Hi Am 14.01.21 um 15:34 schrieb Kieran Bingham:
Hi Thomas, On 14/01/2021 13:26, Thomas Zimmermann wrote:Hi Kieran Am 14.01.21 um 13:51 schrieb Kieran Bingham:Hi Thomas, On 23/11/2020 11:56, Thomas Zimmermann wrote:The new GEM object function drm_gem_cma_mmap() sets the VMA flags and offset as in the old implementation and immediately maps in the buffer's memory pages. Changing CMA helpers to use the GEM object function allows for the removal of the special implementations for mmap and gem_prime_mmap callbacks. The regular functions drm_gem_mmap() and drm_gem_prime_mmap() are now used.I've encountered a memory leak regression in our Renesas R-Car DU tests, and git bisection has led me to this patch (as commit f5ca8eb6f9). Running the tests sequentially, while grepping /proc/meminfo for Cma, it is evident that CMA memory is not released, until exhausted and the allocations fail (seen in [0]) shown by the error report:self.fbs.append(pykms.DumbFramebuffer(self.card, mode.hdisplay, mode.vdisplay, "XR24")) ValueError: DRM_IOCTL_MODE_CREATE_DUMB failed: Cannot allocate memoryFailing tests at f5ca8eb6f9 can be seen at [0], while the tests pass successfully [1] on the commit previous to that (bc2532ab7c2): Reverting f5ca8eb6f9 also produces a successful pass [2] [0] https://paste.ubuntu.com/p/VjPGPgswxR/ # Failed at f5ca8eb6f9 [1] https://paste.ubuntu.com/p/78RRp2WpNR/ # Success at bc2532ab7c2 [2] https://paste.ubuntu.com/p/qJKjZZN2pt/ # Success with revert I don't believe we handle mmap specially in the RCar-DU driver, so I wonder if this issue has hit anyone else as well? Any ideas of a repair without a revert ? Or do we just need to submit a revert?I think we might not be setting the VMA ops and therefore not finalize the BO correctly. Could you please apply the attched (quick-and-dirty) patch and try again?Thanks for the quick response. I can confirm the quick-and-dirty patch resolves the issue: https://paste.ubuntu.com/p/sKDp3dNvwV/ You can add a Tested-by: Kieran Bingham <kieran.bingham+renesas@xxxxxxxxxxxxxxxx>
Great! If you don't mind, I'd also add you in the Reported-by tag.
if it stays like that, but I suspect there might be a better place to initialise the ops rather than in the mmap call itself.
I think that's the fix, basically. We could put such a line as a fall-back somewhere into the DRM core code. But I don't know if this really works with all drivers. Maybe there's one that requires vm_ops to be NULL.
Thanks for reporting this issue and testing quickly. Best regards Thomas
Happy to do further testing as required. -- Regards KieranBest regards ThomasI've yet to fully understand the implications of the patch below. Thanks -- Regards KieranSigned-off-by: Thomas Zimmermann <tzimmermann@xxxxxxx> --- drivers/gpu/drm/drm_file.c | 3 +- drivers/gpu/drm/drm_gem_cma_helper.c | 121 +++++++++------------------ drivers/gpu/drm/pl111/pl111_drv.c | 2 +- drivers/gpu/drm/vc4/vc4_bo.c | 2 +- include/drm/drm_gem_cma_helper.h | 10 +-- 5 files changed, 44 insertions(+), 94 deletions(-) diff --git a/drivers/gpu/drm/drm_file.c b/drivers/gpu/drm/drm_file.c index b50380fa80ce..80886d50d0f1 100644 --- a/drivers/gpu/drm/drm_file.c +++ b/drivers/gpu/drm/drm_file.c @@ -113,8 +113,7 @@ bool drm_dev_needs_global_mutex(struct drm_device *dev) * The memory mapping implementation will vary depending on how the driver * manages memory. Legacy drivers will use the deprecated drm_legacy_mmap() * function, modern drivers should use one of the provided memory-manager - * specific implementations. For GEM-based drivers this is drm_gem_mmap(), and - * for drivers which use the CMA GEM helpers it's drm_gem_cma_mmap(). + * specific implementations. For GEM-based drivers this is drm_gem_mmap(). * * No other file operations are supported by the DRM userspace