bounce_size); > + put_iova_domain(&domain->stream_iovad); > + put_iova_domain(&domain->consistent_iovad); > + vhost_iotlb_free(domain->iotlb); > + vfree(domain->bounce_pages); > + kfree(domain); > + > + return 0; > +} > + > +static const struct file_operations vduse_domain_fops = { > + .mmap = vduse_domain_mmap, > + .release = vduse_domain_release, > +}; > + > +void vduse_domain_destroy(struct vduse_iova_domain *domain) > +{ > + fput(domain->file); > +} > + > +struct vduse_iova_domain * > +vduse_domain_create(unsigned long iova_limit, size_t bounce_size) > +{ > + struct vduse_iova_domain *domain; > + struct file *file; > + unsigned long bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT; > + > + if (iova_limit <= bounce_size) > + return NULL; > + > + domain = kzalloc(sizeof(*domain), GFP_KERNEL); > + if (!domain) > + return NULL; > + > + domain->iotlb = vhost_iotlb_alloc(0, 0); > + if (!domain->iotlb) > + goto err_iotlb; > + > + domain->iova_limit = iova_limit; > + domain->bounce_size = PAGE_ALIGN(bounce_size); > + domain->bounce_pages = vzalloc(bounce_pfns * sizeof(struct page *)); > + if (!domain->bounce_pages) > + goto err_page; > + > + file = anon_inode_getfile("[vduse-domain]", &vduse_domain_fops, > + domain, O_RDWR); > + if (IS_ERR(file)) > + goto err_file; > + > + domain->file = file; > + spin_lock_init(&domain->iotlb_lock); > + init_iova_domain(&domain->stream_iovad, > + IOVA_ALLOC_SIZE, IOVA_START_PFN); > + init_iova_domain(&domain->consistent_iovad, > + PAGE_SIZE, bounce_pfns); > + > + return domain; > +err_file: > + vfree(domain->bounce_pages); > +err_page: > + vhost_iotlb_free(domain->iotlb); > +err_iotlb: > + kfree(domain); > + return NULL; > +} > + > +int vduse_domain_init(void) > +{ > + return iova_cache_get(); > +} > + > +void vduse_domain_exit(void) > +{ > + iova_cache_put(); > +} > diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h > new file mode 100644 > index 000000000000..9c85d8346626 > --- /dev/null > +++ b/drivers/vdpa/vdpa_user/iova_domain.h > @@ -0,0 +1,61 @@ > +/* SPDX-License-Identifier: GPL-2.0-only */ > +/* > + * MMU-based IOMMU implementation > + * > + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > + * > + * Author: Xie Yongji <xieyongji@bytedance.com> > + * > + */ > + > +#ifndef _VDUSE_IOVA_DOMAIN_H > +#define _VDUSE_IOVA_DOMAIN_H > + > +#include <linux/iova.h> > +#include <linux/dma-mapping.h> > +#include <linux/vhost_iotlb.h> > + > +struct vduse_iova_domain { > + struct iova_domain stream_iovad; > + struct iova_domain consistent_iovad; > + struct page **bounce_pages; > + size_t bounce_size; > + unsigned long iova_limit; > + struct vhost_iotlb *iotlb; Sorry if I've asked this before. But what's the reason for maintaing a dedicated IOTLB here? I think we could reuse vduse_dev->iommu since the device can not be used by both virtio and vhost in the same time or use vduse_iova_domain->iotlb for set_map(). Also, since vhost IOTLB support per mapping token (opauqe), can we use that instead of the bounce_pages *? Thanks > + spinlock_t iotlb_lock; > + struct file *file; > +}; > + > +static inline struct file * > +vduse_domain_file(struct vduse_iova_domain *domain) > +{ > + return domain->file; > +} > + > +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, > + struct page *page, unsigned long offset, > + size_t size, enum dma_data_direction dir, > + unsigned long attrs); > + > +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, > + dma_addr_t dma_addr, size_t size, > + enum dma_data_direction dir, unsigned long attrs); > + > +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, > + size_t size, dma_addr_t *dma_addr, > + gfp_t flag, unsigned long attrs); > + > +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, > + void *vaddr, dma_addr_t dma_addr, > + unsigned long attrs); > + > +void vduse_domain_destroy(struct vduse_iova_domain *domain); > + > +struct vduse_iova_domain *vduse_domain_create(unsigned long iova_limit, > + size_t bounce_size); > + > +int vduse_domain_init(void); > + > +void vduse_domain_exit(void); > + > +#endif /* _VDUSE_IOVA_DOMAIN_H */
On Thu, Mar 4, 2021 at 12:21 PM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/2/23 7:50 下午, Xie Yongji wrote: > > This implements a MMU-based IOMMU driver to support mapping > > kernel dma buffer into userspace. The basic idea behind it is > > treating MMU (VA->PA) as IOMMU (IOVA->PA). The driver will set > > up MMU mapping instead of IOMMU mapping for the DMA transfer so > > that the userspace process is able to use its virtual address to > > access the dma buffer in kernel. > > > > And to avoid security issue, a bounce-buffering mechanism is > > introduced to prevent userspace accessing the original buffer > > directly. > > > > Signed-off-by: Xie Yongji <xieyongji@bytedance.com> > > --- > > drivers/vdpa/vdpa_user/iova_domain.c | 486 +++++++++++++++++++++++++++++++++++ > > drivers/vdpa/vdpa_user/iova_domain.h | 61 +++++ > > 2 files changed, 547 insertions(+) > > create mode 100644 drivers/vdpa/vdpa_user/iova_domain.c > > create mode 100644 drivers/vdpa/vdpa_user/iova_domain.h > > > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c > > new file mode 100644 > > index 000000000000..9285d430d486 > > --- /dev/null > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c > > @@ -0,0 +1,486 @@ > > +// SPDX-License-Identifier: GPL-2.0-only > > +/* > > + * MMU-based IOMMU implementation > > + * > > + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > > + * > > + * Author: Xie Yongji <xieyongji@bytedance.com> > > + * > > + */ > > + > > +#include <linux/slab.h> > > +#include <linux/file.h> > > +#include <linux/anon_inodes.h> > > +#include <linux/highmem.h> > > + > > +#include "iova_domain.h" > > + > > +#define IOVA_START_PFN 1 > > +#define IOVA_ALLOC_ORDER 12 > > +#define IOVA_ALLOC_SIZE (1 << IOVA_ALLOC_ORDER) > > + > > +static inline struct page * > > +vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova) > > +{ > > + u64 index = iova >> PAGE_SHIFT; > > + > > + return domain->bounce_pages[index]; > > +} > > + > > +static inline void > > +vduse_domain_set_bounce_page(struct vduse_iova_domain *domain, > > + u64 iova, struct page *page) > > +{ > > + u64 index = iova >> PAGE_SHIFT; > > + > > + domain->bounce_pages[index] = page; > > +} > > + > > +static enum dma_data_direction perm_to_dir(int perm) > > +{ > > + enum dma_data_direction dir; > > + > > + switch (perm) { > > + case VHOST_MAP_WO: > > + dir = DMA_FROM_DEVICE; > > + break; > > + case VHOST_MAP_RO: > > + dir = DMA_TO_DEVICE; > > + break; > > + case VHOST_MAP_RW: > > + dir = DMA_BIDIRECTIONAL; > > + break; > > + default: > > + break; > > + } > > + > > + return dir; > > +} > > + > > +static int dir_to_perm(enum dma_data_direction dir) > > +{ > > + int perm = -EFAULT; > > + > > + switch (dir) { > > + case DMA_FROM_DEVICE: > > + perm = VHOST_MAP_WO; > > + break; > > + case DMA_TO_DEVICE: > > + perm = VHOST_MAP_RO; > > + break; > > + case DMA_BIDIRECTIONAL: > > + perm = VHOST_MAP_RW; > > + break; > > + default: > > + break; > > + } > > + > > + return perm; > > +} > > > Let's move the above two helpers to vhost_iotlb.h so they could be used > by other driver e.g (vpda_sim) > Sure. > > > + > > +static void do_bounce(phys_addr_t orig, void *addr, size_t size, > > + enum dma_data_direction dir) > > +{ > > + unsigned long pfn = PFN_DOWN(orig); > > + > > + if (PageHighMem(pfn_to_page(pfn))) { > > + unsigned int offset = offset_in_page(orig); > > + char *buffer; > > + unsigned int sz = 0; > > + unsigned long flags; > > + > > + while (size) { > > + sz = min_t(size_t, PAGE_SIZE - offset, size); > > + > > + local_irq_save(flags); > > + buffer = kmap_atomic(pfn_to_page(pfn)); > > + if (dir == DMA_TO_DEVICE) > > + memcpy(addr, buffer + offset, sz); > > + else > > + memcpy(buffer + offset, addr, sz); > > + kunmap_atomic(buffer); > > + local_irq_restore(flags); > > > I wonder why we need to deal with highmem and irq flags explicitly like > this. Doesn't kmap_atomic() will take care all of those? > Yes, irq flags is useless here. Will remove it. > > > + > > + size -= sz; > > + pfn++; > > + addr += sz; > > + offset = 0; > > + } > > + } else if (dir == DMA_TO_DEVICE) { > > + memcpy(addr, phys_to_virt(orig), size); > > + } else { > > + memcpy(phys_to_virt(orig), addr, size); > > + } > > +} > > + > > +static struct page * > > +vduse_domain_get_mapping_page(struct vduse_iova_domain *domain, u64 iova) > > +{ > > + u64 start = iova & PAGE_MASK; > > + u64 last = start + PAGE_SIZE - 1; > > + struct vhost_iotlb_map *map; > > + struct page *page = NULL; > > + > > + spin_lock(&domain->iotlb_lock); > > + map = vhost_iotlb_itree_first(domain->iotlb, start, last); > > + if (!map) > > + goto out; > > + > > + page = pfn_to_page((map->addr + iova - map->start) >> PAGE_SHIFT); > > + get_page(page); > > +out: > > + spin_unlock(&domain->iotlb_lock); > > + > > + return page; > > +} > > + > > +static struct page * > > +vduse_domain_alloc_bounce_page(struct vduse_iova_domain *domain, u64 iova) > > +{ > > + u64 start = iova & PAGE_MASK; > > + u64 last = start + PAGE_SIZE - 1; > > + struct vhost_iotlb_map *map; > > + struct page *page = NULL, *new_page = alloc_page(GFP_KERNEL); > > + > > + if (!new_page) > > + return NULL; > > + > > + spin_lock(&domain->iotlb_lock); > > + if (!vhost_iotlb_itree_first(domain->iotlb, start, last)) { > > + __free_page(new_page); > > + goto out; > > + } > > + page = vduse_domain_get_bounce_page(domain, iova); > > + if (page) { > > + get_page(page); > > + __free_page(new_page); > > + goto out; > > + } > > + vduse_domain_set_bounce_page(domain, iova, new_page); > > + get_page(new_page); > > + page = new_page; > > + > > + for (map = vhost_iotlb_itree_first(domain->iotlb, start, last); map; > > + map = vhost_iotlb_itree_next(map, start, last)) { > > + unsigned int src_offset = 0, dst_offset = 0; > > + phys_addr_t src; > > + void *dst; > > + size_t sz; > > + > > + if (perm_to_dir(map->perm) == DMA_FROM_DEVICE) > > + continue; > > + > > + if (start > map->start) > > + src_offset = start - map->start; > > + else > > + dst_offset = map->start - start; > > + > > + src = map->addr + src_offset; > > + dst = page_address(page) + dst_offset; > > + sz = min_t(size_t, map->size - src_offset, > > + PAGE_SIZE - dst_offset); > > + do_bounce(src, dst, sz, DMA_TO_DEVICE); > > + } > > +out: > > + spin_unlock(&domain->iotlb_lock); > > + > > + return page; > > +} > > + > > +static void > > +vduse_domain_free_bounce_pages(struct vduse_iova_domain *domain, > > + u64 iova, size_t size) > > +{ > > + struct page *page; > > + > > + spin_lock(&domain->iotlb_lock); > > + if (WARN_ON(vhost_iotlb_itree_first(domain->iotlb, iova, > > + iova + size - 1))) > > + goto out; > > + > > + while (size > 0) { > > + page = vduse_domain_get_bounce_page(domain, iova); > > + if (page) { > > + vduse_domain_set_bounce_page(domain, iova, NULL); > > + __free_page(page); > > + } > > + size -= PAGE_SIZE; > > + iova += PAGE_SIZE; > > + } > > +out: > > + spin_unlock(&domain->iotlb_lock); > > +} > > + > > +static void vduse_domain_bounce(struct vduse_iova_domain *domain, > > + dma_addr_t iova, phys_addr_t orig, > > + size_t size, enum dma_data_direction dir) > > +{ > > + unsigned int offset = offset_in_page(iova); > > + > > + while (size) { > > + struct page *p = vduse_domain_get_bounce_page(domain, iova); > > + size_t sz = min_t(size_t, PAGE_SIZE - offset, size); > > + > > + WARN_ON(!p && dir == DMA_FROM_DEVICE); > > + > > + if (p) > > + do_bounce(orig, page_address(p) + offset, sz, dir); > > + > > + size -= sz; > > + orig += sz; > > + iova += sz; > > + offset = 0; > > + } > > +} > > + > > +static dma_addr_t vduse_domain_alloc_iova(struct iova_domain *iovad, > > + unsigned long size, unsigned long limit) > > +{ > > + unsigned long shift = iova_shift(iovad); > > + unsigned long iova_len = iova_align(iovad, size) >> shift; > > + unsigned long iova_pfn; > > + > > + if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1))) > > + iova_len = roundup_pow_of_two(iova_len); > > + iova_pfn = alloc_iova_fast(iovad, iova_len, limit >> shift, true); > > + > > + return iova_pfn << shift; > > +} > > + > > +static void vduse_domain_free_iova(struct iova_domain *iovad, > > + dma_addr_t iova, size_t size) > > +{ > > + unsigned long shift = iova_shift(iovad); > > + unsigned long iova_len = iova_align(iovad, size) >> shift; > > + > > + free_iova_fast(iovad, iova >> shift, iova_len); > > +} > > + > > +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, > > + struct page *page, unsigned long offset, > > + size_t size, enum dma_data_direction dir, > > + unsigned long attrs) > > +{ > > + struct iova_domain *iovad = &domain->stream_iovad; > > + unsigned long limit = domain->bounce_size - 1; > > + phys_addr_t pa = page_to_phys(page) + offset; > > + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > > + int ret; > > + > > + if (!iova) > > + return DMA_MAPPING_ERROR; > > + > > + spin_lock(&domain->iotlb_lock); > > + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > > + (u64)iova + size - 1, > > + pa, dir_to_perm(dir)); > > + spin_unlock(&domain->iotlb_lock); > > + if (ret) { > > + vduse_domain_free_iova(iovad, iova, size); > > + return DMA_MAPPING_ERROR; > > + } > > + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) > > + vduse_domain_bounce(domain, iova, pa, size, DMA_TO_DEVICE); > > + > > + return iova; > > +} > > + > > +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, > > + dma_addr_t dma_addr, size_t size, > > + enum dma_data_direction dir, unsigned long attrs) > > +{ > > + struct iova_domain *iovad = &domain->stream_iovad; > > + struct vhost_iotlb_map *map; > > + phys_addr_t pa; > > + > > + spin_lock(&domain->iotlb_lock); > > + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > > + (u64)dma_addr + size - 1); > > + if (WARN_ON(!map)) { > > + spin_unlock(&domain->iotlb_lock); > > + return; > > + } > > + pa = map->addr; > > + vhost_iotlb_map_free(domain->iotlb, map); > > + spin_unlock(&domain->iotlb_lock); > > + > > + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) > > + vduse_domain_bounce(domain, dma_addr, pa, > > + size, DMA_FROM_DEVICE); > > + > > + vduse_domain_free_iova(iovad, dma_addr, size); > > +} > > + > > +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, > > + size_t size, dma_addr_t *dma_addr, > > + gfp_t flag, unsigned long attrs) > > +{ > > + struct iova_domain *iovad = &domain->consistent_iovad; > > + unsigned long limit = domain->iova_limit; > > + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > > + void *orig = alloc_pages_exact(size, flag); > > + int ret; > > + > > + if (!iova || !orig) > > + goto err; > > + > > + spin_lock(&domain->iotlb_lock); > > + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > > + (u64)iova + size - 1, > > + virt_to_phys(orig), VHOST_MAP_RW); > > + spin_unlock(&domain->iotlb_lock); > > + if (ret) > > + goto err; > > + > > + *dma_addr = iova; > > + > > + return orig; > > +err: > > + *dma_addr = DMA_MAPPING_ERROR; > > + if (orig) > > + free_pages_exact(orig, size); > > + if (iova) > > + vduse_domain_free_iova(iovad, iova, size); > > + > > + return NULL; > > +} > > + > > +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, > > + void *vaddr, dma_addr_t dma_addr, > > + unsigned long attrs) > > +{ > > + struct iova_domain *iovad = &domain->consistent_iovad; > > + struct vhost_iotlb_map *map; > > + phys_addr_t pa; > > + > > + spin_lock(&domain->iotlb_lock); > > + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > > + (u64)dma_addr + size - 1); > > + if (WARN_ON(!map)) { > > + spin_unlock(&domain->iotlb_lock); > > + return; > > + } > > + pa = map->addr; > > + vhost_iotlb_map_free(domain->iotlb, map); > > + spin_unlock(&domain->iotlb_lock); > > + > > + vduse_domain_free_iova(iovad, dma_addr, size); > > + free_pages_exact(phys_to_virt(pa), size); > > +} > > + > > +static vm_fault_t vduse_domain_mmap_fault(struct vm_fault *vmf) > > +{ > > + struct vduse_iova_domain *domain = vmf->vma->vm_private_data; > > + unsigned long iova = vmf->pgoff << PAGE_SHIFT; > > + struct page *page; > > + > > + if (!domain) > > + return VM_FAULT_SIGBUS; > > + > > + if (iova < domain->bounce_size) > > + page = vduse_domain_alloc_bounce_page(domain, iova); > > + else > > + page = vduse_domain_get_mapping_page(domain, iova); > > + > > + if (!page) > > + return VM_FAULT_SIGBUS; > > + > > + vmf->page = page; > > + > > + return 0; > > +} > > + > > +static const struct vm_operations_struct vduse_domain_mmap_ops = { > > + .fault = vduse_domain_mmap_fault, > > +}; > > + > > +static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma) > > +{ > > + struct vduse_iova_domain *domain = file->private_data; > > + > > + vma->vm_flags |= VM_DONTDUMP | VM_DONTEXPAND; > > + vma->vm_private_data = domain; > > + vma->vm_ops = &vduse_domain_mmap_ops; > > + > > + return 0; > > +} > > + > > +static int vduse_domain_release(struct inode *inode, struct file *file) > > +{ > > + struct vduse_iova_domain *domain = file->private_data; > > + > > + vduse_domain_free_bounce_pages(domain, 0, domain->bounce_size); > > + put_iova_domain(&domain->stream_iovad); > > + put_iova_domain(&domain->consistent_iovad); > > + vhost_iotlb_free(domain->iotlb); > > + vfree(domain->bounce_pages); > > + kfree(domain); > > + > > + return 0; > > +} > > + > > +static const struct file_operations vduse_domain_fops = { > > + .mmap = vduse_domain_mmap, > > + .release = vduse_domain_release, > > +}; > > + > > +void vduse_domain_destroy(struct vduse_iova_domain *domain) > > +{ > > + fput(domain->file); > > +} > > + > > +struct vduse_iova_domain * > > +vduse_domain_create(unsigned long iova_limit, size_t bounce_size) > > +{ > > + struct vduse_iova_domain *domain; > > + struct file *file; > > + unsigned long bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT; > > + > > + if (iova_limit <= bounce_size) > > + return NULL; > > + > > + domain = kzalloc(sizeof(*domain), GFP_KERNEL); > > + if (!domain) > > + return NULL; > > + > > + domain->iotlb = vhost_iotlb_alloc(0, 0); > > + if (!domain->iotlb) > > + goto err_iotlb; > > + > > + domain->iova_limit = iova_limit; > > + domain->bounce_size = PAGE_ALIGN(bounce_size); > > + domain->bounce_pages = vzalloc(bounce_pfns * sizeof(struct page *)); > > + if (!domain->bounce_pages) > > + goto err_page; > > + > > + file = anon_inode_getfile("[vduse-domain]", &vduse_domain_fops, > > + domain, O_RDWR); > > + if (IS_ERR(file)) > > + goto err_file; > > + > > + domain->file = file; > > + spin_lock_init(&domain->iotlb_lock); > > + init_iova_domain(&domain->stream_iovad, > > + IOVA_ALLOC_SIZE, IOVA_START_PFN); > > + init_iova_domain(&domain->consistent_iovad, > > + PAGE_SIZE, bounce_pfns); > > + > > + return domain; > > +err_file: > > + vfree(domain->bounce_pages); > > +err_page: > > + vhost_iotlb_free(domain->iotlb); > > +err_iotlb: > > + kfree(domain); > > + return NULL; > > +} > > + > > +int vduse_domain_init(void) > > +{ > > + return iova_cache_get(); > > +} > > + > > +void vduse_domain_exit(void) > > +{ > > + iova_cache_put(); > > +} > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h > > new file mode 100644 > > index 000000000000..9c85d8346626 > > --- /dev/null > > +++ b/drivers/vdpa/vdpa_user/iova_domain.h > > @@ -0,0 +1,61 @@ > > +/* SPDX-License-Identifier: GPL-2.0-only */ > > +/* > > + * MMU-based IOMMU implementation > > + * > > + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > > + * > > + * Author: Xie Yongji <xieyongji@bytedance.com> > > + * > > + */ > > + > > +#ifndef _VDUSE_IOVA_DOMAIN_H > > +#define _VDUSE_IOVA_DOMAIN_H > > + > > +#include <linux/iova.h> > > +#include <linux/dma-mapping.h> > > +#include <linux/vhost_iotlb.h> > > + > > +struct vduse_iova_domain { > > + struct iova_domain stream_iovad; > > + struct iova_domain consistent_iovad; > > + struct page **bounce_pages; > > + size_t bounce_size; > > + unsigned long iova_limit; > > + struct vhost_iotlb *iotlb; > > > Sorry if I've asked this before. > > But what's the reason for maintaing a dedicated IOTLB here? I think we > could reuse vduse_dev->iommu since the device can not be used by both > virtio and vhost in the same time or use vduse_iova_domain->iotlb for > set_map(). > The main difference between domain->iotlb and dev->iotlb is the way to deal with bounce buffer. In the domain->iotlb case, bounce buffer needs to be mapped each DMA transfer because we need to get the bounce pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce buffer only needs to be mapped once during initialization, which will be used to tell userspace how to do mmap(). > Also, since vhost IOTLB support per mapping token (opauqe), can we use > that instead of the bounce_pages *? > Sorry, I didn't get you here. Which value do you mean to store in the opaque pointer？ Thanks, Yongji
On 2021/3/4 1:12 下午, Yongji Xie wrote: > On Thu, Mar 4, 2021 at 12:21 PM Jason Wang <jasowang@redhat.com> wrote: >> >> On 2021/2/23 7:50 下午, Xie Yongji wrote: >>> This implements a MMU-based IOMMU driver to support mapping >>> kernel dma buffer into userspace. The basic idea behind it is >>> treating MMU (VA->PA) as IOMMU (IOVA->PA). The driver will set >>> up MMU mapping instead of IOMMU mapping for the DMA transfer so >>> that the userspace process is able to use its virtual address to >>> access the dma buffer in kernel. >>> >>> And to avoid security issue, a bounce-buffering mechanism is >>> introduced to prevent userspace accessing the original buffer >>> directly. >>> >>> Signed-off-by: Xie Yongji <xieyongji@bytedance.com> >>> --- >>> drivers/vdpa/vdpa_user/iova_domain.c | 486 +++++++++++++++++++++++++++++++++++ >>> drivers/vdpa/vdpa_user/iova_domain.h | 61 +++++ >>> 2 files changed, 547 insertions(+) >>> create mode 100644 drivers/vdpa/vdpa_user/iova_domain.c >>> create mode 100644 drivers/vdpa/vdpa_user/iova_domain.h >>> >>> diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c >>> new file mode 100644 >>> index 000000000000..9285d430d486 >>> --- /dev/null >>> +++ b/drivers/vdpa/vdpa_user/iova_domain.c >>> @@ -0,0 +1,486 @@ >>> +// SPDX-License-Identifier: GPL-2.0-only >>> +/* >>> + * MMU-based IOMMU implementation >>> + * >>> + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. >>> + * >>> + * Author: Xie Yongji <xieyongji@bytedance.com> >>> + * >>> + */ >>> + >>> +#include <linux/slab.h> >>> +#include <linux/file.h> >>> +#include <linux/anon_inodes.h> >>> +#include <linux/highmem.h> >>> + >>> +#include "iova_domain.h" >>> + >>> +#define IOVA_START_PFN 1 >>> +#define IOVA_ALLOC_ORDER 12 >>> +#define IOVA_ALLOC_SIZE (1 << IOVA_ALLOC_ORDER) >>> + >>> +static inline struct page * >>> +vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova) >>> +{ >>> + u64 index = iova >> PAGE_SHIFT; >>> + >>> + return domain->bounce_pages[index]; >>> +} >>> + >>> +static inline void >>> +vduse_domain_set_bounce_page(struct vduse_iova_domain *domain, >>> + u64 iova, struct page *page) >>> +{ >>> + u64 index = iova >> PAGE_SHIFT; >>> + >>> + domain->bounce_pages[index] = page; >>> +} >>> + >>> +static enum dma_data_direction perm_to_dir(int perm) >>> +{ >>> + enum dma_data_direction dir; >>> + >>> + switch (perm) { >>> + case VHOST_MAP_WO: >>> + dir = DMA_FROM_DEVICE; >>> + break; >>> + case VHOST_MAP_RO: >>> + dir = DMA_TO_DEVICE; >>> + break; >>> + case VHOST_MAP_RW: >>> + dir = DMA_BIDIRECTIONAL; >>> + break; >>> + default: >>> + break; >>> + } >>> + >>> + return dir; >>> +} >>> + >>> +static int dir_to_perm(enum dma_data_direction dir) >>> +{ >>> + int perm = -EFAULT; >>> + >>> + switch (dir) { >>> + case DMA_FROM_DEVICE: >>> + perm = VHOST_MAP_WO; >>> + break; >>> + case DMA_TO_DEVICE: >>> + perm = VHOST_MAP_RO; >>> + break; >>> + case DMA_BIDIRECTIONAL: >>> + perm = VHOST_MAP_RW; >>> + break; >>> + default: >>> + break; >>> + } >>> + >>> + return perm; >>> +} >> >> Let's move the above two helpers to vhost_iotlb.h so they could be used >> by other driver e.g (vpda_sim) >> > Sure. > >>> + >>> +static void do_bounce(phys_addr_t orig, void *addr, size_t size, >>> + enum dma_data_direction dir) >>> +{ >>> + unsigned long pfn = PFN_DOWN(orig); >>> + >>> + if (PageHighMem(pfn_to_page(pfn))) { >>> + unsigned int offset = offset_in_page(orig); >>> + char *buffer; >>> + unsigned int sz = 0; >>> + unsigned long flags; >>> + >>> + while (size) { >>> + sz = min_t(size_t, PAGE_SIZE - offset, size); >>> + >>> + local_irq_save(flags); >>> + buffer = kmap_atomic(pfn_to_page(pfn)); >>> + if (dir == DMA_TO_DEVICE) >>> + memcpy(addr, buffer + offset, sz); >>> + else >>> + memcpy(buffer + offset, addr, sz); >>> + kunmap_atomic(buffer); >>> + local_irq_restore(flags); >> >> I wonder why we need to deal with highmem and irq flags explicitly like >> this. Doesn't kmap_atomic() will take care all of those? >> > Yes, irq flags is useless here. Will remove