dma-buf: Performance improvements for system heap & a system-uncached implementation
v3 cover dma-buf: Performance improvements for system heap & a system-uncached implementation
Hey All,
So this is another revision of my patch series to performance
optimizations to the dma-buf system heap.
Unfortunately, in working these up, I realized the heap-helpers infrastructure we tried to add to miniimize code duplication is not as generic as we intended. For some heaps it makes sense to deal with page lists, for other heaps it makes more sense to track things with sgtables.
So this series reworks the system heap to use sgtables, and then consolidates the pagelist method from the heap-helpers into the CMA heap. After which the heap-helpers logic is removed (as it is unused). I'd still like to find a better way to avoid some of the logic duplication in implementing the entire dma_buf_ops handlers per heap. But unfortunately that code is tied somewhat to how the buffer's memory is tracked.
After this, the series introduces an optimization that Ørjan Eide implemented for ION that avoids calling sync on attachments that don't have a mapping.
Next, an optimization to use larger order pages for the system heap. This change brings us closer to the current performance of the ION code.
Unfortunately, after submitting the last round, I realized that part of the reason the page-pooling patch I had included was providing such great performance numbers, was because the network page-pool implementation doesn't zero pages that it pulls from the cache. This is very inappropriate for buffers we pass to userland and was what gave it an unfair advantage (almost constant time performance) relative to ION's allocation performance numbers. I added some patches to zero the buffers manually similar to how ION does it, but I found this resulted in basically no performance improvement from the standard page allocator. Thus I've dropped that patch in this series for now.
Unfortunately this means we still have a performance delta from the ION system heap as measured by my microbenchmark, and this delta comes from ION system_heap's use of deferred freeing of pages. So less work is done in the measured interval of the microbenchmark. I'll be looking at adding similar code eventually but I don't want to hold the rest of the patches up on this, as it is still a good improvement over the current code.
I've updated the chart I shared earlier with current numbers (including with the unsubmitted net pagepool implementation, and with a different unsubmitted pagepool implementation borrowed from ION) here:
https://docs.google.com/spreadsheets/d/1-1C8ZQpmkl_0DISkI6z4xelE08MlNAN7oEu34AnO4Ao/edit?usp=sharing
I did add to this series a reworked version of my uncached system heap implementation I was submitting a few weeks back. Since it duplicated a lot of the now reworked system heap code, I realized it would be much simpler to add the functionality to the system_heap implementaiton itself.
While not improving the core allocation performance, the uncached heap allocations do result in *much* improved performance on HiKey960 as it avoids a lot of flushing and invalidating buffers that the cpu doesn't touch often.
Feedback on these would be great!
thanks
-john
New in v3:
* Dropped page-pool patches as after correcting the code to zero buffers, they provided no net performance gain.
* Added system-uncached implementation ontop of reworked system-heap.
* Use the new sgtable mapping functions, in the system and cma code as Suggested-by: Daniel Mentz <[email protected]>
* Cleanup: Use page_size() rather then open-coding it
John Stultz (7):
dma-buf: system_heap: Rework system heap to use sgtables instead of
pagelists
dma-buf: heaps: Move heap-helper logic into the cma_heap
implementation
dma-buf: heaps: Remove heap-helpers code
dma-buf: heaps: Skip sync if not mapped
dma-buf: system_heap: Allocate higher order pages if available
dma-buf: dma-heap: Keep track of the heap device struct
dma-buf: system_heap: Add a system-uncached heap re-using the system
heap
drivers/dma-buf/dma-heap.c | 33 +- drivers/dma-buf/heaps/Makefile | 1 - drivers/dma-buf/heaps/cma_heap.c | 327 +++++++++++++++--- drivers/dma-buf/heaps/heap-helpers.c | 271 --------------- drivers/dma-buf/heaps/heap-helpers.h | 53 --- drivers/dma-buf/heaps/system_heap.c | 480 ++++++++++++++++++++++++--- include/linux/dma-heap.h | 9 + 7 files changed, 741 insertions(+), 433 deletions(-) delete mode 100644 drivers/dma-buf/heaps/heap-helpers.c delete mode 100644 drivers/dma-buf/heaps/heap-helpers.h
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 1/7 dma-buf: system_heap: Rework system heap to use sgtables instead of pagelists
In preparation for some patches to optmize the system heap code, rework the dmabuf exporter to utilize sgtables rather then pageslists for tracking the associated pages.
This will allow for large order page allocations, as well as more efficient page pooling.
In doing so, the system heap stops using the heap-helpers logic which sadly is not quite as generic as I was hoping it to be, so this patch adds heap specific implementations of the dma_buf_ops function handlers.
---
v2:
* Fix locking issue and an unused return value Reported-by:
kernel test robot <[email protected]>
Julia Lawall <[email protected]>
* Make system_heap_buf_ops static Reported-by:
kernel test robot <[email protected]>
v3:
* Use the new sgtable mapping functions, as Suggested-by:
Daniel Mentz <[email protected]>
--- drivers/dma-buf/heaps/system_heap.c | 344 ++++++++++++++++++++++++---- 1 file changed, 298 insertions(+), 46 deletions(-)
drivers/dma-buf/heaps/system_heap.c +298 -46
@@ -3,7 +3,11 @@* DMABUF System heap exporter * * Copyright (C) 2011 Google, Inc. -* Copyright (C) 2019 Linaro Ltd. +* Copyright (C) 2019, 2020 Linaro Ltd. +* +* Portions based off of Andrew Davis' SRAM heap: +* Copyright (C) 2019 Texas Instruments Incorporated - http://www.ti.com/ +* Andrew F. Davis <[email protected]> */ #include <linux/dma-buf.h>
@@ -15,72 +19,321 @@#include <linux/module.h> #include <linux/scatterlist.h> #include <linux/slab.h> -#include <linux/sched/signal.h> -#include <asm/page.h> - -#include "heap-helpers.h" +#include <linux/vmalloc.h> struct dma_heap *sys_heap; -static void system_heap_free (struct heap_helper_buffer *buffer) +struct system_heap_buffer { +struct dma_heap *heap; +struct list_head attachments; +struct mutex lock; +unsigned long len; +struct sg_table sg_table; +int vmap_cnt; +void *vaddr; +}; + +struct dma_heap_attachment { +struct device *dev; +struct sg_table *table; +struct list_head list; +}; + +static struct sg_table *dup_sg_table (struct sg_table *table) { -pgoff_t pg; +struct sg_table *new_table; +int ret, i; +struct scatterlist *sg, *new_sg; + + new_table =kzalloc (sizeof (*new_table),GFP_KERNEL ); +if (!new_table) +return ERR_PTR (-ENOMEM ); + + ret =sg_alloc_table (new_table, table->orig_nents,GFP_KERNEL ); +if (ret) { +kfree (new_table); +return ERR_PTR (-ENOMEM ); + } + + new_sg = new_table->sgl; +for_each_sgtable_sg (table, sg, i) { +sg_set_page (new_sg,sg_page (sg), sg->length, sg->offset); + new_sg =sg_next (new_sg); + } + +return new_table; +} + +static int system_heap_attach (struct dma_buf *dmabuf, +struct dma_buf_attachment *attachment) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; +struct sg_table *table; + + a =kzalloc (sizeof (*a),GFP_KERNEL ); +if (!a) +return -ENOMEM ; + + table =dup_sg_table (&buffer->sg_table); +if (IS_ERR (table)) { +kfree (a); +return -ENOMEM ; + } + + a->table = table; + a->dev = attachment->dev; +INIT_LIST_HEAD (&a->list); + + attachment->priv = a; + +mutex_lock (&buffer->lock); +list_add (&a->list, &buffer->attachments); +mutex_unlock (&buffer->lock); + +return 0 ; +} + +static void system_heap_detatch (struct dma_buf *dmabuf, +struct dma_buf_attachment *attachment) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a = attachment->priv; + +mutex_lock (&buffer->lock); +list_del (&a->list); +mutex_unlock (&buffer->lock); + +sg_free_table (a->table); +kfree (a->table); +kfree (a); +} + +static struct sg_table *system_heap_map_dma_buf (struct dma_buf_attachment *attachment, +enum dma_data_direction direction) +{ +struct dma_heap_attachment *a = attachment->priv; +struct sg_table *table = a->table; +int ret; + + ret =dma_map_sgtable (attachment->dev, table, direction,0 ); +if (ret) +return ERR_PTR (ret); + +return table; +} + +static void system_heap_unmap_dma_buf (struct dma_buf_attachment *attachment, +struct sg_table *table, +enum dma_data_direction direction) +{ +dma_unmap_sgtable (attachment->dev, table, direction,0 ); +} + +static int