API. Overall the * following is an example &file_operations structure:: diff --git a/drivers/gpu/drm/drm_gem_cma_helper.c b/drivers/gpu/drm/drm_gem_cma_helper.c index 6a4ef335ebc9..7942cf05cd93 100644 --- a/drivers/gpu/drm/drm_gem_cma_helper.c +++ b/drivers/gpu/drm/drm_gem_cma_helper.c @@ -38,6 +38,7 @@ static const struct drm_gem_object_funcs drm_gem_cma_default_funcs = { .print_info = drm_gem_cma_print_info, .get_sg_table = drm_gem_cma_get_sg_table, .vmap = drm_gem_cma_vmap, + .mmap = drm_gem_cma_mmap, .vm_ops = &drm_gem_cma_vm_ops, }; @@ -277,62 +278,6 @@ const struct vm_operations_struct drm_gem_cma_vm_ops = { }; EXPORT_SYMBOL_GPL(drm_gem_cma_vm_ops); -static int drm_gem_cma_mmap_obj(struct drm_gem_cma_object *cma_obj, - struct vm_area_struct *vma) -{ - int ret; - - /* - * Clear the VM_PFNMAP flag that was set by drm_gem_mmap(), and set the - * vm_pgoff (used as a fake buffer offset by DRM) to 0 as we want to map - * the whole buffer. - */ - vma->vm_flags &= ~VM_PFNMAP; - vma->vm_pgoff = 0; - - ret = dma_mmap_wc(cma_obj->base.dev->dev, vma, cma_obj->vaddr, - cma_obj->paddr, vma->vm_end - vma->vm_start); - if (ret) - drm_gem_vm_close(vma); - - return ret; -} - -/** - * drm_gem_cma_mmap - memory-map a CMA GEM object - * @filp: file object - * @vma: VMA for the area to be mapped - * - * This function implements an augmented version of the GEM DRM file mmap - * operation for CMA objects: In addition to the usual GEM VMA setup it - * immediately faults in the entire object instead of using on-demaind - * faulting. Drivers which employ the CMA helpers should use this function - * as their ->mmap() handler in the DRM device file's file_operations - * structure. - * - * Instead of directly referencing this function, drivers should use the - * DEFINE_DRM_GEM_CMA_FOPS().macro. - * - * Returns: - * 0 on success or a negative error code on failure. - */ -int drm_gem_cma_mmap(struct file *filp, struct vm_area_struct *vma) -{ - struct drm_gem_cma_object *cma_obj; - struct drm_gem_object *gem_obj; - int ret; - - ret = drm_gem_mmap(filp, vma); - if (ret) - return ret; - - gem_obj = vma->vm_private_data; - cma_obj = to_drm_gem_cma_obj(gem_obj); - - return drm_gem_cma_mmap_obj(cma_obj, vma); -} -EXPORT_SYMBOL_GPL(drm_gem_cma_mmap); - #ifndef CONFIG_MMU /** * drm_gem_cma_get_unmapped_area - propose address for mapping in noMMU cases @@ -500,33 +445,6 @@ drm_gem_cma_prime_import_sg_table(struct drm_device *dev, } EXPORT_SYMBOL_GPL(drm_gem_cma_prime_import_sg_table); -/** - * drm_gem_cma_prime_mmap - memory-map an exported CMA GEM object - * @obj: GEM object - * @vma: VMA for the area to be mapped - * - * This function maps a buffer imported via DRM PRIME into a userspace - * process's address space. Drivers that use the CMA helpers should set this - * as their &drm_driver.gem_prime_mmap callback. - * - * Returns: - * 0 on success or a negative error code on failure. - */ -int drm_gem_cma_prime_mmap(struct drm_gem_object *obj, - struct vm_area_struct *vma) -{ - struct drm_gem_cma_object *cma_obj; - int ret; - - ret = drm_gem_mmap_obj(obj, obj->size, vma); - if (ret < 0) - return ret; - - cma_obj = to_drm_gem_cma_obj(obj); - return drm_gem_cma_mmap_obj(cma_obj, vma); -} -EXPORT_SYMBOL_GPL(drm_gem_cma_prime_mmap); - /** * drm_gem_cma_vmap - map a CMA GEM object into the kernel's virtual * address space @@ -553,6 +471,43 @@ int drm_gem_cma_vmap(struct drm_gem_object *obj, struct dma_buf_map *map) } EXPORT_SYMBOL_GPL(drm_gem_cma_vmap); +/** + * drm_gem_cma_mmap - memory-map an exported CMA GEM object + * @obj: GEM object + * @vma: VMA for the area to be mapped + * + * This function maps a buffer into a userspace process's address space. + * In addition to the usual GEM VMA setup it immediately faults in the entire + * object instead of using