Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

14 participants
2902 discussions

[PATCH v4] dma-buf: heaps: Introduce a new heap for reserved memory

by Maxime Ripard

Some reserved memory regions might have particular memory setup or attributes that make them good candidates for heaps. Let's provide a heap type that will create a new heap for each reserved memory region flagged as such. Signed-off-by: Maxime Ripard <mripard(a)kernel.org> --- Hi, This series is the follow-up of the discussion that John and I had some time ago here: https://lore.kernel.org/all/CANDhNCquJn6bH3KxKf65BWiTYLVqSd9892-xtFDHHqqyrr… The initial problem we were discussing was that I'm currently working on a platform which has a memory layout with ECC enabled. However, enabling the ECC has a number of drawbacks on that platform: lower performance, increased memory usage, etc. So for things like framebuffers, the trade-off isn't great and thus there's a memory region with ECC disabled to allocate from for such use cases. After a suggestion from John, I chose to first start using heap allocations flags to allow for userspace to ask for a particular ECC setup. This is then backed by a new heap type that runs from reserved memory chunks flagged as such, and the existing DT properties to specify the ECC properties. After further discussion, it was considered that flags were not the right solution, and relying on the names of the heaps would be enough to let userspace know the kind of buffer it deals with. Thus, even though the uAPI part of it has been dropped in this second version, we still need a driver to create heaps out of carved-out memory regions. In addition to the original usecase, a similar driver can be found in BSPs from most vendors, so I believe it would be a useful addition to the kernel. I submitted a draft PR to the DT schema for the bindings used in this PR: https://github.com/devicetree-org/dt-schema/pull/138 Let me know what you think, Maxime --- Changes in v4: - Rebased on 6.15-rc7 - Map buffers only when map is actually called, not at allocation time - Deal with restricted-dma-pool and shared-dma-pool - Reword Kconfig options - Properly report dma_map_sgtable failures - Link to v3: https://lore.kernel.org/r/20250407-dma-buf-ecc-heap-v3-0-97cdd36a5f29@kerne… Changes in v3: - Reworked global variable patch - Link to v2: https://lore.kernel.org/r/20250401-dma-buf-ecc-heap-v2-0-043fd006a1af@kerne… Changes in v2: - Add vmap/vunmap operations - Drop ECC flags uapi - Rebase on top of 6.14 - Link to v1: https://lore.kernel.org/r/20240515-dma-buf-ecc-heap-v1-0-54cbbd049511@kerne… --- drivers/dma-buf/heaps/Kconfig | 8 + drivers/dma-buf/heaps/Makefile | 1 + drivers/dma-buf/heaps/carveout_heap.c | 388 ++++++++++++++++++++++++++++++++++ 3 files changed, 397 insertions(+) diff --git a/drivers/dma-buf/heaps/Kconfig b/drivers/dma-buf/heaps/Kconfig index a5eef06c422644e8aadaf5aff2bd9a33c49c1ba3..1ce4f6828d8c06bfdd7bc2e5127707f1778586e6 100644 --- a/drivers/dma-buf/heaps/Kconfig +++ b/drivers/dma-buf/heaps/Kconfig @@ -1,5 +1,13 @@ +config DMABUF_HEAPS_CARVEOUT + bool "DMA-BUF Carveout Heaps" + depends on DMABUF_HEAPS + help + Choose this option to enable the carveout dmabuf heap. The carveout + heap is backed by pages from reserved memory regions flagged as + exportable. If in doubt, say Y. + config DMABUF_HEAPS_SYSTEM bool "DMA-BUF System Heap" depends on DMABUF_HEAPS help Choose this option to enable the system dmabuf heap. The system heap diff --git a/drivers/dma-buf/heaps/Makefile b/drivers/dma-buf/heaps/Makefile index 974467791032ffb8a7aba17b1407d9a19b3f3b44..b734647ad5c84f449106748160258e372f153df2 100644 --- a/drivers/dma-buf/heaps/Makefile +++ b/drivers/dma-buf/heaps/Makefile @@ -1,3 +1,4 @@ # SPDX-License-Identifier: GPL-2.0 +obj-$(CONFIG_DMABUF_HEAPS_CARVEOUT) += carveout_heap.o obj-$(CONFIG_DMABUF_HEAPS_SYSTEM) += system_heap.o obj-$(CONFIG_DMABUF_HEAPS_CMA) += cma_heap.o diff --git a/drivers/dma-buf/heaps/carveout_heap.c b/drivers/dma-buf/heaps/carveout_heap.c new file mode 100644 index 0000000000000000000000000000000000000000..3fac190545bc6f853c18a614f6d89176ed2d7df6 --- /dev/null +++ b/drivers/dma-buf/heaps/carveout_heap.c @@ -0,0 +1,388 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include <linux/dma-buf.h> +#include <linux/dma-heap.h> +#include <linux/genalloc.h> +#include <linux/highmem.h> +#include <linux/of_reserved_mem.h> + +struct carveout_heap_priv { + struct dma_heap *heap; + struct gen_pool *pool; +}; + +struct carveout_heap_buffer_priv { + struct mutex lock; + struct list_head attachments; + + unsigned long num_pages; + struct carveout_heap_priv *heap; + phys_addr_t paddr; + void *vaddr; + unsigned int vmap_cnt; +}; + +struct carveout_heap_attachment { + struct list_head head; + struct sg_table table; + + struct device *dev; + bool mapped; +}; + +static int carveout_heap_attach(struct dma_buf *buf, + struct dma_buf_attachment *attachment) +{ + struct carveout_heap_buffer_priv *priv = buf->priv; + struct carveout_heap_attachment *a; + struct sg_table *sgt; + unsigned long len = priv->num_pages * PAGE_SIZE; + int ret; + + a = kzalloc(sizeof(*a), GFP_KERNEL); + if (!a) + return -ENOMEM; + INIT_LIST_HEAD(&a->head); + a->dev = attachment->dev; + attachment->priv = a; + + sgt = &a->table; + ret = sg_alloc_table(sgt, 1, GFP_KERNEL); + if (ret) + goto err_cleanup_attach; + + sg_set_buf(sgt->sgl, priv->vaddr, len); + + mutex_lock(&priv->lock); + list_add(&a->head, &priv->attachments); + mutex_unlock(&priv->lock); + + return 0; + +err_cleanup_attach: + kfree(a); + return ret; +} + +static void carveout_heap_detach(struct dma_buf *dmabuf, + struct dma_buf_attachment *attachment) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + struct carveout_heap_attachment *a = attachment->priv; + + mutex_lock(&priv->lock); + list_del(&a->head); + mutex_unlock(&priv->lock); + + sg_free_table(&a->table); + kfree(a); +} + +static struct sg_table * +carveout_heap_map_dma_buf(struct dma_buf_attachment *attachment, + enum dma_data_direction direction) +{ + struct carveout_heap_attachment *a = attachment->priv; + struct sg_table *table = &a->table; + int ret; + + ret = dma_map_sgtable(a->dev, table, direction, 0); + if (ret) + return ERR_PTR(ret); + + a->mapped = true; + + return table; +} + +static void carveout_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, + struct sg_table *table, + enum dma_data_direction direction) +{ + struct carveout_heap_attachment *a = attachment->priv; + + a->mapped = false; + dma_unmap_sgtable(a->dev, table, direction, 0); +} + +static int +carveout_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf, + enum dma_data_direction direction) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + struct carveout_heap_attachment *a; + unsigned long len = priv->num_pages * PAGE_SIZE; + + mutex_lock(&priv->lock); + + if (priv->vmap_cnt) + invalidate_kernel_vmap_range(priv->vaddr, len); + + list_for_each_entry(a, &priv->attachments, head) { + if (!a->mapped) + continue; + + dma_sync_sgtable_for_cpu(a->dev, &a->table, direction); + } + + mutex_unlock(&priv->lock); + + return 0; +} + +static int +carveout_heap_dma_buf_end_cpu_access(struct dma_buf *dmabuf, + enum dma_data_direction direction) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + struct carveout_heap_attachment *a; + unsigned long len = priv->num_pages * PAGE_SIZE; + + mutex_lock(&priv->lock); + + if (priv->vmap_cnt) + flush_kernel_vmap_range(priv->vaddr, len); + + list_for_each_entry(a, &priv->attachments, head) { + if (!a->mapped) + continue; + + dma_sync_sgtable_for_device(a->dev, &a->table, direction); + } + + mutex_unlock(&priv->lock); + + return 0; +} + +static int carveout_heap_mmap(struct dma_buf *dmabuf, + struct vm_area_struct *vma) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + unsigned long len = priv->num_pages * PAGE_SIZE; + + return vm_iomap_memory(vma, priv->paddr, len); +} + +static int carveout_heap_vmap(struct dma_buf *dmabuf, struct iosys_map *map) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + unsigned long len = priv->num_pages * PAGE_SIZE; + + mutex_lock(&priv->lock); + + if (!priv->vmap_cnt) { + void *vaddr = memremap(priv->paddr, len, MEMREMAP_WB); + + if (!vaddr) { + mutex_unlock(&priv->lock); + return -ENOMEM; + } + + priv->vaddr = vaddr; + } + + WARN_ON(!priv->vaddr); + iosys_map_set_vaddr(map, priv->vaddr); + priv->vmap_cnt++; + + mutex_unlock(&priv->lock); + + return 0; +} + +static void carveout_heap_vunmap(struct dma_buf *dmabuf, struct iosys_map *map) +{ + struct carveout_heap_buffer_priv *priv = dmabuf->priv; + + mutex_lock(&priv->lock); + + priv->vmap_cnt--; + if (!priv->vmap_cnt) { + memunmap(priv->vaddr); + priv->vaddr = NULL; + } + + mutex_unlock(&priv->lock); + + iosys_map_clear(map); +} + +static void carveout_heap_dma_buf_release(struct dma_buf *buf) +{ + struct carveout_heap_buffer_priv *buffer_priv = buf->priv; + struct carveout_heap_priv *heap_priv = buffer_priv->heap; + unsigned long len = buffer_priv->num_pages * PAGE_SIZE; + + gen_pool_free(heap_priv->pool, buffer_priv->paddr, len); + kfree(buffer_priv); +} + +static const struct dma_buf_ops carveout_heap_buf_ops = { + .attach = carveout_heap_attach, + .detach = carveout_heap_detach, + .map_dma_buf = carveout_heap_map_dma_buf, + .unmap_dma_buf = carveout_heap_unmap_dma_buf, + .begin_cpu_access = carveout_heap_dma_buf_begin_cpu_access, + .end_cpu_access = carveout_heap_dma_buf_end_cpu_access, + .mmap = carveout_heap_mmap, + .vmap = carveout_heap_vmap, + .vunmap = carveout_heap_vunmap, + .release = carveout_heap_dma_buf_release, +}; + +static struct dma_buf *carveout_heap_allocate(struct dma_heap *heap, + unsigned long len, + u32 fd_flags, + u64 heap_flags) +{ + struct carveout_heap_priv *heap_priv = dma_heap_get_drvdata(heap); + struct carveout_heap_buffer_priv *buffer_priv; + DEFINE_DMA_BUF_EXPORT_INFO(exp_info); + struct dma_buf *buf; + phys_addr_t paddr; + /* len is guaranteed to be page-aligned by the framework, so we can use it as is. */ + size_t size = len; + int ret; + + buffer_priv = kzalloc(sizeof(*buffer_priv), GFP_KERNEL); + if (!buffer_priv) + return ERR_PTR(-ENOMEM); + + INIT_LIST_HEAD(&buffer_priv->attachments); + mutex_init(&buffer_priv->lock); + + paddr = gen_pool_alloc(heap_priv->pool, size); + if (!paddr) { + ret = -ENOMEM; + goto err_free_buffer_priv; + } + + buffer_priv->paddr = paddr; + buffer_priv->heap = heap_priv; + buffer_priv->num_pages = size >> PAGE_SHIFT; + + /* create the dmabuf */ + exp_info.exp_name = dma_heap_get_name(heap); + exp_info.ops = &carveout_heap_buf_ops; + exp_info.size = size; + exp_info.flags = fd_flags; + exp_info.priv = buffer_priv; + + buf = dma_buf_export(&exp_info); + if (IS_ERR(buf)) { + ret = PTR_ERR(buf); + goto err_free_buffer; + } + + return buf; + +err_free_buffer: + gen_pool_free(heap_priv->pool, paddr, len); +err_free_buffer_priv: + kfree(buffer_priv); + + return ERR_PTR(ret); +} + +static const struct dma_heap_ops carveout_heap_ops = { + .allocate = carveout_heap_allocate, +}; + +static int __init carveout_heap_setup(struct device_node *node) +{ + struct dma_heap_export_info exp_info = {}; + const struct reserved_mem *rmem; + struct carveout_heap_priv *priv; + struct dma_heap *heap; + struct gen_pool *pool; + int ret; + + rmem = of_reserved_mem_lookup(node); + if (!rmem) + return -EINVAL; + + priv = kzalloc(sizeof(*priv), GFP_KERNEL); + if (!priv) + return -ENOMEM; + + pool = gen_pool_create(PAGE_SHIFT, NUMA_NO_NODE); + if (!pool) { + ret = -ENOMEM; + goto err_cleanup_heap; + } + priv->pool = pool; + + ret = gen_pool_add(pool, rmem->base, rmem->size, NUMA_NO_NODE); + if (ret) + goto err_release_mem_region; + + exp_info.name = node->full_name; + exp_info.ops = &carveout_heap_ops; + exp_info.priv = priv; + + heap = dma_heap_add(&exp_info); + if (IS_ERR(heap)) { + ret = PTR_ERR(heap); + goto err_release_mem_region; + } + priv->heap = heap; + + return 0; + +err_release_mem_region: + gen_pool_destroy(pool); +err_cleanup_heap: + kfree(priv); + return ret; +} + +static int __init carveout_heap_init(void) +{ + struct device_node *rmem_node; + struct device_node *node; + int ret; + + rmem_node = of_find_node_by_path("/reserved-memory"); + if (!rmem_node) + return 0; + + for_each_child_of_node(rmem_node, node) { + /* + * TODO: shared-dma-pools register either a CMA or + * coherent memory region depending on whether the + * reusable property is set. + * + * In order to avoid issues, we would need to use the + * respective allocators if the shared-dma-pool property + * was set. + */ + if (of_device_is_compatible(node, "shared-dma-pool")) { + pr_warn("%pOFn: carveout heap driver doesn't support shared-dma-pools.", + node); + continue; + } + + /* + * TODO: restricted-dma-pools register in swiotlb. In + * order to avoid issues, we would need to use the + * swiotlb if that compatible was set. + */ + if (of_device_is_compatible(node, "restricted-dma-pool")) { + pr_warn("%pOFn: carveout heap driver doesn't support restricted-dma-pools.", + node); + continue; + } + + if (!of_property_read_bool(node, "export")) + continue; + + ret = carveout_heap_setup(node); + if (ret) + return ret; + } + + return 0; +} + +module_init(carveout_heap_init); --- base-commit: a17f109942554a751be2c3b05f898f1b3ae98c78 change-id: 20240515-dma-buf-ecc-heap-28a311d2c94e Best regards, -- Maxime Ripard <mripard(a)kernel.org>

1 month, 1 week

Re: [PATCH v5 08/10] accel/rocket: Add IOCTLs for synchronizing memory accesses

by Lucas Stach

Hi Tomeu, Am Dienstag, dem 20.05.2025 um 12:27 +0200 schrieb Tomeu Vizoso: > The NPU cores have their own access to the memory bus, and this isn't > cache coherent with the CPUs. > > Add IOCTLs so userspace can mark when the caches need to be flushed, and > also when a writer job needs to be waited for before the buffer can be > accessed from the CPU. > > Initially based on the same IOCTLs from the Etnaviv driver. > > v2: > - Don't break UABI by reordering the IOCTL IDs (Jeff Hugo) > > v3: > - Check that padding fields in IOCTLs are zero (Jeff Hugo) > > Signed-off-by: Tomeu Vizoso <tomeu(a)tomeuvizoso.net> > --- > drivers/accel/rocket/rocket_drv.c | 2 + > drivers/accel/rocket/rocket_gem.c | 80 +++++++++++++++++++++++++++++++++++++++ > drivers/accel/rocket/rocket_gem.h | 5 +++ > include/uapi/drm/rocket_accel.h | 37 ++++++++++++++++++ > 4 files changed, 124 insertions(+) > > diff --git a/drivers/accel/rocket/rocket_drv.c b/drivers/accel/rocket/rocket_drv.c > index fef9b93372d3f65c41c1ac35a9bfa0c01ee721a5..c06e66939e6c39909fe08bef3c4f301b07bf8fbf 100644 > --- a/drivers/accel/rocket/rocket_drv.c > +++ b/drivers/accel/rocket/rocket_drv.c > @@ -59,6 +59,8 @@ static const struct drm_ioctl_desc rocket_drm_driver_ioctls[] = { > > ROCKET_IOCTL(CREATE_BO, create_bo), > ROCKET_IOCTL(SUBMIT, submit), > + ROCKET_IOCTL(PREP_BO, prep_bo), > + ROCKET_IOCTL(FINI_BO, fini_bo), > }; > > DEFINE_DRM_ACCEL_FOPS(rocket_accel_driver_fops); > diff --git a/drivers/accel/rocket/rocket_gem.c b/drivers/accel/rocket/rocket_gem.c > index 8a8a7185daac4740081293aae6945c9b2bbeb2dd..cdc5238a93fa5978129dc1ac8ec8de955160dc18 100644 > --- a/drivers/accel/rocket/rocket_gem.c > +++ b/drivers/accel/rocket/rocket_gem.c > @@ -129,3 +129,83 @@ int rocket_ioctl_create_bo(struct drm_device *dev, void *data, struct drm_file * > > return ret; > } > + > +static inline enum dma_data_direction rocket_op_to_dma_dir(u32 op) > +{ > + if (op & ROCKET_PREP_READ) > + return DMA_FROM_DEVICE; > + else if (op & ROCKET_PREP_WRITE) > + return DMA_TO_DEVICE; > + else > + return DMA_BIDIRECTIONAL; > +} This has copied over the bug fixed in etnaviv commit 58979ad6330a ("drm/etnaviv: fix DMA direction handling for cached RW buffers") Regards, Lucas

