On Mon, Aug 10, 2020 at 12:30:42PM +0200, Maarten Lankhorst wrote:
> This reverts commit 7dc8f1143778 ("drm/i915/gem: Drop relocation
> slowpath"). We need the slowpath relocation for taking ww-mutex
> inside the page fault handler, and we will take this mutex when
> pinning all objects.
>
> With this, we have a proper working slowpath again, which
> will allow us to do fault handling with WW locks held.
>
> [mlankhorst: Adjusted for reloc_gpu_flush() changes]
>
> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> Cc: Matthew Auld <matthew.auld@xxxxxxxxx>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@xxxxxxxxxxxxxxx>
So from looking through this it also functionally reverts ef398881d27d
("drm/i915/gem: Limit struct_mutex to eb_reserve"), and well a tiny part
of 003d8b9143a6 ("drm/i915/gem: Only call eb_lookup_vma once during
execbuf ioctl"): The part that adds a flush_workqueue(userptr) in
eb_reserve, which the former patch then fixes up by adding the
dev->struct_mutex lock dropping around it.
The other part is adding a reloc_gpu_flush, comment about that below.
> ---
> .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 263 +++++++++++++++++-
> 1 file changed, 253 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> index 6acbd08f82f0..12a130f92e72 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> @@ -707,7 +707,7 @@ static int eb_reserve(struct i915_execbuffer *eb)
> if (err)
> break;
> }
> - if (!(err == -ENOSPC || err == -EAGAIN))
> + if (err != -ENOSPC)
> break;
>
> /* Resort *all* the objects into priority order */
> @@ -738,13 +738,6 @@ static int eb_reserve(struct i915_execbuffer *eb)
> }
> list_splice_tail(&last, &eb->unbound);
>
> - if (err == -EAGAIN) {
> - mutex_unlock(&eb->i915->drm.struct_mutex);
> - flush_workqueue(eb->i915->mm.userptr_wq);
> - mutex_lock(&eb->i915->drm.struct_mutex);
> - continue;
> - }
> -
> switch (pass++) {
> case 0:
> break;
> @@ -1626,7 +1619,9 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
> * we would try to acquire the struct mutex again. Obviously
> * this is bad and so lockdep complains vehemently.
> */
> - copied = __copy_from_user(r, urelocs, count * sizeof(r[0]));
> + pagefault_disable();
> + copied = __copy_from_user_inatomic(r, urelocs, count * sizeof(r[0]));
> + pagefault_enable();
> if (unlikely(copied)) {
> remain = -EFAULT;
> goto out;
> @@ -1674,6 +1669,251 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
> return remain;
> }
>
> +static int
> +eb_relocate_vma_slow(struct i915_execbuffer *eb, struct eb_vma *ev)
> +{
> + const struct drm_i915_gem_exec_object2 *entry = ev->exec;
> + struct drm_i915_gem_relocation_entry *relocs =
> + u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
> + unsigned int i;
> + int err;
> +
> + for (i = 0; i < entry->relocation_count; i++) {
> + u64 offset = eb_relocate_entry(eb, ev, &relocs[i]);
> +
> + if ((s64)offset < 0) {
> + err = (int)offset;
> + goto err;
> + }
> + }
> + err = 0;
> +err:
> + reloc_cache_reset(&eb->reloc_cache);
> + return err;
> +}
> +
> +static int check_relocations(const struct drm_i915_gem_exec_object2 *entry)
> +{
> + const char __user *addr, *end;
> + unsigned long size;
> + char __maybe_unused c;
> +
> + size = entry->relocation_count;
> + if (size == 0)
> + return 0;
> +
> + if (size > N_RELOC(ULONG_MAX))
> + return -EINVAL;
> +
> + addr = u64_to_user_ptr(entry->relocs_ptr);
> + size *= sizeof(struct drm_i915_gem_relocation_entry);
> + if (!access_ok(addr, size))
> + return -EFAULT;
> +
> + end = addr + size;
> + for (; addr < end; addr += PAGE_SIZE) {
> + int err = __get_user(c, addr);
> + if (err)
> + return err;
> + }
> + return __get_user(c, end - 1);
__get_user and friends considered uncool, since access_ok is considered
uncool. But since this a revert I guess ok.
But might be good to go around and just throw out all the access_ok in
execbuf, and replace them by the preferred functions. But that's maybe for
a follow-up series, when we have time again.
> +}
> +
> +static int eb_copy_relocations(const struct i915_execbuffer *eb)
> +{
> + struct drm_i915_gem_relocation_entry *relocs;
> + const unsigned int count = eb->buffer_count;
> + unsigned int i;
> + int err;
> +
> + for (i = 0; i < count; i++) {
> + const unsigned int nreloc = eb->exec[i].relocation_count;
> + struct drm_i915_gem_relocation_entry __user *urelocs;
> + unsigned long size;
> + unsigned long copied;
> +
> + if (nreloc == 0)
> + continue;
> +
> + err = check_relocations(&eb->exec[i]);
> + if (err)
> + goto err;
> +
> + urelocs = u64_to_user_ptr(eb->exec[i].relocs_ptr);
> + size = nreloc * sizeof(*relocs);
> +
> + relocs = kvmalloc_array(size, 1, GFP_KERNEL);
> + if (!relocs) {
> + err = -ENOMEM;
> + goto err;
> + }
> +
> + /* copy_from_user is limited to < 4GiB */
> + copied = 0;
> + do {
> + unsigned int len =
> + min_t(u64, BIT_ULL(31), size - copied);
> +
> + if (__copy_from_user((char *)relocs + copied,
> + (char __user *)urelocs + copied,
> + len))
> + goto end;
> +
> + copied += len;
> + } while (copied < size);
> +
> + /*
> + * As we do not update the known relocation offsets after
> + * relocating (due to the complexities in lock handling),
> + * we need to mark them as invalid now so that we force the
> + * relocation processing next time. Just in case the target
> + * object is evicted and then rebound into its old
> + * presumed_offset before the next execbuffer - if that
> + * happened we would make the mistake of assuming that the
> + * relocations were valid.