system_heap_dma_buf_begin_cpu_access (struct dma_buf *dmabuf, +enum dma_data_direction direction) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; + +mutex_lock (&buffer->lock); + +if (buffer->vmap_cnt) +invalidate_kernel_vmap_range (buffer->vaddr, buffer->len); + +list_for_each_entry (a, &buffer->attachments, list) { +dma_sync_sgtable_for_cpu (a->dev, a->table, direction); + } +mutex_unlock (&buffer->lock); + +return 0 ; +} -for (pg =0 ; pg < buffer->pagecount; pg++) -__free_page (buffer->pages[pg]); -kfree (buffer->pages); +static int system_heap_dma_buf_end_cpu_access (struct dma_buf *dmabuf, +enum dma_data_direction direction) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; + +mutex_lock (&buffer->lock); + +if (buffer->vmap_cnt) +flush_kernel_vmap_range (buffer->vaddr, buffer->len); + +list_for_each_entry (a, &buffer->attachments, list) { +dma_sync_sgtable_for_device (a->dev, a->table, direction); + } +mutex_unlock (&buffer->lock); + +return 0 ; +} + +static int system_heap_mmap (struct dma_buf *dmabuf,struct vm_area_struct *vma) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct sg_table *table = &buffer->sg_table; +unsigned long addr = vma->vm_start; +struct sg_page_iter piter; +int ret; + +for_each_sgtable_page (table, &piter, vma->vm_pgoff) { +struct page *page =sg_page_iter_page (&piter); + + ret =remap_pfn_range (vma, addr,page_to_pfn (page),PAGE_SIZE , + vma->vm_page_prot); +if (ret) +return ret; + addr +=PAGE_SIZE ; +if (addr >= vma->vm_end) +return 0 ; + } +return 0 ; +} + +static void *system_heap_do_vmap (struct system_heap_buffer *buffer) +{ +struct sg_table *table = &buffer->sg_table; +int npages =PAGE_ALIGN (buffer->len) /PAGE_SIZE ; +struct page **pages =vmalloc (sizeof (struct page *) * npages); +struct page **tmp = pages; +struct sg_page_iter piter; +void *vaddr; + +if (!pages) +return ERR_PTR (-ENOMEM ); + +for_each_sgtable_page (table, &piter,0 ) { +WARN_ON (tmp - pages >= npages); + *tmp++ =sg_page_iter_page (&piter); + } + + vaddr =vmap (pages, npages,VM_MAP ,PAGE_KERNEL ); +vfree (pages); + +if (!vaddr) +return ERR_PTR (-ENOMEM ); + +return vaddr; +} + +static void *system_heap_vmap (struct dma_buf *dmabuf) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +void *vaddr; + +mutex_lock (&buffer->lock); +if (buffer->vmap_cnt) { + buffer->vmap_cnt++; + vaddr = buffer->vaddr; + goto out; + } + + vaddr =system_heap_do_vmap (buffer); +if (IS_ERR (vaddr)) + goto out; + + buffer->vaddr = vaddr; + buffer->vmap_cnt++; +out: +mutex_unlock (&buffer->lock); + +return vaddr; +} + +static void system_heap_vunmap (struct dma_buf *dmabuf,void *vaddr) +{ +struct system_heap_buffer *buffer = dmabuf->priv; + +mutex_lock (&buffer->lock); +if (!--buffer->vmap_cnt) { +vunmap (buffer->vaddr); + buffer->vaddr =NULL ; + } +mutex_unlock (&buffer->lock); +} + +static void system_heap_dma_buf_release (struct dma_buf *dmabuf) +{ +struct system_heap_buffer *buffer = dmabuf->priv; +struct sg_table *table; +struct scatterlist *sg; +int i; + + table = &buffer->sg_table; +for_each_sgtable_sg (table, sg, i) +__free_page (sg_page (sg)); +sg_free_table (table);kfree (buffer); } +static const struct dma_buf_ops system_heap_buf_ops = { + .attach = system_heap_attach, + .detach = system_heap_detatch, + .map_dma_buf = system_heap_map_dma_buf, + .unmap_dma_buf = system_heap_unmap_dma_buf, + .begin_cpu_access = system_heap_dma_buf_begin_cpu_access, + .end_cpu_access = system_heap_dma_buf_end_cpu_access, + .mmap = system_heap_mmap, + .vmap = system_heap_vmap, + .vunmap = system_heap_vunmap, + .release = system_heap_dma_buf_release, +}; +static int system_heap_allocate (struct dma_heap *heap,unsigned long len,unsigned long fd_flags,unsigned long heap_flags) { -struct heap_helper_buffer *helper_buffer; +struct system_heap_buffer *buffer; +DEFINE_DMA_BUF_EXPORT_INFO (exp_info);struct dma_buf *dmabuf; -int ret = -ENOMEM ; +struct sg_table *table; +struct scatterlist *sg; +pgoff_t pagecount;pgoff_t pg; +int i, ret = -ENOMEM ; - helper_buffer =kzalloc (sizeof (*helper_buffer),GFP_KERNEL ); -if (!helper_buffer) + buffer =kzalloc (sizeof (*buffer),GFP_KERNEL ); +if (!buffer)return -ENOMEM ; -init_heap_helper_buffer (helper_buffer, system_heap_free); - helper_buffer->heap = heap; - helper_buffer->size = len; - - helper_buffer->pagecount = len /PAGE_SIZE ; - helper_buffer->pages =kmalloc_array (helper_buffer->pagecount, -sizeof (*helper_buffer->pages), -GFP_KERNEL ); -if (!helper_buffer->pages) { - ret = -ENOMEM ; - goto err0; - } +INIT_LIST_HEAD (&buffer->attachments); +mutex_init (&buffer->lock); + buffer->heap = heap; + buffer->len = len; -for (pg =0 ; pg < helper_buffer->pagecount; pg++) { + table = &buffer->sg_table; + pagecount = len /PAGE_SIZE ; +if (sg_alloc_table (table, pagecount,GFP_KERNEL )) + goto free_buffer; + + sg = table->sgl; +for (pg =0 ; pg < pagecount; pg++) { +struct page *page;/* * Avoid trying to allocate memory if the process -* has been killed by by SIGKILL +* has been killed by SIGKILL */ if (fatal_signal_pending (current)) - goto err1; - - helper_buffer->pages[pg] =alloc_page (GFP_KERNEL | __GFP_ZERO); -if (!helper_buffer->pages[pg]) - goto err1; + goto free_pages; + page =alloc_page (GFP_KERNEL | __GFP_ZERO); +if (!page) + goto free_pages; +sg_set_page (sg, page,page_size (page),0 ); + sg =sg_next (sg); }/* create the dmabuf */ - dmabuf =heap_helper_export_dmabuf (helper_buffer, fd_flags); + exp_info.ops = &system_heap_buf_ops; + exp_info.size = buffer->len; + exp_info.flags = fd_flags; + exp_info.priv = buffer; + dmabuf =dma_buf_export (&exp_info);if (IS_ERR (dmabuf)) { ret =PTR_ERR (dmabuf); - goto err1; + goto free_pages; } - helper_buffer->dmabuf = dmabuf; - ret =dma_buf_fd (dmabuf, fd_flags);if (ret <0 ) {dma_buf_put (dmabuf);
@@ -90,12 +343,12 @@ static int system_heap_allocate(struct dma_heap *heap,return ret; -err1: -while (pg >0 ) -__free_page (helper_buffer->pages[--pg]); -kfree (helper_buffer->pages); -err0: -kfree (helper_buffer); +free_pages: +for_each_sgtable_sg (table, sg, i) +__free_page (sg_page (sg)); +sg_free_table (table); +free_buffer: +kfree (buffer);return ret; }
@@ -107,7 +360,6 @@ static const struct dma_heap_ops system_heap_ops = {static int system_heap_create (void ) {struct dma_heap_export_info exp_info; -int ret =0 ; exp_info.name ="system" ; exp_info.ops = &system_heap_ops;
@@ -115,9 +367,9 @@ static int system_heap_create(void) sys_heap =dma_heap_add (&exp_info);if (IS_ERR (sys_heap)) - ret =PTR_ERR (sys_heap); +return PTR_ERR (sys_heap); -return ret; +return 0 ; }module_init (system_heap_create);MODULE_LICENSE ("GPL v2" );
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 2/7 dma-buf: heaps: Move heap-helper logic into the cma_heap implementation
In reply to John StultzSince the heap-helpers logic ended up not being as generic as hoped, move the heap-helpers dma_buf_ops implementations into the cma_heap directly.
This will allow us to remove the heap_helpers code in a following patch.
---
v2:
* Fix unused return value and locking issue Reported-by:
kernel test robot <[email protected]>
Julia Lawall <[email protected]>
* Make cma_heap_buf_ops static suggested by kernel test robot <[email protected]>
* Fix uninitialized return in cma Reported-by:
kernel test robot <[email protected]>
* Minor cleanups
v3:
* Use the new sgtable mapping functions, as Suggested-by:
Daniel Mentz <[email protected]>
--- drivers/dma-buf/heaps/cma_heap.c | 317 ++++++++++++++++++++++++++----- 1 file changed, 267 insertions(+), 50 deletions(-)
drivers/dma-buf/heaps/cma_heap.c +267 -50