1 month, 1 week

Re: [PATCH v4 05/10] accel/rocket: Add a new driver for Rockchip's NPU

by Krzysztof Kozlowski

On Mon, May 19, 2025 at 03:43:37PM GMT, Tomeu Vizoso wrote: > +#endif > diff --git a/drivers/accel/rocket/rocket_device.c b/drivers/accel/rocket/rocket_device.c > new file mode 100644 > index 0000000000000000000000000000000000000000..bb469ac87d36249157f4ba9d9f7106ad558309e4 > --- /dev/null > +++ b/drivers/accel/rocket/rocket_device.c > @@ -0,0 +1,39 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* Copyright 2024-2025 Tomeu Vizoso <tomeu(a)tomeuvizoso.net> */ > + > +#include <linux/clk.h> > +#include <linux/dev_printk.h> > + > +#include "rocket_device.h" > + > +int rocket_device_init(struct rocket_device *rdev) > +{ > + struct device *dev = rdev->cores[0].dev; > + int err; > + > + rdev->clk_npu = devm_clk_get(dev, "npu"); > + if (IS_ERR(rdev->clk_npu)) { > + err = PTR_ERR(rdev->clk_npu); > + dev_err(dev, "devm_clk_get failed %d for clock npu\n", err); > + return err; > + } That's probe path? so use standard syntax: return dev_err_probe(). One line instead of four. > + > + rdev->pclk = devm_clk_get(dev, "pclk"); > + if (IS_ERR(rdev->pclk)) { > + err = PTR_ERR(rdev->pclk); > + dev_err(dev, "devm_clk_get failed %d for clock pclk\n", err); > + return err; Same here... except that this should be blk API and entire function gets smaller. > + } > + > + /* Initialize core 0 (top) */ > + err = rocket_core_init(&rdev->cores[0]); > + if (err) > + return err; > + > + return 0; > +} ... > +static int rocket_device_runtime_resume(struct device *dev) > +{ > + struct rocket_device *rdev = dev_get_drvdata(dev); > + int core = find_core_for_dev(dev); > + int err = 0; > + > + if (core < 0) > + return -ENODEV; > + > + if (core == 0) { > + err = clk_prepare_enable(rdev->clk_npu); > + if (err) { > + dev_err(dev, "clk_prepare_enable failed %d for clock npu\n", err); > + return err; > + } > + > + err = clk_prepare_enable(rdev->pclk); > + if (err) { > + dev_err(dev, "clk_prepare_enable failed %d for clock pclk\n", err); > + goto error_clk_npu; > + } > + } > + > + err = clk_prepare_enable(rdev->cores[core].a_clk); > + if (err) { > + dev_err(dev, "clk_prepare_enable failed %d for a_clk in core %d\n", err, core); > + goto error_pclk; > + } > + > + err = clk_prepare_enable(rdev->cores[core].h_clk); > + if (err) { > + dev_err(dev, "clk_prepare_enable failed %d for h_clk in core %d\n", err, core); > + goto error_a_clk; > + } All four above calls could be just one call with bulk API. > + > + return 0; > + > +error_a_clk: > + clk_disable_unprepare(rdev->cores[core].a_clk); > + > +error_pclk: > + if (core == 0) > + clk_disable_unprepare(rdev->pclk); > + > +error_clk_npu: > + if (core == 0) > + clk_disable_unprepare(rdev->clk_npu); And all this would be gone... > + > + return err; Best regards, Krzysztof

1 month, 1 week

Re: [PATCH v4 01/10] dt-bindings: npu: rockchip,rknn: Add bindings

by Krzysztof Kozlowski

On Mon, May 19, 2025 at 03:43:33PM GMT, Tomeu Vizoso wrote: > Add the bindings for the Neural Processing Unit IP from Rockchip. > > v2: > - Adapt to new node structure (one node per core, each with its own > IOMMU) > - Several misc. fixes from Sebastian Reichel > > v3: > - Split register block in its constituent subblocks, and only require > the ones that the kernel would ever use (Nicolas Frattaroli) > - Group supplies (Rob Herring) > - Explain the way in which the top core is special (Rob Herring) > > v4: > - Change required node name to npu@ (Rob Herring and Krzysztof Kozlowski) > - Remove unneeded items: (Krzysztof Kozlowski) > - Fix use of minItems/maxItems (Krzysztof Kozlowski) > - Add reg-names to list of required properties (Krzysztof Kozlowski) > - Fix example (Krzysztof Kozlowski) > > Signed-off-by: Tomeu Vizoso <tomeu(a)tomeuvizoso.net> > Signed-off-by: Sebastian Reichel <sebastian.reichel(a)collabora.com> This order of SoB is still odd. You as person sending it should be the last signing person. Are you sure you are using b4 for managing trailers? I would expect it to re-order these on every update and this is already v4. > --- > .../bindings/npu/rockchip,rknn-core.yaml | 149 +++++++++++++++++++++ Filename matching compatible, so rockchip,rk3588-rknn-core.yaml > 1 file changed, 149 insertions(+) > > diff --git a/Documentation/devicetree/bindings/npu/rockchip,rknn-core.yaml b/Documentation/devicetree/bindings/npu/rockchip,rknn-core.yaml > new file mode 100644 > index 0000000000000000000000000000000000000000..fafd0b01da215c7396262012988e364ef07ea137 > --- /dev/null > +++ b/Documentation/devicetree/bindings/npu/rockchip,rknn-core.yaml > @@ -0,0 +1,149 @@ > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) > +%YAML 1.2 > +--- > +$id: http://devicetree.org/schemas/npu/rockchip,rknn-core.yaml# > +$schema: http://devicetree.org/meta-schemas/core.yaml# > + > +title: Neural Processing Unit IP from Rockchip > + > +maintainers: > + - Tomeu Vizoso <tomeu(a)tomeuvizoso.net> > + > +description: > + Rockchip IP for accelerating inference of neural networks, based on NVIDIA's > + open source NVDLA IP. > + > + There is to be a node per each core in the NPU. In Rockchip's design there > + will be one core that is special and needs to be powered on before any of the > + other cores can be used. This special core is called the top core and should > + have the compatible string that corresponds to top cores. > + > +properties: > + $nodename: > + pattern: '^npu@[a-f0-9]+$' > + > + compatible: > + oneOf: Drop... if you followed my advice you would notice it is not necessary. > + - enum: > + - rockchip,rk3588-rknn-core-top > + - enum: > + - rockchip,rk3588-rknn-core My comments were only partially implemented. This syntax is really not readable and not necessary and I asked to make it part of previous enum. This is just one enum: compatible: enum: - foo - bar Best regards, Krzysztof

1 month, 1 week

[PATCH v5 00/40] drm/msm: sparse / "VM_BIND" support

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> Conversion to DRM GPU VA Manager[1], and adding support for Vulkan Sparse Memory[2] in the form of: 1. A new VM_BIND submitqueue type for executing VM MSM_SUBMIT_BO_OP_MAP/ MAP_NULL/UNMAP commands 2. A new VM_BIND ioctl to allow submitting batches of one or more MAP/MAP_NULL/UNMAP commands to a VM_BIND submitqueue I did not implement support for synchronous VM_BIND commands. Since userspace could just immediately wait for the `SUBMIT` to complete, I don't think we need this extra complexity in the kernel. Synchronous/immediate VM_BIND operations could be implemented with a 2nd VM_BIND submitqueue. The corresponding mesa MR: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32533 Changes in v5: - Improved drm/sched enqueue_credit comments, and better define the return from drm_sched_entity_push_job() - Improve DRM_GPUVM_VA_WEAK_REF comments, and additional WARN_ON()s to make it clear that some of the gpuvm functionality is not available in this mode. - Link to v4: https://lore.kernel.org/all/20250514175527.42488-1-robdclark@gmail.com/ Changes in v4: - Various locking/etc fixes - Optimize the pgtable preallocation. If userspace sorts the VM_BIND ops then the kernel detects ops that fall into the same 2MB last level PTD to avoid duplicate page preallocation. - Add way to throttle pushing jobs to the scheduler, to cap the amount of potentially temporary prealloc'd pgtable pages. - Add vm_log to devcoredump for debugging. If the vm_log_shift module param is set, keep a log of the last 1<<vm_log_shift VM updates for easier debugging of faults/crashes. - Link to v3: https://lore.kernel.org/all/20250428205619.227835-1-robdclark@gmail.com/ Changes in v3: - Switched to seperate VM_BIND ioctl. This makes the UABI a bit cleaner, but OTOH the userspace code was cleaner when the end result of either type of VkQueue lead to the same ioctl. So I'm a bit on the fence. - Switched to doing the gpuvm bookkeeping synchronously, and only deferring the pgtable updates. This avoids needing to hold any resv locks in the fence signaling path, resolving the last shrinker related lockdep complaints. OTOH it means userspace can trigger invalid pgtable updates with multiple VM_BIND queues. In this case, we ensure that unmaps happen completely (to prevent userspace from using this to access free'd pages), mark the context as unusable, and move on with life. - Link to v2: https://lore.kernel.org/all/20250319145425.51935-1-robdclark@gmail.com/ Changes in v2: - Dropped Bibek Kumar Patro's arm-smmu patches[3], which have since been merged. - Pre-allocate all the things, and drop HACK patch which disabled shrinker. This includes ensuring that vm_bo objects are allocated up front, pre- allocating VMA objects, and pre-allocating pages used for pgtable updates. The latter utilizes io_pgtable_cfg callbacks for pgtable alloc/free, that were initially added for panthor. - Add back support for BO dumping for devcoredump. - Link to v1 (RFC): https://lore.kernel.org/dri-devel/20241207161651.410556-1-robdclark@gmail.c… [1] https://www.kernel.org/doc/html/next/gpu/drm-mm.html#drm-gpuvm [2] https://docs.vulkan.org/spec/latest/chapters/sparsemem.html [3] https://patchwork.kernel.org/project/linux-arm-kernel/list/?series=909700 Rob Clark (40): drm/gpuvm: Don't require obj lock in destructor path drm/gpuvm: Allow VAs to hold soft reference to BOs drm/gem: Add ww_acquire_ctx support to drm_gem_lru_scan() drm/sched: Add enqueue credit limit iommu/io-pgtable-arm: Add quirk to quiet WARN_ON() drm/msm: Rename msm_file_private -> msm_context drm/msm: Improve msm_context comments drm/msm: Rename msm_gem_address_space -> msm_gem_vm drm/msm: Remove vram carveout support drm/msm: Collapse vma allocation and initialization drm/msm: Collapse vma close and delete drm/msm: Don't close VMAs on purge drm/msm: drm_gpuvm conversion drm/msm: Convert vm locking drm/msm: Use drm_gpuvm types more drm/msm: Split out helper to get iommu prot flags drm/msm: Add mmu support for non-zero offset drm/msm: Add PRR support drm/msm: Rename msm_gem_vma_purge() -> _unmap() drm/msm: Drop queued submits on lastclose() drm/msm: Lazily create context VM drm/msm: Add opt-in for VM_BIND drm/msm: Mark VM as unusable on GPU hangs drm/msm: Add _NO_SHARE flag drm/msm: Crashdump prep for sparse mappings drm/msm: rd dumping prep for sparse mappings drm/msm: Crashdec support for sparse drm/msm: rd dumping support for sparse drm/msm: Extract out syncobj helpers drm/msm: Use DMA_RESV_USAGE_BOOKKEEP/KERNEL drm/msm: Add VM_BIND submitqueue drm/msm: Support IO_PGTABLE_QUIRK_NO_WARN_ON drm/msm: Support pgtable preallocation drm/msm: Split out map/unmap ops drm/msm: Add VM_BIND ioctl drm/msm: Add VM logging for VM_BIND updates drm/msm: Add VMA unmap reason drm/msm: Add mmu prealloc tracepoint drm/msm: use trylock for debugfs drm/msm: Bump UAPI version drivers/gpu/drm/drm_gem.c | 14 +- drivers/gpu/drm/drm_gpuvm.c | 38 +- drivers/gpu/drm/msm/Kconfig | 1 + drivers/gpu/drm/msm/Makefile | 1 + drivers/gpu/drm/msm/adreno/a2xx_gpu.c | 25 +- drivers/gpu/drm/msm/adreno/a2xx_gpummu.c | 5 +- drivers/gpu/drm/msm/adreno/a3xx_gpu.c | 17 +- drivers/gpu/drm/msm/adreno/a4xx_gpu.c | 17 +- drivers/gpu/drm/msm/adreno/a5xx_debugfs.c | 4 +- drivers/gpu/drm/msm/adreno/a5xx_gpu.c | 22 +- drivers/gpu/drm/msm/adreno/a5xx_power.c | 2 +- drivers/gpu/drm/msm/adreno/a5xx_preempt.c | 10 +- drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 32 +- drivers/gpu/drm/msm/adreno/a6xx_gmu.h | 2 +- drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 49 +- drivers/gpu/drm/msm/adreno/a6xx_gpu_state.c | 6 +- drivers/gpu/drm/msm/adreno/a6xx_preempt.c | 10 +- drivers/gpu/drm/msm/adreno/adreno_device.c | 4 - drivers/gpu/drm/msm/adreno/adreno_gpu.c | 99 +- drivers/gpu/drm/msm/adreno/adreno_gpu.h | 23 +- .../drm/msm/disp/dpu1/dpu_encoder_phys_wb.c | 14 +- drivers/gpu/drm/msm/disp/dpu1/dpu_formats.c | 18 +- drivers/gpu/drm/msm/disp/dpu1/dpu_formats.h | 2 +- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 18 +- drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c | 14 +- drivers/gpu/drm/msm/disp/dpu1/dpu_plane.h | 4 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_crtc.c | 6 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_kms.c | 28 +- drivers/gpu/drm/msm/disp/mdp4/mdp4_plane.c | 12 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_crtc.c | 4 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_kms.c | 19 +- drivers/gpu/drm/msm/disp/mdp5/mdp5_plane.c | 12 +- drivers/gpu/drm/msm/dsi/dsi_host.c | 14 +- drivers/gpu/drm/msm/msm_drv.c | 184 +-- drivers/gpu/drm/msm/msm_drv.h | 35 +- drivers/gpu/drm/msm/msm_fb.c | 18 +- drivers/gpu/drm/msm/msm_fbdev.c | 2 +- drivers/gpu/drm/msm/msm_gem.c | 494 +++--- drivers/gpu/drm/msm/msm_gem.h | 247 ++- drivers/gpu/drm/msm/msm_gem_prime.c | 15 + drivers/gpu/drm/msm/msm_gem_shrinker.c | 104 +- drivers/gpu/drm/msm/msm_gem_submit.c | 295 ++-- drivers/gpu/drm/msm/msm_gem_vma.c | 1471 ++++++++++++++++- drivers/gpu/drm/msm/msm_gpu.c | 211 ++- drivers/gpu/drm/msm/msm_gpu.h | 144 +- drivers/gpu/drm/msm/msm_gpu_trace.h | 14 + drivers/gpu/drm/msm/msm_iommu.c | 302 +++- drivers/gpu/drm/msm/msm_kms.c | 18 +- drivers/gpu/drm/msm/msm_kms.h | 2 +- drivers/gpu/drm/msm/msm_mmu.h | 38 +- drivers/gpu/drm/msm/msm_rd.c | 62 +- drivers/gpu/drm/msm/msm_ringbuffer.c | 10 +- drivers/gpu/drm/msm/msm_submitqueue.c | 96 +- drivers/gpu/drm/msm/msm_syncobj.c | 172 ++ drivers/gpu/drm/msm/msm_syncobj.h | 37 + drivers/gpu/drm/scheduler/sched_entity.c | 19 +- drivers/gpu/drm/scheduler/sched_main.c | 3 + drivers/iommu/io-pgtable-arm.c | 27 +- include/drm/drm_gem.h | 10 +- include/drm/drm_gpuvm.h | 19 +- include/drm/gpu_scheduler.h | 24 +- include/linux/io-pgtable.h | 8 + include/uapi/drm/msm_drm.h | 149 +- 63 files changed, 3526 insertions(+), 1250 deletions(-) create mode 100644 drivers/gpu/drm/msm/msm_syncobj.c create mode 100644 drivers/gpu/drm/msm/msm_syncobj.h -- 2.49.0

