This is the fourteenth version of the patch set. Changelog:
v14: * Check return value of dma_fence_wait() * Fix a dma-buf leak in ib_umem_dmabuf_get() * Fix return value type cast for ib_umem_dmabuf_get() * Return -EOPNOTSUPP instead of -EINVAL for unimplemented functions * Remove an unnecessary use of unlikely() * Remove left-over commit message resulted from rebase
v13: https://www.spinics.net/lists/linux-rdma/msg98227.html * Rebase to the latest linux-rdma 'for-next' branch (5.10.0-rc6+) * Check for device on-demand paging capability at the entry point of the new verbs command to avoid calling device's reg_user_mr_dmabuf() method when CONFIG_INFINIBAND_ON_DEMAND_PAGING is diabled.
v12: https://www.spinics.net/lists/linux-rdma/msg97943.html * Move the prototype of function ib_umem_dmabuf_release() to ib_umem.h and remove umem_dmabuf.h * Break a line that is too long
v11: https://www.spinics.net/lists/linux-rdma/msg97860.html * Rework the parameter checking code inside ib_umem_dmabuf_get() * Fix incorrect error handling in the new verbs command handler * Put a duplicated code sequence for checking iova and setting page size into a function * In the invalidation callback, check for if the buffer has been mapped and thus the presence of a valid driver mr is ensured * The patch that checks for dma_virt_ops is dropped because it is no longer needed * The patch that documents that dma-buf size is fixed has landed at: https://cgit.freedesktop.org/drm/drm-misc/commit/?id=476b485be03c and thus is no longer included here * The matching user space patch set is sent separately
v10: https://www.spinics.net/lists/linux-rdma/msg97483.html * Don't map the pages in ib_umem_dmabuf_get(); use the size information of the dma-buf object to validate the umem size instead * Use PAGE_SIZE directly instead of use ib_umem_find_best_pgsz() when the MR is created since the pages have not been mapped yet and dma-buf requires PAGE_SIZE anyway * Always call mlx5_umem_find_best_pgsz() after mapping the pages to verify that the page size requirement is satisfied * Add a patch to document that dma-buf size is fixed
v9: https://www.spinics.net/lists/linux-rdma/msg97432.html * Clean up the code for sg list in-place modification * Prevent dma-buf pages from being mapped multiple times * Map the pages in ib_umem_dmabuf_get() so that inproper values of address/length/iova can be caught early * Check for unsupported flags in the new uverbs command * Add missing uverbs_finalize_uobj_create() * Sort uverbs objects by name * Fix formating issue -- unnecessary alignment of '=' * Unmap pages in mlx5_ib_fence_dmabuf_mr() * Remove address range checking from pagefault_dmabuf_mr()
v8: https://www.spinics.net/lists/linux-rdma/msg97370.html * Modify the dma-buf sg list in place to get a proper umem sg list and restore it before calling dma_buf_unmap_attachment() * Validate the umem sg list with ib_umem_find_best_pgsz() * Remove the logic for slicing the sg list at runtime
v7: https://www.spinics.net/lists/linux-rdma/msg97297.html * Rebase on top of latest mlx5 MR patch series * Slice dma-buf sg list at runtime instead of creating a new list * Preload the buffer page mapping when the MR is created * Move the 'dma_virt_ops' check into dma_buf_dynamic_attach()
v6: https://www.spinics.net/lists/linux-rdma/msg96923.html * Move the dma-buf invalidation callback from the core to the device driver * Move mapping update from work queue to pagefault handler * Add dma-buf based MRs to the xarray of mmkeys so that the pagefault handler can be reached * Update the new driver method and uverbs command signature by changing the paramter 'addr' to 'offset' * Modify the sg list returned from dma_buf_map_attachment() based on the parameters 'offset' and 'length' * Don't import dma-buf if 'dma_virt_ops' is used by the dma device * The patch that clarifies dma-buf sg lists alignment has landed at https://cgit.freedesktop.org/drm/drm-misc/commit/?id=ac80cd17a615 and thus is no longer included with this set
v5: https://www.spinics.net/lists/linux-rdma/msg96786.html * Fix a few warnings reported by kernel test robot: - no previous prototype for function 'ib_umem_dmabuf_release' - no previous prototype for function 'ib_umem_dmabuf_map_pages' - comparison of distinct pointer types in 'check_add_overflow' * Add comment for the wait between getting the dma-buf sg tagle and updating the NIC page table
v4: https://www.spinics.net/lists/linux-rdma/msg96767.html * Add a new ib_device method reg_user_mr_dmabuf() instead of expanding the existing method reg_user_mr() * Use a separate code flow for dma-buf instead of adding special cases to the ODP memory region code path * In invalidation callback, new mapping is updated as whole using work queue instead of being updated in page granularity in the page fault handler * Use dma_resv_get_excl() and dma_fence_wait() to ensure the content of the pages have been moved to the new location before the new mapping is programmed into the NIC * Add code to the ODP page fault handler to check the mapping status * The new access flag added in v3 is removed. * The checking for on-demand paging support in the new uverbs command is removed because it is implied by implementing the new ib_device method * Clarify that dma-buf sg lists are page aligned
v3: https://www.spinics.net/lists/linux-rdma/msg96330.html * Use dma_buf_dynamic_attach() instead of dma_buf_attach() * Use on-demand paging mechanism to avoid pinning the GPU memory * Instead of adding a new parameter to the device method for memory registration, pass all the attributes including the file descriptor as a structure * Define a new access flag for dma-buf based memory region * Check for on-demand paging support in the new uverbs command
v2: https://www.spinics.net/lists/linux-rdma/msg93643.html * The Kconfig option is removed. There is no dependence issue since dma-buf driver is always enabled. * The declaration of new data structure and functions is reorganized to minimize the visibility of the changes. * The new uverbs command now goes through ioctl() instead of write(). * The rereg functionality is removed. * Instead of adding new device method for dma-buf specific registration, existing method is extended to accept an extra parameter. * The correct function is now used for address range checking.
v1: https://www.spinics.net/lists/linux-rdma/msg90720.html * The initial patch set * Implement core functions for importing and mapping dma-buf * Use dma-buf static attach interface * Add two ib_device methods reg_user_mr_fd() and rereg_user_mr_fd() * Add two uverbs commands via the write() interface * Add Kconfig option * Add dma-buf support to mlx5 device
When enabled, an RDMA capable NIC can perform peer-to-peer transactions over PCIe to access the local memory located on another device. This can often lead to better performance than using a system memory buffer for RDMA and copying data between the buffer and device memory.
Current kernel RDMA stack uses get_user_pages() to pin the physical pages backing the user buffer and uses dma_map_sg_attrs() to get the dma addresses for memory access. This usually doesn't work for peer device memory due to the lack of associated page structures.
Several mechanisms exist today to facilitate device memory access.
ZONE_DEVICE is a new zone for device memory in the memory management subsystem. It allows pages from device memory being described with specialized page structures, but what can be done with these page structures may be different from system memory. ZONE_DEVICE is further specialized into multiple memory types, such as one type for PCI p2pmem/p2pdma and one type for HMM.
PCI p2pmem/p2pdma uses ZONE_DEVICE to represent device memory residing in a PCI BAR and provides a set of calls to publish, discover, allocate, and map such memory for peer-to-peer transactions. One feature of the API is that the buffer is allocated by the side that does the DMA transfer. This works well with the storage usage case, but is awkward with GPU-NIC communication, where typically the buffer is allocated by the GPU driver rather than the NIC driver.
Heterogeneous Memory Management (HMM) utilizes mmu_interval_notifier and ZONE_DEVICE to support shared virtual address space and page migration between system memory and device memory. HMM doesn't support pinning device memory because pages located on device must be able to migrate to system memory when accessed by CPU. Peer-to-peer access is currently not supported by HMM.
Dma-buf is a standard mechanism for sharing buffers among different device drivers. The buffer to be shared is exported by the owning driver and imported by the driver that wants to use it. The exporter provides a set of ops that the importer can call to pin and map the buffer. In addition, a file descriptor can be associated with a dma- buf object as the handle that can be passed to user space.
This patch series adds dma-buf importer role to the RDMA driver in attempt to support RDMA using device memory such as GPU VRAM. Dma-buf is chosen for a few reasons: first, the API is relatively simple and allows a lot of flexibility in implementing the buffer manipulation ops. Second, it doesn't require page structure. Third, dma-buf is already supported in many GPU drivers. However, we are aware that existing GPU drivers don't allow pinning device memory via the dma-buf interface. Pinning would simply cause the backing storage to migrate to system RAM. True peer-to-peer access is only possible using dynamic attach, which requires on-demand paging support from the NIC to work. For this reason, this series only works with ODP capable NICs.
This series consists of four patches. The first patch adds the common code for importing dma-buf from a file descriptor and mapping the dma-buf pages. Patch 2 add the new driver method reg_user_mr_dmabuf(). Patch 3 adds a new uverbs command for registering dma-buf based memory region. Patch 4 adds dma-buf support to the mlx5 driver.
Related user space RDMA library changes are provided as a separate patch series.
Jianxin Xiong (4): RDMA/umem: Support importing dma-buf as user memory region RDMA/core: Add device method for registering dma-buf based memory region RDMA/uverbs: Add uverbs command for dma-buf based MR registration RDMA/mlx5: Support dma-buf based userspace memory region
drivers/infiniband/core/Makefile | 2 +- drivers/infiniband/core/device.c | 1 + drivers/infiniband/core/umem.c | 3 + drivers/infiniband/core/umem_dmabuf.c | 174 ++++++++++++++++++++++++++ drivers/infiniband/core/uverbs_std_types_mr.c | 117 ++++++++++++++++- drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++ drivers/infiniband/hw/mlx5/mr.c | 128 ++++++++++++++++++- drivers/infiniband/hw/mlx5/odp.c | 86 ++++++++++++- include/rdma/ib_umem.h | 47 ++++++- include/rdma/ib_verbs.h | 6 +- include/uapi/rdma/ib_user_ioctl_cmds.h | 14 +++ 12 files changed, 582 insertions(+), 16 deletions(-) create mode 100644 drivers/infiniband/core/umem_dmabuf.c
Dma-buf is a standard cross-driver buffer sharing mechanism that can be used to support peer-to-peer access from RDMA devices.
