diff mbox series

[i-g-t,1/2] tests/gem_exec_reloc: Don't filter out addresses when on PPGTT

Message ID 20191023090752.23566-1-janusz.krzysztofik@linux.intel.com (mailing list archive)
State New, archived
Headers show
Series [i-g-t,1/2] tests/gem_exec_reloc: Don't filter out addresses when on PPGTT | expand

Commit Message

Janusz Krzysztofik Oct. 23, 2019, 9:07 a.m. UTC
Commit a355b2d6eb42 ("igt/gem_exec_reloc: Filter out unavailable
addresses for !ppgtt") introduced filtering of addresses possibly
occupied by other users of shared GTT.  Unfortunately, that filtering
is unconditional, no matter if running on old shared GTT or not.  When
running on PPGTT, that may result in errors other than those intended
to be skipped over being silently ignored.

Skip over unavailable addresses only when not running on PPGTT.

Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
---
 tests/i915/gem_exec_reloc.c | 26 ++++++++++++++------------
 1 file changed, 14 insertions(+), 12 deletions(-)

Comments

Chris Wilson Oct. 23, 2019, 9:11 a.m. UTC | #1
Quoting Janusz Krzysztofik (2019-10-23 10:07:51)
> Commit a355b2d6eb42 ("igt/gem_exec_reloc: Filter out unavailable
> addresses for !ppgtt") introduced filtering of addresses possibly
> occupied by other users of shared GTT.  Unfortunately, that filtering
> is unconditional, no matter if running on old shared GTT or not.  When
> running on PPGTT, that may result in errors other than those intended
> to be skipped over being silently ignored.

You forgot the aliasing-ppgtt which has exactly the same issues with
reserved locations (since it is a mirror of the GGTT).
-Chris
Janusz Krzysztofik Oct. 23, 2019, 9:18 a.m. UTC | #2
Hi Chris,

On Wednesday, October 23, 2019 11:11:35 AM CEST Chris Wilson wrote:
> Quoting Janusz Krzysztofik (2019-10-23 10:07:51)
> > Commit a355b2d6eb42 ("igt/gem_exec_reloc: Filter out unavailable
> > addresses for !ppgtt") introduced filtering of addresses possibly
> > occupied by other users of shared GTT.  Unfortunately, that filtering
> > is unconditional, no matter if running on old shared GTT or not.  When
> > running on PPGTT, that may result in errors other than those intended
> > to be skipped over being silently ignored.
> 
> You forgot the aliasing-ppgtt which has exactly the same issues with
> reserved locations (since it is a mirror of the GGTT).

OK, let me check if the aliasing-ppgtt case can be addressed.

Thanks,
Janusz

> -Chris
>
diff mbox series

Patch

diff --git a/tests/i915/gem_exec_reloc.c b/tests/i915/gem_exec_reloc.c
index fdd9661d..8f88826e 100644
--- a/tests/i915/gem_exec_reloc.c
+++ b/tests/i915/gem_exec_reloc.c
@@ -539,12 +539,13 @@  static void basic_range(int fd, unsigned flags)
 		obj[n].offset = (1ull << (i + 12)) - 4096;
 		obj[n].offset = gen8_canonical_address(obj[n].offset);
 		obj[n].flags = EXEC_OBJECT_PINNED | EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
-		gem_write(fd, obj[n].handle, 0, &bbe, sizeof(bbe));
-		execbuf.buffers_ptr = to_user_pointer(&obj[n]);
-		execbuf.buffer_count = 1;
-		if (__gem_execbuf(fd, &execbuf))
-			continue;
-
+		if (!gem_uses_ppgtt(fd)) {
+			gem_write(fd, obj[n].handle, 0, &bbe, sizeof(bbe));
+			execbuf.buffers_ptr = to_user_pointer(&obj[n]);
+			execbuf.buffer_count = 1;
+			if (__gem_execbuf(fd, &execbuf))
+				continue;
+		}
 		igt_debug("obj[%d] handle=%d, address=%llx\n",
 			  n, obj[n].handle, (long long)obj[n].offset);
 
@@ -559,12 +560,13 @@  static void basic_range(int fd, unsigned flags)
 		obj[n].offset = 1ull << (i + 12);
 		obj[n].offset = gen8_canonical_address(obj[n].offset);
 		obj[n].flags = EXEC_OBJECT_PINNED | EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
-		gem_write(fd, obj[n].handle, 0, &bbe, sizeof(bbe));
-		execbuf.buffers_ptr = to_user_pointer(&obj[n]);
-		execbuf.buffer_count = 1;
-		if (__gem_execbuf(fd, &execbuf))
-			continue;
-
+		if (!gem_uses_ppgtt(fd)) {
+			gem_write(fd, obj[n].handle, 0, &bbe, sizeof(bbe));
+			execbuf.buffers_ptr = to_user_pointer(&obj[n]);
+			execbuf.buffer_count = 1;
+			if (__gem_execbuf(fd, &execbuf))
+				continue;
+		}
 		igt_debug("obj[%d] handle=%d, address=%llx\n",
 			  n, obj[n].handle, (long long)obj[n].offset);