@@ -2,76 +2,291 @@/* * DMABUF CMA heap exporter * -* Copyright (C) 2012, 2019 Linaro Ltd. +* Copyright (C) 2012, 2019, 2020 Linaro Ltd. * Author: <[email protected]> for ST-Ericsson. +* +* Also utilizing parts of Andrew Davis' SRAM heap: +* Copyright (C) 2019 Texas Instruments Incorporated - http://www.ti.com/ +* Andrew F. Davis <[email protected]> */ -#include <linux/cma.h> -#include <linux/device.h> #include <linux/dma-buf.h> -#include <linux/dma-heap.h> #include <linux/dma-contiguous.h> +#include <linux/dma-heap.h> +#include <linux/dma-mapping.h> #include <linux/err.h> -#include <linux/errno.h> #include <linux/highmem.h> +#include <linux/io.h> +#include <linux/mm.h> #include <linux/module.h> -#include <linux/slab.h> #include <linux/scatterlist.h> -#include <linux/sched/signal.h> +#include <linux/slab.h> -#include "heap-helpers.h" struct cma_heap {struct dma_heap *heap;struct cma *cma; }; -static void cma_heap_free (struct heap_helper_buffer *buffer) +struct cma_heap_buffer { +struct cma_heap *heap; +struct list_head attachments; +struct mutex lock; +unsigned long len; +struct page *cma_pages; +struct page **pages; +pgoff_t pagecount; +int vmap_cnt; +void *vaddr; +}; + +struct dma_heap_attachment { +struct device *dev; +struct sg_table table; +struct list_head list; +}; + +static int cma_heap_attach (struct dma_buf *dmabuf, +struct dma_buf_attachment *attachment) { -struct cma_heap *cma_heap =dma_heap_get_drvdata (buffer->heap); -unsigned long nr_pages = buffer->pagecount; -struct page *cma_pages = buffer->priv_virt; +struct cma_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; +int ret; -/* free page list */ -kfree (buffer->pages); -/* release memory */ -cma_release (cma_heap->cma, cma_pages, nr_pages); + a =kzalloc (sizeof (*a),GFP_KERNEL ); +if (!a) +return -ENOMEM ; + + ret =sg_alloc_table_from_pages (&a->table, buffer->pages, + buffer->pagecount,0 , + buffer->pagecount <<PAGE_SHIFT , +GFP_KERNEL ); +if (ret) { +kfree (a); +return ret; + } + + a->dev = attachment->dev; +INIT_LIST_HEAD (&a->list); + + attachment->priv = a; + +mutex_lock (&buffer->lock); +list_add (&a->list, &buffer->attachments); +mutex_unlock (&buffer->lock); + +return 0 ; +} + +static void cma_heap_detatch (struct dma_buf *dmabuf, +struct dma_buf_attachment *attachment) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a = attachment->priv; + +mutex_lock (&buffer->lock); +list_del (&a->list); +mutex_unlock (&buffer->lock); + +sg_free_table (&a->table); +kfree (a); +} + +static struct sg_table *cma_heap_map_dma_buf (struct dma_buf_attachment *attachment, +enum dma_data_direction direction) +{ +struct dma_heap_attachment *a = attachment->priv; +struct sg_table *table = &a->table; +int ret; + + ret =dma_map_sgtable (attachment->dev, table, direction,0 ); +if (ret) +return ERR_PTR (-ENOMEM ); +return table; +} + +static void cma_heap_unmap_dma_buf (struct dma_buf_attachment *attachment, +struct sg_table *table, +enum dma_data_direction direction) +{ +dma_unmap_sgtable (attachment->dev, table, direction,0 ); +} + +static int cma_heap_dma_buf_begin_cpu_access (struct dma_buf *dmabuf, +enum dma_data_direction direction) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; + +if (buffer->vmap_cnt) +invalidate_kernel_vmap_range (buffer->vaddr, buffer->len); + +mutex_lock (&buffer->lock); +list_for_each_entry (a, &buffer->attachments, list) { +dma_sync_sgtable_for_cpu (a->dev, &a->table, direction); + } +mutex_unlock (&buffer->lock); + +return 0 ; +} + +static int cma_heap_dma_buf_end_cpu_access (struct dma_buf *dmabuf, +enum dma_data_direction direction) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; +struct dma_heap_attachment *a; + +if (buffer->vmap_cnt) +flush_kernel_vmap_range (buffer->vaddr, buffer->len); + +mutex_lock (&buffer->lock); +list_for_each_entry (a, &buffer->attachments, list) { +dma_sync_sgtable_for_device (a->dev, &a->table, direction); + } +mutex_unlock (&buffer->lock); + +return 0 ; +} + +static vm_fault_t cma_heap_vm_fault (struct vm_fault *vmf) +{ +struct vm_area_struct *vma = vmf->vma; +struct cma_heap_buffer *buffer = vma->vm_private_data; + +if (vmf->pgoff > buffer->pagecount) +return VM_FAULT_SIGBUS ; + + vmf->page = buffer->pages[vmf->pgoff]; +get_page (vmf->page); + +return 0 ; +} + +static const struct vm_operations_struct dma_heap_vm_ops = { + .fault = cma_heap_vm_fault, +}; + +static int cma_heap_mmap (struct dma_buf *dmabuf,struct vm_area_struct *vma) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; + +if ((vma->vm_flags & (VM_SHARED |VM_MAYSHARE )) ==0 ) +return -EINVAL ; + + vma->vm_ops = &dma_heap_vm_ops; + vma->vm_private_data = buffer; + +return 0 ; +} + +static void *cma_heap_do_vmap (struct cma_heap_buffer *buffer) +{ +void *vaddr; + + vaddr =vmap (buffer->pages, buffer->pagecount,VM_MAP ,PAGE_KERNEL ); +if (!vaddr) +return ERR_PTR (-ENOMEM ); + +return vaddr; +} + +static void *cma_heap_vmap (struct dma_buf *dmabuf) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; +void *vaddr; + +mutex_lock (&buffer->lock); +if (buffer->vmap_cnt) { + buffer->vmap_cnt++; + vaddr = buffer->vaddr; + goto out; + } + + vaddr =cma_heap_do_vmap (buffer); +if (IS_ERR (vaddr)) + goto out; + + buffer->vaddr = vaddr; + buffer->vmap_cnt++; +out: +mutex_unlock (&buffer->lock); + +return vaddr; +} + +static void cma_heap_vunmap (struct dma_buf *dmabuf,void *vaddr) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; + +mutex_lock (&buffer->lock); +if (!--buffer->vmap_cnt) { +vunmap (buffer->vaddr); + buffer->vaddr =NULL ; + } +mutex_unlock (&buffer->lock); +} + +static void cma_heap_dma_buf_release (struct dma_buf *dmabuf) +{ +struct cma_heap_buffer *buffer = dmabuf->priv; +struct cma_heap *cma_heap = buffer->heap; + +if (buffer->vmap_cnt >0 ) { +WARN (1 ,"%s: buffer still mapped in the kernel\n" , __func__); +vunmap (buffer->vaddr); + } + +cma_release (cma_heap->cma, buffer->cma_pages, buffer->pagecount);kfree (buffer); } -/* dmabuf heap CMA operations functions */ +static const struct dma_buf_ops cma_heap_buf_ops = { + .attach = cma_heap_attach, + .detach = cma_heap_detatch, + .map_dma_buf = cma_heap_map_dma_buf, + .unmap_dma_buf = cma_heap_unmap_dma_buf, + .begin_cpu_access = cma_heap_dma_buf_begin_cpu_access, + .end_cpu_access = cma_heap_dma_buf_end_cpu_access, + .mmap = cma_heap_mmap, + .vmap = cma_heap_vmap, + .vunmap = cma_heap_vunmap, + .release = cma_heap_dma_buf_release, +}; +static int cma_heap_allocate (struct dma_heap *heap, -unsigned long len, -unsigned long fd_flags, -unsigned long heap_flags) +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags) {struct cma_heap *cma_heap =dma_heap_get_drvdata (heap); -struct heap_helper_buffer *helper_buffer; -struct page *cma_pages; +struct cma_heap_buffer *buffer; +DEFINE_DMA_BUF_EXPORT_INFO (exp_info);size_t size =PAGE_ALIGN (len); -unsigned long nr_pages = size >>PAGE_SHIFT ; +pgoff_t pagecount = size >>PAGE_SHIFT ;unsigned long align =get_order (size); +struct page *cma_pages;struct dma_buf *dmabuf;int ret = -ENOMEM ;pgoff_t pg; -if (align >CONFIG_CMA_ALIGNMENT ) - align =CONFIG_CMA_ALIGNMENT ; - - helper_buffer =kzalloc (sizeof (*helper_buffer),GFP_KERNEL ); -if (!helper_buffer) + buffer =kzalloc (sizeof (*buffer),GFP_KERNEL ); +if (!buffer)return -ENOMEM ; -init_heap_helper_buffer (helper_buffer, cma_heap_free); - helper_buffer->heap = heap; - helper_buffer->size = len; +INIT_LIST_HEAD (&buffer->attachments); +mutex_init (&buffer->lock); + buffer->len = size; + +if (align >CONFIG_CMA_ALIGNMENT ) + align =CONFIG_CMA_ALIGNMENT ; - cma_pages =cma_alloc (cma_heap->cma, nr_pages, align, false); + cma_pages =cma_alloc (cma_heap->cma, pagecount, align, false);if (!cma_pages) - goto free_buf; + goto free_buffer; +/* Clear the cma pages */ if (PageHighMem (cma_pages)) { -unsigned long nr_clear_pages = nr_pages; +unsigned long nr_clear_pages = pagecount;struct page *page = cma_pages;while (nr_clear_pages >0 ) {
@@ -85,7 +300,6 @@ static int cma_heap_allocate(struct dma_heap *heap,*/ if (fatal_signal_pending (current)) goto free_cma; - page++; nr_clear_pages--; }
@@ -93,28 +307,30 @@ static int cma_heap_allocate(struct dma_heap *heap,memset (page_address (cma_pages),0 , size); } - helper_buffer->pagecount = nr_pages; - helper_buffer->pages =kmalloc_array (helper_buffer->pagecount, -sizeof (*helper_buffer->pages), -GFP_KERNEL ); -if (!helper_buffer->pages) { + buffer->pages =kmalloc_array (pagecount,sizeof (*buffer->pages),GFP_KERNEL ); +if (!buffer->pages) { ret = -ENOMEM ; goto free_cma; } -for (pg =0 ; pg < helper_buffer->pagecount; pg++) - helper_buffer->pages[pg] = &cma_pages[pg]; +for (pg =0 ; pg < pagecount; pg++) + buffer->pages[pg] = &cma_pages[pg]; + + buffer->cma_pages = cma_pages; + buffer->heap = cma_heap; + buffer->pagecount = pagecount;/* create the dmabuf */ - dmabuf =heap_helper_export_dmabuf (helper_buffer, fd_flags); + exp_info.ops = &cma_heap_buf_ops; + exp_info.size = buffer->len; + exp_info.flags = fd_flags; + exp_info.priv = buffer; + dmabuf =dma_buf_export (&exp_info);if (IS_ERR (dmabuf)) { ret =PTR_ERR (dmabuf); goto free_pages; } - helper_buffer->dmabuf = dmabuf; - helper_buffer->priv_virt = cma_pages; - ret =dma_buf_fd (dmabuf, fd_flags);if (ret <0 ) {dma_buf_put (dmabuf);
@@ -125,11 +341,12 @@ static int cma_heap_allocate(struct dma_heap *heap,return ret; free_pages: -kfree (helper_buffer->pages); +kfree (buffer->pages); free_cma: -cma_release (cma_heap->cma, cma_pages, nr_pages); -free_buf: -kfree (helper_buffer); +cma_release (cma_heap->cma, cma_pages, pagecount); +free_buffer: +kfree (buffer); +return ret; }
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 3/7 dma-buf: heaps: Remove heap-helpers code
In reply to John StultzThe heap-helpers code was not as generic as initially hoped and it is now not being used, so remove it from the tree.