Device memory exported via dma-buf is associated with a file descriptor. This is passed to the user space as a property associated with the buffer allocation. When the buffer is registered as a memory region, the file descriptor is passed to the RDMA driver along with other parameters.
Implement the common code for importing dma-buf object and mapping dma-buf pages.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch --- drivers/infiniband/core/Makefile | 2 +- drivers/infiniband/core/umem.c | 3 + drivers/infiniband/core/umem_dmabuf.c | 174 ++++++++++++++++++++++++++++++++++ include/rdma/ib_umem.h | 47 ++++++++- 4 files changed, 222 insertions(+), 4 deletions(-) create mode 100644 drivers/infiniband/core/umem_dmabuf.c
diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile index ccf2670..8ab4eea 100644 --- a/drivers/infiniband/core/Makefile +++ b/drivers/infiniband/core/Makefile @@ -40,5 +40,5 @@ ib_uverbs-y := uverbs_main.o uverbs_cmd.o uverbs_marshall.o \ uverbs_std_types_srq.o \ uverbs_std_types_wq.o \ uverbs_std_types_qp.o -ib_uverbs-$(CONFIG_INFINIBAND_USER_MEM) += umem.o +ib_uverbs-$(CONFIG_INFINIBAND_USER_MEM) += umem.o umem_dmabuf.o ib_uverbs-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += umem_odp.o diff --git a/drivers/infiniband/core/umem.c b/drivers/infiniband/core/umem.c index 7ca4112..cc131f8 100644 --- a/drivers/infiniband/core/umem.c +++ b/drivers/infiniband/core/umem.c @@ -2,6 +2,7 @@ * Copyright (c) 2005 Topspin Communications. All rights reserved. * Copyright (c) 2005 Cisco Systems. All rights reserved. * Copyright (c) 2005 Mellanox Technologies. All rights reserved. + * Copyright (c) 2020 Intel Corporation. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU @@ -278,6 +279,8 @@ void ib_umem_release(struct ib_umem *umem) { if (!umem) return; + if (umem->is_dmabuf) + return ib_umem_dmabuf_release(to_ib_umem_dmabuf(umem)); if (umem->is_odp) return ib_umem_odp_release(to_ib_umem_odp(umem));
diff --git a/drivers/infiniband/core/umem_dmabuf.c b/drivers/infiniband/core/umem_dmabuf.c new file mode 100644 index 0000000..3fcbcad --- /dev/null +++ b/drivers/infiniband/core/umem_dmabuf.c @@ -0,0 +1,174 @@ +// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause) +/* + * Copyright (c) 2020 Intel Corporation. All rights reserved. + */ + +#include <linux/dma-buf.h> +#include <linux/dma-resv.h> +#include <linux/dma-mapping.h> + +#include "uverbs.h" + +int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) +{ + struct sg_table *sgt; + struct scatterlist *sg; + struct dma_fence *fence; + unsigned long start, end, cur = 0; + unsigned int nmap = 0; + int i; + + dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); + + if (umem_dmabuf->sgt) + goto wait_fence; + + sgt = dma_buf_map_attachment(umem_dmabuf->attach, DMA_BIDIRECTIONAL); + if (IS_ERR(sgt)) + return PTR_ERR(sgt); + + /* modify the sg list in-place to match umem address and length */ + + start = ALIGN_DOWN(umem_dmabuf->umem.address, PAGE_SIZE); + end = ALIGN(umem_dmabuf->umem.address + umem_dmabuf->umem.length, + PAGE_SIZE); + for_each_sgtable_dma_sg(sgt, sg, i) { + if (start < cur + sg_dma_len(sg) && cur < end) + nmap++; + if (cur <= start && start < cur + sg_dma_len(sg)) { + unsigned long offset = start - cur; + + umem_dmabuf->first_sg = sg; + umem_dmabuf->first_sg_offset = offset; + sg_dma_address(sg) += offset; + sg_dma_len(sg) -= offset; + cur += offset; + } + if (cur < end && end <= cur + sg_dma_len(sg)) { + unsigned long trim = cur + sg_dma_len(sg) - end; + + umem_dmabuf->last_sg = sg; + umem_dmabuf->last_sg_trim = trim; + sg_dma_len(sg) -= trim; + break; + } + cur += sg_dma_len(sg); + } + + umem_dmabuf->umem.sg_head.sgl = umem_dmabuf->first_sg; + umem_dmabuf->umem.sg_head.nents = nmap; + umem_dmabuf->umem.nmap = nmap; + umem_dmabuf->sgt = sgt; + +wait_fence: + /* + * Although the sg list is valid now, the content of the pages + * may be not up-to-date. Wait for the exporter to finish + * the migration. + */ + fence = dma_resv_get_excl(umem_dmabuf->attach->dmabuf->resv); + if (fence) + return dma_fence_wait(fence, false); + + return 0; +} +EXPORT_SYMBOL(ib_umem_dmabuf_map_pages); + +void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf) +{ + dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); + + if (!umem_dmabuf->sgt) + return; + + /* retore the original sg list */ + if (umem_dmabuf->first_sg) { + sg_dma_address(umem_dmabuf->first_sg) -= + umem_dmabuf->first_sg_offset; + sg_dma_len(umem_dmabuf->first_sg) += + umem_dmabuf->first_sg_offset; + umem_dmabuf->first_sg = NULL; + umem_dmabuf->first_sg_offset = 0; + } + if (umem_dmabuf->last_sg) { + sg_dma_len(umem_dmabuf->last_sg) += + umem_dmabuf->last_sg_trim; + umem_dmabuf->last_sg = NULL; + umem_dmabuf->last_sg_trim = 0; + } + + dma_buf_unmap_attachment(umem_dmabuf->attach, umem_dmabuf->sgt, + DMA_BIDIRECTIONAL); + + umem_dmabuf->sgt = NULL; +} +EXPORT_SYMBOL(ib_umem_dmabuf_unmap_pages); + +struct ib_umem *ib_umem_dmabuf_get(struct ib_device *device, + unsigned long offset, size_t size, + int fd, int access, + const struct dma_buf_attach_ops *ops) +{ + struct dma_buf *dmabuf; + struct ib_umem_dmabuf *umem_dmabuf; + struct ib_umem *umem; + unsigned long end; + struct ib_umem *ret = ERR_PTR(-EINVAL); + + if (check_add_overflow(offset, (unsigned long)size, &end)) + return ret; + + if (unlikely(!ops || !ops->move_notify)) + return ret; + + dmabuf = dma_buf_get(fd); + if (IS_ERR(dmabuf)) + return ERR_CAST(dmabuf); + + if (dmabuf->size < end) + goto out_release_dmabuf; + + umem_dmabuf = kzalloc(sizeof(*umem_dmabuf), GFP_KERNEL); + if (!umem_dmabuf) { + ret = ERR_PTR(-ENOMEM); + goto out_release_dmabuf; + } + + umem = &umem_dmabuf->umem; + umem->ibdev = device; + umem->length = size; + umem->address = offset; + umem->writable = ib_access_writable(access); + umem->is_dmabuf = 1; + + if (!ib_umem_num_pages(umem)) + goto out_free_umem; + + umem_dmabuf->attach = dma_buf_dynamic_attach( + dmabuf, + device->dma_device, + ops, + umem_dmabuf); + if (IS_ERR(umem_dmabuf->attach)) { + ret = ERR_CAST(umem_dmabuf->attach); + goto out_free_umem; + } + return umem; + +out_free_umem: + kfree(umem_dmabuf); + +out_release_dmabuf: + dma_buf_put(dmabuf); + return ret; +} +EXPORT_SYMBOL(ib_umem_dmabuf_get); + +void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) +{ + struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf; + + dma_buf_detach(dmabuf, umem_dmabuf->attach); + dma_buf_put(dmabuf); + kfree(umem_dmabuf); +} diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h index 7752211..ec374ec 100644 --- a/include/rdma/ib_umem.h +++ b/include/rdma/ib_umem.h @@ -1,6 +1,7 @@ /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ /* * Copyright (c) 2007 Cisco Systems. All rights reserved. + * Copyright (c) 2020 Intel Corporation. All rights reserved. */
#ifndef IB_UMEM_H @@ -13,6 +14,7 @@
struct ib_ucontext; struct ib_umem_odp; +struct dma_buf_attach_ops;
struct ib_umem { struct ib_device *ibdev; @@ -22,12 +24,29 @@ struct ib_umem { unsigned long address; u32 writable : 1; u32 is_odp : 1; + u32 is_dmabuf : 1; struct work_struct work; struct sg_table sg_head; int nmap; unsigned int sg_nents; };
+struct ib_umem_dmabuf { + struct ib_umem umem; + struct dma_buf_attachment *attach; + struct sg_table *sgt; + struct scatterlist *first_sg; + struct scatterlist *last_sg; + unsigned long first_sg_offset; + unsigned long last_sg_trim; + void *private; +}; + +static inline struct ib_umem_dmabuf *to_ib_umem_dmabuf(struct ib_umem *umem) +{ + return container_of(umem, struct ib_umem_dmabuf, umem); +} + /* Returns the offset of the umem start relative to the first page. */ static inline int ib_umem_offset(struct ib_umem *umem) { @@ -86,6 +105,7 @@ int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, unsigned long ib_umem_find_best_pgsz(struct ib_umem *umem, unsigned long pgsz_bitmap, unsigned long virt); + /** * ib_umem_find_best_pgoff - Find best HW page size * @@ -116,6 +136,14 @@ static inline unsigned long ib_umem_find_best_pgoff(struct ib_umem *umem, dma_addr & pgoff_bitmask); }
+struct ib_umem *ib_umem_dmabuf_get(struct ib_device *device, + unsigned long offset, size_t size, + int fd, int access, + const struct dma_buf_attach_ops *ops); +int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf); +void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf); +void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf); + #else /* CONFIG_INFINIBAND_USER_MEM */
#include <linux/err.h> @@ -124,12 +152,12 @@ static inline struct ib_umem *ib_umem_get(struct ib_device *device, unsigned long addr, size_t size, int access) { - return ERR_PTR(-EINVAL); + return ERR_PTR(-EOPNOTSUPP); } static inline void ib_umem_release(struct ib_umem *umem) { } static inline int ib_umem_copy_from(void *dst, struct ib_umem *umem, size_t offset, size_t length) { - return -EINVAL; + return -EOPNOTSUPP; } static inline unsigned long ib_umem_find_best_pgsz(struct ib_umem *umem, unsigned long pgsz_bitmap, @@ -143,7 +171,20 @@ static inline unsigned long ib_umem_find_best_pgoff(struct ib_umem *umem, { return 0; } +static inline struct ib_umem *ib_umem_dmabuf_get(struct ib_device *device, + unsigned long offset, + size_t size, int fd, + int access, + struct dma_buf_attach_ops *ops) +{ + return ERR_PTR(-EOPNOTSUPP); +} +static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) +{ + return -EOPNOTSUPP; +} +static inline void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf) { } +static inline void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) { }
#endif /* CONFIG_INFINIBAND_USER_MEM */ - #endif /* IB_UMEM_H */
On Tue, Dec 08, 2020 at 02:39:12PM -0800, Jianxin Xiong wrote:
Dma-buf is a standard cross-driver buffer sharing mechanism that can be used to support peer-to-peer access from RDMA devices.