1 month, 1 week

[PATCH v5 35/40] drm/msm: Add VM_BIND ioctl

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> Add a VM_BIND ioctl for binding/unbinding buffers into a VM. This is only supported if userspace has opted in to MSM_PARAM_EN_VM_BIND. Signed-off-by: Rob Clark <robdclark(a)chromium.org> --- drivers/gpu/drm/msm/msm_drv.c | 1 + drivers/gpu/drm/msm/msm_drv.h | 4 +- drivers/gpu/drm/msm/msm_gem.c | 40 +- drivers/gpu/drm/msm/msm_gem.h | 4 + drivers/gpu/drm/msm/msm_gem_submit.c | 22 +- drivers/gpu/drm/msm/msm_gem_vma.c | 1073 +++++++++++++++++++++++++- include/uapi/drm/msm_drm.h | 74 +- 7 files changed, 1185 insertions(+), 33 deletions(-) diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c index 89cb7820064f..bdf775897de8 100644 --- a/drivers/gpu/drm/msm/msm_drv.c +++ b/drivers/gpu/drm/msm/msm_drv.c @@ -791,6 +791,7 @@ static const struct drm_ioctl_desc msm_ioctls[] = { DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_NEW, msm_ioctl_submitqueue_new, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_CLOSE, msm_ioctl_submitqueue_close, DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(MSM_SUBMITQUEUE_QUERY, msm_ioctl_submitqueue_query, DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(MSM_VM_BIND, msm_ioctl_vm_bind, DRM_RENDER_ALLOW), }; static void msm_show_fdinfo(struct drm_printer *p, struct drm_file *file) diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index b0add236cbb3..33240afc6365 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -232,7 +232,9 @@ struct drm_gpuvm *msm_kms_init_vm(struct drm_device *dev); bool msm_use_mmu(struct drm_device *dev); int msm_ioctl_gem_submit(struct drm_device *dev, void *data, - struct drm_file *file); + struct drm_file *file); +int msm_ioctl_vm_bind(struct drm_device *dev, void *data, + struct drm_file *file); #ifdef CONFIG_DEBUG_FS unsigned long msm_gem_shrinker_shrink(struct drm_device *dev, unsigned long nr_to_scan); diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index cf509ca42da0..040f0539baa5 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -233,8 +233,7 @@ static void put_pages(struct drm_gem_object *obj) } } -static struct page **msm_gem_get_pages_locked(struct drm_gem_object *obj, - unsigned madv) +struct page **msm_gem_get_pages_locked(struct drm_gem_object *obj, unsigned madv) { struct msm_gem_object *msm_obj = to_msm_bo(obj); @@ -1036,18 +1035,37 @@ static void msm_gem_free_object(struct drm_gem_object *obj) /* * We need to lock any VMs the object is still attached to, but not * the object itself (see explaination in msm_gem_assert_locked()), - * so just open-code this special case: + * so just open-code this special case. + * + * Note that we skip the dance if we aren't attached to any VM. This + * is load bearing. The driver needs to support two usage models: + * + * 1. Legacy kernel managed VM: Userspace expects the VMA's to be + * implicitly torn down when the object is freed, the VMA's do + * not hold a hard reference to the BO. + * + * 2. VM_BIND, userspace managed VM: The VMA holds a reference to the + * BO. This can be dropped when the VM is closed and it's associated + * VMAs are torn down. (See msm_gem_vm_close()). + * + * In the latter case the last reference to a BO can be dropped while + * we already have the VM locked. It would have already been removed + * from the gpuva list, but lockdep doesn't know that. Or understand + * the differences between the two usage models. */ - drm_exec_init(&exec, 0, 0); - drm_exec_until_all_locked (&exec) { - struct drm_gpuvm_bo *vm_bo; - drm_gem_for_each_gpuvm_bo (vm_bo, obj) { - drm_exec_lock_obj(&exec, drm_gpuvm_resv_obj(vm_bo->vm)); - drm_exec_retry_on_contention(&exec); + if (!list_empty(&obj->gpuva.list)) { + drm_exec_init(&exec, 0, 0); + drm_exec_until_all_locked (&exec) { + struct drm_gpuvm_bo *vm_bo; + drm_gem_for_each_gpuvm_bo (vm_bo, obj) { + drm_exec_lock_obj(&exec, + drm_gpuvm_resv_obj(vm_bo->vm)); + drm_exec_retry_on_contention(&exec); + } } + put_iova_spaces(obj, NULL, true); + drm_exec_fini(&exec); /* drop locks */ } - put_iova_spaces(obj, NULL, true); - drm_exec_fini(&exec); /* drop locks */ if (obj->import_attach) { GEM_WARN_ON(msm_obj->vaddr); diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index 8ad25927c604..bfeb0f584ae5 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -73,6 +73,9 @@ struct msm_gem_vm { /** @mmu: The mmu object which manages the pgtables */ struct msm_mmu *mmu; + /** @mmu_lock: Protects access to the mmu */ + struct mutex mmu_lock; + /** * @pid: For address spaces associated with a specific process, this * will be non-NULL: @@ -205,6 +208,7 @@ int msm_gem_get_and_pin_iova(struct drm_gem_object *obj, struct drm_gpuvm *vm, uint64_t *iova); void msm_gem_unpin_iova(struct drm_gem_object *obj, struct drm_gpuvm *vm); void msm_gem_pin_obj_locked(struct drm_gem_object *obj); +struct page **msm_gem_get_pages_locked(struct drm_gem_object *obj, unsigned madv); struct page **msm_gem_pin_pages_locked(struct drm_gem_object *obj); void msm_gem_unpin_pages_locked(struct drm_gem_object *obj); int msm_gem_dumb_create(struct drm_file *file, struct drm_device *dev, diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index 053e6c65780f..9809918d8eb4 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -193,6 +193,7 @@ static int submit_lookup_objects(struct msm_gem_submit *submit, static int submit_lookup_cmds(struct msm_gem_submit *submit, struct drm_msm_gem_submit *args, struct drm_file *file) { + struct msm_context *ctx = file->driver_priv; unsigned i; size_t sz; int ret = 0; @@ -224,6 +225,20 @@ static int submit_lookup_cmds(struct msm_gem_submit *submit, goto out; } + if (msm_context_is_vmbind(ctx)) { + if (submit_cmd.nr_relocs) { + ret = SUBMIT_ERROR(EINVAL, submit, "nr_relocs must be zero"); + goto out; + } + + if (submit_cmd.submit_idx || submit_cmd.submit_offset) { + ret = SUBMIT_ERROR(EINVAL, submit, "submit_idx/offset must be zero"); + goto out; + } + + submit->cmd[i].iova = submit_cmd.iova; + } + submit->cmd[i].type = submit_cmd.type; submit->cmd[i].size = submit_cmd.size / 4; submit->cmd[i].offset = submit_cmd.submit_offset / 4; @@ -527,6 +542,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, struct msm_syncobj_post_dep *post_deps = NULL; struct drm_syncobj **syncobjs_to_reset = NULL; struct sync_file *sync_file = NULL; + unsigned cmds_to_parse; int out_fence_fd = -1; unsigned i; int ret; @@ -650,7 +666,9 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, if (ret) goto out; - for (i = 0; i < args->nr_cmds; i++) { + cmds_to_parse = msm_context_is_vmbind(ctx) ? 0 : args->nr_cmds; + + for (i = 0; i < cmds_to_parse; i++) { struct drm_gem_object *obj; uint64_t iova; @@ -681,7 +699,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, goto out; } - submit->nr_cmds = i; + submit->nr_cmds = args->nr_cmds; idr_preload(GFP_KERNEL); diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c index a105aed82cae..fe41b7a042c3 100644 --- a/drivers/gpu/drm/msm/msm_gem_vma.c +++ b/drivers/gpu/drm/msm/msm_gem_vma.c @@ -4,9 +4,16 @@ * Author: Rob Clark <robdclark(a)gmail.com> */ +#include "drm/drm_file.h" +#include "drm/msm_drm.h" +#include "linux/file.h" +#include "linux/sync_file.h" + #include "msm_drv.h" #include "msm_gem.h" +#include "msm_gpu.h" #include "msm_mmu.h" +#include "msm_syncobj.h" #define vm_dbg(fmt, ...) pr_debug("%s:%d: "fmt"\n", __func__, __LINE__, ##__VA_ARGS__) @@ -36,6 +43,97 @@ struct msm_vm_unmap_op { uint64_t range; }; +/** + * struct msm_vma_op - A MAP or UNMAP operation + */ +struct msm_vm_op { + /** @op: The operation type */ + enum { + MSM_VM_OP_MAP = 1, + MSM_VM_OP_UNMAP, + } op; + union { + /** @map: Parameters used if op == MSM_VMA_OP_MAP */ + struct msm_vm_map_op map; + /** @unmap: Parameters used if op == MSM_VMA_OP_UNMAP */ + struct msm_vm_unmap_op unmap; + }; + /** @node: list head in msm_vm_bind_job::vm_ops */ + struct list_head node; + + /** + * @obj: backing object for pages to be mapped/unmapped + * + * Async unmap ops, in particular, must hold a reference to the + * original GEM object backing the mapping that will be unmapped. + * But the same can be required in the map path, for example if + * there is not a corresponding unmap op, such as process exit. + * + * This ensures that the pages backing the mapping are not freed + * before the mapping is torn down. + */ + struct drm_gem_object *obj; +}; + +/** + * struct msm_vm_bind_job - Tracking for a VM_BIND ioctl + * + * A table of userspace requested VM updates (MSM_VM_BIND_OP_UNMAP/MAP/MAP_NULL) + * gets applied to the vm, generating a list of VM ops (MSM_VM_OP_MAP/UNMAP) + * which are applied to the pgtables asynchronously. For example a userspace + * requested MSM_VM_BIND_OP_MAP could end up generating both an MSM_VM_OP_UNMAP + * to unmap an existing mapping, and a MSM_VM_OP_MAP to apply the new mapping. + */ +struct msm_vm_bind_job { + /** @base: base class for drm_sched jobs */ + struct drm_sched_job base; + /** @vm: The VM being operated on */ + struct drm_gpuvm *vm; + /** @fence: The fence that is signaled when job completes */ + struct dma_fence *fence; + /** @queue: The queue that the job runs on */ + struct msm_gpu_submitqueue *queue; + /** @prealloc: Tracking for pre-allocated MMU pgtable pages */ + struct msm_mmu_prealloc prealloc; + /** @vm_ops: a list of struct msm_vm_op */ + struct list_head vm_ops; + /** @bos_pinned: are the GEM objects being bound pinned? */ + bool bos_pinned; + /** @nr_ops: the number of userspace requested ops */ + unsigned int nr_ops; + /** + * @ops: the userspace requested ops + * + * The userspace requested ops are copied/parsed and validated + * before we start applying the updates to try to do as much up- + * front error checking as possible, to avoid the VM being in an + * undefined state due to partially executed VM_BIND. + * + * This table also serves to hold a reference to the backing GEM + * objects. + */ + struct msm_vm_bind_op { + uint32_t op; + uint32_t flags; + union { + struct drm_gem_object *obj; + uint32_t handle; + }; + uint64_t obj_offset; + uint64_t iova; + uint64_t range; + } ops[]; +}; + +#define job_foreach_bo(obj, _job) \ + for (unsigned i = 0; i < (_job)->nr_ops; i++) \ + if ((obj = (_job)->ops[i].obj)) + +static inline struct msm_vm_bind_job *to_msm_vm_bind_job(struct drm_sched_job *job) +{ + return container_of(job, struct msm_vm_bind_job, base); +} + static void msm_gem_vm_free(struct drm_gpuvm *gpuvm) { @@ -52,6 +150,9 @@ msm_gem_vm_free(struct drm_gpuvm *gpuvm) static void vm_unmap_op(struct msm_gem_vm *vm, const struct msm_vm_unmap_op *op) { + if (!vm->managed) + lockdep_assert_held(&vm->mmu_lock); + vm_dbg("%p: %016llx %016llx", vm, op->iova, op->iova + op->range); vm->mmu->funcs->unmap(vm->mmu, op->iova, op->range); @@ -60,6 +161,9 @@ vm_unmap_op(struct msm_gem_vm *vm, const struct msm_vm_unmap_op *op) static int vm_map_op(struct msm_gem_vm *vm, const struct msm_vm_map_op *op) { + if (!vm->managed) + lockdep_assert_held(&vm->mmu_lock); + vm_dbg("%p: %016llx %016llx", vm, op->iova, op->iova + op->range); return vm->mmu->funcs->map(vm->mmu, op->iova, op->sgt, op->offset, @@ -69,17 +173,29 @@ vm_map_op(struct msm_gem_vm *vm, const struct msm_vm_map_op *op) /* Actually unmap memory for the vma */ void msm_gem_vma_unmap(struct drm_gpuva *vma) { + struct msm_gem_vm *vm = to_msm_vm(vma->vm); struct msm_gem_vma *msm_vma = to_msm_vma(vma); /* Don't do anything if the memory isn't mapped */ if (!msm_vma->mapped) return; - vm_unmap_op(to_msm_vm(vma->vm), &(struct msm_vm_unmap_op){ + /* + * The mmu_lock is only needed when preallocation is used. But + * in that case we don't need to worry about recursion into + * shrinker + */ + if (!vm->managed) + mutex_lock(&vm->mmu_lock); + + vm_unmap_op(vm, &(struct msm_vm_unmap_op){ .iova = vma->va.addr, .range = vma->va.range, }); + if (!vm->managed) + mutex_unlock(&vm->mmu_lock); + msm_vma->mapped = false; } @@ -87,6 +203,7 @@ void msm_gem_vma_unmap(struct drm_gpuva *vma) int msm_gem_vma_map(struct drm_gpuva *vma, int prot, struct sg_table *sgt) { + struct msm_gem_vm *vm = to_msm_vm(vma->vm); struct msm_gem_vma *msm_vma = to_msm_vma(vma); int ret; @@ -98,6 +215,14 @@ msm_gem_vma_map(struct drm_gpuva *vma, int prot, struct sg_table *sgt) msm_vma->mapped = true; + /* + * The mmu_lock is only needed when preallocation is used. But + * in that case we don't need to worry about recursion into + * shrinker + */ + if (!vm->managed) + mutex_lock(&vm->mmu_lock); + /* * NOTE: iommu/io-pgtable can allocate pages, so we cannot hold * a lock across map/unmap which is also used in the job_run() @@ -107,16 +232,19 @@ msm_gem_vma_map(struct drm_gpuva *vma, int prot, struct sg_table *sgt) * Revisit this if we can come up with a scheme to pre-alloc pages * for the pgtable in map/unmap ops. */ - ret = vm_map_op(to_msm_vm(vma->vm), &(struct msm_vm_map_op){ + ret = vm_map_op(vm, &(struct msm_vm_map_op){ .iova = vma->va.addr, .range = vma->va.range, .offset = vma->gem.offset, .sgt = sgt, .prot = prot, }); - if (ret) { + + if (!vm->managed) + mutex_unlock(&vm->mmu_lock); + + if (ret) msm_vma->mapped = false; - } return ret; } @@ -131,6 +259,9 @@ void msm_gem_vma_close(struct drm_gpuva *vma) drm_gpuvm_resv_assert_held(&vm->base); + if (vma->gem.obj) + msm_gem_assert_locked(vma->gem.obj); + if (vma->va.addr && vm->managed) drm_mm_remove_node(&msm_vma->node); @@ -158,6 +289,7 @@ msm_gem_vma_new(struct drm_gpuvm *gpuvm, struct drm_gem_object *obj, if (vm->managed) { BUG_ON(offset != 0); + BUG_ON(!obj); /* NULL mappings not valid for kernel managed VM */ ret = drm_mm_insert_node_in_range(&vm->mm, &vma->node, obj->size, PAGE_SIZE, 0, range_start, range_end, 0); @@ -169,7 +301,8 @@ msm_gem_vma_new(struct drm_gpuvm *gpuvm, struct drm_gem_object *obj, range_end = range_start + obj->size; } - GEM_WARN_ON((range_end - range_start) > obj->size); + if (obj) + GEM_WARN_ON((range_end - range_start) > obj->size); drm_gpuva_init(&vma->base, range_start, range_end - range_start, obj, offset); vma->mapped = false; @@ -178,6 +311,9 @@ msm_gem_vma_new(struct drm_gpuvm *gpuvm, struct drm_gem_object *obj, if (ret) goto err_free_range; + if (!obj) + return &vma->base; + vm_bo = drm_gpuvm_bo_obtain(&vm->base, obj); if (IS_ERR(vm_bo)) { ret = PTR_ERR(vm_bo); @@ -200,11 +336,297 @@ msm_gem_vma_new(struct drm_gpuvm *gpuvm, struct drm_gem_object *obj, return ERR_PTR(ret); } +static int +msm_gem_vm_bo_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec) +{ + struct drm_gem_object *obj = vm_bo->obj; + struct drm_gpuva *vma; + int ret; + + vm_dbg("validate: %p", obj); + + msm_gem_assert_locked(obj); + + drm_gpuvm_bo_for_each_va (vma, vm_bo) { + ret = msm_gem_pin_vma_locked(obj, vma); + if (ret) + return ret; + } + + return 0; +} + +struct op_arg { + unsigned flags; + struct msm_vm_bind_job *job; +}; + +static void +vm_op_enqueue(struct op_arg *arg, struct msm_vm_op _op) +{ + struct msm_vm_op *op = kmalloc(sizeof(*op), GFP_KERNEL); + *op = _op; + list_add_tail(&op->node, &arg->job->vm_ops); + + if (op->obj) + drm_gem_object_get(op->obj); +} + +static struct drm_gpuva * +vma_from_op(struct op_arg *arg, struct drm_gpuva_op_map *op) +{ + return msm_gem_vma_new(arg->job->vm, op->gem.obj, op->gem.offset, + op->va.addr, op->va.addr + op->va.range); +} + +static int +msm_gem_vm_sm_step_map(struct drm_gpuva_op *op, void *arg) +{ + struct drm_gem_object *obj = op->map.gem.obj; + struct drm_gpuva *vma; + struct sg_table *sgt; + unsigned prot; + + vma = vma_from_op(arg, &op->map); + if (WARN_ON(IS_ERR(vma))) + return PTR_ERR(vma); + + vm_dbg("%p:%p:%p: %016llx %016llx", vma->vm, vma, vma->gem.obj, + vma->va.addr, vma->va.range); + + vma->flags = ((struct op_arg *)arg)->flags; + + if (obj) { + sgt = to_msm_bo(obj)->sgt; + prot = msm_gem_prot(obj); + } else { + sgt = NULL; + prot = IOMMU_READ | IOMMU_WRITE; + } + + vm_op_enqueue(arg, (struct msm_vm_op){ + .op = MSM_VM_OP_MAP, + .map = { + .sgt = sgt, + .iova = vma->va.addr, + .range = vma->va.range, + .offset = vma->gem.offset, + .prot = prot, + }, + .obj = vma->gem.obj, + }); + + to_msm_vma(vma)->mapped = true; + + return 0; +} + +static int +msm_gem_vm_sm_step_remap(struct drm_gpuva_op *op, void *arg) +{ + struct msm_vm_bind_job *job = ((struct op_arg *)arg)->job; + struct drm_gpuvm *vm = job->vm; + struct drm_gpuva *orig_vma = op->remap.unmap->va; + struct drm_gpuva *prev_vma = NULL, *next_vma = NULL; + struct drm_gpuvm_bo *vm_bo = orig_vma->vm_bo; + bool mapped = to_msm_vma(orig_vma)->mapped; + unsigned flags; + + vm_dbg("orig_vma: %p:%p:%p: %016llx %016llx", vm, orig_vma, + orig_vma->gem.obj, orig_vma->va.addr, orig_vma->va.range); + + if (mapped) { + uint64_t unmap_start, unmap_range; + + drm_gpuva_op_remap_to_unmap_range(&op->remap, &unmap_start, &unmap_range); + + vm_op_enqueue(arg, (struct msm_vm_op){ + .op = MSM_VM_OP_UNMAP, + .unmap = { + .iova = unmap_start, + .range = unmap_range, + }, + .obj = orig_vma->gem.obj, + }); + + /* + * Part of this GEM obj is still mapped, but we're going to kill the + * existing VMA and replace it with one or two new ones (ie. two if + * the unmapped range is in the middle of the existing (unmap) VMA). + * So just set the state to unmapped: + */ + to_msm_vma(orig_vma)->mapped = false; + } + + /* + * Hold a ref to the vm_bo between the msm_gem_vma_close() and the + * creation of the new prev/next vma's, in case the vm_bo is tracked + * in the VM's evict list: + */ + if (vm_bo) + drm_gpuvm_bo_get(vm_bo); + + /* + * The prev_vma and/or next_vma are replacing the unmapped vma, and + * therefore should preserve it's flags: + */ + flags = orig_vma->flags; + + msm_gem_vma_close(orig_vma); + + if (op->remap.prev) { + prev_vma = vma_from_op(arg, op->remap.prev); + if (WARN_ON(IS_ERR(prev_vma))) + return PTR_ERR(prev_vma); + + vm_dbg("prev_vma: %p:%p: %016llx %016llx", vm, prev_vma, prev_vma->va.addr, prev_vma->va.range); + to_msm_vma(prev_vma)->mapped = mapped; + prev_vma->flags = flags; + } + + if (op->remap.next) { + next_vma = vma_from_op(arg, op->remap.next); + if (WARN_ON(IS_ERR(next_vma))) + return PTR_ERR(next_vma); + + vm_dbg("next_vma: %p:%p: %016llx %016llx", vm, next_vma, next_vma->va.addr, next_vma->va.range); + to_msm_vma(next_vma)->mapped = mapped; + next_vma->flags = flags; + } + + if (!mapped) + drm_gpuvm_bo_evict(vm_bo, true); + + /* Drop the previous ref: */ + drm_gpuvm_bo_put(vm_bo); + + return 0; +} + +static int +msm_gem_vm_sm_step_unmap(struct drm_gpuva_op *op, void *arg) +{ + struct drm_gpuva *vma = op->unmap.va; + struct msm_gem_vma *msm_vma = to_msm_vma(vma); + + vm_dbg("%p:%p:%p: %016llx %016llx", vma->vm, vma, vma->gem.obj, + vma->va.addr, vma->va.range); + + if (!msm_vma->mapped) + goto out_close; + + vm_op_enqueue(arg, (struct msm_vm_op){ + .op = MSM_VM_OP_UNMAP, + .unmap = { + .iova = vma->va.addr, + .range = vma->va.range, + }, + .obj = vma->gem.obj, + }); + + msm_vma->mapped = false; + +out_close: + msm_gem_vma_close(vma); + + return 0; +} + static const struct drm_gpuvm_ops msm_gpuvm_ops = { .vm_free = msm_gem_vm_free, + .vm_bo_validate = msm_gem_vm_bo_validate, + .sm_step_map = msm_gem_vm_sm_step_map, + .sm_step_remap = msm_gem_vm_sm_step_remap, + .sm_step_unmap = msm_gem_vm_sm_step_unmap, }; +static struct dma_fence * +msm_vma_job_run(struct drm_sched_job *_job) +{ + struct msm_vm_bind_job *job = to_msm_vm_bind_job(_job); + struct msm_gem_vm *vm = to_msm_vm(job->vm); + struct drm_gem_object *obj; + int ret = vm->unusable ? -EINVAL : 0; + + vm_dbg(""); + + mutex_lock(&vm->mmu_lock); + vm->mmu->prealloc = &job->prealloc; + + while (!list_empty(&job->vm_ops)) { + struct msm_vm_op *op = + list_first_entry(&job->vm_ops, struct msm_vm_op, node); + + switch (op->op) { + case MSM_VM_OP_MAP: + /* + * On error, stop trying to map new things.. but we + * still want to process the unmaps (or in particular, + * the drm_gem_object_put()s) + */ + if (!ret) + ret = vm_map_op(vm, &op->map); + break; + case MSM_VM_OP_UNMAP: + vm_unmap_op(vm, &op->unmap); + break; + } + drm_gem_object_put(op->obj); + list_del(&op->node); + kfree(op); + } + + vm->mmu->prealloc = NULL; + mutex_unlock(&vm->mmu_lock); + + /* + * We failed to perform at least _some_ of the pgtable updates, so + * now the VM is in an undefined state. Game over! + */ + if (ret) + vm->unusable = true; + + job_foreach_bo (obj, job) { + msm_gem_lock(obj); + msm_gem_unpin_locked(obj); + msm_gem_unlock(obj); + } + + /* VM_BIND ops are synchronous, so no fence to wait on: */ + return NULL; +} + +static void +msm_vma_job_free(struct drm_sched_job *_job) +{ + struct msm_vm_bind_job *job = to_msm_vm_bind_job(_job); + struct msm_mmu *mmu = to_msm_vm(job->vm)->mmu; + struct drm_gem_object *obj; + + mmu->funcs->prealloc_cleanup(mmu, &job->prealloc); + + drm_sched_job_cleanup(_job); + + job_foreach_bo (obj, job) + drm_gem_object_put(obj); + + msm_submitqueue_put(job->queue); + dma_fence_put(job->fence); + + /* In error paths, we could have unexecuted ops: */ + while (!list_empty(&job->vm_ops)) { + struct msm_vm_op *op = + list_first_entry(&job->vm_ops, struct msm_vm_op, node); + list_del(&op->node); + kfree(op); + } + + kfree(job); +} + static const struct drm_sched_backend_ops msm_vm_bind_ops = { + .run_job = msm_vma_job_run, + .free_job = msm_vma_job_free }; /** @@ -254,6 +676,7 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, .ops = &msm_vm_bind_ops, .num_rqs = 1, .credit_limit = 1, + .enqueue_credit_limit = 1024, .timeout = MAX_SCHEDULE_TIMEOUT, .name = "msm-vm-bind", .dev = drm->dev, @@ -269,6 +692,7 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, drm_gem_object_put(dummy_gem); vm->mmu = mmu; + mutex_init(&vm->mmu_lock); vm->managed = managed; drm_mm_init(&vm->mm, va_start, va_size); @@ -281,7 +705,6 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, err_free_vm: kfree(vm); return ERR_PTR(ret); - } /** @@ -297,6 +720,7 @@ msm_gem_vm_close(struct drm_gpuvm *gpuvm) { struct msm_gem_vm *vm = to_msm_vm(gpuvm); struct drm_gpuva *vma, *tmp; + struct drm_exec exec; /* * For kernel managed VMs, the VMAs are torn down when the handle is @@ -313,22 +737,635 @@ msm_gem_vm_close(struct drm_gpuvm *gpuvm) drm_sched_fini(&vm->sched); /* Tear down any remaining mappings: */ - dma_resv_lock(drm_gpuvm_resv(gpuvm), NULL); - drm_gpuvm_for_each_va_safe (vma, tmp, gpuvm) { - struct drm_gem_object *obj = vma->gem.obj; + drm_exec_init(&exec, 0, 2); + drm_exec_until_all_locked (&exec) { + drm_exec_lock_obj(&exec, drm_gpuvm_resv_obj(gpuvm)); + drm_exec_retry_on_contention(&exec); + + drm_gpuvm_for_each_va_safe (vma, tmp, gpuvm) { + struct drm_gem_object *obj = vma->gem.obj; + + /* + * MSM_BO_NO_SHARE objects share the same resv as the + * VM, in which case the obj is already locked: + */ + if (obj && (obj->resv == drm_gpuvm_resv(gpuvm))) + obj = NULL; + + if (obj) { + drm_exec_lock_obj(&exec, obj); + drm_exec_retry_on_contention(&exec); + } + + msm_gem_vma_unmap(vma); + msm_gem_vma_close(vma); + + if (obj) { + drm_exec_unlock_obj(&exec, obj); + } + } + } + drm_exec_fini(&exec); +} + + +static struct msm_vm_bind_job * +vm_bind_job_create(struct drm_device *dev, struct msm_gpu *gpu, + struct msm_gpu_submitqueue *queue, uint32_t nr_ops) +{ + struct msm_vm_bind_job *job; + uint64_t sz; + int ret; + + sz = struct_size(job, ops, nr_ops); + + if (sz > SIZE_MAX) + return ERR_PTR(-ENOMEM); + + job = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN); + if (!job) + return ERR_PTR(-ENOMEM); + + ret = drm_sched_job_init(&job->base, queue->entity, 1, queue); + if (ret) { + kfree(job); + return ERR_PTR(ret); + } + + job->vm = msm_context_vm(dev, queue->ctx); + job->queue = queue; + INIT_LIST_HEAD(&job->vm_ops); + + return job; +} + +static bool invalid_alignment(uint64_t addr) +{ + /* + * Technically this is about GPU alignment, not CPU alignment. But + * I've not seen any qcom SoC where the SMMU does not support the + * CPU's smallest page size. + */ + return !PAGE_ALIGNED(addr); +} + +static int +lookup_op(struct msm_vm_bind_job *job, const struct drm_msm_vm_bind_op *op) +{ + struct drm_device *dev = job->vm->drm; + int i = job->nr_ops++; + int ret = 0; + + job->ops[i].op = op->op; + job->ops[i].handle = op->handle; + job->ops[i].obj_offset = op->obj_offset; + job->ops[i].iova = op->iova; + job->ops[i].range = op->range; + job->ops[i].flags = op->flags; + + if (op->flags & ~MSM_VM_BIND_OP_FLAGS) + ret = UERR(EINVAL, dev, "invalid flags: %x\n", op->flags); + + if (invalid_alignment(op->iova)) + ret = UERR(EINVAL, dev, "invalid address: %016llx\n", op->iova); + + if (invalid_alignment(op->obj_offset)) + ret = UERR(EINVAL, dev, "invalid bo_offset: %016llx\n", op->obj_offset); + + if (invalid_alignment(op->range)) + ret = UERR(EINVAL, dev, "invalid range: %016llx\n", op->range); + + + /* + * MAP must specify a valid handle. But the handle MBZ for + * UNMAP or MAP_NULL. + */ + if (op->op == MSM_VM_BIND_OP_MAP) { + if (!op->handle) + ret = UERR(EINVAL, dev, "invalid handle\n"); + } else if (op->handle) { + ret = UERR(EINVAL, dev, "handle must be zero\n"); + } + + switch (op->op) { + case MSM_VM_BIND_OP_MAP: + case MSM_VM_BIND_OP_MAP_NULL: + case MSM_VM_BIND_OP_UNMAP: + break; + default: + ret = UERR(EINVAL, dev, "invalid op: %u\n", op->op); + break; + } + + return ret; +} + +/* + * ioctl parsing, parameter validation, and GEM handle lookup + */ +static int +vm_bind_job_lookup_ops(struct msm_vm_bind_job *job, struct drm_msm_vm_bind *args, + struct drm_file *file, int *nr_bos) +{ + struct drm_device *dev = job->vm->drm; + int ret = 0; + int cnt = 0; + + if (args->nr_ops == 1) { + /* Single op case, the op is inlined: */ + ret = lookup_op(job, &args->op); + } else { + for (unsigned i = 0; i < args->nr_ops; i++) { + struct drm_msm_vm_bind_op op; + void __user *userptr = + u64_to_user_ptr(args->ops + (i * sizeof(op))); + + /* make sure we don't have garbage flags, in case we hit + * error path before flags is initialized: + */ + job->ops[i].flags = 0; + + if (copy_from_user(&op, userptr, sizeof(op))) { + ret = -EFAULT; + break; + } + + ret = lookup_op(job, &op); + if (ret) + break; + } + } + + if (ret) { + job->nr_ops = 0; + goto out; + } + + spin_lock(&file->table_lock); + + for (unsigned i = 0; i < args->nr_ops; i++) { + struct drm_gem_object *obj; + + if (!job->ops[i].handle) { + job->ops[i].obj = NULL; + continue; + } + + /* + * normally use drm_gem_object_lookup(), but for bulk lookup + * all under single table_lock just hit object_idr directly: + */ + obj = idr_find(&file->object_idr, job->ops[i].handle); + if (!obj) { + ret = UERR(EINVAL, dev, "invalid handle %u at index %u\n", job->ops[i].handle, i); + goto out_unlock; + } + + drm_gem_object_get(obj); + + job->ops[i].obj = obj; + cnt++; + } + + *nr_bos = cnt; + +out_unlock: + spin_unlock(&file->table_lock); + +out: + return ret; +} + +static void +prealloc_count(struct msm_vm_bind_job *job, + struct msm_vm_bind_op *first, + struct msm_vm_bind_op *last) +{ + struct msm_mmu *mmu = to_msm_vm(job->vm)->mmu; + + if (!first) + return; + + uint64_t start_iova = first->iova; + uint64_t end_iova = last->iova + last->range; + + mmu->funcs->prealloc_count(mmu, &job->prealloc, start_iova, end_iova - start_iova); +} + +static bool +ops_are_same_pte(struct msm_vm_bind_op *first, struct msm_vm_bind_op *next) +{ + /* + * Last level pte covers 2MB.. so we should merge two ops, from + * the PoV of figuring out how much pgtable pages to pre-allocate + * if they land in the same 2MB range: + */ + uint64_t pte_mask = ~(SZ_2M - 1); + return ((first->iova + first->range) & pte_mask) == (next->iova & pte_mask); +} + +/* + * Determine the amount of memory to prealloc for pgtables. For sparse images, + * in