gpu: ion: optimize system heap for non fault buffers

If a buffer's user mappings are not going to be faulted
in it need not be allocated page wise.  We can optimize
this common case by allocating an sglist of larger chunks
rather than creating an entry for each page in the
allocation.

Change-Id: I47814990e55c7bdb7abeaa2af824744b0a97602d
Signed-off-by: Rebecca Schultz Zavin <rebecca@android.com>

diff --git a/drivers/gpu/ion/ion.c b/drivers/gpu/ion/ion.c
index 47d08db885570228dbc5c240f8017fcf1a26628e..f87b97858093b2c097c5ee25f94c9709d87bd661 100644 (file)
--- a/drivers/gpu/ion/ion.c
+++ b/drivers/gpu/ion/ion.c
@@ -100,6 +100,12 @@ struct ion_handle {
        unsigned int kmap_cnt;
 };
 
+bool ion_buffer_fault_user_mappings(struct ion_buffer *buffer)
+{
+        return ((buffer->flags & ION_FLAG_CACHED) &&
+                !(buffer->flags & ION_FLAG_CACHED_NEEDS_SYNC));
+}
+
 /* this function should only be called while dev->lock is held */
 static void ion_buffer_add(struct ion_device *dev,
                           struct ion_buffer *buffer)
@@ -145,6 +151,7 @@ static struct ion_buffer *ion_buffer_create(struct ion_heap *heap,
                return ERR_PTR(-ENOMEM);
 
        buffer->heap = heap;
+       buffer->flags = flags;
        kref_init(&buffer->ref);
 
        ret = heap->ops->allocate(heap, buffer, len, align, flags);
@@ -155,7 +162,6 @@ static struct ion_buffer *ion_buffer_create(struct ion_heap *heap,
 
        buffer->dev = dev;
        buffer->size = len;
-       buffer->flags = flags;
 
        table = heap->ops->map_dma(heap, buffer);
        if (IS_ERR_OR_NULL(table)) {
@@ -164,14 +170,13 @@ static struct ion_buffer *ion_buffer_create(struct ion_heap *heap,
                return ERR_PTR(PTR_ERR(table));
        }
        buffer->sg_table = table;
-       if (buffer->flags & ION_FLAG_CACHED &&
-           !(buffer->flags & ION_FLAG_CACHED_NEEDS_SYNC)) {
+       if (ion_buffer_fault_user_mappings(buffer)) {
                for_each_sg(buffer->sg_table->sgl, sg, buffer->sg_table->nents,
                            i) {
                        if (sg_dma_len(sg) == PAGE_SIZE)
                                continue;
-                       pr_err("%s: cached mappings must have pagewise "
-                              "sg_lists\n", __func__);
+                       pr_err("%s: cached mappings that will be faulted in "
+                              "must have pagewise sg_lists\n", __func__);
                        ret = -EINVAL;
                        goto err;
                }
@@ -764,8 +769,7 @@ static void ion_buffer_sync_for_device(struct ion_buffer *buffer,
        pr_debug("%s: syncing for device %s\n", __func__,
                 dev ? dev_name(dev) : "null");
 
-       if (!(buffer->flags & ION_FLAG_CACHED) ||
-           (buffer->flags & ION_FLAG_CACHED_NEEDS_SYNC))
+       if (!ion_buffer_fault_user_mappings(buffer))
                return;
 
        mutex_lock(&buffer->lock);
@@ -855,8 +859,7 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
                return -EINVAL;
        }
 
-       if (buffer->flags & ION_FLAG_CACHED &&
-           !(buffer->flags & ION_FLAG_CACHED_NEEDS_SYNC)) {
+       if (ion_buffer_fault_user_mappings(buffer)) {
                vma->vm_private_data = buffer;
                vma->vm_ops = &ion_vma_ops;
                ion_vm_open(vma);

diff --git a/drivers/gpu/ion/ion_priv.h b/drivers/gpu/ion/ion_priv.h
index b23972308852edbc57abb404daf49f244534b2b6..449f514261bf941e041ad20f7500367b3a20933d 100644 (file)
--- a/drivers/gpu/ion/ion_priv.h
+++ b/drivers/gpu/ion/ion_priv.h
@@ -130,6 +130,15 @@ struct ion_heap {
        const char *name;
 };
 