--- drivers/dma-buf/heaps/Makefile | 1 - drivers/dma-buf/heaps/heap-helpers.c | 271 --------------------------- drivers/dma-buf/heaps/heap-helpers.h | 53 ------ 3 files changed, 325 deletions(-) delete mode 100644 drivers/dma-buf/heaps/heap-helpers.c delete mode 100644 drivers/dma-buf/heaps/heap-helpers.h
drivers/dma-buf/heaps/Makefile +0 -1
@@ -1,4 +1,3 @@ # SPDX-License-Identifier: GPL-2.0 -obj-y += heap-helpers.o obj-$(CONFIG_DMABUF_HEAPS_SYSTEM) += system_heap.o obj-$(CONFIG_DMABUF_HEAPS_CMA) += cma_heap.o
drivers/dma-buf/heaps/heap-helpers.c deleted +0 -271
271 lines not on this page · raw message
drivers/dma-buf/heaps/heap-helpers.h deleted +0 -53
53 lines not on this page · raw message
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 4/7 dma-buf: heaps: Skip sync if not mapped
In reply to John StultzThis patch is basically a port of Ørjan Eide's similar patch for ION
https://lore.kernel.org/lkml/[email protected]/
Only sync the sg-list of dma-buf heap attachment when the attachment is actually mapped on the device.
dma-bufs may be synced at any time. It can be reached from user space via DMA_BUF_IOCTL_SYNC, so there are no guarantees from callers on when syncs may be attempted, and dma_buf_end_cpu_access() and dma_buf_begin_cpu_access() may not be paired.
Since the sg_list's dma_address isn't set up until the buffer is used on the device, and dma_map_sg() is called on it, the dma_address will be NULL if sync is attempted on the dma-buf before it's mapped on a device.
Before v5.0 (commit 55897af63091 ("dma-direct: merge swiotlb_dma_ops into the dma_direct code")) this was a problem as the dma-api (at least the swiotlb_dma_ops on arm64) would use the potentially invalid dma_address. How that failed depended on how the device handled physical address 0. If 0 was a valid address to physical ram, that page would get flushed a lot, while the actual pages in the buffer would not get synced correctly. While if 0 is an invalid physical address it may cause a fault and trigger a crash.
In v5.0 this was incidentally fixed by commit 55897af63091 ("dma-direct:
merge swiotlb_dma_ops into the dma_direct code"), as this moved the dma-api to use the page pointer in the sg_list, and (for Ion buffers at least) this will always be valid if the sg_list exists at all.
But, this issue is re-introduced in v5.3 with
commit 449fa54d6815 ("dma-direct: correct the physical addr in dma_direct_sync_sg_for_cpu/device") moves the dma-api back to the old behaviour and picks the dma_address that may be invalid.
dma-buf core doesn't ensure that the buffer is mapped on the device, and thus have a valid sg_list, before calling the exporter's begin_cpu_access.
Logic and commit message originally by: Ørjan Eide <[email protected]>
--- drivers/dma-buf/heaps/cma_heap.c | 10 ++++++++++ drivers/dma-buf/heaps/system_heap.c | 10 ++++++++++ 2 files changed, 20 insertions(+)
drivers/dma-buf/heaps/cma_heap.c +10 -0
@@ -44,6 +44,7 @@ struct dma_heap_attachment {struct device *dev;struct sg_table table;struct list_head list; +bool mapped; }; static int cma_heap_attach(struct dma_buf *dmabuf,
@@ -68,6 +69,7 @@ static int cma_heap_attach(struct dma_buf *dmabuf, a->dev = attachment->dev;INIT_LIST_HEAD (&a->list); + a->mapped = false; attachment->priv = a;
@@ -102,6 +104,7 @@ static struct sg_table *cma_heap_map_dma_buf(struct dma_buf_attachment *attachme ret =dma_map_sgtable (attachment->dev, table, direction,0 );if (ret)return ERR_PTR (-ENOMEM ); + a->mapped = true;return table; }
@@ -109,6 +112,9 @@ static void cma_heap_unmap_dma_buf(struct dma_buf_attachment *attachment,struct sg_table *table, enum dma_data_direction direction) { + struct dma_heap_attachment *a = attachment->priv; + + a->mapped = false;dma_unmap_sgtable (attachment->dev, table, direction,0 ); }
@@ -123,6 +129,8 @@ static int cma_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,mutex_lock (&buffer->lock);list_for_each_entry (a, &buffer->attachments, list) { +if (!a->mapped) +continue ;dma_sync_sgtable_for_cpu (a->dev, &a->table, direction); }mutex_unlock (&buffer->lock);
@@ -141,6 +149,8 @@ static int cma_heap_dma_buf_end_cpu_access(struct dma_buf *dmabuf,mutex_lock (&buffer->lock);list_for_each_entry (a, &buffer->attachments, list) { +if (!a->mapped) +continue ;dma_sync_sgtable_for_device (a->dev, &a->table, direction); }mutex_unlock (&buffer->lock);
drivers/dma-buf/heaps/system_heap.c +10 -0
@@ -37,6 +37,7 @@ struct dma_heap_attachment {struct device *dev;struct sg_table *table;struct list_head list; +bool mapped; };static struct sg_table *dup_sg_table (struct sg_table *table)
@@ -84,6 +85,7 @@ static int system_heap_attach(struct dma_buf *dmabuf, a->table = table; a->dev = attachment->dev;INIT_LIST_HEAD (&a->list); + a->mapped = false; attachment->priv = a;
@@ -120,6 +122,7 @@ static struct sg_table *system_heap_map_dma_buf(struct dma_buf_attachment *attacif (ret)return ERR_PTR (ret); + a->mapped = true;return table; }
@@ -127,6 +130,9 @@ static void system_heap_unmap_dma_buf(struct dma_buf_attachment *attachment,struct sg_table *table, enum dma_data_direction direction) { + struct dma_heap_attachment *a = attachment->priv; + + a->mapped = false;dma_unmap_sgtable (attachment->dev, table, direction,0 ); }
@@ -142,6 +148,8 @@ static int system_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,invalidate_kernel_vmap_range (buffer->vaddr, buffer->len);list_for_each_entry (a, &buffer->attachments, list) { +if (!a->mapped) +continue ;dma_sync_sgtable_for_cpu (a->dev, a->table, direction); }mutex_unlock (&buffer->lock);
@@ -161,6 +169,8 @@ static int system_heap_dma_buf_end_cpu_access(struct dma_buf *dmabuf,flush_kernel_vmap_range (buffer->vaddr, buffer->len);list_for_each_entry (a, &buffer->attachments, list) { +if (!a->mapped) +continue ;dma_sync_sgtable_for_device (a->dev, a->table, direction); }mutex_unlock (&buffer->lock);
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 5/7 dma-buf: system_heap: Allocate higher order pages if available
In reply to John StultzWhile the system heap can return non-contiguous pages, try to allocate larger order pages if possible.
This will allow slight performance gains and make implementing page pooling easier.
---
v3:
* Use page_size() rather then opencoding it
--- drivers/dma-buf/heaps/system_heap.c | 83 ++++++++++++++++++++++------- 1 file changed, 65 insertions(+), 18 deletions(-)
drivers/dma-buf/heaps/system_heap.c +65 -18
@@ -40,6 +40,14 @@ struct dma_heap_attachment {bool mapped; }; +#define HIGH_ORDER_GFP (((GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN \ + | __GFP_NORETRY) & ~__GFP_RECLAIM) \ + | __GFP_COMP) +#define LOW_ORDER_GFP (GFP_HIGHUSER | __GFP_ZERO | __GFP_COMP) +static gfp_t order_flags[] = {HIGH_ORDER_GFP ,LOW_ORDER_GFP ,LOW_ORDER_GFP }; +static const unsigned int orders[] = {8 ,4 ,0 }; +#define NUM_ORDERS ARRAY_SIZE(orders) +static struct sg_table *dup_sg_table (struct sg_table *table) {struct sg_table *new_table;
@@ -270,8 +278,11 @@ static void system_heap_dma_buf_release(struct dma_buf *dmabuf)int i; table = &buffer->sg_table; -for_each_sgtable_sg (table, sg, i) -__free_page (sg_page (sg)); +for_each_sg (table->sgl, sg, table->nents, i) { +struct page *page =sg_page (sg); + +__free_pages (page,compound_order (page)); + }sg_free_table (table);kfree (buffer); }
@@ -289,6 +300,26 @@ static const struct dma_buf_ops system_heap_buf_ops = { .release = system_heap_dma_buf_release, }; +static struct page *alloc_largest_available (unsigned long size, +unsigned int max_order) +{ +struct page *page; +int i; + +for (i =0 ; i <NUM_ORDERS ; i++) { +if (size < (PAGE_SIZE << orders[i])) +continue ; +if (max_order < orders[i]) +continue ; + + page =alloc_pages (order_flags[i], orders[i]); +if (!page) +continue ; +return page; + } +return NULL ; +} + static int system_heap_allocate(struct dma_heap *heap, unsigned long len, unsigned long fd_flags,
@@ -296,11 +327,13 @@ static int system_heap_allocate(struct dma_heap *heap, {struct system_heap_buffer *buffer;DEFINE_DMA_BUF_EXPORT_INFO (exp_info); +unsigned long size_remaining = len; +unsigned int max_order = orders[0 ];struct dma_buf *dmabuf;struct sg_table *table;struct scatterlist *sg; -pgoff_t pagecount; -pgoff_t pg; +struct list_head pages; +struct page *page, *tmp_page;int i, ret = -ENOMEM ; buffer =kzalloc (sizeof (*buffer),GFP_KERNEL );
@@ -312,25 +345,35 @@ static int system_heap_allocate(struct dma_heap *heap, buffer->heap = heap; buffer->len = len; - table = &buffer->sg_table; - pagecount = len /PAGE_SIZE ; -if (sg_alloc_table (table, pagecount,GFP_KERNEL )) - goto free_buffer; - - sg = table->sgl; -for (pg =0 ; pg < pagecount; pg++) { -struct page *page; +INIT_LIST_HEAD (&pages); + i =0 ; +while (size_remaining >0 ) {/* * Avoid trying to allocate memory if the process * has been killed by SIGKILL */ if (fatal_signal_pending (current)) - goto free_pages; - page =alloc_page (GFP_KERNEL | __GFP_ZERO); + goto free_buffer; + + page =alloc_largest_available (size_remaining, max_order);if (!page) - goto free_pages; + goto free_buffer; + +list_add_tail (&page->lru, &pages); + size_remaining -=page_size (page); + max_order =compound_order (page); + i++; + } + + table = &buffer->sg_table; +if (sg_alloc_table (table, i,GFP_KERNEL )) + goto free_buffer; + + sg = table->sgl; +list_for_each_entry_safe (page, tmp_page, &pages, lru) {sg_set_page (sg, page,page_size (page),0 ); sg =sg_next (sg); +list_del (&page->lru); }/* create the dmabuf */
@@ -350,14 +393,18 @@ static int system_heap_allocate(struct dma_heap *heap,/* just return, as put will call release and that will free */ return ret; } -return ret; free_pages: -for_each_sgtable_sg (table, sg, i) -__free_page (sg_page (sg)); +for_each_sgtable_sg (table, sg, i) { +struct page *p =sg_page (sg); + +__free_pages (p,compound_order (p)); + }sg_free_table (table); free_buffer: +list_for_each_entry_safe (page, tmp_page, &pages, lru) +__free_pages (page,compound_order (page));kfree (buffer);return ret;
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 6/7 dma-buf: dma-heap: Keep track of the heap device struct
In reply to John StultzKeep track of the heap device struct.