it. > >>> + >>> + size -= sz; >>> + pfn++; >>> + addr += sz; >>> + offset = 0; >>> + } >>> + } else if (dir == DMA_TO_DEVICE) { >>> + memcpy(addr, phys_to_virt(orig), size); >>> + } else { >>> + memcpy(phys_to_virt(orig), addr, size); >>> + } >>> +} >>> + >>> +static struct page * >>> +vduse_domain_get_mapping_page(struct vduse_iova_domain *domain, u64 iova) >>> +{ >>> + u64 start = iova & PAGE_MASK; >>> + u64 last = start + PAGE_SIZE - 1; >>> + struct vhost_iotlb_map *map; >>> + struct page *page = NULL; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + map = vhost_iotlb_itree_first(domain->iotlb, start, last); >>> + if (!map) >>> + goto out; >>> + >>> + page = pfn_to_page((map->addr + iova - map->start) >> PAGE_SHIFT); >>> + get_page(page); >>> +out: >>> + spin_unlock(&domain->iotlb_lock); >>> + >>> + return page; >>> +} >>> + >>> +static struct page * >>> +vduse_domain_alloc_bounce_page(struct vduse_iova_domain *domain, u64 iova) >>> +{ >>> + u64 start = iova & PAGE_MASK; >>> + u64 last = start + PAGE_SIZE - 1; >>> + struct vhost_iotlb_map *map; >>> + struct page *page = NULL, *new_page = alloc_page(GFP_KERNEL); >>> + >>> + if (!new_page) >>> + return NULL; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + if (!vhost_iotlb_itree_first(domain->iotlb, start, last)) { >>> + __free_page(new_page); >>> + goto out; >>> + } >>> + page = vduse_domain_get_bounce_page(domain, iova); >>> + if (page) { >>> + get_page(page); >>> + __free_page(new_page); >>> + goto out; >>> + } >>> + vduse_domain_set_bounce_page(domain, iova, new_page); >>> + get_page(new_page); >>> + page = new_page; >>> + >>> + for (map = vhost_iotlb_itree_first(domain->iotlb, start, last); map; >>> + map = vhost_iotlb_itree_next(map, start, last)) { >>> + unsigned int src_offset = 0, dst_offset = 0; >>> + phys_addr_t src; >>> + void *dst; >>> + size_t sz; >>> + >>> + if (perm_to_dir(map->perm) == DMA_FROM_DEVICE) >>> + continue; >>> + >>> + if (start > map->start) >>> + src_offset = start - map->start; >>> + else >>> + dst_offset = map->start - start; >>> + >>> + src = map->addr + src_offset; >>> + dst = page_address(page) + dst_offset; >>> + sz = min_t(size_t, map->size - src_offset, >>> + PAGE_SIZE - dst_offset); >>> + do_bounce(src, dst, sz, DMA_TO_DEVICE); >>> + } >>> +out: >>> + spin_unlock(&domain->iotlb_lock); >>> + >>> + return page; >>> +} >>> + >>> +static void >>> +vduse_domain_free_bounce_pages(struct vduse_iova_domain *domain, >>> + u64 iova, size_t size) >>> +{ >>> + struct page *page; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + if (WARN_ON(vhost_iotlb_itree_first(domain->iotlb, iova, >>> + iova + size - 1))) >>> + goto out; >>> + >>> + while (size > 0) { >>> + page = vduse_domain_get_bounce_page(domain, iova); >>> + if (page) { >>> + vduse_domain_set_bounce_page(domain, iova, NULL); >>> + __free_page(page); >>> + } >>> + size -= PAGE_SIZE; >>> + iova += PAGE_SIZE; >>> + } >>> +out: >>> + spin_unlock(&domain->iotlb_lock); >>> +} >>> + >>> +static void vduse_domain_bounce(struct vduse_iova_domain *domain, >>> + dma_addr_t iova, phys_addr_t orig, >>> + size_t size, enum dma_data_direction dir) >>> +{ >>> + unsigned int offset = offset_in_page(iova); >>> + >>> + while (size) { >>> + struct page *p = vduse_domain_get_bounce_page(domain, iova); >>> + size_t sz = min_t(size_t, PAGE_SIZE - offset, size); >>> + >>> + WARN_ON(!p && dir == DMA_FROM_DEVICE); >>> + >>> + if (p) >>> + do_bounce(orig, page_address(p) + offset, sz, dir); >>> + >>> + size -= sz; >>> + orig += sz; >>> + iova += sz; >>> + offset = 0; >>> + } >>> +} >>> + >>> +static dma_addr_t vduse_domain_alloc_iova(struct iova_domain *iovad, >>> + unsigned long size, unsigned long limit) >>> +{ >>> + unsigned long shift = iova_shift(iovad); >>> + unsigned long iova_len = iova_align(iovad, size) >> shift; >>> + unsigned long iova_pfn; >>> + >>> + if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1))) >>> + iova_len = roundup_pow_of_two(iova_len); >>> + iova_pfn = alloc_iova_fast(iovad, iova_len, limit >> shift, true); >>> + >>> + return iova_pfn << shift; >>> +} >>> + >>> +static void vduse_domain_free_iova(struct iova_domain *iovad, >>> + dma_addr_t iova, size_t size) >>> +{ >>> + unsigned long shift = iova_shift(iovad); >>> + unsigned long iova_len = iova_align(iovad, size) >> shift; >>> + >>> + free_iova_fast(iovad, iova >> shift, iova_len); >>> +} >>> + >>> +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, >>> + struct page *page, unsigned long offset, >>> + size_t size, enum dma_data_direction dir, >>> + unsigned long attrs) >>> +{ >>> + struct iova_domain *iovad = &domain->stream_iovad; >>> + unsigned long limit = domain->bounce_size - 1; >>> + phys_addr_t pa = page_to_phys(page) + offset; >>> + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); >>> + int ret; >>> + >>> + if (!iova) >>> + return DMA_MAPPING_ERROR; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, >>> + (u64)iova + size - 1, >>> + pa, dir_to_perm(dir)); >>> + spin_unlock(&domain->iotlb_lock); >>> + if (ret) { >>> + vduse_domain_free_iova(iovad, iova, size); >>> + return DMA_MAPPING_ERROR; >>> + } >>> + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) >>> + vduse_domain_bounce(domain, iova, pa, size, DMA_TO_DEVICE); >>> + >>> + return iova; >>> +} >>> + >>> +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, >>> + dma_addr_t dma_addr, size_t size, >>> + enum dma_data_direction dir, unsigned long attrs) >>> +{ >>> + struct iova_domain *iovad = &domain->stream_iovad; >>> + struct vhost_iotlb_map *map; >>> + phys_addr_t pa; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, >>> + (u64)dma_addr + size - 1); >>> + if (WARN_ON(!map)) { >>> + spin_unlock(&domain->iotlb_lock); >>> + return; >>> + } >>> + pa = map->addr; >>> + vhost_iotlb_map_free(domain->iotlb, map); >>> + spin_unlock(&domain->iotlb_lock); >>> + >>> + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) >>> + vduse_domain_bounce(domain, dma_addr, pa, >>> + size, DMA_FROM_DEVICE); >>> + >>> + vduse_domain_free_iova(iovad, dma_addr, size); >>> +} >>> + >>> +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, >>> + size_t size, dma_addr_t *dma_addr, >>> + gfp_t flag, unsigned long attrs) >>> +{ >>> + struct iova_domain *iovad = &domain->consistent_iovad; >>> + unsigned long limit = domain->iova_limit; >>> + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); >>> + void *orig = alloc_pages_exact(size, flag); >>> + int ret; >>> + >>> + if (!iova || !orig) >>> + goto err; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, >>> + (u64)iova + size - 1, >>> + virt_to_phys(orig), VHOST_MAP_RW); >>> + spin_unlock(&domain->iotlb_lock); >>> + if (ret) >>> + goto err; >>> + >>> + *dma_addr = iova; >>> + >>> + return orig; >>> +err: >>> + *dma_addr = DMA_MAPPING_ERROR; >>> + if (orig) >>> + free_pages_exact(orig, size); >>> + if (iova) >>> + vduse_domain_free_iova(iovad, iova, size); >>> + >>> + return NULL; >>> +} >>> + >>> +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, >>> + void *vaddr, dma_addr_t dma_addr, >>> + unsigned long attrs) >>> +{ >>> + struct iova_domain *iovad = &domain->consistent_iovad; >>> + struct vhost_iotlb_map *map; >>> + phys_addr_t pa; >>> + >>> + spin_lock(&domain->iotlb_lock); >>> + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, >>> + (u64)dma_addr + size - 1); >>> + if (WARN_ON(!map)) { >>> + spin_unlock(&domain->iotlb_lock); >>> + return; >>> + } >>> + pa = map->addr; >>> + vhost_iotlb_map_free(domain->iotlb, map); >>> + spin_unlock(&domain->iotlb_lock); >>> + >>> + vduse_domain_free_iova(iovad, dma_addr, size); >>> + free_pages_exact(phys_to_virt(pa), size); >>> +} >>> + >>> +static vm_fault_t vduse_domain_mmap_fault(struct vm_fault *vmf) >>> +{ >>> + struct vduse_iova_domain *domain = vmf->vma->vm_private_data; >>> + unsigned long iova = vmf->pgoff << PAGE_SHIFT; >>> + struct page *page; >>> + >>> + if (!domain) >>> + return VM_FAULT_SIGBUS; >>> + >>> + if (iova < domain->bounce_size) >>> + page = vduse_domain_alloc_bounce_page(domain, iova); >>> + else >>> + page = vduse_domain_get_mapping_page(domain, iova); >>> + >>> + if (!page) >>> + return VM_FAULT_SIGBUS; >>> + >>> + vmf->page = page; >>> + >>> + return 0; >>> +} >>> + >>> +static const struct vm_operations_struct vduse_domain_mmap_ops = { >>> + .fault = vduse_domain_mmap_fault, >>> +}; >>> + >>> +static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma) >>> +{ >>> + struct vduse_iova_domain *domain = file->private_data; >>> + >>> + vma->vm_flags |= VM_DONTDUMP | VM_DONTEXPAND; >>> + vma->vm_private_data = domain; >>> + vma->vm_ops = &vduse_domain_mmap_ops; >>> + >>> + return 0; >>> +} >>> + >>> +static int vduse_domain_release(struct inode *inode, struct file *file) >>> +{ >>> + struct vduse_iova_domain *domain = file->private_data; >>> + >>> + vduse_domain_free_bounce_pages(domain, 0, domain->bounce_size); >>> + put_iova_domain(&domain->stream_iovad); >>> + put_iova_domain(&domain->consistent_iovad); >>> + vhost_iotlb_free(domain->iotlb); >>> + vfree(domain->bounce_pages); >>> + kfree(domain); >>> + >>> + return 0; >>> +} >>> + >>> +static const struct file_operations vduse_domain_fops = { >>> + .mmap = vduse_domain_mmap, >>> + .release = vduse_domain_release, >>> +}; >>> + >>> +void vduse_domain_destroy(struct vduse_iova_domain *domain) >>> +{ >>> + fput(domain->file); >>> +} >>> + >>> +struct vduse_iova_domain * >>> +vduse_domain_create(unsigned long iova_limit, size_t bounce_size) >>> +{ >>> + struct vduse_iova_domain *domain; >>> + struct file *file; >>> + unsigned long bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT; >>> + >>> + if (iova_limit <= bounce_size) >>> + return NULL; >>> + >>> + domain = kzalloc(sizeof(*domain), GFP_KERNEL); >>> + if (!domain) >>> + return NULL; >>> + >>> + domain->iotlb = vhost_iotlb_alloc(0, 0); >>> + if (!domain->iotlb) >>> + goto err_iotlb; >>> + >>> + domain->iova_limit = iova_limit; >>> + domain->bounce_size = PAGE_ALIGN(bounce_size); >>> + domain->bounce_pages = vzalloc(bounce_pfns * sizeof(struct page *)); >>> + if (!domain->bounce_pages) >>> + goto err_page; >>> + >>> + file = anon_inode_getfile("[vduse-domain]", &vduse_domain_fops, >>> + domain, O_RDWR); >>> + if (IS_ERR(file)) >>> + goto err_file; >>> + >>> + domain->file = file; >>> + spin_lock_init(&domain->iotlb_lock); >>> + init_iova_domain(&domain->stream_iovad, >>> + IOVA_ALLOC_SIZE, IOVA_START_PFN); >>> + init_iova_domain(&domain->consistent_iovad, >>> + PAGE_SIZE, bounce_pfns); >>> + >>> + return domain; >>> +err_file: >>> + vfree(domain->bounce_pages); >>> +err_page: >>> + vhost_iotlb_free(domain->iotlb); >>> +err_iotlb: >>> + kfree(domain); >>> + return NULL; >>> +} >>> + >>> +int vduse_domain_init(void) >>> +{ >>> + return iova_cache_get(); >>> +} >>> + >>> +void vduse_domain_exit(void) >>> +{ >>> + iova_cache_put(); >>> +} >>> diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h >>> new file mode 100644 >>> index 000000000000..9c85d8346626 >>> --- /dev/null >>> +++ b/drivers/vdpa/vdpa_user/iova_domain.h >>> @@ -0,0 +1,61 @@ >>> +/* SPDX-License-Identifier: GPL-2.0-only */ >>> +/* >>> + * MMU-based IOMMU implementation >>> + * >>> + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. >>> + * >>> + * Author: Xie Yongji <xieyongji@bytedance.com> >>> + * >>> + */ >>> + >>> +#ifndef _VDUSE_IOVA_DOMAIN_H >>> +#define _VDUSE_IOVA_DOMAIN_H >>> + >>> +#include <linux/iova.h> >>> +#include <linux/dma-mapping.h> >>> +#include <linux/vhost_iotlb.h> >>> + >>> +struct vduse_iova_domain { >>> + struct iova_domain stream_iovad; >>> + struct iova_domain consistent_iovad; >>> + struct page **bounce_pages; >>> + size_t bounce_size; >>> + unsigned long iova_limit; >>> + struct vhost_iotlb *iotlb; >> >> Sorry if I've asked this before. >> >> But what's the reason for maintaing a dedicated IOTLB here? I think we >> could reuse vduse_dev->iommu since the device can not be used by both >> virtio and vhost in the same time or use vduse_iova_domain->iotlb for >> set_map(). >> > The main difference between domain->iotlb and dev->iotlb is the way to > deal with bounce buffer. In the domain->iotlb case, bounce buffer > needs to be mapped each DMA transfer because we need to get the bounce > pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > buffer only needs to be mapped once during initialization, which will > be used to tell userspace how to do mmap(). > >> Also, since vhost IOTLB support per mapping token (opauqe), can we use >> that instead of the bounce_pages *? >> > Sorry, I didn't get you here. Which value do you mean to store in the > opaque pointer？ So I would like to have a way to use a single IOTLB for manage all kinds of mappings. Two possible ideas: 1) map bounce page one by one in vduse_dev_map_page(), in VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then for bounce pages, userspace still only need to map it once and we can maintain the actual mapping by storing the page or pa in the opaque field of IOTLB entry. 2) map bounce page once in vduse_dev_map_page() and store struct page **bounce_pages in the opaque field of this single IOTLB entry. Does this work? Thanks > > Thanks, > Yongji >