on-demand faulting. Drivers that use the CMA + * helpers should set this as their &drm_gem_object_funcs.mmap callback. + * + * Returns: + * 0 on success or a negative error code on failure. + */ +int drm_gem_cma_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) +{ + struct drm_gem_cma_object *cma_obj; + int ret; + + /* + * Clear the VM_PFNMAP flag that was set by drm_gem_mmap(), and set the + * vm_pgoff (used as a fake buffer offset by DRM) to 0 as we want to map + * the whole buffer. + */ + vma->vm_pgoff -= drm_vma_node_start(&obj->vma_node); + vma->vm_flags &= ~VM_PFNMAP; + + cma_obj = to_drm_gem_cma_obj(obj); + + ret = dma_mmap_wc(cma_obj->base.dev->dev, vma, cma_obj->vaddr, + cma_obj->paddr, vma->vm_end - vma->vm_start); + if (ret) + drm_gem_vm_close(vma); + + return ret; +} +EXPORT_SYMBOL_GPL(drm_gem_cma_mmap); + /** * drm_gem_cma_prime_import_sg_table_vmap - PRIME import another driver's * scatter/gather table and get the virtual address of the buffer diff --git a/drivers/gpu/drm/pl111/pl111_drv.c b/drivers/gpu/drm/pl111/pl111_drv.c index 40e6708fbbe2..e4dcaef6c143 100644 --- a/drivers/gpu/drm/pl111/pl111_drv.c +++ b/drivers/gpu/drm/pl111/pl111_drv.c @@ -228,7 +228,7 @@ static const struct drm_driver pl111_drm_driver = { .prime_handle_to_fd = drm_gem_prime_handle_to_fd, .prime_fd_to_handle = drm_gem_prime_fd_to_handle, .gem_prime_import_sg_table = pl111_gem_import_sg_table, - .gem_prime_mmap = drm_gem_cma_prime_mmap, + .gem_prime_mmap = drm_gem_prime_mmap, #if defined(CONFIG_DEBUG_FS) .debugfs_init = pl111_debugfs_init, diff --git a/drivers/gpu/drm/vc4/vc4_bo.c b/drivers/gpu/drm/vc4/vc4_bo.c index 813e6cb3f9af..dc316cb79e00 100644 --- a/drivers/gpu/drm/vc4/vc4_bo.c +++ b/drivers/gpu/drm/vc4/vc4_bo.c @@ -782,7 +782,7 @@ int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma) return -EINVAL; } - return drm_gem_cma_prime_mmap(obj, vma); + return drm_gem_prime_mmap(obj, vma); } int vc4_prime_vmap(struct drm_gem_object *obj, struct dma_buf_map *map) diff --git a/include/drm/drm_gem_cma_helper.h b/include/drm/drm_gem_cma_helper.h index 4680275ab339..0a9711caa3e8 100644 --- a/include/drm/drm_gem_cma_helper.h +++ b/include/drm/drm_gem_cma_helper.h @@ -59,7 +59,7 @@ struct drm_gem_cma_object { .poll = drm_poll,\ .read = drm_read,\ .llseek = noop_llseek,\ - .mmap = drm_gem_cma_mmap,\ + .mmap = drm_gem_mmap,\ DRM_GEM_CMA_UNMAPPED_AREA_FOPS \ } @@ -76,9 +76,6 @@ int drm_gem_cma_dumb_create(struct drm_file *file_priv, struct drm_device *drm, struct drm_mode_create_dumb *args); -/* set vm_flags and we can change the VM attribute to other one at here */ -int drm_gem_cma_mmap(struct file *filp, struct vm_area_struct *vma); - /* allocate physical memory */ struct drm_gem_cma_object *drm_gem_cma_create(struct drm_device *drm, size_t size); @@ -101,9 +98,8 @@ struct drm_gem_object * drm_gem_cma_prime_import_sg_table(struct drm_device *dev, struct dma_buf_attachment *attach, struct sg_table *sgt); -int drm_gem_cma_prime_mmap(struct drm_gem_object *obj, - struct vm_area_struct *vma); int drm_gem_cma_vmap(struct drm_gem_object *obj, struct dma_buf_map *map); +int drm_gem_cma_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma); /** * DRM_GEM_CMA_DRIVER_OPS_WITH_DUMB_CREATE - CMA GEM driver operations @@ -123,7 +119,7 @@ int drm_gem_cma_vmap(struct drm_gem_object *obj, struct dma_buf_map *map); .prime_handle_to_fd = drm_gem_prime_handle_to_fd, \ .prime_fd_to_handle = drm_gem_prime_fd_to_handle, \ .gem_prime_import_sg_table = drm_gem_cma_prime_import_sg_table, \ - .gem_prime_mmap = drm_gem_cma_prime_mmap + .gem_prime_mmap = drm_gem_prime_mmap /** * DRM_GEM_CMA_DRIVER_OPS - CMA GEM driver operations_______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Felix Imendörffer
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