This will be useful for special DMA allocations and actions.
--- drivers/dma-buf/dma-heap.c | 33 +++++++++++++++++++++++++-------- include/linux/dma-heap.h | 9 +++++++++ 2 files changed, 34 insertions(+), 8 deletions(-)
drivers/dma-buf/dma-heap.c +25 -8
@@ -30,6 +30,7 @@* @heap_devt heap device node * @list list head connecting to list of heaps * @heap_cdev heap char device +* @heap_dev heap device struct * * Represents a heap of memory from which buffers can be made. */
@@ -40,6 +41,7 @@ struct dma_heap {dev_t heap_devt;struct list_head list;struct cdev heap_cdev; +struct device *heap_dev; };static LIST_HEAD (heap_list);
@@ -190,10 +192,21 @@ void *dma_heap_get_drvdata(struct dma_heap *heap)return heap->priv; } +/** +* dma_heap_get_dev() - get device struct for the heap +* @heap: DMA-Heap to retrieve device struct from +* +* Returns: +* The device struct for the heap. +*/ +struct device *dma_heap_get_dev (struct dma_heap *heap) +{ +return heap->heap_dev; +} +struct dma_heap *dma_heap_add (const struct dma_heap_export_info *exp_info) {struct dma_heap *heap, *h, *err_ret; -struct device *dev_ret;unsigned int minor;int ret;
@@ -247,16 +260,20 @@ struct dma_heap *dma_heap_add(const struct dma_heap_export_info *exp_info) goto err1; } - dev_ret =device_create (dma_heap_class, -NULL , - heap->heap_devt, -NULL , - heap->name); -if (IS_ERR (dev_ret)) { + heap->heap_dev =device_create (dma_heap_class, +NULL , + heap->heap_devt, +NULL , + heap->name); +if (IS_ERR (heap->heap_dev)) {pr_err ("dma_heap: Unable to create device\n" ); - err_ret =ERR_CAST (dev_ret); + err_ret =ERR_CAST (heap->heap_dev); goto err2; } + +/* Make sure it doesn't disappear on us */ + heap->heap_dev =get_device (heap->heap_dev); +/* Add heap to the list */ mutex_lock (&heap_list_lock);list_add (&heap->list, &heap_list);
include/linux/dma-heap.h +9 -0
@@ -50,6 +50,15 @@ struct dma_heap_export_info {*/ void *dma_heap_get_drvdata (struct dma_heap *heap); +/** +* dma_heap_get_dev() - get device struct for the heap +* @heap: DMA-Heap to retrieve device struct from +* +* Returns: +* The device struct for the heap. +*/ +struct device *dma_heap_get_dev (struct dma_heap *heap); +/** * dma_heap_add - adds a heap to dmabuf heaps * @exp_info: information needed to register this heap
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
PATCH v3 7/7 dma-buf: system_heap: Add a system-uncached heap re-using the system heap
In reply to John StultzThis adds a heap that allocates non-contiguous buffers that are marked as writecombined, so they are not cached by the CPU.
This is useful, as most graphics buffers are usually not touched by the CPU or only written into once by the CPU. So when mapping the buffer over and over between devices, we can skip the CPU syncing, which saves a lot of cache management overhead, greatly improving performance.
For folk using ION, there was a ION_FLAG_CACHED flag, which signaled if the returned buffer should be CPU cacheable or not. With DMA-BUF heaps, we do not yet have such a flag, and by default the current heaps (system and cma) produce CPU cachable buffers. So for folks transitioning from ION to DMA-BUF Heaps, this fills in some of that missing functionality.
There has been a suggestion to make this functionality a flag (DMAHEAP_FLAG_UNCACHED?) on the system heap, similar to how ION used the ION_FLAG_CACHED. But I want to make sure an _UNCACHED flag would truely be a generic attribute across all heaps. So far that has been unclear, so having it as a separate heap seemes better for now. (But I'm open to discussion on this point!)
This is a rework of earlier efforts to add a uncached system heap, done utilizing the exisitng system heap, adding just a bit of logic to handle the uncached case.
Feedback would be very welcome!
Many thanks to Liam Mark for his help to get this working.
Pending opensource users of this code include:
* AOSP HiKey960 gralloc:
- https://android-review.googlesource.com/c/device/linaro/hikey/+/1399519
- Visibly improves performance over the system heap
* AOSP Codec2 (possibly, needs more review):
- https://android-review.googlesource.com/c/platform/frameworks/av/+/1360640/17/media/codec2/vndk/C2DmaBufAllocator.cpp#325
--- drivers/dma-buf/heaps/system_heap.c | 87 ++++++++++++++++++++++++++--- 1 file changed, 79 insertions(+), 8 deletions(-)
drivers/dma-buf/heaps/system_heap.c +79 -8
@@ -22,6 +22,7 @@#include <linux/vmalloc.h> struct dma_heap *sys_heap; +struct dma_heap *sys_uncached_heap;struct system_heap_buffer {struct dma_heap *heap;
@@ -31,6 +32,8 @@ struct system_heap_buffer {struct sg_table sg_table;int vmap_cnt;void *vaddr; + +bool uncached; };struct dma_heap_attachment {
@@ -38,6 +41,8 @@ struct dma_heap_attachment {struct sg_table *table;struct list_head list;bool mapped; + +bool uncached; };#define HIGH_ORDER_GFP (((GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN \
@@ -94,7 +99,7 @@ static int system_heap_attach(struct dma_buf *dmabuf, a->dev = attachment->dev;INIT_LIST_HEAD (&a->list); a->mapped = false; - + a->uncached = buffer->uncached; attachment->priv = a;mutex_lock (&buffer->lock);
@@ -124,9 +129,13 @@ static struct sg_table *system_heap_map_dma_buf(struct dma_buf_attachment *attac {struct dma_heap_attachment *a = attachment->priv;struct sg_table *table = a->table; +int attr =0 ;int ret; - ret =dma_map_sgtable (attachment->dev, table, direction,0 ); +if (a->uncached) + attr =DMA_ATTR_SKIP_CPU_SYNC ; + + ret =dma_map_sgtable (attachment->dev, table, direction, attr);if (ret)return ERR_PTR (ret);
@@ -139,9 +148,12 @@ static void system_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, enum dma_data_direction direction) {struct dma_heap_attachment *a = attachment->priv; +int attr =0 ; +if (a->uncached) + attr =DMA_ATTR_SKIP_CPU_SYNC ; a->mapped = false; -dma_unmap_sgtable (attachment->dev, table, direction,0 ); +dma_unmap_sgtable (attachment->dev, table, direction, attr); } static int system_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,
@@ -150,6 +162,9 @@ static int system_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,struct system_heap_buffer *buffer = dmabuf->priv;struct dma_heap_attachment *a; +if (buffer->uncached) +return 0 ; +mutex_lock (&buffer->lock);if (buffer->vmap_cnt)
@@ -171,6 +186,9 @@ static int system_heap_dma_buf_end_cpu_access(struct dma_buf *dmabuf,struct system_heap_buffer *buffer = dmabuf->priv;struct dma_heap_attachment *a; +if (buffer->uncached) +return 0 ; +mutex_lock (&buffer->lock);if (buffer->vmap_cnt)
@@ -194,6 +212,9 @@ static int system_heap_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)struct sg_page_iter piter;int ret; +if (buffer->uncached) + vma->vm_page_prot =pgprot_writecombine (vma->vm_page_prot); +for_each_sgtable_page (table, &piter, vma->vm_pgoff) {struct page *page =sg_page_iter_page (&piter);
@@ -215,8 +236,12 @@ static void *system_heap_do_vmap(struct system_heap_buffer *buffer)struct page **pages =vmalloc (sizeof (struct page *) * npages);struct page **tmp = pages;struct sg_page_iter piter; +pgprot_t pgprot =PAGE_KERNEL ;void *vaddr; +if (buffer->uncached) + pgprot =pgprot_writecombine (PAGE_KERNEL ); +if (!pages)return ERR_PTR (-ENOMEM );
@@ -225,7 +250,7 @@ static void *system_heap_do_vmap(struct system_heap_buffer *buffer) *tmp++ =sg_page_iter_page (&piter); } - vaddr =vmap (pages, npages,VM_MAP ,PAGE_KERNEL ); + vaddr =vmap (pages, npages,VM_MAP , pgprot);vfree (pages);if (!vaddr)
@@ -278,6 +303,10 @@ static void system_heap_dma_buf_release(struct dma_buf *dmabuf)int i; table = &buffer->sg_table; +/* Unmap the uncached buffers from the heap device (pairs with the map on allocate) */ +if (buffer->uncached) +dma_unmap_sgtable (dma_heap_get_dev (buffer->heap), table,DMA_BIDIRECTIONAL ,0 ); +for_each_sg (table->sgl, sg, table->nents, i) {struct page *page =sg_page (sg);
@@ -320,10 +349,11 @@ static struct page *alloc_largest_available(unsigned long size,return NULL ; } -static int system_heap_allocate (struct dma_heap *heap, -unsigned long len, -unsigned long fd_flags, -unsigned long heap_flags) +static int system_heap_do_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags, +bool uncached) {struct system_heap_buffer *buffer;DEFINE_DMA_BUF_EXPORT_INFO (exp_info);
@@ -344,6 +374,7 @@ static int system_heap_allocate(struct dma_heap *heap,mutex_init (&buffer->lock); buffer->heap = heap; buffer->len = len; + buffer->uncached = uncached;INIT_LIST_HEAD (&pages); i =0 ;
@@ -393,6 +424,16 @@ static int system_heap_allocate(struct dma_heap *heap,/* just return, as put will call release and that will free */ return ret; } + +/* +* For uncached buffers, we need to initially flush cpu cache, since +* the __GFP_ZERO on the allocation means the zeroing was done by the +* cpu and thus it is likely cached. Map (and implicitly flush) it out +* now so we don't get corruption later on. +*/ +if (buffer->uncached) +dma_map_sgtable (dma_heap_get_dev (heap), table,DMA_BIDIRECTIONAL ,0 ); +return ret; free_pages:
@@ -410,10 +451,30 @@ static int system_heap_allocate(struct dma_heap *heap,return ret; } +static int system_heap_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags) +{ +return system_heap_do_allocate (heap, len, fd_flags, heap_flags, false); +} +static const struct dma_heap_ops system_heap_ops = { .allocate = system_heap_allocate, }; +static int system_uncached_heap_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags) +{ +return system_heap_do_allocate (heap, len, fd_flags, heap_flags, true); +} + +static const struct dma_heap_ops system_uncached_heap_ops = { + .allocate = system_uncached_heap_allocate, +}; +static int system_heap_create (void ) {struct dma_heap_export_info exp_info;
@@ -426,6 +487,16 @@ static int system_heap_create(void)if (IS_ERR (sys_heap))return PTR_ERR (sys_heap); + exp_info.name ="system-uncached" ; + exp_info.ops = &system_uncached_heap_ops; + exp_info.priv =NULL ; + + sys_uncached_heap =dma_heap_add (&exp_info); +if (IS_ERR (sys_uncached_heap)) +return PTR_ERR (sys_heap); + +dma_coerce_mask_and_coherent (dma_heap_get_dev (sys_uncached_heap),DMA_BIT_MASK (64 )); +return 0 ; }module_init (system_heap_create);
Signature · 6 lines
2.17.1 _______________________________________________ dri-devel mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
How is this going to deal with VIVT caches?