Device memory exported via dma-buf is associated with a file descriptor. This is passed to the user space as a property associated with the buffer allocation. When the buffer is registered as a memory region, the file descriptor is passed to the RDMA driver along with other parameters.
Implement the common code for importing dma-buf object and mapping dma-buf pages.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/core/Makefile | 2 +- drivers/infiniband/core/umem.c | 3 + drivers/infiniband/core/umem_dmabuf.c | 174 ++++++++++++++++++++++++++++++++++ include/rdma/ib_umem.h | 47 ++++++++- 4 files changed, 222 insertions(+), 4 deletions(-) create mode 100644 drivers/infiniband/core/umem_dmabuf.c
<...>
+int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) +{
- struct sg_table *sgt;
- struct scatterlist *sg;
- struct dma_fence *fence;
- unsigned long start, end, cur = 0;
- unsigned int nmap = 0;
- int i;
- dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
- if (umem_dmabuf->sgt)
goto wait_fence;- sgt = dma_buf_map_attachment(umem_dmabuf->attach, DMA_BIDIRECTIONAL);
- if (IS_ERR(sgt))
return PTR_ERR(sgt);- /* modify the sg list in-place to match umem address and length */
- start = ALIGN_DOWN(umem_dmabuf->umem.address, PAGE_SIZE);
- end = ALIGN(umem_dmabuf->umem.address + umem_dmabuf->umem.length,
PAGE_SIZE);- for_each_sgtable_dma_sg(sgt, sg, i) {
if (start < cur + sg_dma_len(sg) && cur < end)nmap++;if (cur <= start && start < cur + sg_dma_len(sg)) {unsigned long offset = start - cur;umem_dmabuf->first_sg = sg;umem_dmabuf->first_sg_offset = offset;sg_dma_address(sg) += offset;sg_dma_len(sg) -= offset;cur += offset;}if (cur < end && end <= cur + sg_dma_len(sg)) {unsigned long trim = cur + sg_dma_len(sg) - end;umem_dmabuf->last_sg = sg;umem_dmabuf->last_sg_trim = trim;sg_dma_len(sg) -= trim;break;}cur += sg_dma_len(sg);- }
- umem_dmabuf->umem.sg_head.sgl = umem_dmabuf->first_sg;
- umem_dmabuf->umem.sg_head.nents = nmap;
- umem_dmabuf->umem.nmap = nmap;
- umem_dmabuf->sgt = sgt;
+wait_fence:
- /*
* Although the sg list is valid now, the content of the pages* may be not up-to-date. Wait for the exporter to finish* the migration.*/- fence = dma_resv_get_excl(umem_dmabuf->attach->dmabuf->resv);
- if (fence)
return dma_fence_wait(fence, false);
You called to dma_buf_map_attachment() earlier in this function, so if you return an error here, the dma_buf won't be unmapped in pagefault_dmabuf_mr()
Thanks
-----Original Message----- From: Leon Romanovsky leon@kernel.org Sent: Tuesday, December 08, 2020 10:31 PM To: Xiong, Jianxin jianxin.xiong@intel.com Cc: linux-rdma@vger.kernel.org; dri-devel@lists.freedesktop.org; Doug Ledford dledford@redhat.com; Jason Gunthorpe jgg@ziepe.ca; Sumit Semwal sumit.semwal@linaro.org; Christian Koenig christian.koenig@amd.com; Vetter, Daniel daniel.vetter@intel.com Subject: Re: [PATCH v14 1/4] RDMA/umem: Support importing dma-buf as user memory region
On Tue, Dec 08, 2020 at 02:39:12PM -0800, Jianxin Xiong wrote:
Dma-buf is a standard cross-driver buffer sharing mechanism that can be used to support peer-to-peer access from RDMA devices.
Device memory exported via dma-buf is associated with a file descriptor. This is passed to the user space as a property associated with the buffer allocation. When the buffer is registered as a memory region, the file descriptor is passed to the RDMA driver along with other parameters.
Implement the common code for importing dma-buf object and mapping dma-buf pages.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/core/Makefile | 2 +- drivers/infiniband/core/umem.c | 3 + drivers/infiniband/core/umem_dmabuf.c | 174 ++++++++++++++++++++++++++++++++++ include/rdma/ib_umem.h | 47 ++++++++- 4 files changed, 222 insertions(+), 4 deletions(-) create mode 100644 drivers/infiniband/core/umem_dmabuf.c
<...>
+int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) {
- struct sg_table *sgt;
- struct scatterlist *sg;
- struct dma_fence *fence;
- unsigned long start, end, cur = 0;
- unsigned int nmap = 0;
- int i;
- dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
- if (umem_dmabuf->sgt)
goto wait_fence;- sgt = dma_buf_map_attachment(umem_dmabuf->attach, DMA_BIDIRECTIONAL);
- if (IS_ERR(sgt))
return PTR_ERR(sgt);- /* modify the sg list in-place to match umem address and length */
- start = ALIGN_DOWN(umem_dmabuf->umem.address, PAGE_SIZE);
- end = ALIGN(umem_dmabuf->umem.address + umem_dmabuf->umem.length,
PAGE_SIZE);- for_each_sgtable_dma_sg(sgt, sg, i) {
if (start < cur + sg_dma_len(sg) && cur < end)nmap++;if (cur <= start && start < cur + sg_dma_len(sg)) {unsigned long offset = start - cur;umem_dmabuf->first_sg = sg;umem_dmabuf->first_sg_offset = offset;sg_dma_address(sg) += offset;sg_dma_len(sg) -= offset;cur += offset;}if (cur < end && end <= cur + sg_dma_len(sg)) {unsigned long trim = cur + sg_dma_len(sg) - end;umem_dmabuf->last_sg = sg;umem_dmabuf->last_sg_trim = trim;sg_dma_len(sg) -= trim;break;}cur += sg_dma_len(sg);- }
- umem_dmabuf->umem.sg_head.sgl = umem_dmabuf->first_sg;
- umem_dmabuf->umem.sg_head.nents = nmap;
- umem_dmabuf->umem.nmap = nmap;
- umem_dmabuf->sgt = sgt;
+wait_fence:
- /*
* Although the sg list is valid now, the content of the pages* may be not up-to-date. Wait for the exporter to finish* the migration.*/- fence = dma_resv_get_excl(umem_dmabuf->attach->dmabuf->resv);
- if (fence)
return dma_fence_wait(fence, false);You called to dma_buf_map_attachment() earlier in this function, so if you return an error here, the dma_buf won't be unmapped in pagefault_dmabuf_mr()
Yes, that's by design. Next time ib_unen_dmabuf_map_pages() is called, dma_buf_map_attachment() will be skipped and dma_fence_wait() is retried.