On Fri, Mar 5, 2021 at 11:36 AM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/4 1:12 下午, Yongji Xie wrote: > > On Thu, Mar 4, 2021 at 12:21 PM Jason Wang <jasowang@redhat.com> wrote: > >> > >> On 2021/2/23 7:50 下午, Xie Yongji wrote: > >>> This implements a MMU-based IOMMU driver to support mapping > >>> kernel dma buffer into userspace. The basic idea behind it is > >>> treating MMU (VA->PA) as IOMMU (IOVA->PA). The driver will set > >>> up MMU mapping instead of IOMMU mapping for the DMA transfer so > >>> that the userspace process is able to use its virtual address to > >>> access the dma buffer in kernel. > >>> > >>> And to avoid security issue, a bounce-buffering mechanism is > >>> introduced to prevent userspace accessing the original buffer > >>> directly. > >>> > >>> Signed-off-by: Xie Yongji <xieyongji@bytedance.com> > >>> --- > >>> drivers/vdpa/vdpa_user/iova_domain.c | 486 +++++++++++++++++++++++++++++++++++ > >>> drivers/vdpa/vdpa_user/iova_domain.h | 61 +++++ > >>> 2 files changed, 547 insertions(+) > >>> create mode 100644 drivers/vdpa/vdpa_user/iova_domain.c > >>> create mode 100644 drivers/vdpa/vdpa_user/iova_domain.h > >>> > >>> diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c > >>> new file mode 100644 > >>> index 000000000000..9285d430d486 > >>> --- /dev/null > >>> +++ b/drivers/vdpa/vdpa_user/iova_domain.c > >>> @@ -0,0 +1,486 @@ > >>> +// SPDX-License-Identifier: GPL-2.0-only > >>> +/* > >>> + * MMU-based IOMMU implementation > >>> + * > >>> + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > >>> + * > >>> + * Author: Xie Yongji <xieyongji@bytedance.com> > >>> + * > >>> + */ > >>> + > >>> +#include <linux/slab.h> > >>> +#include <linux/file.h> > >>> +#include <linux/anon_inodes.h> > >>> +#include <linux/highmem.h> > >>> + > >>> +#include "iova_domain.h" > >>> + > >>> +#define IOVA_START_PFN 1 > >>> +#define IOVA_ALLOC_ORDER 12 > >>> +#define IOVA_ALLOC_SIZE (1 << IOVA_ALLOC_ORDER) > >>> + > >>> +static inline struct page * > >>> +vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova) > >>> +{ > >>> + u64 index = iova >> PAGE_SHIFT; > >>> + > >>> + return domain->bounce_pages[index]; > >>> +} > >>> + > >>> +static inline void > >>> +vduse_domain_set_bounce_page(struct vduse_iova_domain *domain, > >>> + u64 iova, struct page *page) > >>> +{ > >>> + u64 index = iova >> PAGE_SHIFT; > >>> + > >>> + domain->bounce_pages[index] = page; > >>> +} > >>> + > >>> +static enum dma_data_direction perm_to_dir(int perm) > >>> +{ > >>> + enum dma_data_direction dir; > >>> + > >>> + switch (perm) { > >>> + case VHOST_MAP_WO: > >>> + dir = DMA_FROM_DEVICE; > >>> + break; > >>> + case VHOST_MAP_RO: > >>> + dir = DMA_TO_DEVICE; > >>> + break; > >>> + case VHOST_MAP_RW: > >>> + dir = DMA_BIDIRECTIONAL; > >>> + break; > >>> + default: > >>> + break; > >>> + } > >>> + > >>> + return dir; > >>> +} > >>> + > >>> +static int dir_to_perm(enum dma_data_direction dir) > >>> +{ > >>> + int perm = -EFAULT; > >>> + > >>> + switch (dir) { > >>> + case DMA_FROM_DEVICE: > >>> + perm = VHOST_MAP_WO; > >>> + break; > >>> + case DMA_TO_DEVICE: > >>> + perm = VHOST_MAP_RO; > >>> + break; > >>> + case DMA_BIDIRECTIONAL: > >>> + perm = VHOST_MAP_RW; > >>> + break; > >>> + default: > >>> + break; > >>> + } > >>> + > >>> + return perm; > >>> +} > >> > >> Let's move the above two helpers to vhost_iotlb.h so they could be used > >> by other driver e.g (vpda_sim) > >> > > Sure. > > > >>> + > >>> +static void do_bounce(phys_addr_t orig, void *addr, size_t size, > >>> + enum dma_data_direction dir) > >>> +{ > >>> + unsigned long pfn = PFN_DOWN(orig); > >>> + > >>> + if (PageHighMem(pfn_to_page(pfn))) { > >>> + unsigned int offset = offset_in_page(orig); > >>> + char *buffer; > >>> + unsigned int sz = 0; > >>> + unsigned long flags; > >>> + > >>> + while (size) { > >>> + sz = min_t(size_t, PAGE_SIZE - offset, size); > >>> + > >>> + local_irq_save(flags); > >>> + buffer = kmap_atomic(pfn_to_page(pfn)); > >>> + if (dir == DMA_TO_DEVICE) > >>> + memcpy(addr, buffer + offset, sz); > >>> + else > >>> + memcpy(buffer + offset, addr, sz); > >>> + kunmap_atomic(buffer); > >>> + local_irq_restore(flags); > >> > >> I wonder why we need to deal with highmem and irq flags explicitly like > >> this. Doesn't kmap_atomic() will take care all of those? > >> > > Yes, irq flags is useless here. Will remove it. > > > >>> + > >>> + size -= sz; > >>> + pfn++; > >>> + addr += sz; > >>> + offset = 0; > >>> + } > >>> + } else if (dir == DMA_TO_DEVICE) { > >>> + memcpy(addr, phys_to_virt(orig), size); > >>> + } else { > >>> + memcpy(phys_to_virt(orig), addr, size); > >>> + } > >>> +} > >>> + > >>> +static struct page * > >>> +vduse_domain_get_mapping_page(struct vduse_iova_domain *domain, u64 iova) > >>> +{ > >>> + u64 start = iova & PAGE_MASK; > >>> + u64 last = start + PAGE_SIZE - 1; > >>> + struct vhost_iotlb_map *map; > >>> + struct page *page = NULL; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + map = vhost_iotlb_itree_first(domain->iotlb, start, last); > >>> + if (!map) > >>> + goto out; > >>> + > >>> + page = pfn_to_page((map->addr + iova - map->start) >> PAGE_SHIFT); > >>> + get_page(page); > >>> +out: > >>> + spin_unlock(&domain->iotlb_lock); > >>> + > >>> + return page; > >>> +} > >>> + > >>> +static struct page * > >>> +vduse_domain_alloc_bounce_page(struct vduse_iova_domain *domain, u64 iova) > >>> +{ > >>> + u64 start = iova & PAGE_MASK; > >>> + u64 last = start + PAGE_SIZE - 1; > >>> + struct vhost_iotlb_map *map; > >>> + struct page *page = NULL, *new_page = alloc_page(GFP_KERNEL); > >>> + > >>> + if (!new_page) > >>> + return NULL; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + if (!vhost_iotlb_itree_first(domain->iotlb, start, last)) { > >>> + __free_page(new_page); > >>> + goto out; > >>> + } > >>> + page = vduse_domain_get_bounce_page(domain, iova); > >>> + if (page) { > >>> + get_page(page); > >>> + __free_page(new_page); > >>> + goto out; > >>> + } > >>> + vduse_domain_set_bounce_page(domain, iova, new_page); > >>> + get_page(new_page); > >>> + page = new_page; > >>> + > >>> + for (map = vhost_iotlb_itree_first(domain->iotlb, start, last); map; > >>> + map = vhost_iotlb_itree_next(map, start, last)) { > >>> + unsigned int src_offset = 0, dst_offset = 0; > >>> + phys_addr_t src; > >>> + void *dst; > >>> + size_t sz; > >>> + > >>> + if (perm_to_dir(map->perm) == DMA_FROM_DEVICE) > >>> + continue; > >>> + > >>> + if (start > map->start) > >>> + src_offset = start - map->start; > >>> + else > >>> + dst_offset = map->start - start; > >>> + > >>> + src = map->addr + src_offset; > >>> + dst = page_address(page) + dst_offset; > >>> + sz = min_t(size_t, map->size - src_offset, > >>> + PAGE_SIZE - dst_offset); > >>> + do_bounce(src, dst, sz, DMA_TO_DEVICE); > >>> + } > >>> +out: > >>> + spin_unlock(&domain->iotlb_lock); > >>> + > >>> + return page; > >>> +} > >>> + > >>> +static void > >>> +vduse_domain_free_bounce_pages(struct vduse_iova_domain *domain, > >>> + u64 iova, size_t size) > >>> +{ > >>> + struct page *page; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + if (WARN_ON(vhost_iotlb_itree_first(domain->iotlb, iova, > >>> + iova + size - 1))) > >>> + goto out; > >>> + > >>> + while (size > 0) { > >>> + page = vduse_domain_get_bounce_page(domain, iova); > >>> + if (page) { > >>> + vduse_domain_set_bounce_page(domain, iova, NULL); > >>> + __free_page(page); > >>> + } > >>> + size -= PAGE_SIZE; > >>> + iova += PAGE_SIZE; > >>> + } > >>> +out: > >>> + spin_unlock(&domain->iotlb_lock); > >>> +} > >>> + > >>> +static void vduse_domain_bounce(struct vduse_iova_domain *domain, > >>> + dma_addr_t iova, phys_addr_t orig, > >>> + size_t size, enum dma_data_direction dir) > >>> +{ > >>> + unsigned int offset = offset_in_page(iova); > >>> + > >>> + while (size) { > >>> + struct page *p = vduse_domain_get_bounce_page(domain, iova); > >>> + size_t sz = min_t(size_t, PAGE_SIZE - offset, size); > >>> + > >>> + WARN_ON(!p && dir == DMA_FROM_DEVICE); > >>> + > >>> + if (p) > >>> + do_bounce(orig, page_address(p) + offset, sz, dir); > >>> + > >>> + size -= sz; > >>> + orig += sz; > >>> + iova += sz; > >>> + offset = 0; > >>> + } > >>> +} > >>> + > >>> +static dma_addr_t vduse_domain_alloc_iova(struct iova_domain *iovad, > >>> + unsigned long size, unsigned long limit) > >>> +{ > >>> + unsigned long shift = iova_shift(iovad); > >>> + unsigned long iova_len = iova_align(iovad, size) >> shift; > >>> + unsigned long iova_pfn; > >>> + > >>> + if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1))) > >>> + iova_len = roundup_pow_of_two(iova_len); > >>> + iova_pfn = alloc_iova_fast(iovad, iova_len, limit >> shift, true); > >>> + > >>> + return iova_pfn << shift; > >>> +} > >>> + > >>> +static void vduse_domain_free_iova(struct iova_domain *iovad, > >>> + dma_addr_t iova, size_t size) > >>> +{ > >>> + unsigned long shift = iova_shift(iovad); > >>> + unsigned long iova_len = iova_align(iovad, size) >> shift; > >>> + > >>> + free_iova_fast(iovad, iova >> shift, iova_len); > >>> +} > >>> + > >>> +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, > >>> + struct page *page, unsigned long offset, > >>> + size_t size, enum dma_data_direction dir, > >>> + unsigned long attrs) > >>> +{ > >>> + struct iova_domain *iovad = &domain->stream_iovad; > >>> + unsigned long limit = domain->bounce_size - 1; > >>> + phys_addr_t pa = page_to_phys(page) + offset; > >>> + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > >>> + int ret; > >>> + > >>> + if (!iova) > >>> + return DMA_MAPPING_ERROR; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > >>> + (u64)iova + size - 1, > >>> + pa, dir_to_perm(dir)); > >>> + spin_unlock(&domain->iotlb_lock); > >>> + if (ret) { > >>> + vduse_domain_free_iova(iovad, iova, size); > >>> + return DMA_MAPPING_ERROR; > >>> + } > >>> + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) > >>> + vduse_domain_bounce(domain, iova, pa, size, DMA_TO_DEVICE); > >>> + > >>> + return iova; > >>> +} > >>> + > >>> +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, > >>> + dma_addr_t dma_addr, size_t size, > >>> + enum dma_data_direction dir, unsigned long attrs) > >>> +{ > >>> + struct iova_domain *iovad = &domain->stream_iovad; > >>> + struct vhost_iotlb_map *map; > >>> + phys_addr_t pa; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > >>> + (u64)dma_addr + size - 1); > >>> + if (WARN_ON(!map)) { > >>> + spin_unlock(&domain->iotlb_lock); > >>> + return; > >>> + } > >>> + pa = map->addr; > >>> + vhost_iotlb_map_free(domain->iotlb, map); > >>> + spin_unlock(&domain->iotlb_lock); > >>> + > >>> + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) > >>> + vduse_domain_bounce(domain, dma_addr, pa, > >>> + size, DMA_FROM_DEVICE); > >>> + > >>> + vduse_domain_free_iova(iovad, dma_addr, size); > >>> +} > >>> + > >>> +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, > >>> + size_t size, dma_addr_t *dma_addr, > >>> + gfp_t flag, unsigned long attrs) > >>> +{ > >>> + struct iova_domain *iovad = &domain->consistent_iovad; > >>> + unsigned long limit = domain->iova_limit; > >>> + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); > >>> + void *orig = alloc_pages_exact(size, flag); > >>> + int ret; > >>> + > >>> + if (!iova || !orig) > >>> + goto err; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, > >>> + (u64)iova + size - 1, > >>> + virt_to_phys(orig), VHOST_MAP_RW); > >>> + spin_unlock(&domain->iotlb_lock); > >>> + if (ret) > >>> + goto err; > >>> + > >>> + *dma_addr = iova; > >>> + > >>> + return orig; > >>> +err: > >>> + *dma_addr = DMA_MAPPING_ERROR; > >>> + if (orig) > >>> + free_pages_exact(orig, size); > >>> + if (iova) > >>> + vduse_domain_free_iova(iovad, iova, size); > >>> + > >>> + return NULL; > >>> +} > >>> + > >>> +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, > >>> + void *vaddr, dma_addr_t dma_addr, > >>> + unsigned long attrs) > >>> +{ > >>> + struct iova_domain *iovad = &domain->consistent_iovad; > >>> + struct vhost_iotlb_map *map; > >>> + phys_addr_t pa; > >>> + > >>> + spin_lock(&domain->iotlb_lock); > >>> + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, > >>> + (u64)dma_addr + size - 1); > >>> + if (WARN_ON(!map)) { > >>> + spin_unlock(&domain->iotlb_lock); > >>> + return; > >>> + } > >>> + pa = map->addr; > >>> + vhost_iotlb_map_free(domain->iotlb, map); > >>> + spin_unlock(&domain->iotlb_lock); > >>> + > >>> + vduse_domain_free_iova(iovad, dma_addr, size); > >>> + free_pages_exact(phys_to_virt(pa), size); > >>> +} > >>> + > >>> +static vm_fault_t vduse_domain_mmap_fault(struct vm_fault *vmf) > >>> +{ > >>> + struct vduse_iova_domain *domain = vmf->vma->vm_private_data; > >>> + unsigned long iova = vmf->pgoff << PAGE_SHIFT; > >>> + struct page *page; > >>> + > >>> + if (!domain) > >>> + return VM_FAULT_SIGBUS; > >>> + > >>> + if (iova < domain->bounce_size) > >>> + page = vduse_domain_alloc_bounce_page(domain, iova); > >>> + else > >>> + page = vduse_domain_get_mapping_page(domain, iova); > >>> + > >>> + if (!page) > >>> + return VM_FAULT_SIGBUS; > >>> + > >>> + vmf->page = page; > >>> + > >>> + return 0; > >>> +} > >>> + > >>> +static const struct vm_operations_struct vduse_domain_mmap_ops = { > >>> + .fault = vduse_domain_mmap_fault, > >>> +}; > >>> + > >>> +static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma) > >>> +{ > >>> + struct vduse_iova_domain *domain = file->private_data; > >>> + > >>> + vma->vm_flags |= VM_DONTDUMP | VM_DONTEXPAND; > >>> + vma->vm_private_data = domain; > >>> + vma->vm_ops = &vduse_domain_mmap_ops; > >>> + > >>> + return 0; > >>> +} > >>> + > >>> +static int vduse_domain_release(struct inode *inode, struct file *file) > >>> +{ > >>> + struct vduse_iova_domain *domain = file->private_data; > >>> + > >>> + vduse_domain_free_bounce_pages(domain, 0, domain->bounce_size); > >>> + put_iova_domain(&domain->stream_iovad); > >>> + put_iova_domain(&domain->consistent_iovad); > >>> + vhost_iotlb_free(domain->iotlb); > >>> + vfree(domain->bounce_pages); > >>> + kfree(domain); > >>> + > >>> + return 0; > >>> +} > >>> + > >>> +static const struct file_operations vduse_domain_fops = { > >>> + .mmap = vduse_domain_mmap, > >>> + .release = vduse_domain_release, > >>> +}; > >>> + > >>> +void vduse_domain_destroy(struct vduse_iova_domain *domain) > >>> +{ > >>> + fput(domain->file); > >>> +} > >>> + > >>> +struct vduse_iova_domain * > >>> +vduse_domain_create(unsigned long iova_limit, size_t bounce_size) > >>> +{ > >>> + struct vduse_iova_domain *domain; > >>> + struct file *file; > >>> + unsigned long bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT; > >>> + > >>> + if (iova_limit <= bounce_size) > >>> + return NULL; > >>> + > >>> + domain = kzalloc(sizeof(*domain), GFP_KERNEL); > >>> + if (!domain) > >>> + return NULL; > >>> + > >>> + domain->iotlb = vhost_iotlb_alloc(0, 0); > >>> + if (!domain->iotlb) > >>> + goto err_iotlb; > >>> + > >>> + domain->iova_limit = iova_limit; > >>> + domain->bounce_size = PAGE_ALIGN(bounce_size); > >>> + domain->bounce_pages = vzalloc(bounce_pfns * sizeof(struct page *)); > >>> + if (!domain->bounce_pages) > >>> + goto err_page; > >>> + > >>> + file = anon_inode_getfile("[vduse-domain]", &vduse_domain_fops, > >>> + domain, O_RDWR); > >>> + if (IS_ERR(file)) > >>> + goto err_file; > >>> + > >>> + domain->file = file; > >>> + spin_lock_init(&domain->iotlb_lock); > >>> + init_iova_domain(&domain->stream_iovad, > >>> + IOVA_ALLOC_SIZE, IOVA_START_PFN); > >>> + init_iova_domain(&domain->consistent_iovad, > >>> + PAGE_SIZE, bounce_pfns); > >>> + > >>> + return domain; > >>> +err_file: > >>> + vfree(domain->bounce_pages); > >>> +err_page: > >>> + vhost_iotlb_free(domain->iotlb); > >>> +err_iotlb: > >>> + kfree(domain); > >>> + return NULL; > >>> +} > >>> + > >>> +int vduse_domain_init(void) > >>> +{ > >>> + return iova_cache_get(); > >>> +} > >>> + > >>> +void vduse_domain_exit(void) > >>> +{ > >>> + iova_cache_put(); > >>> +} > >>> diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h > >>> new file mode 100644 > >>> index 000000000000..9c85d8346626 > >>> --- /dev/null > >>> +++ b/drivers/vdpa/vdpa_user/iova_domain.h > >>> @@ -0,0 +1,61 @@ > >>> +/* SPDX-License-Identifier: GPL-2.0-only */ > >>> +/* > >>> + * MMU-based IOMMU implementation > >>> + * > >>> + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. > >>> + * > >>> + * Author: Xie Yongji <xieyongji@bytedance.com> > >>> + * > >>> + */ > >>> + > >>> +#ifndef _VDUSE_IOVA_DOMAIN_H > >>> +#define _VDUSE_IOVA_DOMAIN_H > >>> + > >>> +#include <linux/iova.h> > >>> +#include <linux/dma-mapping.h> > >>> +#include <linux/vhost_iotlb.h> > >>> + > >>> +struct vduse_iova_domain { > >>> + struct iova_domain stream_iovad; > >>> + struct iova_domain consistent_iovad; > >>> + struct page **bounce_pages; > >>> + size_t bounce_size; > >>> + unsigned long iova_limit; > >>> + struct vhost_iotlb *iotlb; > >> > >> Sorry if I've asked this before. > >> > >> But what's the reason for maintaing a dedicated IOTLB here? I think we > >> could reuse vduse_dev->iommu since the device can not be used by both > >> virtio and vhost in the same time or use vduse_iova_domain->iotlb for > >> set_map(). > >> > > The main difference between domain->iotlb and dev->iotlb is the way to > > deal with bounce buffer. In the domain->iotlb case, bounce buffer > > needs to be mapped each DMA transfer because we need to get the bounce > > pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > > buffer only needs to be mapped once during initialization, which will > > be used to tell userspace how to do mmap(). > > > >> Also, since vhost IOTLB support per mapping token (opauqe), can we use > >> that instead of the bounce_pages *? > >> > > Sorry, I didn't get you here. Which value do you mean to store in the > > opaque pointer？ > > > So I would like to have a way to use a single IOTLB for manage all kinds > of mappings. Two possible ideas: > > 1) map bounce page one by one in vduse_dev_map_page(), in > VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > for bounce pages, userspace still only need to map it once and we can > maintain the actual mapping by storing the page or pa in the opaque > field of IOTLB entry. Looks like userspace still needs to unmap the old region and map a new region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > 2) map bounce page once in vduse_dev_map_page() and store struct page > **bounce_pages in the opaque field of this single IOTLB entry. > We can get the struct page **bounce_pages from vduse_iova_domain. Why do we need to store it in the opaque field? Should the opaque field be used to store vdpa_map_file? And I think it works. One problem is we need to find a place to store the original DMA buffer's address and size. I think we can modify the array of bounce_pages for this purpose. Thanks, Yongji
On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/5 2:15 下午, Yongji Xie wrote: > > Sorry if I've asked this before. > > But what's the reason for maintaing a dedicated IOTLB here? I think we > could reuse vduse_dev->iommu since the device can not be used by both > virtio and vhost in the same time or use vduse_iova_domain->iotlb for > set_map(). > > The main difference between domain->iotlb and dev->iotlb is the way to > deal with bounce buffer. In the domain->iotlb case, bounce buffer > needs to be mapped each DMA transfer because we need to get the bounce > pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > buffer only needs to be mapped once during initialization, which will > be used to tell userspace how to do mmap(). > > Also, since vhost IOTLB support per mapping token (opauqe), can we use > that instead of the bounce_pages *? > > Sorry, I didn't get you here. Which value do you mean to store in the > opaque pointer？ > > So I would like to have a way to use a single IOTLB for manage all kinds > of mappings. Two possible ideas: > > 1) map bounce page one by one in vduse_dev_map_page(), in > VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > for bounce pages, userspace still only need to map it once and we can > maintain the actual mapping by storing the page or pa in the opaque > field of IOTLB entry. > > Looks like userspace still needs to unmap the old region and map a new > region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > > > I don't get here. Can you give an example? > For example, userspace needs to process two I/O requests (one page per request). To process the first request, userspace uses VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap it. To process the second request, userspace uses VDUSE_IOTLB_GET_FD ioctl to query the new iova region and map a new region (0 ~ 8192). Then userspace needs to traverse the list of iova regions and unmap the old region (0 ~ 4096). Looks like this is a little complex. Thanks, Yongji
On 2021/3/5 3:13 下午, Yongji Xie wrote: > On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: >> >> On 2021/3/5 2:15 下午, Yongji Xie wrote: >> >> Sorry if I've asked this before. >> >> But what's the reason for maintaing a dedicated IOTLB here? I think we >> could reuse vduse_dev->iommu since the device can not be used by both >> virtio and vhost in the same time or use vduse_iova_domain->iotlb for >> set_map(). >> >> The main difference between domain->iotlb and dev->iotlb is the way to >> deal with bounce buffer. In the domain->iotlb case, bounce buffer >> needs to be mapped each DMA transfer because we need to get the bounce >> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce >> buffer only needs to be mapped once during initialization, which will >> be used to tell userspace how to do mmap(). >> >> Also, since vhost IOTLB support per mapping token (opauqe), can we use >> that instead of the bounce_pages *? >> >> Sorry, I didn't get you here. Which value do you mean to store in the >> opaque pointer？ >> >> So I would like to have a way to use a single IOTLB for manage all kinds >> of mappings. Two possible ideas: >> >> 1) map bounce page one by one in vduse_dev_map_page(), in >> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then >> for bounce pages, userspace still only need to map it once and we can >> maintain the actual mapping by storing the page or pa in the opaque >> field of IOTLB entry. >> >> Looks like userspace still needs to unmap the old region and map a new >> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. >> >> >> I don't get here. Can you give an example? >> > For example, userspace needs to process two I/O requests (one page per > request). To process the first request, userspace uses > VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap > it. I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum range as far as they are backed by the same fd. In the case of bounce page, it's bascially the whole range of bounce buffer? Thanks > To process the second request, userspace uses VDUSE_IOTLB_GET_FD > ioctl to query the new iova region and map a new region (0 ~ 8192). > Then userspace needs to traverse the list of iova regions and unmap > the old region (0 ~ 4096). Looks like this is a little complex. > > Thanks, > Yongji >