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
Quoted 2 lines · Christoph Hellwig
How is this going to deal with VIVT caches?
Hrm. That's a good question. I'm not sure I totally have my head
around it but, I guess we could make sure to call invalidate_kernel_vmap_range() in begin_cpu_access() and flush_kernel_vmap_range() in end_cpu_access() rather then exiting out early as we do now?
Unless you have better guidance?
Worse case we could check CONFIG_CPU_CACHE_VIVT and not register the system-uncached heap.
thanks
-john
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
kernel test robot
build botRe: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
In reply to John StultzI love your patch! Perhaps something to improve:
Hi John,
I love your patch! Perhaps something to improve:
[auto build test WARNING on linux/master]
[also build test WARNING on tegra-drm/drm/tegra/for-next linus/master v5.9-rc8] [cannot apply to next-20201002]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/0day-ci/linux/commits/John-Stultz/dma-buf-Performance-improvements-for-system-heap-a-system-uncached-implementation/20201003-120520 base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git bcf876870b95592b52519ed4aafcf9d95999bc9c config: i386-randconfig-s032-20201005 (attached as .config) compiler: gcc-9 (Debian 9.3.0-15) 9.3.0 reproduce: # apt-get install sparse # sparse version: v0.6.2-201-g24bdaac6-dirty # https://github.com/0day-ci/linux/commit/553f4e0fafc5b3b4542ea0c7bc21a70fce504c7f git remote add linux-review https://github.com/0day-ci/linux git fetch --no-tags linux-review John-Stultz/dma-buf-Performance-improvements-for-system-heap-a-system-uncached-implementation/20201003-120520 git checkout 553f4e0fafc5b3b4542ea0c7bc21a70fce504c7f # save the attached .config to linux build tree make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=i386
If you fix the issue, kindly add following tag as appropriate
echo
echo "sparse warnings: (new ones prefixed by >>)" echo
>> drivers/dma-buf/heaps/system_heap.c:25:17: sparse: sparse: symbol 'sys_uncached_heap' was not declared. Should it be static?
Please review and possibly fold the followup patch.
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
kernel test robot
build bot[RFC PATCH] dma-buf: system_heap: sys_uncached_heap can be static
In reply to John StultzSigned-off-by: kernel test robot <[email protected]>
--- system_heap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/dma-buf/heaps/system_heap.c +1 -1
@@ -22,7 +22,7 @@#include <linux/vmalloc.h> struct dma_heap *sys_heap; -struct dma_heap *sys_uncached_heap; +static struct dma_heap *sys_uncached_heap;struct system_heap_buffer {struct dma_heap *heap;
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
In reply to John StultzHi John,
url: https://github.com/0day-ci/linux/commits/John-Stultz/dma-buf-Performance-improvements-for-system-heap-a-system-uncached-implementation/20201003-120520 base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git bcf876870b95592b52519ed4aafcf9d95999bc9c config: i386-randconfig-m021-20201002 (attached as .config) compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
If you fix the issue, kindly add following tag as appropriate
smatch warnings:drivers/dma-buf/heaps/system_heap.c:496 system_heap_create() warn: passing zero to 'PTR_ERR'
vim +/PTR_ERR +496 drivers/dma-buf/heaps/system_heap.c
efa04fefebbd724 John Stultz 2019-12-03 478 static int system_heap_create(void)
efa04fefebbd724 John Stultz 2019-12-03 479 {
efa04fefebbd724 John Stultz 2019-12-03 480 struct dma_heap_export_info exp_info;
efa04fefebbd724 John Stultz 2019-12-03 481
263e38f82cbb35b Andrew F. Davis 2019-12-16 482 exp_info.name = "system";
efa04fefebbd724 John Stultz 2019-12-03 483 exp_info.ops = &system_heap_ops;
efa04fefebbd724 John Stultz 2019-12-03 484 exp_info.priv = NULL;
efa04fefebbd724 John Stultz 2019-12-03 485
efa04fefebbd724 John Stultz 2019-12-03 486 sys_heap = dma_heap_add(&exp_info);
efa04fefebbd724 John Stultz 2019-12-03 487 if (IS_ERR(sys_heap))
a2e10cdd2e4d12a John Stultz 2020-10-03 488 return PTR_ERR(sys_heap);
efa04fefebbd724 John Stultz 2019-12-03 489
553f4e0fafc5b3b John Stultz 2020-10-03 490 exp_info.name = "system-uncached";
553f4e0fafc5b3b John Stultz 2020-10-03 491 exp_info.ops = &system_uncached_heap_ops;
553f4e0fafc5b3b John Stultz 2020-10-03 492 exp_info.priv = NULL;
553f4e0fafc5b3b John Stultz 2020-10-03 493
553f4e0fafc5b3b John Stultz 2020-10-03 494 sys_uncached_heap = dma_heap_add(&exp_info);
553f4e0fafc5b3b John Stultz 2020-10-03 495 if (IS_ERR(sys_uncached_heap))
553f4e0fafc5b3b John Stultz 2020-10-03 @496 return PTR_ERR(sys_heap);
^^^^^^^^^^^^^^^^^This should be return PTR_ERR(sys_uncached_heap);
---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
Quoted 37 lines · Dan Carpenter
Hi John,
url: https://github.com/0day-ci/linux/commits/John-Stultz/dma-buf-Performance-improvements-for-system-heap-a-system-uncached-implementation/20201003-120520 base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git bcf876870b95592b52519ed4aafcf9d95999bc9c config: i386-randconfig-m021-20201002 (attached as .config) compiler: gcc-9 (Debian 9.3.0-15) 9.3.0If you fix the issue, kindly add following tag as appropriate
Reported-by kernel test robot <[email protected]>Reported-by Dan Carpenter <[email protected]>smatch warnings:
drivers/dma-buf/heaps/system_heap.c:496system_heap_create()warn: passing zero to 'PTR_ERR'vim +/PTR_ERR +496
drivers/dma-buf/heaps/system_heap.cefa04fefebbd724 John Stultz 2019-12-03 478 static int system_heap_create(void) efa04fefebbd724 John Stultz 2019-12-03 479 { efa04fefebbd724 John Stultz 2019-12-03 480 struct dma_heap_export_info exp_info; efa04fefebbd724 John Stultz 2019-12-03 481 263e38f82cbb35b Andrew F. Davis 2019-12-16 482 exp_info.name = "system"; efa04fefebbd724 John Stultz 2019-12-03 483 exp_info.ops = &system_heap_ops; efa04fefebbd724 John Stultz 2019-12-03 484 exp_info.priv = NULL; efa04fefebbd724 John Stultz 2019-12-03 485 efa04fefebbd724 John Stultz 2019-12-03 486 sys_heap = dma_heap_add(&exp_info); efa04fefebbd724 John Stultz 2019-12-03 487 if (IS_ERR(sys_heap)) a2e10cdd2e4d12a John Stultz 2020-10-03 488 return PTR_ERR(sys_heap); efa04fefebbd724 John Stultz 2019-12-03 489 553f4e0fafc5b3b John Stultz 2020-10-03 490 exp_info.name = "system-uncached"; 553f4e0fafc5b3b John Stultz 2020-10-03 491 exp_info.ops = &system_uncached_heap_ops; 553f4e0fafc5b3b John Stultz 2020-10-03 492 exp_info.priv = NULL; 553f4e0fafc5b3b John Stultz 2020-10-03 493 553f4e0fafc5b3b John Stultz 2020-10-03 494 sys_uncached_heap = dma_heap_add(&exp_info); 553f4e0fafc5b3b John Stultz 2020-10-03 495 if (IS_ERR(sys_uncached_heap)) 553f4e0fafc5b3b John Stultz 2020-10-03 @496 return PTR_ERR(sys_heap); ^^^^^^^^^^^^^^^^^
Quoted 1 line
This should be return
PTR_ERR(sys_uncached_heap);
Oh nice! Very impressed that the tool caught that! Thanks so much for the report! I'll fix it up here shortly. -john
Brian Starkey
reviewerRe: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
In reply to John StultzQuoted 159 lines · John Stultz
This adds a heap that allocates non-contiguous buffers that are marked as writecombined, so they are not cached by the CPU.