Thanks
Dma-buf based memory region requires one extra parameter and is processed quite differently. Adding a separate method allows clean separation from regular memory regions.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch Reviewed-by: Leon Romanovsky leonro@nvidia.com --- drivers/infiniband/core/device.c | 1 + include/rdma/ib_verbs.h | 6 +++++- 2 files changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c index 3ab1ede..23f7440 100644 --- a/drivers/infiniband/core/device.c +++ b/drivers/infiniband/core/device.c @@ -2677,6 +2677,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops) SET_DEVICE_OP(dev_ops, read_counters); SET_DEVICE_OP(dev_ops, reg_dm_mr); SET_DEVICE_OP(dev_ops, reg_user_mr); + SET_DEVICE_OP(dev_ops, reg_user_mr_dmabuf); SET_DEVICE_OP(dev_ops, req_ncomp_notif); SET_DEVICE_OP(dev_ops, req_notify_cq); SET_DEVICE_OP(dev_ops, rereg_user_mr); diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 7bee8ab..fa3882b 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -2,7 +2,7 @@ /* * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved. * Copyright (c) 2004 Infinicon Corporation. All rights reserved. - * Copyright (c) 2004 Intel Corporation. All rights reserved. + * Copyright (c) 2004, 2020 Intel Corporation. All rights reserved. * Copyright (c) 2004 Topspin Corporation. All rights reserved. * Copyright (c) 2004 Voltaire Corporation. All rights reserved. * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. @@ -2433,6 +2433,10 @@ struct ib_device_ops { struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int mr_access_flags, struct ib_udata *udata); + struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, int fd, + int mr_access_flags, + struct ib_udata *udata); int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, u64 virt_addr, int mr_access_flags, struct ib_pd *pd, struct ib_udata *udata);
Implement a new uverbs ioctl method for memory registration with file descriptor as an extra parameter.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch Reviewed-by: Leon Romanovsky leonro@nvidia.com --- drivers/infiniband/core/uverbs_std_types_mr.c | 117 +++++++++++++++++++++++++- include/uapi/rdma/ib_user_ioctl_cmds.h | 14 +++ 2 files changed, 129 insertions(+), 2 deletions(-)
diff --git a/drivers/infiniband/core/uverbs_std_types_mr.c b/drivers/infiniband/core/uverbs_std_types_mr.c index dc58564..4660c76 100644 --- a/drivers/infiniband/core/uverbs_std_types_mr.c +++ b/drivers/infiniband/core/uverbs_std_types_mr.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2018, Mellanox Technologies inc. All rights reserved. + * Copyright (c) 2020, Intel Corporation. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU @@ -182,6 +183,86 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QUERY_MR)( return IS_UVERBS_COPY_ERR(ret) ? ret : 0; }
+static int UVERBS_HANDLER(UVERBS_METHOD_REG_DMABUF_MR)( + struct uverbs_attr_bundle *attrs) +{ + struct ib_uobject *uobj = + uverbs_attr_get_uobject(attrs, UVERBS_ATTR_REG_DMABUF_MR_HANDLE); + struct ib_pd *pd = + uverbs_attr_get_obj(attrs, UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE); + struct ib_device *ib_dev = pd->device; + + u64 offset, length, virt_addr; + u32 fd, access_flags; + struct ib_mr *mr; + int ret; + + if (!ib_dev->ops.reg_user_mr_dmabuf) + return -EOPNOTSUPP; + + if (!(pd->device->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING)) + return -EOPNOTSUPP; + + ret = uverbs_copy_from(&offset, attrs, + UVERBS_ATTR_REG_DMABUF_MR_OFFSET); + if (ret) + return ret; + + ret = uverbs_copy_from(&length, attrs, + UVERBS_ATTR_REG_DMABUF_MR_LENGTH); + if (ret) + return ret; + + ret = uverbs_copy_from(&virt_addr, attrs, + UVERBS_ATTR_REG_DMABUF_MR_IOVA); + if (ret) + return ret; + + ret = uverbs_copy_from(&fd, attrs, + UVERBS_ATTR_REG_DMABUF_MR_FD); + if (ret) + return ret; + + ret = uverbs_get_flags32(&access_flags, attrs, + UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + IB_ACCESS_LOCAL_WRITE | + IB_ACCESS_REMOTE_READ | + IB_ACCESS_REMOTE_WRITE | + IB_ACCESS_REMOTE_ATOMIC | + IB_ACCESS_RELAXED_ORDERING); + if (ret) + return ret; + + ret = ib_check_mr_access(access_flags); + if (ret) + return ret; + + mr = pd->device->ops.reg_user_mr_dmabuf(pd, offset, length, virt_addr, + fd, access_flags, + &attrs->driver_udata); + if (IS_ERR(mr)) + return PTR_ERR(mr); + + mr->device = pd->device; + mr->pd = pd; + mr->type = IB_MR_TYPE_USER; + mr->uobject = uobj; + atomic_inc(&pd->usecnt); + + uobj->object = mr; + + uverbs_finalize_uobj_create(attrs, UVERBS_ATTR_REG_DMABUF_MR_HANDLE); + + ret = uverbs_copy_to(attrs, UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY, + &mr->lkey, sizeof(mr->lkey)); + if (ret) + return ret; + + ret = uverbs_copy_to(attrs, UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY, + &mr->rkey, sizeof(mr->rkey)); + return ret; +} + DECLARE_UVERBS_NAMED_METHOD( UVERBS_METHOD_ADVISE_MR, UVERBS_ATTR_IDR(UVERBS_ATTR_ADVISE_MR_PD_HANDLE, @@ -247,6 +328,37 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QUERY_MR)( UVERBS_ATTR_TYPE(u32), UA_MANDATORY));
+DECLARE_UVERBS_NAMED_METHOD( + UVERBS_METHOD_REG_DMABUF_MR, + UVERBS_ATTR_IDR(UVERBS_ATTR_REG_DMABUF_MR_HANDLE, + UVERBS_OBJECT_MR, + UVERBS_ACCESS_NEW, + UA_MANDATORY), + UVERBS_ATTR_IDR(UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE, + UVERBS_OBJECT_PD, + UVERBS_ACCESS_READ, + UA_MANDATORY), + UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_OFFSET, + UVERBS_ATTR_TYPE(u64), + UA_MANDATORY), + UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_LENGTH, + UVERBS_ATTR_TYPE(u64), + UA_MANDATORY), + UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_IOVA, + UVERBS_ATTR_TYPE(u64), + UA_MANDATORY), + UVERBS_ATTR_PTR_IN(UVERBS_ATTR_REG_DMABUF_MR_FD, + UVERBS_ATTR_TYPE(u32), + UA_MANDATORY), + UVERBS_ATTR_FLAGS_IN(UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + enum ib_access_flags), + UVERBS_ATTR_PTR_OUT(UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY, + UVERBS_ATTR_TYPE(u32), + UA_MANDATORY), + UVERBS_ATTR_PTR_OUT(UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY, + UVERBS_ATTR_TYPE(u32), + UA_MANDATORY)); + DECLARE_UVERBS_NAMED_METHOD_DESTROY( UVERBS_METHOD_MR_DESTROY, UVERBS_ATTR_IDR(UVERBS_ATTR_DESTROY_MR_HANDLE, @@ -257,10 +369,11 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QUERY_MR)( DECLARE_UVERBS_NAMED_OBJECT( UVERBS_OBJECT_MR, UVERBS_TYPE_ALLOC_IDR(uverbs_free_mr), + &UVERBS_METHOD(UVERBS_METHOD_ADVISE_MR), &UVERBS_METHOD(UVERBS_METHOD_DM_MR_REG), &UVERBS_METHOD(UVERBS_METHOD_MR_DESTROY), - &UVERBS_METHOD(UVERBS_METHOD_ADVISE_MR), - &UVERBS_METHOD(UVERBS_METHOD_QUERY_MR)); + &UVERBS_METHOD(UVERBS_METHOD_QUERY_MR), + &UVERBS_METHOD(UVERBS_METHOD_REG_DMABUF_MR));
const struct uapi_definition uverbs_def_obj_mr[] = { UAPI_DEF_CHAIN_OBJ_TREE_NAMED(UVERBS_OBJECT_MR, diff --git a/include/uapi/rdma/ib_user_ioctl_cmds.h b/include/uapi/rdma/ib_user_ioctl_cmds.h index 7968a18..dafc7eb 100644 --- a/include/uapi/rdma/ib_user_ioctl_cmds.h +++ b/include/uapi/rdma/ib_user_ioctl_cmds.h @@ -1,5 +1,6 @@ /* * Copyright (c) 2018, Mellanox Technologies inc. All rights reserved. + * Copyright (c) 2020, Intel Corporation. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU @@ -251,6 +252,7 @@ enum uverbs_methods_mr { UVERBS_METHOD_MR_DESTROY, UVERBS_METHOD_ADVISE_MR, UVERBS_METHOD_QUERY_MR, + UVERBS_METHOD_REG_DMABUF_MR, };
enum uverbs_attrs_mr_destroy_ids { @@ -272,6 +274,18 @@ enum uverbs_attrs_query_mr_cmd_attr_ids { UVERBS_ATTR_QUERY_MR_RESP_IOVA, };
+enum uverbs_attrs_reg_dmabuf_mr_cmd_attr_ids { + UVERBS_ATTR_REG_DMABUF_MR_HANDLE, + UVERBS_ATTR_REG_DMABUF_MR_PD_HANDLE, + UVERBS_ATTR_REG_DMABUF_MR_OFFSET, + UVERBS_ATTR_REG_DMABUF_MR_LENGTH, + UVERBS_ATTR_REG_DMABUF_MR_IOVA, + UVERBS_ATTR_REG_DMABUF_MR_FD, + UVERBS_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + UVERBS_ATTR_REG_DMABUF_MR_RESP_LKEY, + UVERBS_ATTR_REG_DMABUF_MR_RESP_RKEY, +}; + enum uverbs_attrs_create_counters_cmd_attr_ids { UVERBS_ATTR_CREATE_COUNTERS_HANDLE, };
Implement the new driver method 'reg_user_mr_dmabuf'. Utilize the core functions to import dma-buf based memory region and update the mappings.
