Hi,
On Thu, 2016-02-11 at 11:40 +0000, Tvrtko Ursulin wrote:
>
> On 04/02/16 09:30, ankitprasad.r.sharma@xxxxxxxxx wrote:
> > From: Ankitprasad Sharma <ankitprasad.r.sharma@xxxxxxxxx>
> >
> > This patch adds support for extending the pread/pwrite functionality
> > for objects not backed by shmem. The access will be made through
> > gtt interface. This will cover objects backed by stolen memory as well
> > as other non-shmem backed objects.
> >
> > v2: Drop locks around slow_user_access, prefault the pages before
> > access (Chris)
> >
> > v3: Rebased to the latest drm-intel-nightly (Ankit)
> >
> > v4: Moved page base & offset calculations outside the copy loop,
> > corrected data types for size and offset variables, corrected if-else
> > braces format (Tvrtko/kerneldocs)
> >
> > v5: Enabled pread/pwrite for all non-shmem backed objects including
> > without tiling restrictions (Ankit)
> >
> > v6: Using pwrite_fast for non-shmem backed objects as well (Chris)
> >
> > v7: Updated commit message, Renamed i915_gem_gtt_read to i915_gem_gtt_copy,
> > added pwrite slow path for non-shmem backed objects (Chris/Tvrtko)
> >
> > v8: Updated v7 commit message, mutex unlock around pwrite slow path for
> > non-shmem backed objects (Tvrtko)
> >
> > v9: Corrected check during pread_ioctl, to avoid shmem_pread being
> > called for non-shmem backed objects (Tvrtko)
> >
> > v10: Moved the write_domain check to needs_clflush and tiling mode check
> > to pwrite_fast (Chris)
> >
> > v11: Use pwrite_fast fallback for all objects (shmem and non-shmem backed),
> > call fast_user_write regardless of pagefault in previous iteration
> >
> > v12: Use page-by-page copy for slow user access too (Chris)
> >
> > v13: Handled EFAULT, Avoid use of WARN_ON, put_fence only if whole obj
> > pinned (Chris)
> >
> > Testcase: igt/gem_stolen, igt/gem_pread, igt/gem_pwrite
> >
> > Signed-off-by: Ankitprasad Sharma <ankitprasad.r.sharma@xxxxxxxxx>
> > ---
> > drivers/gpu/drm/i915/i915_gem.c | 211 ++++++++++++++++++++++++++++++++++------
> > 1 file changed, 179 insertions(+), 32 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> > index ed8ae5d..40f2906 100644
> > --- a/drivers/gpu/drm/i915/i915_gem.c
> > +++ b/drivers/gpu/drm/i915/i915_gem.c
> > @@ -55,6 +55,9 @@ static bool cpu_cache_is_coherent(struct drm_device *dev,
> >
> > static bool cpu_write_needs_clflush(struct drm_i915_gem_object *obj)
> > {
> > + if (obj->base.write_domain == I915_GEM_DOMAIN_CPU)
> > + return false;
> > +
> > if (!cpu_cache_is_coherent(obj->base.dev, obj->cache_level))
> > return true;
> >
> > @@ -646,6 +649,141 @@ shmem_pread_slow(struct page *page, int shmem_page_offset, int page_length,
> > return ret ? - EFAULT : 0;
> > }
> >
> > +static inline uint64_t
> > +slow_user_access(struct io_mapping *mapping,
> > + uint64_t page_base, int page_offset,
> > + char __user *user_data,
> > + int length, bool pwrite)
> > +{
>
> Return type and length should be unsigned long to match
> __copy_to/from_user prototypes.
>
> > + void __iomem *ioaddr;
> > + void *vaddr;
> > + uint64_t unwritten;
> > +
> > + ioaddr = io_mapping_map_wc(mapping, page_base);
> > + /* We can use the cpu mem copy function because this is X86. */
> > + vaddr = (void __force *)ioaddr + page_offset;
> > + if (pwrite)
> > + unwritten = __copy_from_user(vaddr, user_data, length);
> > + else
> > + unwritten = __copy_to_user(user_data, vaddr, length);
> > +
> > + io_mapping_unmap(ioaddr);
> > + return unwritten;
> > +}
> > +
> > +static int
> > +i915_gem_gtt_pread(struct drm_device *dev,
> > + struct drm_i915_gem_object *obj, uint64_t size,
> > + uint64_t data_offset, uint64_t data_ptr)
> > +{
> > + struct drm_i915_private *dev_priv = dev->dev_private;
> > + struct drm_mm_node node;
> > + char __user *user_data;
> > + uint64_t remain;
> > + uint64_t offset;
> > + int ret = 0;
>
> No need to initialize.