This is useful, as most graphics buffers are usually not touched by the CPU or only written into once by the CPU. So when mapping the buffer over and over between devices, we can skip the CPU syncing, which saves a lot of cache management overhead, greatly improving performance.
For folk using ION, there was a
ION_FLAG_CACHEDflag, which signaled if the returned buffer should be CPU cacheable or not. With DMA-BUF heaps, we do not yet have such a flag, and by default the current heaps (system and cma) produce CPU cachable buffers. So for folks transitioning from ION to DMA-BUF Heaps, this fills in some of that missing functionality.There has been a suggestion to make this functionality a flag (
DMAHEAP_FLAG_UNCACHED?) on the system heap, similar to how ION used theION_FLAG_CACHED. But I want to make sure an_UNCACHEDflag would truely be a generic attribute across all heaps. So far that has been unclear, so having it as a separate heap seemes better for now. (But I'm open to discussion on this point!)This is a rework of earlier efforts to add a uncached system heap, done utilizing the exisitng system heap, adding just a bit of logic to handle the uncached case.
Feedback would be very welcome!
Many thanks to Liam Mark for his help to get this working.
Pending opensource users of this code include:
* AOSP HiKey960 gralloc:
- https://android-review.googlesource.com/c/device/linaro/hikey/+/1399519
- Visibly improves performance over the system heap
* AOSP Codec2 (possibly, needs more review):
- https://android-review.googlesource.com/c/platform/frameworks/av/+/1360640/17/media/codec2/vndk/C2DmaBufAllocator.cpp#325Cc Sumit Semwal <[email protected]>Cc Liam Mark <[email protected]>Cc Laura Abbott <[email protected]>Cc Brian Starkey <[email protected]>Cc Hridya Valsaraju <[email protected]>Cc Suren Baghdasaryan <[email protected]>Cc Sandeep Patil <[email protected]>Cc Daniel Mentz <[email protected]>Cc Chris Goldsworthy <[email protected]>Cc �rjan Eide <[email protected]>Cc Robin Murphy <[email protected]>Cc Ezequiel Garcia <[email protected]>Cc Simon Ser <[email protected]>Cc James Jones <[email protected]>Signed-off-by John Stultz <[email protected]>--- drivers/dma-buf/heaps/system_heap.c | 87 ++++++++++++++++++++++++++--- 1 file changed, 79 insertions(+), 8 deletions(-)drivers/dma-buf/heaps/system_heap.c +24
-3@@ -22,6 +22,7 @@#include <linux/vmalloc.h> struct dma_heap *sys_heap; +struct dma_heap *sys_uncached_heap;struct system_heap_buffer {struct dma_heap *heap;@@ -31,6 +32,8 @@ struct system_heap_buffer {struct sg_table sg_table;int vmap_cnt;void *vaddr; + +bool uncached; };struct dma_heap_attachment {@@ -38,6 +41,8 @@ struct dma_heap_attachment {struct sg_table *table;struct list_head list;bool mapped; + +bool uncached; };#define HIGH_ORDER_GFP (((GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN \@@ -94,7 +99,7 @@ static int system_heap_attach(struct dma_buf *dmabuf, a->dev = attachment->dev;INIT_LIST_HEAD (&a->list); a->mapped = false; - + a->uncached = buffer->uncached; attachment->priv = a;mutex_lock (&buffer->lock);@@ -124,9 +129,13 @@ static struct sg_table *system_heap_map_dma_buf(struct dma_buf_attachment *attac {struct dma_heap_attachment *a = attachment->priv;struct sg_table *table = a->table; +int attr =0 ;int ret; - ret =dma_map_sgtable (attachment->dev, table, direction,0 ); +if (a->uncached) + attr =DMA_ATTR_SKIP_CPU_SYNC ; + + ret =dma_map_sgtable (attachment->dev, table, direction, attr);if (ret)return ERR_PTR (ret);@@ -139,9 +148,12 @@ static void system_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, enum dma_data_direction direction) {struct dma_heap_attachment *a = attachment->priv; +int attr =0 ; +if (a->uncached) + attr =DMA_ATTR_SKIP_CPU_SYNC ; a->mapped = false; -dma_unmap_sgtable (attachment->dev, table, direction,0 ); +dma_unmap_sgtable (attachment->dev, table, direction, attr); } static int system_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,@@ -150,6 +162,9 @@ static int system_heap_dma_buf_begin_cpu_access(struct dma_buf *dmabuf,struct system_heap_buffer *buffer = dmabuf->priv;struct dma_heap_attachment *a; +if (buffer->uncached) +return 0 ; +mutex_lock (&buffer->lock);if (buffer->vmap_cnt)@@ -171,6 +186,9 @@ static int system_heap_dma_buf_end_cpu_access(struct dma_buf *dmabuf,struct system_heap_buffer *buffer = dmabuf->priv;struct dma_heap_attachment *a; +if (buffer->uncached) +return 0 ; +mutex_lock (&buffer->lock);if (buffer->vmap_cnt)@@ -194,6 +212,9 @@ static int system_heap_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)struct sg_page_iter piter;int ret; +if (buffer->uncached) + vma->vm_page_prot =pgprot_writecombine (vma->vm_page_prot); +for_each_sgtable_page (table, &piter, vma->vm_pgoff) {struct page *page =sg_page_iter_page (&piter);
drivers/dma-buf/heaps/system_heap.c static void *system_heap_do_vmap(struct system_heap_buffer *buffer) · line 243
Quoted 8 lines
drivers/dma-buf/heaps/system_heap.c +3
-0@@ -215,8 +236,12 @@ static void *system_heap_do_vmap(struct system_heap_buffer *buffer)struct page **pages =vmalloc (sizeof (struct page *) * npages);struct page **tmp = pages;struct sg_page_iter piter; +pgprot_t pgprot =PAGE_KERNEL ;void *vaddr; +if (buffer->uncached) + pgprot =pgprot_writecombine (PAGE_KERNEL );
I think this should go after the 'if (!pages)' check. Best to get the allocation failure check as close to the allocation as possible IMO.
Quoted 51 lines
drivers/dma-buf/heaps/system_heap.c +12
-5+if (!pages)return ERR_PTR (-ENOMEM );@@ -225,7 +250,7 @@ static void *system_heap_do_vmap(struct system_heap_buffer *buffer) *tmp++ =sg_page_iter_page (&piter); } - vaddr =vmap (pages, npages,VM_MAP ,PAGE_KERNEL ); + vaddr =vmap (pages, npages,VM_MAP , pgprot);vfree (pages);if (!vaddr)@@ -278,6 +303,10 @@ static void system_heap_dma_buf_release(struct dma_buf *dmabuf)int i; table = &buffer->sg_table; +/* Unmap the uncached buffers from the heap device (pairs with the map on allocate) */ +if (buffer->uncached) +dma_unmap_sgtable (dma_heap_get_dev (buffer->heap), table,DMA_BIDIRECTIONAL ,0 ); +for_each_sg (table->sgl, sg, table->nents, i) {struct page *page =sg_page (sg);@@ -320,10 +349,11 @@ static struct page *alloc_largest_available(unsigned long size,return NULL ; } -static int system_heap_allocate (struct dma_heap *heap, -unsigned long len, -unsigned long fd_flags, -unsigned long heap_flags) +static int system_heap_do_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags, +bool uncached) {struct system_heap_buffer *buffer;DEFINE_DMA_BUF_EXPORT_INFO (exp_info);@@ -344,6 +374,7 @@ static int system_heap_allocate(struct dma_heap *heap,mutex_init (&buffer->lock); buffer->heap = heap; buffer->len = len; + buffer->uncached = uncached;INIT_LIST_HEAD (&pages); i =0 ;@@ -393,6 +424,16 @@ static int system_heap_allocate(struct dma_heap *heap,/* just return, as put will call release and that will free */ return ret;
drivers/dma-buf/heaps/system_heap.c static int system_heap_allocate(struct dma_heap *heap, · line 435
Quoted 10 lines
drivers/dma-buf/heaps/system_heap.c +9
-0} + +/* +* For uncached buffers, we need to initially flush cpu cache, since +* the __GFP_ZERO on the allocation means the zeroing was done by the +* cpu and thus it is likely cached. Map (and implicitly flush) it out +* now so we don't get corruption later on. +*/ +if (buffer->uncached) +dma_map_sgtable (dma_heap_get_dev (heap), table,DMA_BIDIRECTIONAL ,0 );
Do we have to keep this mapping around for the entire lifetime of the buffer?
Also, this problem (and solution) keeps lingering around. It really feels like there should be a better way to solve "clean the linear mapping all the way to DRAM", but I don't know what that should be.
Quoted 37 lines
drivers/dma-buf/heaps/system_heap.c +21
-0+return ret; free_pages:@@ -410,10 +451,30 @@ static int system_heap_allocate(struct dma_heap *heap,return ret; } +static int system_heap_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags) +{ +return system_heap_do_allocate (heap, len, fd_flags, heap_flags, false); +} +static const struct dma_heap_ops system_heap_ops = { .allocate = system_heap_allocate, }; +static int system_uncached_heap_allocate (struct dma_heap *heap, +unsigned long len, +unsigned long fd_flags, +unsigned long heap_flags) +{ +return system_heap_do_allocate (heap, len, fd_flags, heap_flags, true); +} + +static const struct dma_heap_ops system_uncached_heap_ops = { + .allocate = system_uncached_heap_allocate, +}; +static int system_heap_create (void ) {struct dma_heap_export_info exp_info;@@ -426,6 +487,16 @@ static int system_heap_create(void)if (IS_ERR (sys_heap))
drivers/dma-buf/heaps/system_heap.c static int system_heap_create(void) · line 497
Quoted 10 lines
drivers/dma-buf/heaps/system_heap.c +8
-0return PTR_ERR (sys_heap); + exp_info.name ="system-uncached" ; + exp_info.ops = &system_uncached_heap_ops; + exp_info.priv =NULL ; + + sys_uncached_heap =dma_heap_add (&exp_info); +if (IS_ERR (sys_uncached_heap)) +return PTR_ERR (sys_heap); +
In principle, there's a race here between the heap getting registered to sysfs and the dma_mask getting updated.