Add code to handle dma-buf related page fault.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch --- drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++++ drivers/infiniband/hw/mlx5/mr.c | 128 +++++++++++++++++++++++++++++++++-- drivers/infiniband/hw/mlx5/odp.c | 86 +++++++++++++++++++++-- 4 files changed, 225 insertions(+), 9 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 4a054eb..c025746 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -1,6 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB /* * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved. + * Copyright (c) 2020, Intel Corporation. All rights reserved. */
#include <linux/debugfs.h> @@ -4069,6 +4070,7 @@ static int mlx5_ib_enable_driver(struct ib_device *dev) .query_srq = mlx5_ib_query_srq, .query_ucontext = mlx5_ib_query_ucontext, .reg_user_mr = mlx5_ib_reg_user_mr, + .reg_user_mr_dmabuf = mlx5_ib_reg_user_mr_dmabuf, .req_notify_cq = mlx5_ib_arm_cq, .rereg_user_mr = mlx5_ib_rereg_user_mr, .resize_cq = mlx5_ib_resize_cq, diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 718e59f..6f4d1b4 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1,6 +1,7 @@ /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ /* * Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved. + * Copyright (c) 2020, Intel Corporation. All rights reserved. */
#ifndef MLX5_IB_H @@ -704,6 +705,12 @@ static inline bool is_odp_mr(struct mlx5_ib_mr *mr) mr->umem->is_odp; }
+static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr) +{ + return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem && + mr->umem->is_dmabuf; +} + struct mlx5_ib_mw { struct ib_mw ibmw; struct mlx5_core_mkey mmkey; @@ -1239,6 +1246,10 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_udata *udata); +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, + u64 length, u64 virt_addr, + int fd, int access_flags, + struct ib_udata *udata); int mlx5_ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, @@ -1249,11 +1260,13 @@ int mlx5_ib_advise_mr(struct ib_pd *pd, int mlx5_ib_dealloc_mw(struct ib_mw *mw); int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags); struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, struct ib_udata *udata, int access_flags); void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr); +void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr); int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_pd *pd, struct ib_udata *udata); @@ -1341,6 +1354,7 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, struct ib_sge *sg_list, u32 num_sge); int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable); +int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr); #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) { @@ -1366,6 +1380,10 @@ static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) { return -EOPNOTSUPP; } +static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{ + return -EOPNOTSUPP; +} #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
extern const struct mmu_interval_notifier_ops mlx5_mn_ops; diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index b6116f6..e3be1f5 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1,5 +1,6 @@ /* * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved. + * Copyright (c) 2020, Intel Corporation. All rights reserved. * * This software is available to you under a choice of one of two * licenses. You may choose to be licensed under the terms of the GNU @@ -36,6 +37,8 @@ #include <linux/debugfs.h> #include <linux/export.h> #include <linux/delay.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h> #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <rdma/ib_verbs.h> @@ -957,6 +960,16 @@ static struct mlx5_cache_ent *mr_cache_ent_from_order(struct mlx5_ib_dev *dev, return &cache->ent[order]; }
+static unsigned int mlx5_umem_dmabuf_default_pgsz(struct ib_umem *umem, + u64 iova) +{ + if ((iova ^ umem->address) & (PAGE_SIZE - 1)) + return 0; + + umem->iova = iova; + return PAGE_SIZE; +} + static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags) @@ -966,7 +979,11 @@ static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct mlx5_ib_mr *mr; unsigned int page_size;
- page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova); + if (umem->is_dmabuf) + page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova); + else + page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, + 0, iova); if (WARN_ON(!page_size)) return ERR_PTR(-EINVAL); ent = mr_cache_ent_from_order( @@ -1212,8 +1229,10 @@ int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
/* * Send the DMA list to the HW for a normal MR using UMR. + * Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP + * flag may be used. */ -static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) { struct mlx5_ib_dev *dev = mr->dev; struct device *ddev = &dev->mdev->pdev->dev; @@ -1255,6 +1274,10 @@ static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) cur_mtt->ptag = cpu_to_be64(rdma_block_iter_dma_address(&biter) | MLX5_IB_MTT_PRESENT); + + if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP)) + cur_mtt->ptag = 0; + cur_mtt++; }
@@ -1291,8 +1314,11 @@ static struct mlx5_ib_mr *reg_create(struct ib_mr *ibmr, struct ib_pd *pd, int err; bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg));
- page_size = - mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova); + if (umem->is_dmabuf) + page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova); + else + page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, + 0, iova); if (WARN_ON(!page_size)) return ERR_PTR(-EINVAL);
@@ -1572,6 +1598,96 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, return ERR_PTR(err); }
+static void mlx5_ib_dmabuf_invalidate_cb(struct dma_buf_attachment *attach) +{ + struct ib_umem_dmabuf *umem_dmabuf = attach->importer_priv; + struct mlx5_ib_mr *mr = umem_dmabuf->private; + + dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); + + if (!umem_dmabuf->sgt) + return; + + mlx5_ib_update_mr_pas(mr, MLX5_IB_UPD_XLT_ZAP); + ib_umem_dmabuf_unmap_pages(umem_dmabuf); +} + +static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = { + .allow_peer2peer = 1, + .move_notify = mlx5_ib_dmabuf_invalidate_cb, +}; + +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, + int fd, int access_flags, + struct ib_udata *udata) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct mlx5_ib_mr *mr = NULL; + struct ib_umem *umem; + int err; + + if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM)) + return ERR_PTR(-EOPNOTSUPP); + + mlx5_ib_dbg(dev, + "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n", + offset, virt_addr, length, fd, access_flags); + + if (!mlx5_ib_can_load_pas_with_umr(dev, length)) + return ERR_PTR(-EINVAL); + + umem = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, access_flags, + &mlx5_ib_dmabuf_attach_ops); + if (IS_ERR(umem)) { + mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem)); + return ERR_PTR(PTR_ERR(umem)); + } + + mr = alloc_mr_from_cache(pd, umem, virt_addr, access_flags); + if (IS_ERR(mr)) + mr = NULL; + + if (!mr) { + mutex_lock(&dev->slow_path_mutex); + mr = reg_create(NULL, pd, umem, virt_addr, access_flags, + false); + mutex_unlock(&dev->slow_path_mutex); + } + + if (IS_ERR(mr)) { + err = PTR_ERR(mr); + goto error; + } + + mlx5_ib_dbg(dev, "mkey 0x%x\n", mr->mmkey.key); + + mr->umem = umem; + atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages); + set_mr_fields(dev, mr, length, access_flags); + + to_ib_umem_dmabuf(umem)->private = mr; + init_waitqueue_head(&mr->q_deferred_work); + atomic_set(&mr->num_deferred_work, 0); + err = xa_err(xa_store(&dev->odp_mkeys, + mlx5_base_mkey(mr->mmkey.key), &mr->mmkey, + GFP_KERNEL)); + if (err) { + dereg_mr(dev, mr); + return ERR_PTR(err); + } + + err = mlx5_ib_init_dmabuf_mr(mr); + if (err) { + dereg_mr(dev, mr); + return ERR_PTR(err); + } + return &mr->ibmr; +error: + ib_umem_release(umem); + return ERR_PTR(err); +} + /** * mlx5_mr_cache_invalidate - Fence all DMA on the MR * @mr: The MR to fence @@ -1640,7 +1756,7 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, if (!mr->umem) return -EINVAL;
- if (is_odp_mr(mr)) + if (is_odp_mr(mr) || is_dmabuf_mr(mr)) return -EOPNOTSUPP;
if (flags & IB_MR_REREG_TRANS) { @@ -1804,6 +1920,8 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) /* Stop all DMA */ if (is_odp_mr(mr)) mlx5_ib_fence_odp_mr(mr); + else if (is_dmabuf_mr(mr)) + mlx5_ib_fence_dmabuf_mr(mr); else clean_mr(dev, mr);
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 5c853ec..35d6770 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -33,6 +33,8 @@ #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <linux/kernel.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h>
#include "mlx5_ib.h" #include "cmd.h" @@ -664,6 +666,37 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr) dma_fence_odp_mr(mr); }
+/** + * mlx5_ib_fence_dmabuf_mr - Stop all access to the dmabuf MR + * @mr: to fence + * + * On return no parallel threads will be touching this MR and no DMA will be + * active. + */ +void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr) +{ + struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); + + /* Prevent new page faults and prefetch requests from succeeding */ + xa_erase(&mr->dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key)); + + /* Wait for all running page-fault handlers to finish. */ + synchronize_srcu(&mr->dev->odp_srcu); + + wait_event(mr->q_deferred_work, !atomic_read(&mr->num_deferred_work)); + + dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL); + mlx5_mr_cache_invalidate(mr); + umem_dmabuf->private = NULL; + ib_umem_dmabuf_unmap_pages(umem_dmabuf); + dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); + + if (!mr->cache_ent) { + mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey); + WARN_ON(mr->descs); + } +} + #define MLX5_PF_FLAGS_DOWNGRADE BIT(1) #define MLX5_PF_FLAGS_SNAPSHOT BIT(2) #define MLX5_PF_FLAGS_ENABLE BIT(3) @@ -797,6 +830,41 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr, return ret; }
+static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt, + u32 *bytes_mapped, u32 flags) +{ + struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); + u32 xlt_flags = 0; + int err; + unsigned int page_size; + + if (flags & MLX5_PF_FLAGS_ENABLE) + xlt_flags |= MLX5_IB_UPD_XLT_ENABLE; + + dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL); + err = ib_umem_dmabuf_map_pages(umem_dmabuf); + if (!err) { + page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc, + log_page_size, 0, + umem_dmabuf->umem.iova); + if (unlikely(page_size < PAGE_SIZE)) { + ib_umem_dmabuf_unmap_pages(umem_dmabuf); + err = -EINVAL; + } else { + err = mlx5_ib_update_mr_pas(mr, xlt_flags); + } + } + dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); + + if (err) + return err; + + if (bytes_mapped) + *bytes_mapped += bcnt; + + return ib_umem_num_pages(mr->umem); +} + /* * Returns: * -EFAULT: The io_virt->bcnt is not within the MR, it covers pages that are @@ -815,6 +883,9 @@ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt, if (unlikely(io_virt < mr->mmkey.iova)) return -EFAULT;
+ if (mr->umem->is_dmabuf) + return pagefault_dmabuf_mr(mr, bcnt, bytes_mapped, flags); + if (!odp->is_implicit_odp) { u64 user_va;
@@ -845,6 +916,16 @@ int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) return ret >= 0 ? 0 : ret; }
+int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{ + int ret; + + ret = pagefault_dmabuf_mr(mr, mr->umem->length, NULL, + MLX5_PF_FLAGS_ENABLE); + + return ret >= 0 ? 0 : ret; +} + struct pf_frame { struct pf_frame *next; u32 key; @@ -1747,7 +1828,6 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_core_mkey *mmkey; - struct ib_umem_odp *odp; struct mlx5_ib_mr *mr;
lockdep_assert_held(&dev->odp_srcu); @@ -1761,11 +1841,9 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) if (mr->ibmr.pd != pd) return NULL;
- odp = to_ib_umem_odp(mr->umem); - /* prefetch with write-access must be supported by the MR */ if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE && - !odp->umem.writable) + !mr->umem->writable) return NULL;
return mr;
On Tue, Dec 08, 2020 at 02:39:15PM -0800, Jianxin Xiong wrote:
Implement the new driver method 'reg_user_mr_dmabuf'. Utilize the core functions to import dma-buf based memory region and update the mappings.