>
> > +
> > + ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE);
> > + if (ret) {
> > + ret = insert_mappable_node(dev_priv, &node, PAGE_SIZE);
> > + if (ret)
> > + goto out;
> > +
> > + ret = i915_gem_object_get_pages(obj);
> > + if (ret) {
> > + remove_mappable_node(&node);
> > + goto out;
> > + }
> > +
> > + i915_gem_object_pin_pages(obj);
> > + } else {
> > + node.start = i915_gem_obj_ggtt_offset(obj);
> > + node.allocated = false;
> > + ret = i915_gem_object_put_fence(obj);
> > + if (ret)
> > + goto out_unpin;
> > + }
> > +
> > + ret = i915_gem_object_set_to_gtt_domain(obj, false);
> > + if (ret)
> > + goto out_unpin;
> > +
> > + user_data = to_user_ptr(data_ptr);
> > + remain = size;
> > + offset = i915_gem_obj_ggtt_offset(obj) + data_offset;
>
> This is wrong for the page-by-page (no pin) mode.
>
> > +
> > + mutex_unlock(&dev->struct_mutex);
> > + if (likely(!i915.prefault_disable)) {
> > + ret = fault_in_multipages_writeable(user_data, remain);
> > + if (ret) {
> > + mutex_lock(&dev->struct_mutex);
> > + goto out_unpin;
> > + }
> > + }
> > +
> > + while (remain > 0) {
> > + /* Operation in this page
> > + *
> > + * page_base = page offset within aperture
> > + * page_offset = offset within page
> > + * page_length = bytes to copy for this page
> > + */
> > + u32 page_base = node.start;
> > + unsigned page_offset = offset_in_page(offset);
> > + unsigned page_length = PAGE_SIZE - page_offset;
> > + page_length = remain < page_length ? remain : page_length;
> > + if (node.allocated) {
> > + wmb();
> > + dev_priv->gtt.base.insert_page(&dev_priv->gtt.base,
> > + i915_gem_object_get_dma_address(obj, offset >> PAGE_SHIFT),
> > + node.start,
> > + I915_CACHE_NONE, 0);
> > + wmb();
> > + } else {
> > + page_base += offset & PAGE_MASK;
> > + }
> > + /* This is a slow read/write as it tries to read from
> > + * and write to user memory which may result into page
> > + * faults, and so we cannot perform this under struct_mutex.
> > + */
> > + if (slow_user_access(dev_priv->gtt.mappable, page_base,
> > + page_offset, user_data,
> > + page_length, false)) {
> > + ret = -EFAULT;
> > + break;
> > + }
>
> Read does not want to try the fast access first, equivalent to pwrite ?
>
> > +
> > + remain -= page_length;
> > + user_data += page_length;
> > + offset += page_length;
> > + }
> > +
> > + mutex_lock(&dev->struct_mutex);
> > + if (ret == 0 && (obj->base.read_domains & I915_GEM_DOMAIN_GTT) == 0) {
> > + /* The user has modified the object whilst we tried
> > + * reading from it, and we now have no idea what domain
> > + * the pages should be in. As we have just been touching
> > + * them directly, flush everything back to the GTT
> > + * domain.
> > + */
> > + ret = i915_gem_object_set_to_gtt_domain(obj, false);
>
> How can this be OK? It suggests user could have moved the object to the
> CPU domain, calls pread and it gets silently moved to the GTT domain?
>
I agree to this comment, we can add a check if the object is not in
CPU_DOMAIN then only go on with gtt_pread else return the shmem_pread
error.
> > + }
> > +
> > +out_unpin:
> > + if (node.allocated) {
> > + wmb();
> > + dev_priv->gtt.base.clear_range(&dev_priv->gtt.base,
> > + node.start, node.size,
> > + true);
> > + i915_gem_object_unpin_pages(obj);
> > + remove_mappable_node(&node);
> > + } else {
> > + i915_gem_object_ggtt_unpin(obj);
> > + }
> > +out:
> > + return ret;
> > +}
> > +
> > static int
> > i915_gem_shmem_pread(struct drm_device *dev,
> > struct drm_i915_gem_object *obj,
> > @@ -661,6 +799,9 @@ i915_gem_shmem_pread(struct drm_device *dev,
> > int needs_clflush = 0;
> > struct sg_page_iter sg_iter;
> >
> > + if (!obj->base.filp)
> > + return -ENODEV;
> > +
> > user_data = to_user_ptr(args->data_ptr);
> > remain = args->size;
> >
> > @@ -769,18 +910,15 @@ i915_gem_pread_ioctl(struct drm_device *dev, void *data,
> > goto out;
> > }
> >
> > - /* prime objects have no backing filp to GEM pread/pwrite
> > - * pages from.
> > - */
> > - if (!obj->base.filp) {
> > - ret = -EINVAL;
> > - goto out;
> > - }
> > -
> > trace_i915_gem_object_pread(obj, args->offset, args->size);
> >
> > ret = i915_gem_shmem_pread(dev, obj, args, file);
> >
> > + /* pread for non shmem backed objects */
> > + if (ret == -EFAULT || ret == -ENODEV)
> > + ret = i915_gem_gtt_pread(dev, obj, args->size,
> > + args->offset, args->data_ptr);
> > +
> > out:
> > drm_gem_object_unreference(&obj->base);
> > unlock:
> > @@ -821,10 +959,15 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915,
> > struct drm_i915_gem_pwrite *args,
> > struct drm_file *file)
> > {
> > + struct drm_device *dev = obj->base.dev;
> > struct drm_mm_node node;
> > uint64_t remain, offset;
> > char __user *user_data;
> > int ret;
> > + bool hit_slow_path = false;
> > +
> > + if (obj->tiling_mode != I915_TILING_NONE)
> > + return -EFAULT;
> >
> > ret = i915_gem_obj_ggtt_pin(obj, 0, PIN_MAPPABLE | PIN_NONBLOCK);
> > if (ret) {
> > @@ -842,16 +985,15 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915,
> > } else {
> > node.start = i915_gem_obj_ggtt_offset(obj);
> > node.allocated = false;
> > + ret = i915_gem_object_put_fence(obj);
> > + if (ret)
> > + goto out_unpin;
> > }
> >
> > ret = i915_gem_object_set_to_gtt_domain(obj, true);
> > if (ret)
> > goto out_unpin;
> >
> > - ret = i915_gem_object_put_fence(obj);
> > - if (ret)
> > - goto out_unpin;
> > -
> > intel_fb_obj_invalidate(obj, ORIGIN_GTT);
> > obj->dirty = true;
> >
> > @@ -870,24 +1012,36 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915,
> > unsigned page_length = PAGE_SIZE - page_offset;
> > page_length = remain < page_length ? remain : page_length;
> > if (node.allocated) {
> > - wmb();
> > + wmb(); /* flush the write before we modify the GGTT */
> > i915->gtt.base.insert_page(&i915->gtt.base,
> > i915_gem_object_get_dma_address(obj, offset >> PAGE_SHIFT),
> > node.start,
> > I915_CACHE_NONE,
> > 0);
> > - wmb();
> > + wmb(); /* flush modifications to the GGTT (insert_page) */
> > } else {
> > page_base += offset & PAGE_MASK;
> > }
> > /* If we get a fault while copying data, then (presumably) our
> > * source page isn't available. Return the error and we'll
> > * retry in the slow path.
> > + * If the object is non-shmem backed, we retry again with the
> > + * path that handles page fault.
> > */
> > if (fast_user_write(i915->gtt.mappable, page_base,
> > page_offset, user_data, page_length)) {
> > - ret = -EFAULT;
> > - goto out_flush;
> > + hit_slow_path = true;
> > + mutex_unlock(&dev->struct_mutex);
> > + if (slow_user_access(i915->gtt.mappable,
> > + page_base,
> > + page_offset, user_data,
> > + page_length, true)) {
> > + ret = -EFAULT;
> > + mutex_lock(&dev->struct_mutex);
> > + goto out_flush;
> > + }
>
> I think the function now be called i915_gem_gtt_pwrite.
>
> Would it also need the same pre-fault as in i915_gem_gtt_pread ?
>
> > +
> > + mutex_lock(&dev->struct_mutex);
> > }
> >
> > remain -= page_length;
> > @@ -896,6 +1050,9 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_private *i915,
> > }
> >
> > out_flush:
> > + if (hit_slow_path)
> > + WARN_ON(i915_gem_object_set_to_gtt_domain(obj, true));
>
> I suppose this is for the same reason as in pread, maybe duplicate the
> comment here?
>
I do not think that this is same as pread, as we only call gtt_pwrite if
the object is not in CPU_DOMAIN.
> Regards,
>
> Tvrtko
>
Thanks,
Ankit
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