I don't think it would cause a problem in practice, but with the API as it is, there's no way to avoid it - so I wonder if the dma_heap_get_dev() accessor isn't the right API design.
Cheers,
-Brian
Quoted 8 lines
drivers/dma-buf/heaps/system_heap.c +2
-0+dma_coerce_mask_and_coherent (dma_heap_get_dev (sys_uncached_heap),DMA_BIT_MASK (64 )); +return 0 ; }module_init (system_heap_create);Signature · 1 line
2.17.1
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
Quoted 8 lines · Brian Starkey
Quoted 7 lines · John Stultz
drivers/dma-buf/heaps/system_heap.c +1
-0@@ -215,8 +236,12 @@ static void *system_heap_do_vmap(struct system_heap_buffer *buffer)struct page **pages =vmalloc (sizeof (struct page *) * npages);struct page **tmp = pages;struct sg_page_iter piter; +pgprot_t pgprot =PAGE_KERNEL ;void *vaddr;
Quoted 5 lines
drivers/dma-buf/heaps/system_heap.c static void *system_heap_do_vmap(struct system_heap_buffer *buffer) · line 243
Quoted 2 lines
drivers/dma-buf/heaps/system_heap.c +2
-0+if (buffer->uncached) + pgprot =pgprot_writecombine (PAGE_KERNEL );I think this should go after the 'if (!pages)' check. Best to get the allocation failure check as close to the allocation as possible IMO.
Sounds good. Changed in my tree.
Quoted 8 lines
Quoted 8 lines
drivers/dma-buf/heaps/system_heap.c +4
-0@@ -393,6 +424,16 @@ static int system_heap_allocate(struct dma_heap *heap,/* just return, as put will call release and that will free */ return ret; } + +/* +* For uncached buffers, we need to initially flush cpu cache, since +* the __GFP_ZERO on the allocation means the zeroing was done by the
Quoted 8 lines
drivers/dma-buf/heaps/system_heap.c static int system_heap_allocate(struct dma_heap *heap, · line 435
Quoted 5 lines
drivers/dma-buf/heaps/system_heap.c +5
-0+* cpu and thus it is likely cached. Map (and implicitly flush) it out +* now so we don't get corruption later on. +*/ +if (buffer->uncached) +dma_map_sgtable (dma_heap_get_dev (heap), table,DMA_BIDIRECTIONAL ,0 );Do we have to keep this mapping around for the entire lifetime of the buffer?
Yea, I guess we can just map and unmap it right there. It will look a little absurd, but that sort of aligns with your next point.
Quoted 3 lines
Also, this problem (and solution) keeps lingering around. It really feels like there should be a better way to solve "clean the linear mapping all the way to DRAM", but I don't know what that should be.
Yea, something better here would be nice...
Quoted 16 lines
drivers/dma-buf/heaps/system_heap.c static int system_heap_create(void) · line 497
Quoted 12 lines
drivers/dma-buf/heaps/system_heap.c +8
-0@@ -426,6 +487,16 @@ static int system_heap_create(void)if (IS_ERR (sys_heap))return PTR_ERR (sys_heap); + exp_info.name ="system-uncached" ; + exp_info.ops = &system_uncached_heap_ops; + exp_info.priv =NULL ; + + sys_uncached_heap =dma_heap_add (&exp_info); +if (IS_ERR (sys_uncached_heap)) +return PTR_ERR (sys_heap); +In principle, there's a race here between the heap getting registered to sysfs and the
dma_maskgetting updated.
Quoted 3 lines
I don't think it would cause a problem in practice, but with the API as it is, there's no way to avoid it - so I wonder if the
dma_heap_get_dev()accessor isn't the right API design.
Hrm. I guess to address your concern we would need split dma_heap_add() into something like:
dma_heap_create()
dma_heap_add()
Which breaks the creation of the heap with the registering it to the subsystem, so some attributes can be tweaked inbetween?
I'll see about taking a stab at this, but I'll probably submit my updated series sooner with this un-addressed so I can get some further review.
thanks
-john
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
Quoted 22 lines · John Stultz
Quoted 17 lines · Brian Starkey
drivers/dma-buf/heaps/system_heap.c static int system_heap_create(void) · line 497
Quoted 12 lines · John Stultz
drivers/dma-buf/heaps/system_heap.c +8
-0@@ -426,6 +487,16 @@ static int system_heap_create(void)if (IS_ERR (sys_heap))return PTR_ERR (sys_heap); + exp_info.name ="system-uncached" ; + exp_info.ops = &system_uncached_heap_ops; + exp_info.priv =NULL ; + + sys_uncached_heap =dma_heap_add (&exp_info); +if (IS_ERR (sys_uncached_heap)) +return PTR_ERR (sys_heap); +In principle, there's a race here between the heap getting registered to sysfs and the
dma_maskgetting updated.Quoted 3 lines
I don't think it would cause a problem in practice, but with the API as it is, there's no way to avoid it - so I wonder if the
dma_heap_get_dev()accessor isn't the right API design.
Quoted 7 lines
Hrm. I guess to address your concern we would need split
dma_heap_add()into something like:
dma_heap_create()
dma_heap_add()Which breaks the creation of the heap with the registering it to the subsystem, so some attributes can be tweaked inbetween?
Looking at this some more, this approach isn't going to work. We create the device and then we call dma_coerce_mask_and_coherent() on it, but as soon as the device is created it seems possible for userland to directly access it. Again, though, as you mentioned this isn't terribly likely in practice.
The best thing I can think of for now is to have the uncached heap's allocate pointer initially point to a dummy function that returns EBUSY and then after we update the dma mask then we can set it to the real alloc. I'll go with that for now, but let me know if you have other suggestions.
thanks
-john
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: [PATCH v3 7/7] dma-buf: system_heap: Add a system-uncached heap re-using the system heap
In reply to John StultzQuoted 20 lines · John Stultz
Quoted 17 lines · Brian Starkey
drivers/dma-buf/heaps/system_heap.c static int system_heap_allocate(struct dma_heap *heap, · line 435
Quoted 13 lines · John Stultz
drivers/dma-buf/heaps/system_heap.c +9
-0@@ -393,6 +424,16 @@ static int system_heap_allocate(struct dma_heap *heap,/* just return, as put will call release and that will free */ return ret; } + +/* +* For uncached buffers, we need to initially flush cpu cache, since +* the __GFP_ZERO on the allocation means the zeroing was done by the +* cpu and thus it is likely cached. Map (and implicitly flush) it out +* now so we don't get corruption later on. +*/ +if (buffer->uncached) +dma_map_sgtable (dma_heap_get_dev (heap), table,DMA_BIDIRECTIONAL ,0 );Do we have to keep this mapping around for the entire lifetime of the buffer?
Quoted 8 lines
Yea, I guess we can just map and unmap it right there. It will look a little absurd, but that sort of aligns with your next point.
Quoted 3 lines
Also, this problem (and solution) keeps lingering around. It really feels like there should be a better way to solve "clean the linear mapping all the way to DRAM", but I don't know what that should be.
Yea, something better here would be nice...
In ION, we had a little helper function named ion_buffer_prep_noncached that called arch_dma_prep_coherent() on all sg entries like so
for_each_sg(table->sgl, sg, table->orig_nents, i)
arch_dma_prep_coherent(sg_page(sg), sg->length);
Would that help?
Brian Starkey
reviewerHi John,
Quoted 1 line · John Stultz
Hey All,
...
Quoted 6 lines
I did add to this series a reworked version of my uncached system heap implementation I was submitting a few weeks back. Since it duplicated a lot of the now reworked system heap code, I realized it would be much simpler to add the functionality to the
system_heapimplementaiton itself.
That looks like a neat approach to me. Referencing your previous thread, I like the separate heap (as you have done), rather than a generic "cached"/"noncached" flag on all heaps.
Quoted 7 lines
While not improving the core allocation performance, the uncached heap allocations do result in *much* improved performance on HiKey960 as it avoids a lot of flushing and invalidating buffers that the cpu doesn't touch often.
Feedback on these would be great!
Minor nit: s/detatch/detach/ on both heaps, but other than that you can add my r-b to patches 1-5.
As you've said, it does feel like there's some room for de-duplication, but that will be easier to work out once the implementations settle.
I've a couple of comments for the uncached heap, but I'm not confident I understand the implications of having the non-cached alias enough to say if it looks OK or not.
Cheers!
-Brian
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Quoted 15 lines · Brian Starkey
Hi John,
Quoted 1 line · John Stultz
Hey All,
...
Quoted 6 lines
I did add to this series a reworked version of my uncached system heap implementation I was submitting a few weeks back. Since it duplicated a lot of the now reworked system heap code, I realized it would be much simpler to add the functionality to the
system_heapimplementaiton itself.
Quoted 4 lines
That looks like a neat approach to me. Referencing your previous thread, I like the separate heap (as you have done), rather than a generic "cached"/"noncached" flag on all heaps.
Sounds good! I really appreciate the feedback on this.
Quoted 9 lines
Quoted 6 lines
While not improving the core allocation performance, the uncached heap allocations do result in *much* improved performance on HiKey960 as it avoids a lot of flushing and invalidating buffers that the cpu doesn't touch often.
Feedback on these would be great!
Minor nit: s/detatch/detach/ on both heaps, but other than that you can add my r-b to patches 1-5.
Doh! Thanks for the spelling catch! Thanks again!
Quoted 7 lines
As you've said, it does feel like there's some room for de-duplication, but that will be easier to work out once the implementations settle.
I've a couple of comments for the uncached heap, but I'm not confident I understand the implications of having the non-cached alias enough to say if it looks OK or not.
Thanks so much!
-john