Add code to handle dma-buf related page fault.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++++ drivers/infiniband/hw/mlx5/mr.c | 128 +++++++++++++++++++++++++++++++++-- drivers/infiniband/hw/mlx5/odp.c | 86 +++++++++++++++++++++-- 4 files changed, 225 insertions(+), 9 deletions(-)
<...>
- umem = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, access_flags,
&mlx5_ib_dmabuf_attach_ops);- if (IS_ERR(umem)) {
mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem));return ERR_PTR(PTR_ERR(umem));
return ERR_CAST(umem);
- }
<...>
- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- err = ib_umem_dmabuf_map_pages(umem_dmabuf);
- if (!err) {
page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc,log_page_size, 0,umem_dmabuf->umem.iova);if (unlikely(page_size < PAGE_SIZE)) {ib_umem_dmabuf_unmap_pages(umem_dmabuf);err = -EINVAL;} else {err = mlx5_ib_update_mr_pas(mr, xlt_flags);}- }
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
Let's write this section in kernel coding style, please
dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL); err = ib_umem_dmabuf_map_pages(umem_dmabuf); if (err) { dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); return err; } .....
Thanks
-----Original Message----- From: Leon Romanovsky leon@kernel.org Sent: Tuesday, December 08, 2020 10:45 PM To: Xiong, Jianxin jianxin.xiong@intel.com Cc: linux-rdma@vger.kernel.org; dri-devel@lists.freedesktop.org; Doug Ledford dledford@redhat.com; Jason Gunthorpe jgg@ziepe.ca; Sumit Semwal sumit.semwal@linaro.org; Christian Koenig christian.koenig@amd.com; Vetter, Daniel daniel.vetter@intel.com Subject: Re: [PATCH v14 4/4] RDMA/mlx5: Support dma-buf based userspace memory region
On Tue, Dec 08, 2020 at 02:39:15PM -0800, Jianxin Xiong wrote:
Implement the new driver method 'reg_user_mr_dmabuf'. Utilize the core functions to import dma-buf based memory region and update the mappings.
Add code to handle dma-buf related page fault.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++++ drivers/infiniband/hw/mlx5/mr.c | 128 +++++++++++++++++++++++++++++++++-- drivers/infiniband/hw/mlx5/odp.c | 86 +++++++++++++++++++++-- 4 files changed, 225 insertions(+), 9 deletions(-)
<...>
- umem = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, access_flags,
&mlx5_ib_dmabuf_attach_ops);- if (IS_ERR(umem)) {
mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem));return ERR_PTR(PTR_ERR(umem));return ERR_CAST(umem);
- }
<...>
- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- err = ib_umem_dmabuf_map_pages(umem_dmabuf);
- if (!err) {
page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc,log_page_size, 0,umem_dmabuf->umem.iova);if (unlikely(page_size < PAGE_SIZE)) {ib_umem_dmabuf_unmap_pages(umem_dmabuf);err = -EINVAL;} else {err = mlx5_ib_update_mr_pas(mr, xlt_flags);}- }
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
Let's write this section in kernel coding style, please
dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL); err = ib_umem_dmabuf_map_pages(umem_dmabuf); if (err) { dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); return err; } .....
Thanks a lot. Will fix these.
On 12/9/2020 12:39 AM, Jianxin Xiong wrote:
Implement the new driver method 'reg_user_mr_dmabuf'. Utilize the core functions to import dma-buf based memory region and update the mappings.
Add code to handle dma-buf related page fault.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++++ drivers/infiniband/hw/mlx5/mr.c | 128 +++++++++++++++++++++++++++++++++-- drivers/infiniband/hw/mlx5/odp.c | 86 +++++++++++++++++++++-- 4 files changed, 225 insertions(+), 9 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 4a054eb..c025746 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -1,6 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB /*
- Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
*/
#include <linux/debugfs.h>
@@ -4069,6 +4070,7 @@ static int mlx5_ib_enable_driver(struct ib_device *dev) .query_srq = mlx5_ib_query_srq, .query_ucontext = mlx5_ib_query_ucontext, .reg_user_mr = mlx5_ib_reg_user_mr,
- .reg_user_mr_dmabuf = mlx5_ib_reg_user_mr_dmabuf, .req_notify_cq = mlx5_ib_arm_cq, .rereg_user_mr = mlx5_ib_rereg_user_mr, .resize_cq = mlx5_ib_resize_cq,
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 718e59f..6f4d1b4 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1,6 +1,7 @@ /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ /*
- Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
*/
#ifndef MLX5_IB_H
@@ -704,6 +705,12 @@ static inline bool is_odp_mr(struct mlx5_ib_mr *mr) mr->umem->is_odp; }
+static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
mr->umem->is_dmabuf;+}
Didn't we agree that IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) should be checked earlier
upon mlx5_ib_reg_user_mr_dmabuf () to fully prevent the functionality when ODP is not supported ? see note on previous versions.
struct mlx5_ib_mw { struct ib_mw ibmw; struct mlx5_core_mkey mmkey; @@ -1239,6 +1246,10 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_udata *udata); +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
u64 length, u64 virt_addr,int fd, int access_flags,struct ib_udata *udata);@@ -1249,11 +1260,13 @@ int mlx5_ib_advise_mr(struct ib_pd *pd, int mlx5_ib_dealloc_mw(struct ib_mw *mw); int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags); struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, struct ib_udata *udata, int access_flags); void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr); +void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr); int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_pd *pd, struct ib_udata *udata); @@ -1341,6 +1354,7 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, struct ib_sge *sg_list, u32 num_sge); int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable); +int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr); #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) { @@ -1366,6 +1380,10 @@ static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) { return -EOPNOTSUPP; } +static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- return -EOPNOTSUPP;
+} #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
extern const struct mmu_interval_notifier_ops mlx5_mn_ops; diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index b6116f6..e3be1f5 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1,5 +1,6 @@ /*
- Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
- This software is available to you under a choice of one of two
- licenses. You may choose to be licensed under the terms of the GNU
@@ -36,6 +37,8 @@ #include <linux/debugfs.h> #include <linux/export.h> #include <linux/delay.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h> #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <rdma/ib_verbs.h> @@ -957,6 +960,16 @@ static struct mlx5_cache_ent *mr_cache_ent_from_order(struct mlx5_ib_dev *dev, return &cache->ent[order]; }
+static unsigned int mlx5_umem_dmabuf_default_pgsz(struct ib_umem *umem,
u64 iova)+{
- if ((iova ^ umem->address) & (PAGE_SIZE - 1))
return 0;
Also here, see my note from previous versions, can user space trigger this by supplying some invalid input ?
If so, this may cause a WARN_ON() in the caller (i.e.alloc_mr_from_cache) that we should prevent from being triggered by user space application.
- umem->iova = iova;
- return PAGE_SIZE;
+}
- static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags)
@@ -966,7 +979,11 @@ static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct mlx5_ib_mr *mr; unsigned int page_size;
- page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova);
- if (umem->is_dmabuf)
page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova);- else
page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size,0, iova);@@ -1212,8 +1229,10 @@ int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
/*
- Send the DMA list to the HW for a normal MR using UMR.
- Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP
*/
- flag may be used.
-static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) { struct mlx5_ib_dev *dev = mr->dev; struct device *ddev = &dev->mdev->pdev->dev; @@ -1255,6 +1274,10 @@ static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) cur_mtt->ptag = cpu_to_be64(rdma_block_iter_dma_address(&biter) | MLX5_IB_MTT_PRESENT);
if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP))cur_mtt->ptag = 0;- cur_mtt++; }
@@ -1291,8 +1314,11 @@ static struct mlx5_ib_mr *reg_create(struct ib_mr *ibmr, struct ib_pd *pd, int err; bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg));
- page_size =
mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova);
- if (umem->is_dmabuf)
page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova);- else
page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size,0, iova);@@ -1572,6 +1598,96 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, return ERR_PTR(err); }
+static void mlx5_ib_dmabuf_invalidate_cb(struct dma_buf_attachment *attach) +{
- struct ib_umem_dmabuf *umem_dmabuf = attach->importer_priv;
- struct mlx5_ib_mr *mr = umem_dmabuf->private;
- dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
- if (!umem_dmabuf->sgt)
return;- mlx5_ib_update_mr_pas(mr, MLX5_IB_UPD_XLT_ZAP);
- ib_umem_dmabuf_unmap_pages(umem_dmabuf);
+}
+static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = {
- .allow_peer2peer = 1,
- .move_notify = mlx5_ib_dmabuf_invalidate_cb,
+};
+struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset,
u64 length, u64 virt_addr,int fd, int access_flags,struct ib_udata *udata)+{
- struct mlx5_ib_dev *dev = to_mdev(pd->device);
- struct mlx5_ib_mr *mr = NULL;
- struct ib_umem *umem;
- int err;
- if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM))
return ERR_PTR(-EOPNOTSUPP);- mlx5_ib_dbg(dev,
"offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n",offset, virt_addr, length, fd, access_flags);- if (!mlx5_ib_can_load_pas_with_umr(dev, length))
return ERR_PTR(-EINVAL);- umem = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, access_flags,
&mlx5_ib_dmabuf_attach_ops);- if (IS_ERR(umem)) {
mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem));return ERR_PTR(PTR_ERR(umem));- }
- mr = alloc_mr_from_cache(pd, umem, virt_addr, access_flags);
- if (IS_ERR(mr))
mr = NULL;- if (!mr) {
mutex_lock(&dev->slow_path_mutex);mr = reg_create(NULL, pd, umem, virt_addr, access_flags,false);mutex_unlock(&dev->slow_path_mutex);- }
- if (IS_ERR(mr)) {
err = PTR_ERR(mr);goto error;- }
- mlx5_ib_dbg(dev, "mkey 0x%x\n", mr->mmkey.key);
- mr->umem = umem;
- atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages);
- set_mr_fields(dev, mr, length, access_flags);
- to_ib_umem_dmabuf(umem)->private = mr;
- init_waitqueue_head(&mr->q_deferred_work);
- atomic_set(&mr->num_deferred_work, 0);
- err = xa_err(xa_store(&dev->odp_mkeys,
mlx5_base_mkey(mr->mmkey.key), &mr->mmkey,GFP_KERNEL));- if (err) {
dereg_mr(dev, mr);return ERR_PTR(err);- }
- err = mlx5_ib_init_dmabuf_mr(mr);
- if (err) {
dereg_mr(dev, mr);return ERR_PTR(err);- }
- return &mr->ibmr;
+error:
- ib_umem_release(umem);
- return ERR_PTR(err);
+}
- /**
- mlx5_mr_cache_invalidate - Fence all DMA on the MR
- @mr: The MR to fence
@@ -1640,7 +1756,7 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, if (!mr->umem) return -EINVAL;
- if (is_odp_mr(mr))
if (is_odp_mr(mr) || is_dmabuf_mr(mr)) return -EOPNOTSUPP;
if (flags & IB_MR_REREG_TRANS) {
@@ -1804,6 +1920,8 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) /* Stop all DMA */ if (is_odp_mr(mr)) mlx5_ib_fence_odp_mr(mr);
- else if (is_dmabuf_mr(mr))
mlx5_ib_fence_dmabuf_mr(mr);diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 5c853ec..35d6770 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -33,6 +33,8 @@ #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <linux/kernel.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h>
#include "mlx5_ib.h" #include "cmd.h" @@ -664,6 +666,37 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr) dma_fence_odp_mr(mr); }
+/**
- mlx5_ib_fence_dmabuf_mr - Stop all access to the dmabuf MR
- @mr: to fence
- On return no parallel threads will be touching this MR and no DMA will be
- active.