particular, userspace plays some tricks with the order of page mappings + * to get the desired swizzle pattern, resulting in a large # of tiny MAP ops. + * So detect when multiple MAP operations are physically contiguous, and count + * them as a single mapping. Otherwise the prealloc_count() will not realize + * they can share pagetable pages and vastly overcount. + */ +static void +vm_bind_prealloc_count(struct msm_vm_bind_job *job) +{ + struct msm_vm_bind_op *first = NULL, *last = NULL; + + for (int i = 0; i < job->nr_ops; i++) { + struct msm_vm_bind_op *op = &job->ops[i]; + + /* We only care about MAP/MAP_NULL: */ + if (op->op == MSM_VM_BIND_OP_UNMAP) + continue; + + /* + * If op is contiguous with last in the current range, then + * it becomes the new last in the range and we continue + * looping: + */ + if (last && ops_are_same_pte(last, op)) { + last = op; + continue; + } + + /* + * If op is not contiguous with the current range, flush + * the current range and start anew: + */ + prealloc_count(job, first, last); + first = last = op; + } + + /* Flush the remaining range: */ + prealloc_count(job, first, last); + + job->base.enqueue_credits = job->prealloc.count; +} + +/* + * Lock VM and GEM objects + */ +static int +vm_bind_job_lock_objects(struct msm_vm_bind_job *job, struct drm_exec *exec) +{ + struct drm_gem_object *obj; + int ret; + + /* Lock VM and objects: */ + drm_exec_until_all_locked(exec) { + ret = drm_exec_lock_obj(exec, drm_gpuvm_resv_obj(job->vm)); + drm_exec_retry_on_contention(exec); + if (ret) + return ret; + + job_foreach_bo (obj, job) { + ret = drm_exec_prepare_obj(exec, obj, 1); + drm_exec_retry_on_contention(exec); + if (ret) + return ret; + } + } + + return 0; +} + +/* + * Pin GEM objects, ensuring that we have backing pages. Pinning will move + * the object to the pinned LRU so that the shrinker knows to first consider + * other objects for evicting. + */ +static int +vm_bind_job_pin_objects(struct msm_vm_bind_job *job) +{ + struct drm_gem_object *obj; + + /* + * First loop, before holding the LRU lock, avoids holding the + * LRU lock while calling msm_gem_pin_vma_locked (which could + * trigger get_pages()) + */ + job_foreach_bo (obj, job) { + struct page **pages; + + pages = msm_gem_get_pages_locked(obj, MSM_MADV_WILLNEED); + if (IS_ERR(pages)) + return PTR_ERR(pages); + } + + struct msm_drm_private *priv = job->vm->drm->dev_private; + + /* + * A second loop while holding the LRU lock (a) avoids acquiring/dropping + * the LRU lock for each individual bo, while (b) avoiding holding the + * LRU lock while calling msm_gem_pin_vma_locked() (which could trigger + * get_pages() which could trigger reclaim.. and if we held the LRU lock + * could trigger deadlock with the shrinker). + */ + mutex_lock(&priv->lru.lock); + job_foreach_bo (obj, job) + msm_gem_pin_obj_locked(obj); + mutex_unlock(&priv->lru.lock); + + job->bos_pinned = true; + + return 0; +} + +/* + * Unpin GEM objects. Normally this is done after the bind job is run. + */ +static void +vm_bind_job_unpin_objects(struct msm_vm_bind_job *job) +{ + struct drm_gem_object *obj; + + if (!job->bos_pinned) + return; + + job_foreach_bo (obj, job) + msm_gem_unpin_locked(obj); - if (obj && obj->resv != drm_gpuvm_resv(gpuvm)) { - drm_gem_object_get(obj); - msm_gem_lock(obj); + job->bos_pinned = false; +} + +/* + * Pre-allocate pgtable memory, and translate the VM bind requests into a + * sequence of pgtable updates to be applied asynchronously. + */ +static int +vm_bind_job_prepare(struct msm_vm_bind_job *job) +{ + struct msm_gem_vm *vm = to_msm_vm(job->vm); + struct msm_mmu *mmu = vm->mmu; + int ret; + + ret = mmu->funcs->prealloc_allocate(mmu, &job->prealloc); + if (ret) + return ret; + + for (unsigned i = 0; i < job->nr_ops; i++) { + const struct msm_vm_bind_op *op = &job->ops[i]; + struct op_arg arg = { + .job = job, + }; + + switch (op->op) { + case MSM_VM_BIND_OP_UNMAP: + ret = drm_gpuvm_sm_unmap(job->vm, &arg, op->iova, + op->obj_offset); + break; + case MSM_VM_BIND_OP_MAP: + if (op->flags & MSM_VM_BIND_OP_DUMP) + arg.flags |= MSM_VMA_DUMP; + fallthrough; + case MSM_VM_BIND_OP_MAP_NULL: + ret = drm_gpuvm_sm_map(job->vm, &arg, op->iova, + op->range, op->obj, op->obj_offset); + break; + default: + /* + * lookup_op() should have already thrown an error for + * invalid ops + */ + BUG_ON("unreachable"); } - msm_gem_vma_unmap(vma); - msm_gem_vma_close(vma); + if (ret) { + /* + * If we've already started modifying the vm, we can't + * adequetly describe to userspace the intermediate + * state the vm is in. So throw up our hands! + */ + if (i > 0) + vm->unusable = true; + return ret; + } + } + + return 0; +} + +/* + * Attach fences to the GEM objects being bound. This will signify to + * the shrinker that they are busy even after dropping the locks (ie. + * drm_exec_fini()) + */ +static void +vm_bind_job_attach_fences(struct msm_vm_bind_job *job) +{ + for (unsigned i = 0; i < job->nr_ops; i++) { + struct drm_gem_object *obj = job->ops[i].obj; + + if (!obj) + continue; + + dma_resv_add_fence(obj->resv, job->fence, + DMA_RESV_USAGE_KERNEL); + } +} + +int +msm_ioctl_vm_bind(struct drm_device *dev, void *data, struct drm_file *file) +{ + struct msm_drm_private *priv = dev->dev_private; + struct drm_msm_vm_bind *args = data; + struct msm_context *ctx = file->driver_priv; + struct msm_vm_bind_job *job = NULL; + struct msm_gpu *gpu = priv->gpu; + struct msm_gpu_submitqueue *queue; + struct msm_syncobj_post_dep *post_deps = NULL; + struct drm_syncobj **syncobjs_to_reset = NULL; + struct sync_file *sync_file = NULL; + struct dma_fence *fence; + int out_fence_fd = -1; + int ret, nr_bos = 0; + unsigned i; + + if (!gpu) + return -ENXIO; + + /* + * Maybe we could allow just UNMAP ops? OTOH userspace should just + * immediately close the device file and all will be torn down. + */ + if (to_msm_vm(ctx->vm)->unusable) + return UERR(EPIPE, dev, "context is unusable"); + + /* + * Technically, you cannot create a VM_BIND submitqueue in the first + * place, if you haven't opted in to VM_BIND context. But it is + * cleaner / less confusing, to check this case directly. + */ + if (!msm_context_is_vmbind(ctx)) + return UERR(EINVAL, dev, "context does not support vmbind"); + + if (args->flags & ~MSM_VM_BIND_FLAGS) + return UERR(EINVAL, dev, "invalid flags"); - if (obj && obj->resv != drm_gpuvm_resv(gpuvm)) { - msm_gem_unlock(obj); - drm_gem_object_put(obj); + queue = msm_submitqueue_get(ctx, args->queue_id); + if (!queue) + return -ENOENT; + + if (!(queue->flags & MSM_SUBMITQUEUE_VM_BIND)) { + ret = UERR(EINVAL, dev, "Invalid queue type"); + goto out_post_unlock; + } + + if (args->flags & MSM_VM_BIND_FENCE_FD_OUT) { + out_fence_fd = get_unused_fd_flags(O_CLOEXEC); + if (out_fence_fd < 0) { + ret = out_fence_fd; + goto out_post_unlock; } } - dma_resv_unlock(drm_gpuvm_resv(gpuvm)); + + job = vm_bind_job_create(dev, gpu, queue, args->nr_ops); + if (IS_ERR(job)) { + ret = PTR_ERR(job); + goto out_post_unlock; + } + + ret = mutex_lock_interruptible(&queue->lock); + if (ret) + goto out_post_unlock; + + if (args->flags & MSM_VM_BIND_FENCE_FD_IN) { + struct dma_fence *in_fence; + + in_fence = sync_file_get_fence(args->fence_fd); + + if (!in_fence) { + ret = UERR(EINVAL, dev, "invalid in-fence"); + goto out_unlock; + } + + ret = drm_sched_job_add_dependency(&job->base, in_fence); + if (ret) + goto out_unlock; + } + + if (args->in_syncobjs > 0) { + syncobjs_to_reset = msm_syncobj_parse_deps(dev, &job->base, + file, args->in_syncobjs, + args->nr_in_syncobjs, + args->syncobj_stride); + if (IS_ERR(syncobjs_to_reset)) { + ret = PTR_ERR(syncobjs_to_reset); + goto out_unlock; + } + } + + if (args->out_syncobjs > 0) { + post_deps = msm_syncobj_parse_post_deps(dev, file, + args->out_syncobjs, + args->nr_out_syncobjs, + args->syncobj_stride); + if (IS_ERR(post_deps)) { + ret = PTR_ERR(post_deps); + goto out_unlock; + } + } + + ret = vm_bind_job_lookup_ops(job, args, file, &nr_bos); + if (ret) + goto out_unlock; + + vm_bind_prealloc_count(job); + + struct drm_exec exec; + unsigned flags = DRM_EXEC_IGNORE_DUPLICATES | DRM_EXEC_INTERRUPTIBLE_WAIT; + drm_exec_init(&exec, flags, nr_bos + 1); + + ret = vm_bind_job_lock_objects(job, &exec); + if (ret) + goto out; + + ret = vm_bind_job_pin_objects(job); + if (ret) + goto out; + + ret = vm_bind_job_prepare(job); + if (ret) + goto out; + + drm_sched_job_arm(&job->base); + + job->fence = dma_fence_get(&job->base.s_fence->finished); + + if (args->flags & MSM_VM_BIND_FENCE_FD_OUT) { + sync_file = sync_file_create(job->fence); + if (!sync_file) { + ret = -ENOMEM; + } else { + fd_install(out_fence_fd, sync_file->file); + args->fence_fd = out_fence_fd; + } + } + + if (ret) + goto out; + + vm_bind_job_attach_fences(job); + + /* + * The job can be free'd (and fence unref'd) at any point after + * drm_sched_entity_push_job(), so we need to hold our own ref + */ + fence = dma_fence_get(job->fence); + + ret = drm_sched_entity_push_job(&job->base); + + msm_syncobj_reset(syncobjs_to_reset, args->nr_in_syncobjs); + msm_syncobj_process_post_deps(post_deps, args->nr_out_syncobjs, fence); + + dma_fence_put(fence); + +out: + if (ret) + vm_bind_job_unpin_objects(job); + + drm_exec_fini(&exec); +out_unlock: + mutex_unlock(&queue->lock); +out_post_unlock: + if (ret && (out_fence_fd >= 0)) { + put_unused_fd(out_fence_fd); + if (sync_file) + fput(sync_file->file); + } + + if (!IS_ERR_OR_NULL(job)) { + if (ret) + msm_vma_job_free(&job->base); + } else { + /* + * If the submit hasn't yet taken ownership of the queue + * then we need to drop the reference ourself: + */ + msm_submitqueue_put(queue); + } + + if (!IS_ERR_OR_NULL(post_deps)) { + for (i = 0; i < args->nr_out_syncobjs; ++i) { + kfree(post_deps[i].chain); + drm_syncobj_put(post_deps[i].syncobj); + } + kfree(post_deps); + } + + if (!IS_ERR_OR_NULL(syncobjs_to_reset)) { + for (i = 0; i < args->nr_in_syncobjs; ++i) { + if (syncobjs_to_reset[i]) + drm_syncobj_put(syncobjs_to_reset[i]); + } + kfree(syncobjs_to_reset); + } + + return ret; } diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 6d6cd1219926..5c67294edc95 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -272,7 +272,10 @@ struct drm_msm_gem_submit_cmd { __u32 size; /* in, cmdstream size */ __u32 pad; __u32 nr_relocs; /* in, number of submit_reloc's */ - __u64 relocs; /* in, ptr to array of submit_reloc's */ + union { + __u64 relocs; /* in, ptr to array of submit_reloc's */ + __u64 iova; /* cmdstream address (for VM_BIND contexts) */ + }; }; /* Each buffer referenced elsewhere in the cmdstream submit (ie. the @@ -339,7 +342,74 @@ struct drm_msm_gem_submit { __u32 nr_out_syncobjs; /* in, number of entries in out_syncobj. */ __u32 syncobj_stride; /* in, stride of syncobj arrays. */ __u32 pad; /*in, reserved for future use, always 0. */ +}; + +#define MSM_VM_BIND_OP_UNMAP 0 +#define MSM_VM_BIND_OP_MAP 1 +#define MSM_VM_BIND_OP_MAP_NULL 2 + +#define MSM_VM_BIND_OP_DUMP 1 +#define MSM_VM_BIND_OP_FLAGS ( \ + MSM_VM_BIND_OP_DUMP | \ + 0) +/** + * struct drm_msm_vm_bind_op - bind/unbind op to run + */ +struct drm_msm_vm_bind_op { + /** @op: one of MSM_VM_BIND_OP_x */ + __u32 op; + /** @handle: GEM object handle, MBZ for UNMAP or MAP_NULL */ + __u32 handle; + /** @obj_offset: Offset into GEM object, MBZ for UNMAP or MAP_NULL */ + __u64 obj_offset; + /** @iova: Address to operate on */ + __u64 iova; + /** @range: Number of bites to to map/unmap */ + __u64 range; + /** @flags: Bitmask of MSM_VM_BIND_OP_FLAG_x */ + __u32 flags; + /** @pad: MBZ */ + __u32 pad; +}; + +#define MSM_VM_BIND_FENCE_FD_IN 0x00000001 +#define MSM_VM_BIND_FENCE_FD_OUT 0x00000002 +#define MSM_VM_BIND_FLAGS ( \ + MSM_VM_BIND_FENCE_FD_IN | \ + MSM_VM_BIND_FENCE_FD_OUT | \ + 0) + +/** + * struct drm_msm_vm_bind - Input of &DRM_IOCTL_MSM_VM_BIND + */ +struct drm_msm_vm_bind { + /** @flags: in, bitmask of MSM_VM_BIND_x */ + __u32 flags; + /** @nr_ops: the number of bind ops in this ioctl */ + __u32 nr_ops; + /** @fence_fd: in/out fence fd (see MSM_VM_BIND_FENCE_FD_IN/OUT) */ + __s32 fence_fd; + /** @queue_id: in, submitqueue id */ + __u32 queue_id; + /** @in_syncobjs: in, ptr to array of drm_msm_gem_syncobj */ + __u64 in_syncobjs; + /** @out_syncobjs: in, ptr to array of drm_msm_gem_syncobj */ + __u64 out_syncobjs; + /** @nr_in_syncobjs: in, number of entries in in_syncobj */ + __u32 nr_in_syncobjs; + /** @nr_out_syncobjs: in, number of entries in out_syncobj */ + __u32 nr_out_syncobjs; + /** @syncobj_stride: in, stride of syncobj arrays */ + __u32 syncobj_stride; + /** @op_stride: sizeof each struct drm_msm_vm_bind_op in @ops */ + __u32 op_stride; + union { + /** @op: used if num_ops == 1 */ + struct drm_msm_vm_bind_op op; + /** @ops: userptr to array of drm_msm_vm_bind_op if num_ops > 1 */ + __u64 ops; + }; }; #define MSM_WAIT_FENCE_BOOST 0x00000001 @@ -435,6 +505,7 @@ struct drm_msm_submitqueue_query { #define DRM_MSM_SUBMITQUEUE_NEW 0x0A #define DRM_MSM_SUBMITQUEUE_CLOSE 0x0B #define DRM_MSM_SUBMITQUEUE_QUERY 0x0C +#define DRM_MSM_VM_BIND 0x0D #define DRM_IOCTL_MSM_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_GET_PARAM, struct drm_msm_param) #define DRM_IOCTL_MSM_SET_PARAM DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SET_PARAM, struct drm_msm_param) @@ -448,6 +519,7 @@ struct drm_msm_submitqueue_query { #define DRM_IOCTL_MSM_SUBMITQUEUE_NEW DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_NEW, struct drm_msm_submitqueue) #define DRM_IOCTL_MSM_SUBMITQUEUE_CLOSE DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_CLOSE, __u32) #define DRM_IOCTL_MSM_SUBMITQUEUE_QUERY DRM_IOW (DRM_COMMAND_BASE + DRM_MSM_SUBMITQUEUE_QUERY, struct drm_msm_submitqueue_query) +#define DRM_IOCTL_MSM_VM_BIND DRM_IOWR(DRM_COMMAND_BASE + DRM_MSM_VM_BIND, struct drm_msm_vm_bind) #if defined(__cplusplus) } -- 2.49.0