+/**
+ * ion_buffer_fault_user_mappings - fault in user mappings of this buffer
+ * @buffer:            buffer
+ *
+ * indicates whether userspace mappings of this buffer will be faulted
+ * in, this can affect how buffers are allocated from the heap.
+ */
+bool ion_buffer_fault_user_mappings(struct ion_buffer *buffer);
+
 /**
  * ion_device_create - allocates and returns an ion device
  * @custom_ioctl:      arch specific ioctl function if applicable

diff --git a/drivers/gpu/ion/ion_system_heap.c b/drivers/gpu/ion/ion_system_heap.c
index ca6de04f6c29ff202ec7510471ee051e22449105..310c4f66cfa2ba58da7c160be9a7735efb3eb1b9 100644 (file)
--- a/drivers/gpu/ion/ion_system_heap.c
+++ b/drivers/gpu/ion/ion_system_heap.c
@@ -31,7 +31,8 @@ struct page_info {
        struct list_head list;
 };
 
-static struct page_info *alloc_largest_available(unsigned long size)
+static struct page_info *alloc_largest_available(unsigned long size,
+                                                bool split_pages)
 {
        static unsigned int orders[] = {8, 4, 0};
        struct page *page;
@@ -45,7 +46,8 @@ static struct page_info *alloc_largest_available(unsigned long size)
                                   __GFP_NOWARN | __GFP_NORETRY, orders[i]);
                if (!page)
                        continue;
-               split_page(page, orders[i]);
+               if (split_pages)
+                       split_page(page, orders[i]);
                info = kmalloc(sizeof(struct page_info *), GFP_KERNEL);
                info->page = page;
                info->order = orders[i];
@@ -64,35 +66,49 @@ static int ion_system_heap_allocate(struct ion_heap *heap,
        int ret;
        struct list_head pages;
        struct page_info *info, *tmp_info;
-       int i;
+       int i = 0;
        long size_remaining = PAGE_ALIGN(size);
+       bool split_pages = ion_buffer_fault_user_mappings(buffer);
+
 
        INIT_LIST_HEAD(&pages);
        while (size_remaining > 0) {
-               info = alloc_largest_available(size_remaining);
+               info = alloc_largest_available(size_remaining, split_pages);
                if (!info)
                        goto err;
                list_add_tail(&info->list, &pages);
                size_remaining -= (1 << info->order) * PAGE_SIZE;
+               i++;
        }
 
        table = kmalloc(sizeof(struct sg_table), GFP_KERNEL);
        if (!table)
                goto err;
 
-       ret = sg_alloc_table(table, PAGE_ALIGN(size) / PAGE_SIZE, GFP_KERNEL);
+       if (split_pages)
+               ret = sg_alloc_table(table, PAGE_ALIGN(size) / PAGE_SIZE,
+                                    GFP_KERNEL);
+       else
+               ret = sg_alloc_table(table, i, GFP_KERNEL);
+
        if (ret)
                goto err1;
 
        sg = table->sgl;
        list_for_each_entry_safe(info, tmp_info, &pages, list) {
                struct page *page = info->page;
-               for (i = 0; i < (1 << info->order); i++) {
-                       sg_set_page(sg, page + i, PAGE_SIZE, 0);
+
+               if (split_pages) {
+                       for (i = 0; i < (1 << info->order); i++) {
+                               sg_set_page(sg, page + i, PAGE_SIZE, 0);
+                               sg = sg_next(sg);
+                       }
+               } else {
+                       sg_set_page(sg, page, (1 << info->order) * PAGE_SIZE,
+                                   0);
                        sg = sg_next(sg);
                }
                list_del(&info->list);
-               memset(info, 0, sizeof(struct page_info));
                kfree(info);
        }
 
@@ -105,8 +121,12 @@ err1:
        kfree(table);
 err:
        list_for_each_entry(info, &pages, list) {
-               for (i = 0; i < (1 << info->order); i++)
-                       __free_page(info->page + i);
+               if (split_pages)
+                       for (i = 0; i < (1 << info->order); i++)
+                               __free_page(info->page + i);
+               else
+                       __free_pages(info->page, info->order);
+
                kfree(info);
        }
        return -ENOMEM;