- */
+void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem);
- /* Prevent new page faults and prefetch requests from succeeding */
- xa_erase(&mr->dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key));
- /* Wait for all running page-fault handlers to finish. */
- synchronize_srcu(&mr->dev->odp_srcu);
- wait_event(mr->q_deferred_work, !atomic_read(&mr->num_deferred_work));
- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- mlx5_mr_cache_invalidate(mr);
- umem_dmabuf->private = NULL;
- ib_umem_dmabuf_unmap_pages(umem_dmabuf);
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
- if (!mr->cache_ent) {
mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey);WARN_ON(mr->descs);- }
+}
- #define MLX5_PF_FLAGS_DOWNGRADE BIT(1) #define MLX5_PF_FLAGS_SNAPSHOT BIT(2) #define MLX5_PF_FLAGS_ENABLE BIT(3)
@@ -797,6 +830,41 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr, return ret; }
+static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt,
u32 *bytes_mapped, u32 flags)+{
- struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem);
- u32 xlt_flags = 0;
- int err;
- unsigned int page_size;
- if (flags & MLX5_PF_FLAGS_ENABLE)
xlt_flags |= MLX5_IB_UPD_XLT_ENABLE;- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- err = ib_umem_dmabuf_map_pages(umem_dmabuf);
- if (!err) {
page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc,log_page_size, 0,umem_dmabuf->umem.iova);if (unlikely(page_size < PAGE_SIZE)) {ib_umem_dmabuf_unmap_pages(umem_dmabuf);err = -EINVAL;} else {err = mlx5_ib_update_mr_pas(mr, xlt_flags);}- }
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
- if (err)
return err;- if (bytes_mapped)
*bytes_mapped += bcnt;- return ib_umem_num_pages(mr->umem);
+}
- /*
- Returns:
- -EFAULT: The io_virt->bcnt is not within the MR, it covers pages that are
@@ -815,6 +883,9 @@ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt, if (unlikely(io_virt < mr->mmkey.iova)) return -EFAULT;
- if (mr->umem->is_dmabuf)
return pagefault_dmabuf_mr(mr, bcnt, bytes_mapped, flags);- if (!odp->is_implicit_odp) { u64 user_va;
@@ -845,6 +916,16 @@ int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) return ret >= 0 ? 0 : ret; }
+int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- int ret;
- ret = pagefault_dmabuf_mr(mr, mr->umem->length, NULL,
MLX5_PF_FLAGS_ENABLE);- return ret >= 0 ? 0 : ret;
+}
- struct pf_frame { struct pf_frame *next; u32 key;
@@ -1747,7 +1828,6 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_core_mkey *mmkey;
struct ib_umem_odp *odp; struct mlx5_ib_mr *mr;
lockdep_assert_held(&dev->odp_srcu);
@@ -1761,11 +1841,9 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) if (mr->ibmr.pd != pd) return NULL;
- odp = to_ib_umem_odp(mr->umem);
- /* prefetch with write-access must be supported by the MR */ if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE &&
!odp->umem.writable)
!mr->umem->writable)return NULL;
return mr;
-----Original Message----- From: Yishai Hadas yishaih@nvidia.com Sent: Tuesday, December 08, 2020 11:13 PM To: Xiong, Jianxin jianxin.xiong@intel.com Cc: Doug Ledford dledford@redhat.com; Jason Gunthorpe jgg@ziepe.ca; Leon Romanovsky leon@kernel.org; Sumit Semwal sumit.semwal@linaro.org; Christian Koenig christian.koenig@amd.com; Vetter, Daniel daniel.vetter@intel.com; linux- rdma@vger.kernel.org; dri-devel@lists.freedesktop.org; Yishai Hadas yishaih@nvidia.com Subject: Re: [PATCH v14 4/4] RDMA/mlx5: Support dma-buf based userspace memory region
On 12/9/2020 12:39 AM, Jianxin Xiong wrote:
Implement the new driver method 'reg_user_mr_dmabuf'. Utilize the core functions to import dma-buf based memory region and update the mappings.
Add code to handle dma-buf related page fault.
Signed-off-by: Jianxin Xiong jianxin.xiong@intel.com Reviewed-by: Sean Hefty sean.hefty@intel.com Acked-by: Michael J. Ruhl michael.j.ruhl@intel.com Acked-by: Christian Koenig christian.koenig@amd.com Acked-by: Daniel Vetter daniel.vetter@ffwll.ch
drivers/infiniband/hw/mlx5/main.c | 2 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++++ drivers/infiniband/hw/mlx5/mr.c | 128 +++++++++++++++++++++++++++++++++-- drivers/infiniband/hw/mlx5/odp.c | 86 +++++++++++++++++++++-- 4 files changed, 225 insertions(+), 9 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 4a054eb..c025746 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -1,6 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB /*
- Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
*/
#include <linux/debugfs.h>
@@ -4069,6 +4070,7 @@ static int mlx5_ib_enable_driver(struct ib_device *dev) .query_srq = mlx5_ib_query_srq, .query_ucontext = mlx5_ib_query_ucontext, .reg_user_mr = mlx5_ib_reg_user_mr,
- .reg_user_mr_dmabuf = mlx5_ib_reg_user_mr_dmabuf, .req_notify_cq = mlx5_ib_arm_cq, .rereg_user_mr = mlx5_ib_rereg_user_mr, .resize_cq = mlx5_ib_resize_cq,
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index 718e59f..6f4d1b4 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1,6 +1,7 @@ /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ /*
- Copyright (c) 2013-2020, Mellanox Technologies inc. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
*/
#ifndef MLX5_IB_H
@@ -704,6 +705,12 @@ static inline bool is_odp_mr(struct mlx5_ib_mr *mr) mr->umem->is_odp; }
+static inline bool is_dmabuf_mr(struct mlx5_ib_mr *mr) {
- return IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) && mr->umem &&
mr->umem->is_dmabuf;+}
Didn't we agree that IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING) should be checked earlier
upon mlx5_ib_reg_user_mr_dmabuf () to fully prevent the functionality when ODP is not supported ? see note on previous versions.
Yes, it's now checked at the entry point of the uverbs command for registering dmabuf mr (see the following lines extracted from patch 0003 in this set). The check is on the device's on-demand paging cap, which is reset when CONFIG_INFINIBAND_ON_DEMAND_PAGING is not enabled.
+ if (!(pd->device->attrs.device_cap_flags & IB_DEVICE_ON_DEMAND_PAGING)) + return -EOPNOTSUPP;
struct mlx5_ib_mw { struct ib_mw ibmw; struct mlx5_core_mkey mmkey; @@ -1239,6 +1246,10 @@ int mlx5_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_udata *udata); +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start,
u64 length, u64 virt_addr,int fd, int access_flags,struct ib_udata *udata);@@ -1249,11 +1260,13 @@ int mlx5_ib_advise_mr(struct ib_pd *pd, int mlx5_ib_dealloc_mw(struct ib_mw *mw); int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags); struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd, struct ib_udata *udata, int access_flags); void mlx5_ib_free_implicit_mr(struct mlx5_ib_mr *mr); void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr); +void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr); int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, u64 length, u64 virt_addr, int access_flags, struct ib_pd *pd, struct ib_udata *udata); @@ -1341,6 +1354,7 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, struct ib_sge *sg_list, u32 num_sge); int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable); +int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr); #else /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */ static inline void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev) { @@ -1366,6 +1380,10 @@ static inline int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) { return -EOPNOTSUPP; } +static inline int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- return -EOPNOTSUPP;
+} #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */
extern const struct mmu_interval_notifier_ops mlx5_mn_ops; diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index b6116f6..e3be1f5 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1,5 +1,6 @@ /*
- Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
- Copyright (c) 2020, Intel Corporation. All rights reserved.
- This software is available to you under a choice of one of two
- licenses. You may choose to be licensed under the terms of the GNU
@@ -36,6 +37,8 @@ #include <linux/debugfs.h> #include <linux/export.h> #include <linux/delay.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h> #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <rdma/ib_verbs.h> @@ -957,6 +960,16 @@ static struct mlx5_cache_ent *mr_cache_ent_from_order(struct mlx5_ib_dev *dev, return &cache->ent[order]; }
+static unsigned int mlx5_umem_dmabuf_default_pgsz(struct ib_umem *umem,
u64 iova)+{
- if ((iova ^ umem->address) & (PAGE_SIZE - 1))
return 0;Also here, see my note from previous versions, can user space trigger this by supplying some invalid input ?
If so, this may cause a WARN_ON() in the caller (i.e.alloc_mr_from_cache) that we should prevent from being triggered by user space application.
I missed this one. Yes, this can be triggered by user input and the WARN_ON needs to be dropped.