1 month, 1 week

[PATCH v5 31/40] drm/msm: Add VM_BIND submitqueue

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> This submitqueue type isn't tied to a hw ringbuffer, but instead executes on the CPU for performing async VM_BIND ops. Signed-off-by: Rob Clark <robdclark(a)chromium.org> --- drivers/gpu/drm/msm/msm_gem.h | 12 +++++ drivers/gpu/drm/msm/msm_gem_submit.c | 60 +++++++++++++++++++--- drivers/gpu/drm/msm/msm_gem_vma.c | 71 +++++++++++++++++++++++++++ drivers/gpu/drm/msm/msm_gpu.h | 3 ++ drivers/gpu/drm/msm/msm_submitqueue.c | 67 +++++++++++++++++++------ include/uapi/drm/msm_drm.h | 9 +++- 6 files changed, 197 insertions(+), 25 deletions(-) diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index f7b85084e228..c1581bd4b5fd 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -53,6 +53,13 @@ struct msm_gem_vm { /** @base: Inherit from drm_gpuvm. */ struct drm_gpuvm base; + /** + * @sched: Scheduler used for asynchronous VM_BIND request. + * + * Unused for kernel managed VMs (where all operations are synchronous). + */ + struct drm_gpu_scheduler sched; + /** * @mm: Memory management for kernel managed VA allocations * @@ -71,6 +78,9 @@ struct msm_gem_vm { */ struct pid *pid; + /** @last_fence: Fence for last pending work scheduled on the VM */ + struct dma_fence *last_fence; + /** @faults: the number of GPU hangs associated with this address space */ int faults; @@ -100,6 +110,8 @@ struct drm_gpuvm * msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, u64 va_start, u64 va_size, bool managed); +void msm_gem_vm_close(struct drm_gpuvm *gpuvm); + struct msm_fence_context; #define MSM_VMA_DUMP (DRM_GPUVA_USERBITS << 0) diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index bfb8c5ac1f1e..053e6c65780f 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -4,6 +4,7 @@ * Author: Rob Clark <robdclark(a)gmail.com> */ +#include <linux/dma-fence-unwrap.h> #include <linux/file.h> #include <linux/sync_file.h> #include <linux/uaccess.h> @@ -258,30 +259,43 @@ static int submit_lookup_cmds(struct msm_gem_submit *submit, static int submit_lock_objects(struct msm_gem_submit *submit) { unsigned flags = DRM_EXEC_IGNORE_DUPLICATES | DRM_EXEC_INTERRUPTIBLE_WAIT; + struct drm_exec *exec = &submit->exec; int ret; -// TODO need to add vm_bind path which locks vm resv + external objs drm_exec_init(&submit->exec, flags, submit->nr_bos); + if (msm_context_is_vmbind(submit->queue->ctx)) { + drm_exec_until_all_locked (&submit->exec) { + ret = drm_gpuvm_prepare_vm(submit->vm, exec, 1); + drm_exec_retry_on_contention(exec); + if (ret) + return ret; + + ret = drm_gpuvm_prepare_objects(submit->vm, exec, 1); + drm_exec_retry_on_contention(exec); + if (ret) + return ret; + } + + return 0; + } + drm_exec_until_all_locked (&submit->exec) { ret = drm_exec_lock_obj(&submit->exec, drm_gpuvm_resv_obj(submit->vm)); drm_exec_retry_on_contention(&submit->exec); if (ret) - goto error; + return ret; for (unsigned i = 0; i < submit->nr_bos; i++) { struct drm_gem_object *obj = submit->bos[i].obj; ret = drm_exec_prepare_obj(&submit->exec, obj, 1); drm_exec_retry_on_contention(&submit->exec); if (ret) - goto error; + return ret; } } return 0; - -error: - return ret; } static int submit_fence_sync(struct msm_gem_submit *submit) @@ -366,9 +380,18 @@ static void submit_unpin_objects(struct msm_gem_submit *submit) static void submit_attach_object_fences(struct msm_gem_submit *submit) { - int i; + struct msm_gem_vm *vm = to_msm_vm(submit->vm); + struct dma_fence *last_fence; + + if (msm_context_is_vmbind(submit->queue->ctx)) { + drm_gpuvm_resv_add_fence(submit->vm, &submit->exec, + submit->user_fence, + DMA_RESV_USAGE_BOOKKEEP, + DMA_RESV_USAGE_BOOKKEEP); + return; + } - for (i = 0; i < submit->nr_bos; i++) { + for (unsigned i = 0; i < submit->nr_bos; i++) { struct drm_gem_object *obj = submit->bos[i].obj; if (submit->bos[i].flags & MSM_SUBMIT_BO_WRITE) @@ -378,6 +401,10 @@ static void submit_attach_object_fences(struct msm_gem_submit *submit) dma_resv_add_fence(obj->resv, submit->user_fence, DMA_RESV_USAGE_READ); } + + last_fence = vm->last_fence; + vm->last_fence = dma_fence_unwrap_merge(submit->user_fence, last_fence); + dma_fence_put(last_fence); } static int submit_bo(struct msm_gem_submit *submit, uint32_t idx, @@ -532,6 +559,11 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, if (!queue) return -ENOENT; + if (queue->flags & MSM_SUBMITQUEUE_VM_BIND) { + ret = UERR(EINVAL, dev, "Invalid queue type"); + goto out_post_unlock; + } + ring = gpu->rb[queue->ring_nr]; if (args->flags & MSM_SUBMIT_FENCE_FD_OUT) { @@ -721,6 +753,18 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, submit_attach_object_fences(submit); + if (msm_context_is_vmbind(ctx)) { + /* + * If we are not using VM_BIND, submit_pin_vmas() will validate + * just the BOs attached to the submit. In that case we don't + * need to validate the _entire_ vm, because userspace tracked + * what BOs are associated with the submit. + */ + ret = drm_gpuvm_validate(submit->vm, &submit->exec); + if (ret) + goto out; + } + /* The scheduler owns a ref now: */ msm_gem_submit_get(submit); diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c index 72667316df51..73baa9451ada 100644 --- a/drivers/gpu/drm/msm/msm_gem_vma.c +++ b/drivers/gpu/drm/msm/msm_gem_vma.c @@ -16,6 +16,7 @@ msm_gem_vm_free(struct drm_gpuvm *gpuvm) drm_mm_takedown(&vm->mm); if (vm->mmu) vm->mmu->funcs->destroy(vm->mmu); + dma_fence_put(vm->last_fence); put_pid(vm->pid); kfree(vm); } @@ -154,6 +155,9 @@ static const struct drm_gpuvm_ops msm_gpuvm_ops = { .vm_free = msm_gem_vm_free, }; +static const struct drm_sched_backend_ops msm_vm_bind_ops = { +}; + /** * msm_gem_vm_create() - Create and initialize a &msm_gem_vm * @drm: the drm device @@ -196,6 +200,21 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, goto err_free_vm; } + if (!managed) { + struct drm_sched_init_args args = { + .ops = &msm_vm_bind_ops, + .num_rqs = 1, + .credit_limit = 1, + .timeout = MAX_SCHEDULE_TIMEOUT, + .name = "msm-vm-bind", + .dev = drm->dev, + }; + + ret = drm_sched_init(&vm->sched, &args); + if (ret) + goto err_free_dummy; + } + drm_gpuvm_init(&vm->base, name, flags, drm, dummy_gem, va_start, va_size, 0, 0, &msm_gpuvm_ops); drm_gem_object_put(dummy_gem); @@ -207,8 +226,60 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, return &vm->base; +err_free_dummy: + drm_gem_object_put(dummy_gem); + err_free_vm: kfree(vm); return ERR_PTR(ret); } + +/** + * msm_gem_vm_close() - Close a VM + * @gpuvm: The VM to close + * + * Called when the drm device file is closed, to tear down VM related resources + * (which will drop refcounts to GEM objects that were still mapped into the + * VM at the time). + */ +void +msm_gem_vm_close(struct drm_gpuvm *gpuvm) +{ + struct msm_gem_vm *vm = to_msm_vm(gpuvm); + struct drm_gpuva *vma, *tmp; + + /* + * For kernel managed VMs, the VMAs are torn down when the handle is + * closed, so nothing more to do. + */ + if (vm->managed) + return; + + if (vm->last_fence) + dma_fence_wait(vm->last_fence, false); + + /* Kill the scheduler now, so we aren't racing with it for cleanup: */ + drm_sched_stop(&vm->sched, NULL); + drm_sched_fini(&vm->sched); + + /* Tear down any remaining mappings: */ + dma_resv_lock(drm_gpuvm_resv(gpuvm), NULL); + drm_gpuvm_for_each_va_safe (vma, tmp, gpuvm) { + struct drm_gem_object *obj = vma->gem.obj; + + if (obj && obj->resv != drm_gpuvm_resv(gpuvm)) { + drm_gem_object_get(obj); + msm_gem_lock(obj); + } + + msm_gem_vma_unmap(vma); + msm_gem_vma_close(vma); + + if (obj && obj->resv != drm_gpuvm_resv(gpuvm)) { + msm_gem_unlock(obj); + drm_gem_object_put(obj); + } + } + dma_resv_unlock(drm_gpuvm_resv(gpuvm)); +} diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h index 448ebf721bd8..9cbf155ff222 100644 --- a/drivers/gpu/drm/msm/msm_gpu.h +++ b/drivers/gpu/drm/msm/msm_gpu.h @@ -570,6 +570,9 @@ struct msm_gpu_submitqueue { struct mutex lock; struct kref ref; struct drm_sched_entity *entity; + + /** @_vm_bind_entity: used for @entity pointer for VM_BIND queues */ + struct drm_sched_entity _vm_bind_entity[0]; }; struct msm_gpu_state_bo { diff --git a/drivers/gpu/drm/msm/msm_submitqueue.c b/drivers/gpu/drm/msm/msm_submitqueue.c index 8ced49c7557b..8617a82cd6b3 100644 --- a/drivers/gpu/drm/msm/msm_submitqueue.c +++ b/drivers/gpu/drm/msm/msm_submitqueue.c @@ -72,6 +72,9 @@ void msm_submitqueue_destroy(struct kref *kref) idr_destroy(&queue->fence_idr); + if (queue->entity == &queue->_vm_bind_entity[0]) + drm_sched_entity_destroy(queue->entity); + msm_context_put(queue->ctx); kfree(queue); @@ -102,7 +105,7 @@ struct msm_gpu_submitqueue *msm_submitqueue_get(struct msm_context *ctx, void msm_submitqueue_close(struct msm_context *ctx) { - struct msm_gpu_submitqueue *entry, *tmp; + struct msm_gpu_submitqueue *queue, *tmp; if (!ctx) return; @@ -111,10 +114,17 @@ void msm_submitqueue_close(struct msm_context *ctx) * No lock needed in close and there won't * be any more user ioctls coming our way */ - list_for_each_entry_safe(entry, tmp, &ctx->submitqueues, node) { - list_del(&entry->node); - msm_submitqueue_put(entry); + list_for_each_entry_safe(queue, tmp, &ctx->submitqueues, node) { + if (queue->entity == &queue->_vm_bind_entity[0]) + drm_sched_entity_flush(queue->entity, MAX_WAIT_SCHED_ENTITY_Q_EMPTY); + list_del(&queue->node); + msm_submitqueue_put(queue); } + + if (!ctx->vm) + return; + + msm_gem_vm_close(ctx->vm); } static struct drm_sched_entity * @@ -160,8 +170,6 @@ int msm_submitqueue_create(struct drm_device *drm, struct msm_context *ctx, struct msm_drm_private *priv = drm->dev_private; struct msm_gpu_submitqueue *queue; enum drm_sched_priority sched_prio; - extern int enable_preemption; - bool preemption_supported; unsigned ring_nr; int ret; @@ -171,26 +179,53 @@ int msm_submitqueue_create(struct drm_device *drm, struct msm_context *ctx, if (!priv->gpu) return -ENODEV; - preemption_supported = priv->gpu->nr_rings == 1 && enable_preemption != 0; + if (flags & MSM_SUBMITQUEUE_VM_BIND) { + unsigned sz; - if (flags & MSM_SUBMITQUEUE_ALLOW_PREEMPT && preemption_supported) - return -EINVAL; + /* Not allowed for kernel managed VMs (ie. kernel allocs VA) */ + if (!msm_context_is_vmbind(ctx)) + return -EINVAL; - ret = msm_gpu_convert_priority(priv->gpu, prio, &ring_nr, &sched_prio); - if (ret) - return ret; + if (prio) + return -EINVAL; + + sz = struct_size(queue, _vm_bind_entity, 1); + queue = kzalloc(sz, GFP_KERNEL); + } else { + extern int enable_preemption; + bool preemption_supported = + priv->gpu->nr_rings == 1 && enable_preemption != 0; + + if (flags & MSM_SUBMITQUEUE_ALLOW_PREEMPT && preemption_supported) + return -EINVAL; - queue = kzalloc(sizeof(*queue), GFP_KERNEL); + ret = msm_gpu_convert_priority(priv->gpu, prio, &ring_nr, &sched_prio); + if (ret) + return ret; + + queue = kzalloc(sizeof(*queue), GFP_KERNEL); + } if (!queue) return -ENOMEM; kref_init(&queue->ref); queue->flags = flags; - queue->ring_nr = ring_nr; - queue->entity = get_sched_entity(ctx, priv->gpu->rb[ring_nr], - ring_nr, sched_prio); + if (flags & MSM_SUBMITQUEUE_VM_BIND) { + struct drm_gpu_scheduler *sched = &to_msm_vm(msm_context_vm(drm, ctx))->sched; + + queue->entity = &queue->_vm_bind_entity[0]; + + drm_sched_entity_init(queue->entity, DRM_SCHED_PRIORITY_KERNEL, + &sched, 1, NULL); + } else { + queue->ring_nr = ring_nr; + + queue->entity = get_sched_entity(ctx, priv->gpu->rb[ring_nr], + ring_nr, sched_prio); + } + if (IS_ERR(queue->entity)) { ret = PTR_ERR(queue->entity); kfree(queue); diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 2c2fc4b284d0..6d6cd1219926 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -385,12 +385,19 @@ struct drm_msm_gem_madvise { /* * Draw queues allow the user to set specific submission parameter. Command * submissions specify a specific submitqueue to use. ID 0 is reserved for - * backwards compatibility as a "default" submitqueue + * backwards compatibility as a "default" submitqueue. + * + * Because VM_BIND async updates happen on the CPU, they must run on a + * virtual queue created with the flag MSM_SUBMITQUEUE_VM_BIND. If we had + * a way to do pgtable updates on the GPU, we could drop this restriction. */ #define MSM_SUBMITQUEUE_ALLOW_PREEMPT 0x00000001 +#define MSM_SUBMITQUEUE_VM_BIND 0x00000002 /* virtual queue for VM_BIND ops */ + #define MSM_SUBMITQUEUE_FLAGS ( \ MSM_SUBMITQUEUE_ALLOW_PREEMPT | \ + MSM_SUBMITQUEUE_VM_BIND | \ 0) /* -- 2.49.0