> + */
> + if (!user_access_begin(urelocs, size))
> + goto end;
> +
> + for (copied = 0; copied < nreloc; copied++)
> + unsafe_put_user(-1,
> + &urelocs[copied].presumed_offset,
> + end_user);
> + user_access_end();
> +
> + eb->exec[i].relocs_ptr = (uintptr_t)relocs;
> + }
> +
> + return 0;
> +
> +end_user:
> + user_access_end();
> +end:
> + kvfree(relocs);
> + err = -EFAULT;
> +err:
> + while (i--) {
> + relocs = u64_to_ptr(typeof(*relocs), eb->exec[i].relocs_ptr);
> + if (eb->exec[i].relocation_count)
> + kvfree(relocs);
> + }
> + return err;
> +}
> +
> +static int eb_prefault_relocations(const struct i915_execbuffer *eb)
> +{
> + const unsigned int count = eb->buffer_count;
> + unsigned int i;
> +
> + for (i = 0; i < count; i++) {
> + int err;
> +
> + err = check_relocations(&eb->exec[i]);
> + if (err)
> + return err;
> + }
> +
> + return 0;
> +}
> +
> +static noinline int eb_relocate_slow(struct i915_execbuffer *eb)
> +{
> + bool have_copy = false;
> + struct eb_vma *ev;
> + int err = 0;
> +
> +repeat:
> + if (signal_pending(current)) {
> + err = -ERESTARTSYS;
> + goto out;
> + }
> +
> + /*
> + * We take 3 passes through the slowpatch.
> + *
> + * 1 - we try to just prefault all the user relocation entries and
> + * then attempt to reuse the atomic pagefault disabled fast path again.
> + *
> + * 2 - we copy the user entries to a local buffer here outside of the
> + * local and allow ourselves to wait upon any rendering before
> + * relocations
> + *
> + * 3 - we already have a local copy of the relocation entries, but
> + * were interrupted (EAGAIN) whilst waiting for the objects, try again.
> + */
> + if (!err) {
> + err = eb_prefault_relocations(eb);
> + } else if (!have_copy) {
> + err = eb_copy_relocations(eb);
> + have_copy = err == 0;
> + } else {
> + cond_resched();
> + err = 0;
> + }
> +
> + flush_workqueue(eb->i915->mm.userptr_wq);
> +
> + if (err)
> + goto out;
> +
> + err = mutex_lock_interruptible(&eb->i915->drm.struct_mutex);
> + if (err)
> + goto out;
> +
> + list_for_each_entry(ev, &eb->relocs, reloc_link) {
> + if (!have_copy) {
> + pagefault_disable();
> + err = eb_relocate_vma(eb, ev);
> + pagefault_enable();
> + if (err)
> + break;
> + } else {
> + err = eb_relocate_vma_slow(eb, ev);
> + if (err)
> + break;
> + }
> + }
> +
> + reloc_gpu_flush(&eb->reloc_cache);
So maybe my baseline is all off, but I'm not understanding why this needs
to be added here. Both eb_relocate_vma_slow and eb_relocate_vma have a
reloc_gpu_flush at the end. So no idea why we need to flush once more here
...
Assuming this all works without this line here (which is also not
justified by the revert or the functional revert I spotted), and the
commit message fixed up:
Reviewed-by: Daniel Vetter <daniel.vetter@xxxxxxxx>
> + mutex_unlock(&eb->i915->drm.struct_mutex);
> + if (err && !have_copy)
> + goto repeat;
> +
> + if (err)
> + goto err;
> +
> + /*
> + * Leave the user relocations as are, this is the painfully slow path,
> + * and we want to avoid the complication of dropping the lock whilst
> + * having buffers reserved in the aperture and so causing spurious
> + * ENOSPC for random operations.
> + */
> +
> +err:
> + if (err == -EAGAIN)
> + goto repeat;
> +
> +out:
> + if (have_copy) {
> + const unsigned int count = eb->buffer_count;
> + unsigned int i;
> +
> + for (i = 0; i < count; i++) {
> + const struct drm_i915_gem_exec_object2 *entry =
> + &eb->exec[i];
> + struct drm_i915_gem_relocation_entry *relocs;
> +
> + if (!entry->relocation_count)
> + continue;
> +
> + relocs = u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
> + kvfree(relocs);
> + }
> + }
> +
> + return err;
> +}
> +
> static int eb_relocate(struct i915_execbuffer *eb)
> {
> int err;
> @@ -1695,8 +1935,11 @@ static int eb_relocate(struct i915_execbuffer *eb)
> list_for_each_entry(ev, &eb->relocs, reloc_link) {
> err = eb_relocate_vma(eb, ev);
> if (err)
> - return err;
> + break;
> }
> +
> + if (err)
> + return eb_relocate_slow(eb);
> }
>
> return 0;
> --
> 2.28.0
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