On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/5 3:13 下午, Yongji Xie wrote: > > On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: > >> > >> On 2021/3/5 2:15 下午, Yongji Xie wrote: > >> > >> Sorry if I've asked this before. > >> > >> But what's the reason for maintaing a dedicated IOTLB here? I think we > >> could reuse vduse_dev->iommu since the device can not be used by both > >> virtio and vhost in the same time or use vduse_iova_domain->iotlb for > >> set_map(). > >> > >> The main difference between domain->iotlb and dev->iotlb is the way to > >> deal with bounce buffer. In the domain->iotlb case, bounce buffer > >> needs to be mapped each DMA transfer because we need to get the bounce > >> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > >> buffer only needs to be mapped once during initialization, which will > >> be used to tell userspace how to do mmap(). > >> > >> Also, since vhost IOTLB support per mapping token (opauqe), can we use > >> that instead of the bounce_pages *? > >> > >> Sorry, I didn't get you here. Which value do you mean to store in the > >> opaque pointer？ > >> > >> So I would like to have a way to use a single IOTLB for manage all kinds > >> of mappings. Two possible ideas: > >> > >> 1) map bounce page one by one in vduse_dev_map_page(), in > >> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > >> for bounce pages, userspace still only need to map it once and we can > >> maintain the actual mapping by storing the page or pa in the opaque > >> field of IOTLB entry. > >> > >> Looks like userspace still needs to unmap the old region and map a new > >> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > >> > >> > >> I don't get here. Can you give an example? > >> > > For example, userspace needs to process two I/O requests (one page per > > request). To process the first request, userspace uses > > VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap > > it. > > > I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum > range as far as they are backed by the same fd. > But now the bounce page is mapped one by one. The second page (4096 ~ 8192) might not be mapped when userspace is processing the first request. So the maximum range is 0 ~ 4096 at that time. Thanks, Yongji
On 2021/3/5 3:59 下午, Yongji Xie wrote: > On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: >> >> On 2021/3/5 3:13 下午, Yongji Xie wrote: >>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: >>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: >>>> >>>> Sorry if I've asked this before. >>>> >>>> But what's the reason for maintaing a dedicated IOTLB here? I think we >>>> could reuse vduse_dev->iommu since the device can not be used by both >>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for >>>> set_map(). >>>> >>>> The main difference between domain->iotlb and dev->iotlb is the way to >>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer >>>> needs to be mapped each DMA transfer because we need to get the bounce >>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce >>>> buffer only needs to be mapped once during initialization, which will >>>> be used to tell userspace how to do mmap(). >>>> >>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use >>>> that instead of the bounce_pages *? >>>> >>>> Sorry, I didn't get you here. Which value do you mean to store in the >>>> opaque pointer？ >>>> >>>> So I would like to have a way to use a single IOTLB for manage all kinds >>>> of mappings. Two possible ideas: >>>> >>>> 1) map bounce page one by one in vduse_dev_map_page(), in >>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then >>>> for bounce pages, userspace still only need to map it once and we can >>>> maintain the actual mapping by storing the page or pa in the opaque >>>> field of IOTLB entry. >>>> >>>> Looks like userspace still needs to unmap the old region and map a new >>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. >>>> >>>> >>>> I don't get here. Can you give an example? >>>> >>> For example, userspace needs to process two I/O requests (one page per >>> request). To process the first request, userspace uses >>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap >>> it. >> >> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum >> range as far as they are backed by the same fd. >> > But now the bounce page is mapped one by one. The second page (4096 ~ > 8192) might not be mapped when userspace is processing the first > request. So the maximum range is 0 ~ 4096 at that time. > > Thanks, > Yongji A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return the whole bounce map range which is setup in vduse_dev_map_page()? So my understanding is that usersapce may choose to map all its range via mmap(). So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here 'map' means using multiple itree entries instead of a single one). Then in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) until the range is backed by a different file. With this, there's no userspace visible changes and there's no need for the domain->iotlb? Thanks >
On Mon, Mar 8, 2021 at 11:17 AM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/5 3:59 下午, Yongji Xie wrote: > > On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: > >> > >> On 2021/3/5 3:13 下午, Yongji Xie wrote: > >>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: > >>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: > >>>> > >>>> Sorry if I've asked this before. > >>>> > >>>> But what's the reason for maintaing a dedicated IOTLB here? I think we > >>>> could reuse vduse_dev->iommu since the device can not be used by both > >>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for > >>>> set_map(). > >>>> > >>>> The main difference between domain->iotlb and dev->iotlb is the way to > >>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer > >>>> needs to be mapped each DMA transfer because we need to get the bounce > >>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > >>>> buffer only needs to be mapped once during initialization, which will > >>>> be used to tell userspace how to do mmap(). > >>>> > >>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use > >>>> that instead of the bounce_pages *? > >>>> > >>>> Sorry, I didn't get you here. Which value do you mean to store in the > >>>> opaque pointer？ > >>>> > >>>> So I would like to have a way to use a single IOTLB for manage all kinds > >>>> of mappings. Two possible ideas: > >>>> > >>>> 1) map bounce page one by one in vduse_dev_map_page(), in > >>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > >>>> for bounce pages, userspace still only need to map it once and we can > >>>> maintain the actual mapping by storing the page or pa in the opaque > >>>> field of IOTLB entry. > >>>> > >>>> Looks like userspace still needs to unmap the old region and map a new > >>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > >>>> > >>>> > >>>> I don't get here. Can you give an example? > >>>> > >>> For example, userspace needs to process two I/O requests (one page per > >>> request). To process the first request, userspace uses > >>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap > >>> it. > >> > >> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum > >> range as far as they are backed by the same fd. > >> > > But now the bounce page is mapped one by one. The second page (4096 ~ > > 8192) might not be mapped when userspace is processing the first > > request. So the maximum range is 0 ~ 4096 at that time. > > > > Thanks, > > Yongji > > > A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return > the whole bounce map range which is setup in vduse_dev_map_page()? So my > understanding is that usersapce may choose to map all its range via mmap(). > Yes. > So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here > 'map' means using multiple itree entries instead of a single one). Then > in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) > until the range is backed by a different file. > > With this, there's no userspace visible changes and there's no need for > the domain->iotlb? > In this case, I wonder what range can be obtained if userspace calls VDUSE_IOTLB_GET_FD when the first I/O (e.g. 4K) occurs. [0, 4K] or [0, 64M]? In current implementation, userspace will map [0, 64M]. Thanks, Yongji
On 2021/3/8 11:45 上午, Yongji Xie wrote: > On Mon, Mar 8, 2021 at 11:17 AM Jason Wang <jasowang@redhat.com> wrote: >> >> On 2021/3/5 3:59 下午, Yongji Xie wrote: >>> On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: >>>> On 2021/3/5 3:13 下午, Yongji Xie wrote: >>>>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: >>>>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: >>>>>> >>>>>> Sorry if I've asked this before. >>>>>> >>>>>> But what's the reason for maintaing a dedicated IOTLB here? I think we >>>>>> could reuse vduse_dev->iommu since the device can not be used by both >>>>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for >>>>>> set_map(). >>>>>> >>>>>> The main difference between domain->iotlb and dev->iotlb is the way to >>>>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer >>>>>> needs to be mapped each DMA transfer because we need to get the bounce >>>>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce >>>>>> buffer only needs to be mapped once during initialization, which will >>>>>> be used to tell userspace how to do mmap(). >>>>>> >>>>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use >>>>>> that instead of the bounce_pages *? >>>>>> >>>>>> Sorry, I didn't get you here. Which value do you mean to store in the >>>>>> opaque pointer？ >>>>>> >>>>>> So I would like to have a way to use a single IOTLB for manage all kinds >>>>>> of mappings. Two possible ideas: >>>>>> >>>>>> 1) map bounce page one by one in vduse_dev_map_page(), in >>>>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then >>>>>> for bounce pages, userspace still only need to map it once and we can >>>>>> maintain the actual mapping by storing the page or pa in the opaque >>>>>> field of IOTLB entry. >>>>>> >>>>>> Looks like userspace still needs to unmap the old region and map a new >>>>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. >>>>>> >>>>>> >>>>>> I don't get here. Can you give an example? >>>>>> >>>>> For example, userspace needs to process two I/O requests (one page per >>>>> request). To process the first request, userspace uses >>>>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap >>>>> it. >>>> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum >>>> range as far as they are backed by the same fd. >>>> >>> But now the bounce page is mapped one by one. The second page (4096 ~ >>> 8192) might not be mapped when userspace is processing the first >>> request. So the maximum range is 0 ~ 4096 at that time. >>> >>> Thanks, >>> Yongji >> >> A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return >> the whole bounce map range which is setup in vduse_dev_map_page()? So my >> understanding is that usersapce may choose to map all its range via mmap(). >> > Yes. > >> So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here >> 'map' means using multiple itree entries instead of a single one). Then >> in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) >> until the range is backed by a different file. >> >> With this, there's no userspace visible changes and there's no need for >> the domain->iotlb? >> > In this case, I wonder what range can be obtained if userspace calls > VDUSE_IOTLB_GET_FD when the first I/O (e.g. 4K) occurs. [0, 4K] or [0, > 64M]? In current implementation, userspace will map [0, 64M]. It should still be [0, 64M). Do you see any issue? Thanks > > Thanks, > Yongji >