- umem->iova = iova;
- return PAGE_SIZE;
+}
- static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags)
@@ -966,7 +979,11 @@ static struct mlx5_ib_mr *alloc_mr_from_cache(struct ib_pd *pd, struct mlx5_ib_mr *mr; unsigned int page_size;
- page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova);
- if (umem->is_dmabuf)
page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova);- else
page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size,0, iova);@@ -1212,8 +1229,10 @@ int mlx5_ib_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages,
/*
- Send the DMA list to the HW for a normal MR using UMR.
- Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP
*/
- flag may be used.
-static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) { struct mlx5_ib_dev *dev = mr->dev; struct device *ddev = &dev->mdev->pdev->dev; @@ -1255,6 +1274,10 @@ static int mlx5_ib_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) cur_mtt->ptag = cpu_to_be64(rdma_block_iter_dma_address(&biter) | MLX5_IB_MTT_PRESENT);
if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP))cur_mtt->ptag = 0;- cur_mtt++; }
@@ -1291,8 +1314,11 @@ static struct mlx5_ib_mr *reg_create(struct ib_mr *ibmr, struct ib_pd *pd, int err; bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg));
- page_size =
mlx5_umem_find_best_pgsz(umem, mkc, log_page_size, 0, iova);
- if (umem->is_dmabuf)
page_size = mlx5_umem_dmabuf_default_pgsz(umem, iova);- else
page_size = mlx5_umem_find_best_pgsz(umem, mkc, log_page_size,0, iova);@@ -1572,6 +1598,96 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, return ERR_PTR(err); }
+static void mlx5_ib_dmabuf_invalidate_cb(struct dma_buf_attachment *attach) +{
- struct ib_umem_dmabuf *umem_dmabuf = attach->importer_priv;
- struct mlx5_ib_mr *mr = umem_dmabuf->private;
- dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv);
- if (!umem_dmabuf->sgt)
return;- mlx5_ib_update_mr_pas(mr, MLX5_IB_UPD_XLT_ZAP);
- ib_umem_dmabuf_unmap_pages(umem_dmabuf);
+}
+static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = {
- .allow_peer2peer = 1,
- .move_notify = mlx5_ib_dmabuf_invalidate_cb,
+};
+struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset,
u64 length, u64 virt_addr,int fd, int access_flags,struct ib_udata *udata)+{
- struct mlx5_ib_dev *dev = to_mdev(pd->device);
- struct mlx5_ib_mr *mr = NULL;
- struct ib_umem *umem;
- int err;
- if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM))
return ERR_PTR(-EOPNOTSUPP);- mlx5_ib_dbg(dev,
"offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n",offset, virt_addr, length, fd, access_flags);- if (!mlx5_ib_can_load_pas_with_umr(dev, length))
return ERR_PTR(-EINVAL);- umem = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, access_flags,
&mlx5_ib_dmabuf_attach_ops);- if (IS_ERR(umem)) {
mlx5_ib_dbg(dev, "umem get failed (%ld)\n", PTR_ERR(umem));return ERR_PTR(PTR_ERR(umem));- }
- mr = alloc_mr_from_cache(pd, umem, virt_addr, access_flags);
- if (IS_ERR(mr))
mr = NULL;- if (!mr) {
mutex_lock(&dev->slow_path_mutex);mr = reg_create(NULL, pd, umem, virt_addr, access_flags,false);mutex_unlock(&dev->slow_path_mutex);- }
- if (IS_ERR(mr)) {
err = PTR_ERR(mr);goto error;- }
- mlx5_ib_dbg(dev, "mkey 0x%x\n", mr->mmkey.key);
- mr->umem = umem;
- atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages);
- set_mr_fields(dev, mr, length, access_flags);
- to_ib_umem_dmabuf(umem)->private = mr;
- init_waitqueue_head(&mr->q_deferred_work);
- atomic_set(&mr->num_deferred_work, 0);
- err = xa_err(xa_store(&dev->odp_mkeys,
mlx5_base_mkey(mr->mmkey.key), &mr->mmkey,GFP_KERNEL));- if (err) {
dereg_mr(dev, mr);return ERR_PTR(err);- }
- err = mlx5_ib_init_dmabuf_mr(mr);
- if (err) {
dereg_mr(dev, mr);return ERR_PTR(err);- }
- return &mr->ibmr;
+error:
- ib_umem_release(umem);
- return ERR_PTR(err);
+}
- /**
- mlx5_mr_cache_invalidate - Fence all DMA on the MR
- @mr: The MR to fence
@@ -1640,7 +1756,7 @@ int mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, if (!mr->umem) return -EINVAL;
- if (is_odp_mr(mr))
if (is_odp_mr(mr) || is_dmabuf_mr(mr)) return -EOPNOTSUPP;
if (flags & IB_MR_REREG_TRANS) {
@@ -1804,6 +1920,8 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr) /* Stop all DMA */ if (is_odp_mr(mr)) mlx5_ib_fence_odp_mr(mr);
- else if (is_dmabuf_mr(mr))
mlx5_ib_fence_dmabuf_mr(mr);diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index 5c853ec..35d6770 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -33,6 +33,8 @@ #include <rdma/ib_umem.h> #include <rdma/ib_umem_odp.h> #include <linux/kernel.h> +#include <linux/dma-buf.h> +#include <linux/dma-resv.h>
#include "mlx5_ib.h" #include "cmd.h" @@ -664,6 +666,37 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr) dma_fence_odp_mr(mr); }
+/**
- mlx5_ib_fence_dmabuf_mr - Stop all access to the dmabuf MR
- @mr: to fence
- On return no parallel threads will be touching this MR and no DMA will be
- active.
- */
+void mlx5_ib_fence_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem);
- /* Prevent new page faults and prefetch requests from succeeding */
- xa_erase(&mr->dev->odp_mkeys, mlx5_base_mkey(mr->mmkey.key));
- /* Wait for all running page-fault handlers to finish. */
- synchronize_srcu(&mr->dev->odp_srcu);
- wait_event(mr->q_deferred_work, !atomic_read(&mr->num_deferred_work));
- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- mlx5_mr_cache_invalidate(mr);
- umem_dmabuf->private = NULL;
- ib_umem_dmabuf_unmap_pages(umem_dmabuf);
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
- if (!mr->cache_ent) {
mlx5_core_destroy_mkey(mr->dev->mdev, &mr->mmkey);WARN_ON(mr->descs);- }
+}
- #define MLX5_PF_FLAGS_DOWNGRADE BIT(1) #define MLX5_PF_FLAGS_SNAPSHOT BIT(2) #define MLX5_PF_FLAGS_ENABLE BIT(3)
@@ -797,6 +830,41 @@ static int pagefault_implicit_mr(struct mlx5_ib_mr *imr, return ret; }
+static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt,
u32 *bytes_mapped, u32 flags)+{
- struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem);
- u32 xlt_flags = 0;
- int err;
- unsigned int page_size;
- if (flags & MLX5_PF_FLAGS_ENABLE)
xlt_flags |= MLX5_IB_UPD_XLT_ENABLE;- dma_resv_lock(umem_dmabuf->attach->dmabuf->resv, NULL);
- err = ib_umem_dmabuf_map_pages(umem_dmabuf);
- if (!err) {
page_size = mlx5_umem_find_best_pgsz(&umem_dmabuf->umem, mkc,log_page_size, 0,umem_dmabuf->umem.iova);if (unlikely(page_size < PAGE_SIZE)) {ib_umem_dmabuf_unmap_pages(umem_dmabuf);err = -EINVAL;} else {err = mlx5_ib_update_mr_pas(mr, xlt_flags);}- }
- dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
- if (err)
return err;- if (bytes_mapped)
*bytes_mapped += bcnt;- return ib_umem_num_pages(mr->umem);
+}
- /*
- Returns:
- -EFAULT: The io_virt->bcnt is not within the MR, it covers pages that are
@@ -815,6 +883,9 @@ static int pagefault_mr(struct mlx5_ib_mr *mr, u64 io_virt, size_t bcnt, if (unlikely(io_virt < mr->mmkey.iova)) return -EFAULT;
- if (mr->umem->is_dmabuf)
return pagefault_dmabuf_mr(mr, bcnt, bytes_mapped, flags);- if (!odp->is_implicit_odp) { u64 user_va;
@@ -845,6 +916,16 @@ int mlx5_ib_init_odp_mr(struct mlx5_ib_mr *mr, bool enable) return ret >= 0 ? 0 : ret; }
+int mlx5_ib_init_dmabuf_mr(struct mlx5_ib_mr *mr) +{
- int ret;
- ret = pagefault_dmabuf_mr(mr, mr->umem->length, NULL,
MLX5_PF_FLAGS_ENABLE);- return ret >= 0 ? 0 : ret;
+}
- struct pf_frame { struct pf_frame *next; u32 key;
@@ -1747,7 +1828,6 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_core_mkey *mmkey;
struct ib_umem_odp *odp; struct mlx5_ib_mr *mr;
lockdep_assert_held(&dev->odp_srcu);
@@ -1761,11 +1841,9 @@ static void destroy_prefetch_work(struct prefetch_mr_work *work) if (mr->ibmr.pd != pd) return NULL;
- odp = to_ib_umem_odp(mr->umem);
- /* prefetch with write-access must be supported by the MR */ if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE &&
!odp->umem.writable)
!mr->umem->writable)return NULL;
return mr;
On Tue, Dec 08, 2020 at 02:39:11PM -0800, Jianxin Xiong wrote:
This is the fourteenth version of the patch set. Changelog:
v14:
- Check return value of dma_fence_wait()
- Fix a dma-buf leak in ib_umem_dmabuf_get()
- Fix return value type cast for ib_umem_dmabuf_get()
- Return -EOPNOTSUPP instead of -EINVAL for unimplemented functions
- Remove an unnecessary use of unlikely()
- Remove left-over commit message resulted from rebase
You should probably slow down a bit, v13 needed rebasing too.
Jason
dri-devel@lists.freedesktop.org
-
Jason Gunthorpe -
Jianxin Xiong -
Leon Romanovsky -
Xiong, Jianxin
-
Yishai Hadas