1 month, 1 week

[PATCH v5 29/40] drm/msm: Extract out syncobj helpers

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> We'll be re-using these for the VM_BIND ioctl. Also, rename a few things in the uapi header to reflect that syncobj use is not specific to the submit ioctl. Signed-off-by: Rob Clark <robdclark(a)chromium.org> --- drivers/gpu/drm/msm/Makefile | 1 + drivers/gpu/drm/msm/msm_gem_submit.c | 192 ++------------------------- drivers/gpu/drm/msm/msm_syncobj.c | 172 ++++++++++++++++++++++++ drivers/gpu/drm/msm/msm_syncobj.h | 37 ++++++ include/uapi/drm/msm_drm.h | 26 ++-- 5 files changed, 235 insertions(+), 193 deletions(-) create mode 100644 drivers/gpu/drm/msm/msm_syncobj.c create mode 100644 drivers/gpu/drm/msm/msm_syncobj.h diff --git a/drivers/gpu/drm/msm/Makefile b/drivers/gpu/drm/msm/Makefile index 5df20cbeafb8..8af34f87e0c8 100644 --- a/drivers/gpu/drm/msm/Makefile +++ b/drivers/gpu/drm/msm/Makefile @@ -128,6 +128,7 @@ msm-y += \ msm_rd.o \ msm_ringbuffer.o \ msm_submitqueue.o \ + msm_syncobj.o \ msm_gpu_tracepoints.o \ msm-$(CONFIG_DRM_FBDEV_EMULATION) += msm_fbdev.o diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index f282d691087f..bfb8c5ac1f1e 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -16,6 +16,7 @@ #include "msm_gpu.h" #include "msm_gem.h" #include "msm_gpu_trace.h" +#include "msm_syncobj.h" /* For userspace errors, use DRM_UT_DRIVER.. so that userspace can enable * error msgs for debugging, but we don't spam dmesg by default @@ -486,173 +487,6 @@ void msm_submit_retire(struct msm_gem_submit *submit) } } -struct msm_submit_post_dep { - struct drm_syncobj *syncobj; - uint64_t point; - struct dma_fence_chain *chain; -}; - -static struct drm_syncobj **msm_parse_deps(struct msm_gem_submit *submit, - struct drm_file *file, - uint64_t in_syncobjs_addr, - uint32_t nr_in_syncobjs, - size_t syncobj_stride) -{ - struct drm_syncobj **syncobjs = NULL; - struct drm_msm_gem_submit_syncobj syncobj_desc = {0}; - int ret = 0; - uint32_t i, j; - - syncobjs = kcalloc(nr_in_syncobjs, sizeof(*syncobjs), - GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); - if (!syncobjs) - return ERR_PTR(-ENOMEM); - - for (i = 0; i < nr_in_syncobjs; ++i) { - uint64_t address = in_syncobjs_addr + i * syncobj_stride; - - if (copy_from_user(&syncobj_desc, - u64_to_user_ptr(address), - min(syncobj_stride, sizeof(syncobj_desc)))) { - ret = -EFAULT; - break; - } - - if (syncobj_desc.point && - !drm_core_check_feature(submit->dev, DRIVER_SYNCOBJ_TIMELINE)) { - ret = SUBMIT_ERROR(EOPNOTSUPP, submit, "syncobj timeline unsupported"); - break; - } - - if (syncobj_desc.flags & ~MSM_SUBMIT_SYNCOBJ_FLAGS) { - ret = SUBMIT_ERROR(EINVAL, submit, "invalid syncobj flags: %x", syncobj_desc.flags); - break; - } - - ret = drm_sched_job_add_syncobj_dependency(&submit->base, file, - syncobj_desc.handle, syncobj_desc.point); - if (ret) - break; - - if (syncobj_desc.flags & MSM_SUBMIT_SYNCOBJ_RESET) { - syncobjs[i] = - drm_syncobj_find(file, syncobj_desc.handle); - if (!syncobjs[i]) { - ret = SUBMIT_ERROR(EINVAL, submit, "invalid syncobj handle: %u", i); - break; - } - } - } - - if (ret) { - for (j = 0; j <= i; ++j) { - if (syncobjs[j]) - drm_syncobj_put(syncobjs[j]); - } - kfree(syncobjs); - return ERR_PTR(ret); - } - return syncobjs; -} - -static void msm_reset_syncobjs(struct drm_syncobj **syncobjs, - uint32_t nr_syncobjs) -{ - uint32_t i; - - for (i = 0; syncobjs && i < nr_syncobjs; ++i) { - if (syncobjs[i]) - drm_syncobj_replace_fence(syncobjs[i], NULL); - } -} - -static struct msm_submit_post_dep *msm_parse_post_deps(struct drm_device *dev, - struct drm_file *file, - uint64_t syncobjs_addr, - uint32_t nr_syncobjs, - size_t syncobj_stride) -{ - struct msm_submit_post_dep *post_deps; - struct drm_msm_gem_submit_syncobj syncobj_desc = {0}; - int ret = 0; - uint32_t i, j; - - post_deps = kcalloc(nr_syncobjs, sizeof(*post_deps), - GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); - if (!post_deps) - return ERR_PTR(-ENOMEM); - - for (i = 0; i < nr_syncobjs; ++i) { - uint64_t address = syncobjs_addr + i * syncobj_stride; - - if (copy_from_user(&syncobj_desc, - u64_to_user_ptr(address), - min(syncobj_stride, sizeof(syncobj_desc)))) { - ret = -EFAULT; - break; - } - - post_deps[i].point = syncobj_desc.point; - - if (syncobj_desc.flags) { - ret = UERR(EINVAL, dev, "invalid syncobj flags"); - break; - } - - if (syncobj_desc.point) { - if (!drm_core_check_feature(dev, - DRIVER_SYNCOBJ_TIMELINE)) { - ret = UERR(EOPNOTSUPP, dev, "syncobj timeline unsupported"); - break; - } - - post_deps[i].chain = dma_fence_chain_alloc(); - if (!post_deps[i].chain) { - ret = -ENOMEM; - break; - } - } - - post_deps[i].syncobj = - drm_syncobj_find(file, syncobj_desc.handle); - if (!post_deps[i].syncobj) { - ret = UERR(EINVAL, dev, "invalid syncobj handle"); - break; - } - } - - if (ret) { - for (j = 0; j <= i; ++j) { - dma_fence_chain_free(post_deps[j].chain); - if (post_deps[j].syncobj) - drm_syncobj_put(post_deps[j].syncobj); - } - - kfree(post_deps); - return ERR_PTR(ret); - } - - return post_deps; -} - -static void msm_process_post_deps(struct msm_submit_post_dep *post_deps, - uint32_t count, struct dma_fence *fence) -{ - uint32_t i; - - for (i = 0; post_deps && i < count; ++i) { - if (post_deps[i].chain) { - drm_syncobj_add_point(post_deps[i].syncobj, - post_deps[i].chain, - fence, post_deps[i].point); - post_deps[i].chain = NULL; - } else { - drm_syncobj_replace_fence(post_deps[i].syncobj, - fence); - } - } -} - int msm_ioctl_gem_submit(struct drm_device *dev, void *data, struct drm_file *file) { @@ -663,7 +497,7 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, struct msm_gpu *gpu = priv->gpu; struct msm_gpu_submitqueue *queue; struct msm_ringbuffer *ring; - struct msm_submit_post_dep *post_deps = NULL; + struct msm_syncobj_post_dep *post_deps = NULL; struct drm_syncobj **syncobjs_to_reset = NULL; struct sync_file *sync_file = NULL; int out_fence_fd = -1; @@ -740,10 +574,10 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, } if (args->flags & MSM_SUBMIT_SYNCOBJ_IN) { - syncobjs_to_reset = msm_parse_deps(submit, file, - args->in_syncobjs, - args->nr_in_syncobjs, - args->syncobj_stride); + syncobjs_to_reset = msm_syncobj_parse_deps(dev, &submit->base, + file, args->in_syncobjs, + args->nr_in_syncobjs, + args->syncobj_stride); if (IS_ERR(syncobjs_to_reset)) { ret = PTR_ERR(syncobjs_to_reset); goto out_unlock; @@ -751,10 +585,10 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, } if (args->flags & MSM_SUBMIT_SYNCOBJ_OUT) { - post_deps = msm_parse_post_deps(dev, file, - args->out_syncobjs, - args->nr_out_syncobjs, - args->syncobj_stride); + post_deps = msm_syncobj_parse_post_deps(dev, file, + args->out_syncobjs, + args->nr_out_syncobjs, + args->syncobj_stride); if (IS_ERR(post_deps)) { ret = PTR_ERR(post_deps); goto out_unlock; @@ -897,10 +731,8 @@ int msm_ioctl_gem_submit(struct drm_device *dev, void *data, args->fence = submit->fence_id; queue->last_fence = submit->fence_id; - msm_reset_syncobjs(syncobjs_to_reset, args->nr_in_syncobjs); - msm_process_post_deps(post_deps, args->nr_out_syncobjs, - submit->user_fence); - + msm_syncobj_reset(syncobjs_to_reset, args->nr_in_syncobjs); + msm_syncobj_process_post_deps(post_deps, args->nr_out_syncobjs, submit->user_fence); out: submit_cleanup(submit, !!ret); diff --git a/drivers/gpu/drm/msm/msm_syncobj.c b/drivers/gpu/drm/msm/msm_syncobj.c new file mode 100644 index 000000000000..4baa9f522c54 --- /dev/null +++ b/drivers/gpu/drm/msm/msm_syncobj.c @@ -0,0 +1,172 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* Copyright (C) 2020 Google, Inc */ + +#include "drm/drm_drv.h" + +#include "msm_drv.h" +#include "msm_syncobj.h" + +struct drm_syncobj ** +msm_syncobj_parse_deps(struct drm_device *dev, + struct drm_sched_job *job, + struct drm_file *file, + uint64_t in_syncobjs_addr, + uint32_t nr_in_syncobjs, + size_t syncobj_stride) +{ + struct drm_syncobj **syncobjs = NULL; + struct drm_msm_syncobj syncobj_desc = {0}; + int ret = 0; + uint32_t i, j; + + syncobjs = kcalloc(nr_in_syncobjs, sizeof(*syncobjs), + GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); + if (!syncobjs) + return ERR_PTR(-ENOMEM); + + for (i = 0; i < nr_in_syncobjs; ++i) { + uint64_t address = in_syncobjs_addr + i * syncobj_stride; + + if (copy_from_user(&syncobj_desc, + u64_to_user_ptr(address), + min(syncobj_stride, sizeof(syncobj_desc)))) { + ret = -EFAULT; + break; + } + + if (syncobj_desc.point && + !drm_core_check_feature(dev, DRIVER_SYNCOBJ_TIMELINE)) { + ret = UERR(EOPNOTSUPP, dev, "syncobj timeline unsupported"); + break; + } + + if (syncobj_desc.flags & ~MSM_SYNCOBJ_FLAGS) { + ret = UERR(EINVAL, dev, "invalid syncobj flags: %x", syncobj_desc.flags); + break; + } + + ret = drm_sched_job_add_syncobj_dependency(job, file, + syncobj_desc.handle, + syncobj_desc.point); + if (ret) + break; + + if (syncobj_desc.flags & MSM_SYNCOBJ_RESET) { + syncobjs[i] = drm_syncobj_find(file, syncobj_desc.handle); + if (!syncobjs[i]) { + ret = UERR(EINVAL, dev, "invalid syncobj handle: %u", i); + break; + } + } + } + + if (ret) { + for (j = 0; j <= i; ++j) { + if (syncobjs[j]) + drm_syncobj_put(syncobjs[j]); + } + kfree(syncobjs); + return ERR_PTR(ret); + } + return syncobjs; +} + +void +msm_syncobj_reset(struct drm_syncobj **syncobjs, uint32_t nr_syncobjs) +{ + uint32_t i; + + for (i = 0; syncobjs && i < nr_syncobjs; ++i) { + if (syncobjs[i]) + drm_syncobj_replace_fence(syncobjs[i], NULL); + } +} + +struct msm_syncobj_post_dep * +msm_syncobj_parse_post_deps(struct drm_device *dev, + struct drm_file *file, + uint64_t syncobjs_addr, + uint32_t nr_syncobjs, + size_t syncobj_stride) +{ + struct msm_syncobj_post_dep *post_deps; + struct drm_msm_syncobj syncobj_desc = {0}; + int ret = 0; + uint32_t i, j; + + post_deps = kcalloc(nr_syncobjs, sizeof(*post_deps), + GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY); + if (!post_deps) + return ERR_PTR(-ENOMEM); + + for (i = 0; i < nr_syncobjs; ++i) { + uint64_t address = syncobjs_addr + i * syncobj_stride; + + if (copy_from_user(&syncobj_desc, + u64_to_user_ptr(address), + min(syncobj_stride, sizeof(syncobj_desc)))) { + ret = -EFAULT; + break; + } + + post_deps[i].point = syncobj_desc.point; + + if (syncobj_desc.flags) { + ret = UERR(EINVAL, dev, "invalid syncobj flags"); + break; + } + + if (syncobj_desc.point) { + if (!drm_core_check_feature(dev, + DRIVER_SYNCOBJ_TIMELINE)) { + ret = UERR(EOPNOTSUPP, dev, "syncobj timeline unsupported"); + break; + } + + post_deps[i].chain = dma_fence_chain_alloc(); + if (!post_deps[i].chain) { + ret = -ENOMEM; + break; + } + } + + post_deps[i].syncobj = + drm_syncobj_find(file, syncobj_desc.handle); + if (!post_deps[i].syncobj) { + ret = UERR(EINVAL, dev, "invalid syncobj handle"); + break; + } + } + + if (ret) { + for (j = 0; j <= i; ++j) { + dma_fence_chain_free(post_deps[j].chain); + if (post_deps[j].syncobj) + drm_syncobj_put(post_deps[j].syncobj); + } + + kfree(post_deps); + return ERR_PTR(ret); + } + + return post_deps; +} + +void +msm_syncobj_process_post_deps(struct msm_syncobj_post_dep *post_deps, + uint32_t count, struct dma_fence *fence) +{ + uint32_t i; + + for (i = 0; post_deps && i < count; ++i) { + if (post_deps[i].chain) { + drm_syncobj_add_point(post_deps[i].syncobj, + post_deps[i].chain, + fence, post_deps[i].point); + post_deps[i].chain = NULL; + } else { + drm_syncobj_replace_fence(post_deps[i].syncobj, + fence); + } + } +} diff --git a/drivers/gpu/drm/msm/msm_syncobj.h b/drivers/gpu/drm/msm/msm_syncobj.h new file mode 100644 index 000000000000..bcaa15d01da0 --- /dev/null +++ b/drivers/gpu/drm/msm/msm_syncobj.h @@ -0,0 +1,37 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* Copyright (C) 2020 Google, Inc */ + +#ifndef __MSM_GEM_SYNCOBJ_H__ +#define __MSM_GEM_SYNCOBJ_H__ + +#include "drm/drm_device.h" +#include "drm/drm_syncobj.h" +#include "drm/gpu_scheduler.h" + +struct msm_syncobj_post_dep { + struct drm_syncobj *syncobj; + uint64_t point; + struct dma_fence_chain *chain; +}; + +struct drm_syncobj ** +msm_syncobj_parse_deps(struct drm_device *dev, + struct drm_sched_job *job, + struct drm_file *file, + uint64_t in_syncobjs_addr, + uint32_t nr_in_syncobjs, + size_t syncobj_stride); + +void msm_syncobj_reset(struct drm_syncobj **syncobjs, uint32_t nr_syncobjs); + +struct msm_syncobj_post_dep * +msm_syncobj_parse_post_deps(struct drm_device *dev, + struct drm_file *file, + uint64_t syncobjs_addr, + uint32_t nr_syncobjs, + size_t syncobj_stride); + +void msm_syncobj_process_post_deps(struct msm_syncobj_post_dep *post_deps, + uint32_t count, struct dma_fence *fence); + +#endif /* __MSM_GEM_SYNCOBJ_H__ */ diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index 1bccc347945c..2c2fc4b284d0 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -220,6 +220,17 @@ struct drm_msm_gem_cpu_fini { * Cmdstream Submission: */ +#define MSM_SYNCOBJ_RESET 0x00000001 /* Reset syncobj after wait. */ +#define MSM_SYNCOBJ_FLAGS ( \ + MSM_SYNCOBJ_RESET | \ + 0) + +struct drm_msm_syncobj { + __u32 handle; /* in, syncobj handle. */ + __u32 flags; /* in, from MSM_SUBMIT_SYNCOBJ_FLAGS */ + __u64 point; /* in, timepoint for timeline syncobjs. */ +}; + /* The value written into the cmdstream is logically: * * ((relocbuf->gpuaddr + reloc_offset) << shift) | or @@ -309,17 +320,6 @@ struct drm_msm_gem_submit_bo { MSM_SUBMIT_FENCE_SN_IN | \ 0) -#define MSM_SUBMIT_SYNCOBJ_RESET 0x00000001 /* Reset syncobj after wait. */ -#define MSM_SUBMIT_SYNCOBJ_FLAGS ( \ - MSM_SUBMIT_SYNCOBJ_RESET | \ - 0) - -struct drm_msm_gem_submit_syncobj { - __u32 handle; /* in, syncobj handle. */ - __u32 flags; /* in, from MSM_SUBMIT_SYNCOBJ_FLAGS */ - __u64 point; /* in, timepoint for timeline syncobjs. */ -}; - /* Each cmdstream submit consists of a table of buffers involved, and * one or more cmdstream buffers. This allows for conditional execution * (context-restore), and IB buffers needed for per tile/bin draw cmds. @@ -333,8 +333,8 @@ struct drm_msm_gem_submit { __u64 cmds; /* in, ptr to array of submit_cmd's */ __s32 fence_fd; /* in/out fence fd (see MSM_SUBMIT_FENCE_FD_IN/OUT) */ __u32 queueid; /* in, submitqueue id */ - __u64 in_syncobjs; /* in, ptr to array of drm_msm_gem_submit_syncobj */ - __u64 out_syncobjs; /* in, ptr to array of drm_msm_gem_submit_syncobj */ + __u64 in_syncobjs; /* in, ptr to array of drm_msm_syncobj */ + __u64 out_syncobjs; /* in, ptr to array of drm_msm_syncobj */ __u32 nr_in_syncobjs; /* in, number of entries in in_syncobj */ __u32 nr_out_syncobjs; /* in, number of entries in out_syncobj. */ __u32 syncobj_stride; /* in, stride of syncobj arrays. */ -- 2.49.0

1 month, 1 week

[PATCH v5 24/40] drm/msm: Add _NO_SHARE flag

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> Buffers that are not shared between contexts can share a single resv object. This way drm_gpuvm will not track them as external objects, and submit-time validating overhead will be O(1) for all N non-shared BOs, instead of O(n). Signed-off-by: Rob Clark <robdclark(a)chromium.org> --- drivers/gpu/drm/msm/msm_drv.h | 1 + drivers/gpu/drm/msm/msm_gem.c | 23 +++++++++++++++++++++++ drivers/gpu/drm/msm/msm_gem_prime.c | 15 +++++++++++++++ include/uapi/drm/msm_drm.h | 14 ++++++++++++++ 4 files changed, 53 insertions(+) diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h index b77fd2c531c3..b0add236cbb3 100644 --- a/drivers/gpu/drm/msm/msm_drv.h +++ b/drivers/gpu/drm/msm/msm_drv.h @@ -246,6 +246,7 @@ int msm_gem_prime_vmap(struct drm_gem_object *obj, struct iosys_map *map); void msm_gem_prime_vunmap(struct drm_gem_object *obj, struct iosys_map *map); struct drm_gem_object *msm_gem_prime_import_sg_table(struct drm_device *dev, struct dma_buf_attachment *attach, struct sg_table *sg); +struct dma_buf *msm_gem_prime_export(struct drm_gem_object *obj, int flags); int msm_gem_prime_pin(struct drm_gem_object *obj); void msm_gem_prime_unpin(struct drm_gem_object *obj); diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 36b9e9eefc3c..65ec99526f82 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -532,6 +532,9 @@ static int get_and_pin_iova_range_locked(struct drm_gem_object *obj, msm_gem_assert_locked(obj); + if (to_msm_bo(obj)->flags & MSM_BO_NO_SHARE) + return -EINVAL; + vma = get_vma_locked(obj, vm, range_start, range_end); if (IS_ERR(vma)) return PTR_ERR(vma); @@ -1060,6 +1063,16 @@ static void msm_gem_free_object(struct drm_gem_object *obj) put_pages(obj); } + if (obj->resv != &obj->_resv) { + struct drm_gem_object *r_obj = + container_of(obj->resv, struct drm_gem_object, _resv); + + BUG_ON(!(msm_obj->flags & MSM_BO_NO_SHARE)); + + /* Drop reference we hold to shared resv obj: */ + drm_gem_object_put(r_obj); + } + drm_gem_object_release(obj); kfree(msm_obj->metadata); @@ -1092,6 +1105,15 @@ int msm_gem_new_handle(struct drm_device *dev, struct drm_file *file, if (name) msm_gem_object_set_name(obj, "%s", name); + if (flags & MSM_BO_NO_SHARE) { + struct msm_context *ctx = file->driver_priv; + struct drm_gem_object *r_obj = drm_gpuvm_resv_obj(ctx->vm); + + drm_gem_object_get(r_obj); + + obj->resv = r_obj->resv; + } + ret = drm_gem_handle_create(file, obj, handle); /* drop reference from allocate - handle holds it now */ @@ -1124,6 +1146,7 @@ static const struct drm_gem_object_funcs msm_gem_object_funcs = { .free = msm_gem_free_object, .open = msm_gem_open, .close = msm_gem_close, + .export = msm_gem_prime_export, .pin = msm_gem_prime_pin, .unpin = msm_gem_prime_unpin, .get_sg_table = msm_gem_prime_get_sg_table, diff --git a/drivers/gpu/drm/msm/msm_gem_prime.c b/drivers/gpu/drm/msm/msm_gem_prime.c index ee267490c935..1a6d8099196a 100644 --- a/drivers/gpu/drm/msm/msm_gem_prime.c +++ b/drivers/gpu/drm/msm/msm_gem_prime.c @@ -16,6 +16,9 @@ struct sg_table *msm_gem_prime_get_sg_table(struct drm_gem_object *obj) struct msm_gem_object *msm_obj = to_msm_bo(obj); int npages = obj->size >> PAGE_SHIFT; + if (msm_obj->flags & MSM_BO_NO_SHARE) + return ERR_PTR(-EINVAL); + if (WARN_ON(!msm_obj->pages)) /* should have already pinned! */ return ERR_PTR(-ENOMEM); @@ -45,6 +48,15 @@ struct drm_gem_object *msm_gem_prime_import_sg_table(struct drm_device *dev, return msm_gem_import(dev, attach->dmabuf, sg); } + +struct dma_buf *msm_gem_prime_export(struct drm_gem_object *obj, int flags) +{ + if (to_msm_bo(obj)->flags & MSM_BO_NO_SHARE) + return ERR_PTR(-EPERM); + + return drm_gem_prime_export(obj, flags); +} + int msm_gem_prime_pin(struct drm_gem_object *obj) { struct page **pages; @@ -53,6 +65,9 @@ int msm_gem_prime_pin(struct drm_gem_object *obj) if (obj->import_attach) return 0; + if (to_msm_bo(obj)->flags & MSM_BO_NO_SHARE) + return -EINVAL; + pages = msm_gem_pin_pages_locked(obj); if (IS_ERR(pages)) ret = PTR_ERR(pages); diff --git a/include/uapi/drm/msm_drm.h b/include/uapi/drm/msm_drm.h index b974f5a24dbc..1bccc347945c 100644 --- a/include/uapi/drm/msm_drm.h +++ b/include/uapi/drm/msm_drm.h @@ -140,6 +140,19 @@ struct drm_msm_param { #define MSM_BO_SCANOUT 0x00000001 /* scanout capable */ #define MSM_BO_GPU_READONLY 0x00000002 +/* Private buffers do not need to be explicitly listed in the SUBMIT + * ioctl, unless referenced by a drm_msm_gem_submit_cmd. Private + * buffers may NOT be imported/exported or used for scanout (or any + * other situation where buffers can be indefinitely pinned, but + * cases other than scanout are all kernel owned BOs which are not + * visible to userspace). + * + * In exchange for those constraints, all private BOs associated with + * a single context (drm_file) share a single dma_resv, and if there + * has been no eviction since the last submit, there are no per-BO + * bookeeping to do, significantly cutting the SUBMIT overhead. + */ +#define MSM_BO_NO_SHARE 0x00000004 #define MSM_BO_CACHE_MASK 0x000f0000 /* cache modes */ #define MSM_BO_CACHED 0x00010000 @@ -149,6 +162,7 @@ struct drm_msm_param { #define MSM_BO_FLAGS (MSM_BO_SCANOUT | \ MSM_BO_GPU_READONLY | \ + MSM_BO_NO_SHARE | \ MSM_BO_CACHE_MASK) struct drm_msm_gem_new { -- 2.49.0