On Mon, Mar 8, 2021 at 11:52 AM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/8 11:45 上午, Yongji Xie wrote: > > On Mon, Mar 8, 2021 at 11:17 AM Jason Wang <jasowang@redhat.com> wrote: > >> > >> On 2021/3/5 3:59 下午, Yongji Xie wrote: > >>> On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: > >>>> On 2021/3/5 3:13 下午, Yongji Xie wrote: > >>>>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: > >>>>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: > >>>>>> > >>>>>> Sorry if I've asked this before. > >>>>>> > >>>>>> But what's the reason for maintaing a dedicated IOTLB here? I think we > >>>>>> could reuse vduse_dev->iommu since the device can not be used by both > >>>>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for > >>>>>> set_map(). > >>>>>> > >>>>>> The main difference between domain->iotlb and dev->iotlb is the way to > >>>>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer > >>>>>> needs to be mapped each DMA transfer because we need to get the bounce > >>>>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > >>>>>> buffer only needs to be mapped once during initialization, which will > >>>>>> be used to tell userspace how to do mmap(). > >>>>>> > >>>>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use > >>>>>> that instead of the bounce_pages *? > >>>>>> > >>>>>> Sorry, I didn't get you here. Which value do you mean to store in the > >>>>>> opaque pointer？ > >>>>>> > >>>>>> So I would like to have a way to use a single IOTLB for manage all kinds > >>>>>> of mappings. Two possible ideas: > >>>>>> > >>>>>> 1) map bounce page one by one in vduse_dev_map_page(), in > >>>>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > >>>>>> for bounce pages, userspace still only need to map it once and we can > >>>>>> maintain the actual mapping by storing the page or pa in the opaque > >>>>>> field of IOTLB entry. > >>>>>> > >>>>>> Looks like userspace still needs to unmap the old region and map a new > >>>>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > >>>>>> > >>>>>> > >>>>>> I don't get here. Can you give an example? > >>>>>> > >>>>> For example, userspace needs to process two I/O requests (one page per > >>>>> request). To process the first request, userspace uses > >>>>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap > >>>>> it. > >>>> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum > >>>> range as far as they are backed by the same fd. > >>>> > >>> But now the bounce page is mapped one by one. The second page (4096 ~ > >>> 8192) might not be mapped when userspace is processing the first > >>> request. So the maximum range is 0 ~ 4096 at that time. > >>> > >>> Thanks, > >>> Yongji > >> > >> A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return > >> the whole bounce map range which is setup in vduse_dev_map_page()? So my > >> understanding is that usersapce may choose to map all its range via mmap(). > >> > > Yes. > > > >> So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here > >> 'map' means using multiple itree entries instead of a single one). Then > >> in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) > >> until the range is backed by a different file. > >> > >> With this, there's no userspace visible changes and there's no need for > >> the domain->iotlb? > >> > > In this case, I wonder what range can be obtained if userspace calls > > VDUSE_IOTLB_GET_FD when the first I/O (e.g. 4K) occurs. [0, 4K] or [0, > > 64M]? In current implementation, userspace will map [0, 64M]. > > > It should still be [0, 64M). Do you see any issue? > Does it mean we still need to map the whole bounce buffer into itree (dev->iommu) at initialization? Thanks, Yongji
On 2021/3/8 1:05 下午, Yongji Xie wrote: > On Mon, Mar 8, 2021 at 11:52 AM Jason Wang <jasowang@redhat.com> wrote: >> >> On 2021/3/8 11:45 上午, Yongji Xie wrote: >>> On Mon, Mar 8, 2021 at 11:17 AM Jason Wang <jasowang@redhat.com> wrote: >>>> On 2021/3/5 3:59 下午, Yongji Xie wrote: >>>>> On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: >>>>>> On 2021/3/5 3:13 下午, Yongji Xie wrote: >>>>>>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: >>>>>>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: >>>>>>>> >>>>>>>> Sorry if I've asked this before. >>>>>>>> >>>>>>>> But what's the reason for maintaing a dedicated IOTLB here? I think we >>>>>>>> could reuse vduse_dev->iommu since the device can not be used by both >>>>>>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for >>>>>>>> set_map(). >>>>>>>> >>>>>>>> The main difference between domain->iotlb and dev->iotlb is the way to >>>>>>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer >>>>>>>> needs to be mapped each DMA transfer because we need to get the bounce >>>>>>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce >>>>>>>> buffer only needs to be mapped once during initialization, which will >>>>>>>> be used to tell userspace how to do mmap(). >>>>>>>> >>>>>>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use >>>>>>>> that instead of the bounce_pages *? >>>>>>>> >>>>>>>> Sorry, I didn't get you here. Which value do you mean to store in the >>>>>>>> opaque pointer？ >>>>>>>> >>>>>>>> So I would like to have a way to use a single IOTLB for manage all kinds >>>>>>>> of mappings. Two possible ideas: >>>>>>>> >>>>>>>> 1) map bounce page one by one in vduse_dev_map_page(), in >>>>>>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then >>>>>>>> for bounce pages, userspace still only need to map it once and we can >>>>>>>> maintain the actual mapping by storing the page or pa in the opaque >>>>>>>> field of IOTLB entry. >>>>>>>> >>>>>>>> Looks like userspace still needs to unmap the old region and map a new >>>>>>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. >>>>>>>> >>>>>>>> >>>>>>>> I don't get here. Can you give an example? >>>>>>>> >>>>>>> For example, userspace needs to process two I/O requests (one page per >>>>>>> request). To process the first request, userspace uses >>>>>>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap >>>>>>> it. >>>>>> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum >>>>>> range as far as they are backed by the same fd. >>>>>> >>>>> But now the bounce page is mapped one by one. The second page (4096 ~ >>>>> 8192) might not be mapped when userspace is processing the first >>>>> request. So the maximum range is 0 ~ 4096 at that time. >>>>> >>>>> Thanks, >>>>> Yongji >>>> A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return >>>> the whole bounce map range which is setup in vduse_dev_map_page()? So my >>>> understanding is that usersapce may choose to map all its range via mmap(). >>>> >>> Yes. >>> >>>> So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here >>>> 'map' means using multiple itree entries instead of a single one). Then >>>> in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) >>>> until the range is backed by a different file. >>>> >>>> With this, there's no userspace visible changes and there's no need for >>>> the domain->iotlb? >>>> >>> In this case, I wonder what range can be obtained if userspace calls >>> VDUSE_IOTLB_GET_FD when the first I/O (e.g. 4K) occurs. [0, 4K] or [0, >>> 64M]? In current implementation, userspace will map [0, 64M]. >> >> It should still be [0, 64M). Do you see any issue? >> > Does it mean we still need to map the whole bounce buffer into itree > (dev->iommu) at initialization? It's your choice I think, the point is to use a single IOTLB for maintaining mappings of all types of pages (bounce, coherent, or shared). Thanks > > Thanks, > Yongji >
On Mon, Mar 8, 2021 at 3:04 PM Jason Wang <jasowang@redhat.com> wrote: > > > On 2021/3/8 1:05 下午, Yongji Xie wrote: > > On Mon, Mar 8, 2021 at 11:52 AM Jason Wang <jasowang@redhat.com> wrote: > >> > >> On 2021/3/8 11:45 上午, Yongji Xie wrote: > >>> On Mon, Mar 8, 2021 at 11:17 AM Jason Wang <jasowang@redhat.com> wrote: > >>>> On 2021/3/5 3:59 下午, Yongji Xie wrote: > >>>>> On Fri, Mar 5, 2021 at 3:27 PM Jason Wang <jasowang@redhat.com> wrote: > >>>>>> On 2021/3/5 3:13 下午, Yongji Xie wrote: > >>>>>>> On Fri, Mar 5, 2021 at 2:52 PM Jason Wang <jasowang@redhat.com> wrote: > >>>>>>>> On 2021/3/5 2:15 下午, Yongji Xie wrote: > >>>>>>>> > >>>>>>>> Sorry if I've asked this before. > >>>>>>>> > >>>>>>>> But what's the reason for maintaing a dedicated IOTLB here? I think we > >>>>>>>> could reuse vduse_dev->iommu since the device can not be used by both > >>>>>>>> virtio and vhost in the same time or use vduse_iova_domain->iotlb for > >>>>>>>> set_map(). > >>>>>>>> > >>>>>>>> The main difference between domain->iotlb and dev->iotlb is the way to > >>>>>>>> deal with bounce buffer. In the domain->iotlb case, bounce buffer > >>>>>>>> needs to be mapped each DMA transfer because we need to get the bounce > >>>>>>>> pages by an IOVA during DMA unmapping. In the dev->iotlb case, bounce > >>>>>>>> buffer only needs to be mapped once during initialization, which will > >>>>>>>> be used to tell userspace how to do mmap(). > >>>>>>>> > >>>>>>>> Also, since vhost IOTLB support per mapping token (opauqe), can we use > >>>>>>>> that instead of the bounce_pages *? > >>>>>>>> > >>>>>>>> Sorry, I didn't get you here. Which value do you mean to store in the > >>>>>>>> opaque pointer？ > >>>>>>>> > >>>>>>>> So I would like to have a way to use a single IOTLB for manage all kinds > >>>>>>>> of mappings. Two possible ideas: > >>>>>>>> > >>>>>>>> 1) map bounce page one by one in vduse_dev_map_page(), in > >>>>>>>> VDUSE_IOTLB_GET_FD, try to merge the result if we had the same fd. Then > >>>>>>>> for bounce pages, userspace still only need to map it once and we can > >>>>>>>> maintain the actual mapping by storing the page or pa in the opaque > >>>>>>>> field of IOTLB entry. > >>>>>>>> > >>>>>>>> Looks like userspace still needs to unmap the old region and map a new > >>>>>>>> region (size is changed) with the fd in each VDUSE_IOTLB_GET_FD ioctl. > >>>>>>>> > >>>>>>>> > >>>>>>>> I don't get here. Can you give an example? > >>>>>>>> > >>>>>>> For example, userspace needs to process two I/O requests (one page per > >>>>>>> request). To process the first request, userspace uses > >>>>>>> VDUSE_IOTLB_GET_FD ioctl to query the iova region (0 ~ 4096) and mmap > >>>>>>> it. > >>>>>> I think in this case we should let VDUSE_IOTLB_GET_FD return the maximum > >>>>>> range as far as they are backed by the same fd. > >>>>>> > >>>>> But now the bounce page is mapped one by one. The second page (4096 ~ > >>>>> 8192) might not be mapped when userspace is processing the first > >>>>> request. So the maximum range is 0 ~ 4096 at that time. > >>>>> > >>>>> Thanks, > >>>>> Yongji > >>>> A question, if I read the code correctly, VDUSE_IOTLB_GET_FD will return > >>>> the whole bounce map range which is setup in vduse_dev_map_page()? So my > >>>> understanding is that usersapce may choose to map all its range via mmap(). > >>>> > >>> Yes. > >>> > >>>> So if we 'map' bounce page one by one in vduse_dev_map_page(). (Here > >>>> 'map' means using multiple itree entries instead of a single one). Then > >>>> in the VDUSE_IOTLB_GET_FD we can keep traversing itree (dev->iommu) > >>>> until the range is backed by a different file. > >>>> > >>>> With this, there's no userspace visible changes and there's no need for > >>>> the domain->iotlb? > >>>> > >>> In this case, I wonder what range can be obtained if userspace calls > >>> VDUSE_IOTLB_GET_FD when the first I/O (e.g. 4K) occurs. [0, 4K] or [0, > >>> 64M]? In current implementation, userspace will map [0, 64M]. > >> > >> It should still be [0, 64M). Do you see any issue? > >> > > Does it mean we still need to map the whole bounce buffer into itree > > (dev->iommu) at initialization? > > > It's your choice I think, the point is to use a single IOTLB for > maintaining mappings of all types of pages (bounce, coherent, or shared). > OK, got it. Thanks, Yongji
diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c new file mode 100644 index 000000000000..9285d430d486 --- /dev/null +++ b/drivers/vdpa/vdpa_user/iova_domain.c @@ -0,0 +1,486 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * MMU-based IOMMU implementation + * + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. + * + * Author: Xie Yongji <xieyongji@bytedance.com> + * + */ + +#include <linux/slab.h> +#include <linux/file.h> +#include <linux/anon_inodes.h> +#include <linux/highmem.h> + +#include "iova_domain.h" + +#define IOVA_START_PFN 1 +#define IOVA_ALLOC_ORDER 12 +#define IOVA_ALLOC_SIZE (1 << IOVA_ALLOC_ORDER) + +static inline struct page * +vduse_domain_get_bounce_page(struct vduse_iova_domain *domain, u64 iova) +{ + u64 index = iova >> PAGE_SHIFT; + + return domain->bounce_pages[index]; +} + +static inline void +vduse_domain_set_bounce_page(struct vduse_iova_domain *domain, + u64 iova, struct page *page) +{ + u64 index = iova >> PAGE_SHIFT; + + domain->bounce_pages[index] = page; +} + +static enum dma_data_direction perm_to_dir(int perm) +{ + enum dma_data_direction dir; + + switch (perm) { + case VHOST_MAP_WO: + dir = DMA_FROM_DEVICE; + break; + case VHOST_MAP_RO: + dir = DMA_TO_DEVICE; + break; + case VHOST_MAP_RW: + dir = DMA_BIDIRECTIONAL; + break; + default: + break; + } + + return dir; +} + +static int dir_to_perm(enum dma_data_direction dir) +{ + int perm = -EFAULT; + + switch (dir) { + case DMA_FROM_DEVICE: + perm = VHOST_MAP_WO; + break; + case DMA_TO_DEVICE: + perm = VHOST_MAP_RO; + break; + case DMA_BIDIRECTIONAL: + perm = VHOST_MAP_RW; + break; + default: + break; + } + + return perm; +} + +static void do_bounce(phys_addr_t orig, void *addr, size_t size, + enum dma_data_direction dir) +{ + unsigned long pfn = PFN_DOWN(orig); + + if (PageHighMem(pfn_to_page(pfn))) { + unsigned int offset = offset_in_page(orig); + char *buffer; + unsigned int sz = 0; + unsigned long flags; + + while (size) { + sz = min_t(size_t, PAGE_SIZE - offset, size); + + local_irq_save(flags); + buffer = kmap_atomic(pfn_to_page(pfn)); + if (dir == DMA_TO_DEVICE) + memcpy(addr, buffer + offset, sz); + else + memcpy(buffer + offset, addr, sz); + kunmap_atomic(buffer); + local_irq_restore(flags); + + size -= sz; + pfn++; + addr += sz; + offset = 0; + } + } else if (dir == DMA_TO_DEVICE) { + memcpy(addr, phys_to_virt(orig), size); + } else { + memcpy(phys_to_virt(orig), addr, size); + } +} + +static struct page * +vduse_domain_get_mapping_page(struct vduse_iova_domain *domain, u64 iova) +{ + u64 start = iova & PAGE_MASK; + u64 last = start + PAGE_SIZE - 1; + struct vhost_iotlb_map *map; + struct page *page = NULL; + + spin_lock(&domain->iotlb_lock); + map = vhost_iotlb_itree_first(domain->iotlb, start, last); + if (!map) + goto out; + + page = pfn_to_page((map->addr + iova - map->start) >> PAGE_SHIFT); + get_page(page); +out: + spin_unlock(&domain->iotlb_lock); + + return page; +} + +static struct page * +vduse_domain_alloc_bounce_page(struct vduse_iova_domain *domain, u64 iova) +{ + u64 start = iova & PAGE_MASK; + u64 last = start + PAGE_SIZE - 1; + struct vhost_iotlb_map *map; + struct page *page = NULL, *new_page = alloc_page(GFP_KERNEL); + + if (!new_page) + return NULL; + + spin_lock(&domain->iotlb_lock); + if (!vhost_iotlb_itree_first(domain->iotlb, start, last)) { + __free_page(new_page); + goto out; + } + page = vduse_domain_get_bounce_page(domain, iova); + if (page) { + get_page(page); + __free_page(new_page); + goto out; + } + vduse_domain_set_bounce_page(domain, iova, new_page); + get_page(new_page); + page = new_page; + + for (map = vhost_iotlb_itree_first(domain->iotlb, start, last); map; + map = vhost_iotlb_itree_next(map, start, last)) { + unsigned int src_offset = 0, dst_offset = 0; + phys_addr_t src; + void *dst; + size_t sz; + + if (perm_to_dir(map->perm) == DMA_FROM_DEVICE) + continue; + + if (start > map->start) + src_offset = start - map->start; + else + dst_offset = map->start - start; + + src = map->addr + src_offset; + dst = page_address(page) + dst_offset; + sz = min_t(size_t, map->size - src_offset, + PAGE_SIZE - dst_offset); + do_bounce(src, dst, sz, DMA_TO_DEVICE); + } +out: + spin_unlock(&domain->iotlb_lock); + + return page; +} + +static void +vduse_domain_free_bounce_pages(struct vduse_iova_domain *domain, + u64 iova, size_t size) +{ + struct page *page; + + spin_lock(&domain->iotlb_lock); + if (WARN_ON(vhost_iotlb_itree_first(domain->iotlb, iova, + iova + size - 1))) + goto out; + + while (size > 0) { + page = vduse_domain_get_bounce_page(domain, iova); + if (page) { + vduse_domain_set_bounce_page(domain, iova, NULL); + __free_page(page); + } + size -= PAGE_SIZE; + iova += PAGE_SIZE; + } +out: + spin_unlock(&domain->iotlb_lock); +} + +static void vduse_domain_bounce(struct vduse_iova_domain *domain, + dma_addr_t iova, phys_addr_t orig, + size_t size, enum dma_data_direction dir) +{ + unsigned int offset = offset_in_page(iova); + + while (size) { + struct page *p = vduse_domain_get_bounce_page(domain, iova); + size_t sz = min_t(size_t, PAGE_SIZE - offset, size); + + WARN_ON(!p && dir == DMA_FROM_DEVICE); + + if (p) + do_bounce(orig, page_address(p) + offset, sz, dir); + + size -= sz; + orig += sz; + iova += sz; + offset = 0; + } +} + +static dma_addr_t vduse_domain_alloc_iova(struct iova_domain *iovad, + unsigned long size, unsigned long limit) +{ + unsigned long shift = iova_shift(iovad); + unsigned long iova_len = iova_align(iovad, size) >> shift; + unsigned long iova_pfn; + + if (iova_len < (1 << (IOVA_RANGE_CACHE_MAX_SIZE - 1))) + iova_len = roundup_pow_of_two(iova_len); + iova_pfn = alloc_iova_fast(iovad, iova_len, limit >> shift, true); + + return iova_pfn << shift; +} + +static void vduse_domain_free_iova(struct iova_domain *iovad, + dma_addr_t iova, size_t size) +{ + unsigned long shift = iova_shift(iovad); + unsigned long iova_len = iova_align(iovad, size) >> shift; + + free_iova_fast(iovad, iova >> shift, iova_len); +} + +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, + struct page *page, unsigned long offset, + size_t size, enum dma_data_direction dir, + unsigned long attrs) +{ + struct iova_domain *iovad = &domain->stream_iovad; + unsigned long limit = domain->bounce_size - 1; + phys_addr_t pa = page_to_phys(page) + offset; + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); + int ret; + + if (!iova) + return DMA_MAPPING_ERROR; + + spin_lock(&domain->iotlb_lock); + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, + (u64)iova + size - 1, + pa, dir_to_perm(dir)); + spin_unlock(&domain->iotlb_lock); + if (ret) { + vduse_domain_free_iova(iovad, iova, size); + return DMA_MAPPING_ERROR; + } + if (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL) + vduse_domain_bounce(domain, iova, pa, size, DMA_TO_DEVICE); + + return iova; +} + +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, + dma_addr_t dma_addr, size_t size, + enum dma_data_direction dir, unsigned long attrs) +{ + struct iova_domain *iovad = &domain->stream_iovad; + struct vhost_iotlb_map *map; + phys_addr_t pa; + + spin_lock(&domain->iotlb_lock); + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, + (u64)dma_addr + size - 1); + if (WARN_ON(!map)) { + spin_unlock(&domain->iotlb_lock); + return; + } + pa = map->addr; + vhost_iotlb_map_free(domain->iotlb, map); + spin_unlock(&domain->iotlb_lock); + + if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) + vduse_domain_bounce(domain, dma_addr, pa, + size, DMA_FROM_DEVICE); + + vduse_domain_free_iova(iovad, dma_addr, size); +} + +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, + size_t size, dma_addr_t *dma_addr, + gfp_t flag, unsigned long attrs) +{ + struct iova_domain *iovad = &domain->consistent_iovad; + unsigned long limit = domain->iova_limit; + dma_addr_t iova = vduse_domain_alloc_iova(iovad, size, limit); + void *orig = alloc_pages_exact(size, flag); + int ret; + + if (!iova || !orig) + goto err; + + spin_lock(&domain->iotlb_lock); + ret = vhost_iotlb_add_range(domain->iotlb, (u64)iova, + (u64)iova + size - 1, + virt_to_phys(orig), VHOST_MAP_RW); + spin_unlock(&domain->iotlb_lock); + if (ret) + goto err; + + *dma_addr = iova; + + return orig; +err: + *dma_addr = DMA_MAPPING_ERROR; + if (orig) + free_pages_exact(orig, size); + if (iova) + vduse_domain_free_iova(iovad, iova, size); + + return NULL; +} + +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, + void *vaddr, dma_addr_t dma_addr, + unsigned long attrs) +{ + struct iova_domain *iovad = &domain->consistent_iovad; + struct vhost_iotlb_map *map; + phys_addr_t pa; + + spin_lock(&domain->iotlb_lock); + map = vhost_iotlb_itree_first(domain->iotlb, (u64)dma_addr, + (u64)dma_addr + size - 1); + if (WARN_ON(!map)) { + spin_unlock(&domain->iotlb_lock); + return; + } + pa = map->addr; + vhost_iotlb_map_free(domain->iotlb, map); + spin_unlock(&domain->iotlb_lock); + + vduse_domain_free_iova(iovad, dma_addr, size); + free_pages_exact(phys_to_virt(pa), size); +} + +static vm_fault_t vduse_domain_mmap_fault(struct vm_fault *vmf) +{ + struct vduse_iova_domain *domain = vmf->vma->vm_private_data; + unsigned long iova = vmf->pgoff << PAGE_SHIFT; + struct page *page; + + if (!domain) + return VM_FAULT_SIGBUS; + + if (iova < domain->bounce_size) + page = vduse_domain_alloc_bounce_page(domain, iova); + else + page = vduse_domain_get_mapping_page(domain, iova); + + if (!page) + return VM_FAULT_SIGBUS; + + vmf->page = page; + + return 0; +} + +static const struct vm_operations_struct vduse_domain_mmap_ops = { + .fault = vduse_domain_mmap_fault, +}; + +static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma) +{ + struct vduse_iova_domain *domain = file->private_data; + + vma->vm_flags |= VM_DONTDUMP | VM_DONTEXPAND; + vma->vm_private_data = domain; + vma->vm_ops = &vduse_domain_mmap_ops; + + return 0; +} + +static int vduse_domain_release(struct inode *inode, struct file *file) +{ + struct vduse_iova_domain *domain = file->private_data; + + vduse_domain_free_bounce_pages(domain, 0, domain->bounce_size); + put_iova_domain(&domain->stream_iovad); + put_iova_domain(&domain->consistent_iovad); + vhost_iotlb_free(domain->iotlb); + vfree(domain->bounce_pages); + kfree(domain); + + return 0; +} + +static const struct file_operations vduse_domain_fops = { + .mmap = vduse_domain_mmap, + .release = vduse_domain_release, +}; + +void vduse_domain_destroy(struct vduse_iova_domain *domain) +{ + fput(domain->file); +} + +struct vduse_iova_domain * +vduse_domain_create(unsigned long iova_limit, size_t bounce_size) +{ + struct vduse_iova_domain *domain; + struct file *file; + unsigned long bounce_pfns = PAGE_ALIGN(bounce_size) >> PAGE_SHIFT; + + if (iova_limit <= bounce_size) + return NULL; + + domain = kzalloc(sizeof(*domain), GFP_KERNEL); + if (!domain) + return NULL; + + domain->iotlb = vhost_iotlb_alloc(0, 0); + if (!domain->iotlb) + goto err_iotlb; + + domain->iova_limit = iova_limit; + domain->bounce_size = PAGE_ALIGN(bounce_size); + domain->bounce_pages = vzalloc(bounce_pfns * sizeof(struct page *)); + if (!domain->bounce_pages) + goto err_page; + + file = anon_inode_getfile("[vduse-domain]", &vduse_domain_fops, + domain, O_RDWR); + if (IS_ERR(file)) + goto err_file; + + domain->file = file; + spin_lock_init(&domain->iotlb_lock); + init_iova_domain(&domain->stream_iovad, + IOVA_ALLOC_SIZE, IOVA_START_PFN); + init_iova_domain(&domain->consistent_iovad, + PAGE_SIZE, bounce_pfns); + + return domain; +err_file: + vfree(domain->bounce_pages); +err_page: + vhost_iotlb_free(domain->iotlb); +err_iotlb: + kfree(domain); + return NULL; +} + +int vduse_domain_init(void) +{ + return iova_cache_get(); +} + +void vduse_domain_exit(void) +{ + iova_cache_put(); +} diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h new file mode 100644 index 000000000000..9c85d8346626 --- /dev/null +++ b/drivers/vdpa/vdpa_user/iova_domain.h @@ -0,0 +1,61 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * MMU-based IOMMU implementation + * + * Copyright (C) 2020 Bytedance Inc. and/or its affiliates. All rights reserved. + * + * Author: Xie Yongji <xieyongji@bytedance.com> + * + */ + +#ifndef _VDUSE_IOVA_DOMAIN_H +#define _VDUSE_IOVA_DOMAIN_H + +#include <linux/iova.h> +#include <linux/dma-mapping.h> +#include <linux/vhost_iotlb.h> + +struct vduse_iova_domain { + struct iova_domain stream_iovad; + struct iova_domain consistent_iovad; + struct page **bounce_pages; + size_t bounce_size; + unsigned long iova_limit; + struct vhost_iotlb *iotlb; + spinlock_t iotlb_lock; + struct file *file; +}; + +static inline struct file * +vduse_domain_file(struct vduse_iova_domain *domain) +{ + return domain->file; +} + +dma_addr_t vduse_domain_map_page(struct vduse_iova_domain *domain, + struct page *page, unsigned long offset, + size_t size, enum dma_data_direction dir, + unsigned long attrs); + +void vduse_domain_unmap_page(struct vduse_iova_domain *domain, + dma_addr_t dma_addr, size_t size, + enum dma_data_direction dir, unsigned long attrs); + +void *vduse_domain_alloc_coherent(struct vduse_iova_domain *domain, + size_t size, dma_addr_t *dma_addr, + gfp_t flag, unsigned long attrs); + +void vduse_domain_free_coherent(struct vduse_iova_domain *domain, size_t size, + void *vaddr, dma_addr_t dma_addr, + unsigned long attrs); + +void vduse_domain_destroy(struct vduse_iova_domain *domain); + +struct vduse_iova_domain *vduse_domain_create(unsigned long iova_limit, + size_t bounce_size); + +int vduse_domain_init(void); + +void vduse_domain_exit(void); + +#endif /* _VDUSE_IOVA_DOMAIN_H */
This implements a MMU-based IOMMU driver to support mapping kernel dma buffer into userspace. The basic idea behind it is treating MMU (VA->PA) as IOMMU (IOVA->PA). The driver will set up MMU mapping instead of IOMMU mapping for the DMA transfer so that the userspace process is able to use its virtual address to access the dma buffer in kernel. And to avoid security issue, a bounce-buffering mechanism is introduced to prevent userspace accessing the original buffer directly. Signed-off-by: Xie Yongji <xieyongji@bytedance.com> --- drivers/vdpa/vdpa_user/iova_domain.c | 486 +++++++++++++++++++++++++++++++++++ drivers/vdpa/vdpa_user/iova_domain.h | 61 +++++ 2 files changed, 547 insertions(+) create mode 100644 drivers/vdpa/vdpa_user/iova_domain.c create mode 100644 drivers/vdpa/vdpa_user/iova_domain.h