1 month, 1 week

[PATCH v5 14/40] drm/msm: Convert vm locking

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> Convert to using the gpuvm's r_obj for serializing access to the VM. This way we can use the drm_exec helper for dealing with deadlock detection and backoff. This will let us deal with upcoming locking order conflicts with the VM_BIND implmentation (ie. in some scenarious we need to acquire the obj lock first, for ex. to iterate all the VMs an obj is bound in, and in other scenarious we need to acquire the VM lock first). Signed-off-by: Rob Clark <robdclark(a)chromium.org> --- drivers/gpu/drm/msm/msm_gem.c | 35 ++++++++--- drivers/gpu/drm/msm/msm_gem.h | 37 ++++++++++-- drivers/gpu/drm/msm/msm_gem_shrinker.c | 80 +++++++++++++++++++++++--- drivers/gpu/drm/msm/msm_gem_submit.c | 9 ++- drivers/gpu/drm/msm/msm_gem_vma.c | 27 ++++----- 5 files changed, 150 insertions(+), 38 deletions(-) diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c index 3b7db3b3f763..b7055805a5dd 100644 --- a/drivers/gpu/drm/msm/msm_gem.c +++ b/drivers/gpu/drm/msm/msm_gem.c @@ -52,6 +52,7 @@ static void put_iova_spaces(struct drm_gem_object *obj, struct drm_gpuvm *vm, bo static void msm_gem_close(struct drm_gem_object *obj, struct drm_file *file) { struct msm_context *ctx = file->driver_priv; + struct drm_exec exec; update_ctx_mem(file, -obj->size); @@ -70,9 +71,9 @@ static void msm_gem_close(struct drm_gem_object *obj, struct drm_file *file) dma_resv_wait_timeout(obj->resv, DMA_RESV_USAGE_READ, false, msecs_to_jiffies(1000)); - msm_gem_lock(obj); + msm_gem_lock_vm_and_obj(&exec, obj, ctx->vm); put_iova_spaces(obj, &ctx->vm->base, true); - msm_gem_unlock(obj); + drm_exec_fini(&exec); /* drop locks */ } /* @@ -538,11 +539,12 @@ int msm_gem_get_and_pin_iova_range(struct drm_gem_object *obj, struct msm_gem_vm *vm, uint64_t *iova, u64 range_start, u64 range_end) { + struct drm_exec exec; int ret; - msm_gem_lock(obj); + msm_gem_lock_vm_and_obj(&exec, obj, vm); ret = get_and_pin_iova_range_locked(obj, vm, iova, range_start, range_end); - msm_gem_unlock(obj); + drm_exec_fini(&exec); /* drop locks */ return ret; } @@ -562,16 +564,17 @@ int msm_gem_get_iova(struct drm_gem_object *obj, struct msm_gem_vm *vm, uint64_t *iova) { struct msm_gem_vma *vma; + struct drm_exec exec; int ret = 0; - msm_gem_lock(obj); + msm_gem_lock_vm_and_obj(&exec, obj, vm); vma = get_vma_locked(obj, vm, 0, U64_MAX); if (IS_ERR(vma)) { ret = PTR_ERR(vma); } else { *iova = vma->base.va.addr; } - msm_gem_unlock(obj); + drm_exec_fini(&exec); /* drop locks */ return ret; } @@ -600,9 +603,10 @@ static int clear_iova(struct drm_gem_object *obj, int msm_gem_set_iova(struct drm_gem_object *obj, struct msm_gem_vm *vm, uint64_t iova) { + struct drm_exec exec; int ret = 0; - msm_gem_lock(obj); + msm_gem_lock_vm_and_obj(&exec, obj, vm); if (!iova) { ret = clear_iova(obj, vm); } else { @@ -615,7 +619,7 @@ int msm_gem_set_iova(struct drm_gem_object *obj, ret = -EBUSY; } } - msm_gem_unlock(obj); + drm_exec_fini(&exec); /* drop locks */ return ret; } @@ -1007,12 +1011,27 @@ static void msm_gem_free_object(struct drm_gem_object *obj) struct msm_gem_object *msm_obj = to_msm_bo(obj); struct drm_device *dev = obj->dev; struct msm_drm_private *priv = dev->dev_private; + struct drm_exec exec; mutex_lock(&priv->obj_lock); list_del(&msm_obj->node); mutex_unlock(&priv->obj_lock); + /* + * We need to lock any VMs the object is still attached to, but not + * the object itself (see explaination in msm_gem_assert_locked()), + * so just open-code this special case: + */ + drm_exec_init(&exec, 0, 0); + drm_exec_until_all_locked (&exec) { + struct drm_gpuvm_bo *vm_bo; + drm_gem_for_each_gpuvm_bo (vm_bo, obj) { + drm_exec_lock_obj(&exec, drm_gpuvm_resv_obj(vm_bo->vm)); + drm_exec_retry_on_contention(&exec); + } + } put_iova_spaces(obj, NULL, true); + drm_exec_fini(&exec); /* drop locks */ if (obj->import_attach) { GEM_WARN_ON(msm_obj->vaddr); diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h index f7f7e7910754..36a846e9b943 100644 --- a/drivers/gpu/drm/msm/msm_gem.h +++ b/drivers/gpu/drm/msm/msm_gem.h @@ -62,12 +62,6 @@ struct msm_gem_vm { */ struct drm_mm mm; - /** @mm_lock: protects @mm node allocation/removal */ - struct spinlock mm_lock; - - /** @vm_lock: protects gpuvm insert/remove/traverse */ - struct mutex vm_lock; - /** @mmu: The mmu object which manages the pgtables */ struct msm_mmu *mmu; @@ -246,6 +240,37 @@ msm_gem_unlock(struct drm_gem_object *obj) dma_resv_unlock(obj->resv); } +/** + * msm_gem_lock_vm_and_obj() - Helper to lock an obj + VM + * @exec: the exec context helper which will be initalized + * @obj: the GEM object to lock + * @vm: the VM to lock + * + * Operations which modify a VM frequently need to lock both the VM and + * the object being mapped/unmapped/etc. This helper uses drm_exec to + * acquire both locks, dealing with potential deadlock/backoff scenarios + * which arise when multiple locks are involved. + */ +static inline int +msm_gem_lock_vm_and_obj(struct drm_exec *exec, + struct drm_gem_object *obj, + struct msm_gem_vm *vm) +{ + int ret = 0; + + drm_exec_init(exec, 0, 2); + drm_exec_until_all_locked (exec) { + ret = drm_exec_lock_obj(exec, drm_gpuvm_resv_obj(&vm->base)); + if (!ret && (obj->resv != drm_gpuvm_resv(&vm->base))) + ret = drm_exec_lock_obj(exec, obj); + drm_exec_retry_on_contention(exec); + if (GEM_WARN_ON(ret)) + break; + } + + return ret; +} + static inline void msm_gem_assert_locked(struct drm_gem_object *obj) { diff --git a/drivers/gpu/drm/msm/msm_gem_shrinker.c b/drivers/gpu/drm/msm/msm_gem_shrinker.c index de185fc34084..5faf6227584a 100644 --- a/drivers/gpu/drm/msm/msm_gem_shrinker.c +++ b/drivers/gpu/drm/msm/msm_gem_shrinker.c @@ -43,6 +43,75 @@ msm_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc) return count; } +static bool +with_vm_locks(struct ww_acquire_ctx *ticket, + void (*fn)(struct drm_gem_object *obj), + struct drm_gem_object *obj) +{ + /* + * Track last locked entry for for unwinding locks in error and + * success paths + */ + struct drm_gpuvm_bo *vm_bo, *last_locked = NULL; + int ret = 0; + + drm_gem_for_each_gpuvm_bo (vm_bo, obj) { + struct dma_resv *resv = drm_gpuvm_resv(vm_bo->vm); + + if (resv == obj->resv) + continue; + + ret = dma_resv_lock(resv, ticket); + + /* + * Since we already skip the case when the VM and obj + * share a resv (ie. _NO_SHARE objs), we don't expect + * to hit a double-locking scenario... which the lock + * unwinding cannot really cope with. + */ + WARN_ON(ret == -EALREADY); + + /* + * Don't bother with slow-lock / backoff / retry sequence, + * if we can't get the lock just give up and move on to + * the next object. + */ + if (ret) + goto out_unlock; + + /* + * Hold a ref to prevent the vm_bo from being freed + * and removed from the obj's gpuva list, as that would + * would result in missing the unlock below + */ + drm_gpuvm_bo_get(vm_bo); + + last_locked = vm_bo; + } + + fn(obj); + +out_unlock: + if (last_locked) { + drm_gem_for_each_gpuvm_bo (vm_bo, obj) { + struct dma_resv *resv = drm_gpuvm_resv(vm_bo->vm); + + if (resv == obj->resv) + continue; + + dma_resv_unlock(resv); + + /* Drop the ref taken while locking: */ + drm_gpuvm_bo_put(vm_bo); + + if (last_locked == vm_bo) + break; + } + } + + return ret == 0; +} + static bool purge(struct drm_gem_object *obj, struct ww_acquire_ctx *ticket) { @@ -52,9 +121,7 @@ purge(struct drm_gem_object *obj, struct ww_acquire_ctx *ticket) if (msm_gem_active(obj)) return false; - msm_gem_purge(obj); - - return true; + return with_vm_locks(ticket, msm_gem_purge, obj); } static bool @@ -66,9 +133,7 @@ evict(struct drm_gem_object *obj, struct ww_acquire_ctx *ticket) if (msm_gem_active(obj)) return false; - msm_gem_evict(obj); - - return true; + return with_vm_locks(ticket, msm_gem_evict, obj); } static bool @@ -100,6 +165,7 @@ static unsigned long msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) { struct msm_drm_private *priv = shrinker->private_data; + struct ww_acquire_ctx ticket; struct { struct drm_gem_lru *lru; bool (*shrink)(struct drm_gem_object *obj, struct ww_acquire_ctx *ticket); @@ -124,7 +190,7 @@ msm_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc) drm_gem_lru_scan(stages[i].lru, nr, &stages[i].remaining, stages[i].shrink, - NULL); + &ticket); nr -= stages[i].freed; freed += stages[i].freed; remaining += stages[i].remaining; diff --git a/drivers/gpu/drm/msm/msm_gem_submit.c b/drivers/gpu/drm/msm/msm_gem_submit.c index 86791a854c42..6924d03026ba 100644 --- a/drivers/gpu/drm/msm/msm_gem_submit.c +++ b/drivers/gpu/drm/msm/msm_gem_submit.c @@ -256,11 +256,18 @@ static int submit_lookup_cmds(struct msm_gem_submit *submit, /* This is where we make sure all the bo's are reserved and pin'd: */ static int submit_lock_objects(struct msm_gem_submit *submit) { + unsigned flags = DRM_EXEC_IGNORE_DUPLICATES | DRM_EXEC_INTERRUPTIBLE_WAIT; int ret; - drm_exec_init(&submit->exec, DRM_EXEC_INTERRUPTIBLE_WAIT, submit->nr_bos); +// TODO need to add vm_bind path which locks vm resv + external objs + drm_exec_init(&submit->exec, flags, submit->nr_bos); drm_exec_until_all_locked (&submit->exec) { + ret = drm_exec_lock_obj(&submit->exec, + drm_gpuvm_resv_obj(&submit->vm->base)); + drm_exec_retry_on_contention(&submit->exec); + if (ret) + goto error; for (unsigned i = 0; i < submit->nr_bos; i++) { struct drm_gem_object *obj = submit->bos[i].obj; ret = drm_exec_prepare_obj(&submit->exec, obj, 1); diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c index d1621761ef36..e294e7f6e723 100644 --- a/drivers/gpu/drm/msm/msm_gem_vma.c +++ b/drivers/gpu/drm/msm/msm_gem_vma.c @@ -92,15 +92,13 @@ void msm_gem_vma_close(struct msm_gem_vma *vma) GEM_WARN_ON(vma->mapped); - spin_lock(&vm->mm_lock); + drm_gpuvm_resv_assert_held(&vm->base); + if (vma->base.va.addr) drm_mm_remove_node(&vma->node); - spin_unlock(&vm->mm_lock); - mutex_lock(&vm->vm_lock); drm_gpuva_remove(&vma->base); drm_gpuva_unlink(&vma->base); - mutex_unlock(&vm->vm_lock); kfree(vma); } @@ -114,16 +112,16 @@ msm_gem_vma_new(struct msm_gem_vm *vm, struct drm_gem_object *obj, struct msm_gem_vma *vma; int ret; + drm_gpuvm_resv_assert_held(&vm->base); + vma = kzalloc(sizeof(*vma), GFP_KERNEL); if (!vma) return ERR_PTR(-ENOMEM); if (vm->managed) { - spin_lock(&vm->mm_lock); ret = drm_mm_insert_node_in_range(&vm->mm, &vma->node, obj->size, PAGE_SIZE, 0, range_start, range_end, 0); - spin_unlock(&vm->mm_lock); if (ret) goto err_free_vma; @@ -137,9 +135,7 @@ msm_gem_vma_new(struct msm_gem_vm *vm, struct drm_gem_object *obj, drm_gpuva_init(&vma->base, range_start, range_end - range_start, obj, 0); vma->mapped = false; - mutex_lock(&vm->vm_lock); ret = drm_gpuva_insert(&vm->base, &vma->base); - mutex_unlock(&vm->vm_lock); if (ret) goto err_free_range; @@ -149,18 +145,14 @@ msm_gem_vma_new(struct msm_gem_vm *vm, struct drm_gem_object *obj, goto err_va_remove; } - mutex_lock(&vm->vm_lock); drm_gpuvm_bo_extobj_add(vm_bo); drm_gpuva_link(&vma->base, vm_bo); - mutex_unlock(&vm->vm_lock); GEM_WARN_ON(drm_gpuvm_bo_put(vm_bo)); return vma; err_va_remove: - mutex_lock(&vm->vm_lock); drm_gpuva_remove(&vma->base); - mutex_unlock(&vm->vm_lock); err_free_range: if (vm->managed) drm_mm_remove_node(&vma->node); @@ -191,7 +183,13 @@ struct msm_gem_vm * msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, u64 va_start, u64 va_size, bool managed) { - enum drm_gpuvm_flags flags = managed ? DRM_GPUVM_VA_WEAK_REF : 0; + /* + * We mostly want to use DRM_GPUVM_RESV_PROTECTED, except that + * makes drm_gpuvm_bo_evict() a no-op for extobjs (ie. we loose + * tracking that an extobj is evicted) :facepalm: + */ + enum drm_gpuvm_flags flags = + (managed ? DRM_GPUVM_VA_WEAK_REF : 0); struct msm_gem_vm *vm; struct drm_gem_object *dummy_gem; int ret = 0; @@ -213,9 +211,6 @@ msm_gem_vm_create(struct drm_device *drm, struct msm_mmu *mmu, const char *name, va_start, va_size, 0, 0, &msm_gpuvm_ops); drm_gem_object_put(dummy_gem); - spin_lock_init(&vm->mm_lock); - mutex_init(&vm->vm_lock); - vm->mmu = mmu; vm->managed = managed; -- 2.49.0

1 month, 1 week

← Newer
1
...
6
7
8
9
10
11
12
...
291
Older →

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig