[PATCH v7/v3 21/22] gpu: host1x: Use RESTART_W to skip timed out jobs on Tegra186+

From: Mikko Perttunen
Date: Mon Jun 27 2022 - 10:21:38 EST


From: Mikko Perttunen <mperttunen@xxxxxxxxxx>

When MLOCK enforcement is enabled, the 0-word write currently done
is rejected by the hardware outside of an MLOCK region. As such,
on these chips, which also have the newer, more convenient RESTART_W
opcode, use that instead to skip over the timed out job.

Signed-off-by: Mikko Perttunen <mperttunen@xxxxxxxxxx>
---
drivers/gpu/host1x/cdma.c | 19 +++++++++++++++++--
1 file changed, 17 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/host1x/cdma.c b/drivers/gpu/host1x/cdma.c
index 765e5aa64eb6..bb1f3c746be4 100644
--- a/drivers/gpu/host1x/cdma.c
+++ b/drivers/gpu/host1x/cdma.c
@@ -457,9 +457,24 @@ void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
* to offset 0xbad. This does nothing but
* has a easily detected signature in debug
* traces.
+ *
+ * On systems with MLOCK enforcement enabled,
+ * the above 0 word writes would fall foul of
+ * the enforcement. As such, in the first slot
+ * put a RESTART_W opcode to the beginning
+ * of the next job. We don't use this for older
+ * chips since those only support the RESTART
+ * opcode with inconvenient alignment requirements.
*/
- mapped[2*slot+0] = 0x1bad0000;
- mapped[2*slot+1] = 0x1bad0000;
+ if (i == 0 && host1x->info->has_wide_gather) {
+ unsigned int next_job = (job->first_get/8 + job->num_slots)
+ % HOST1X_PUSHBUFFER_SLOTS;
+ mapped[2*slot+0] = (0xd << 28) | (next_job * 2);
+ mapped[2*slot+1] = 0x0;
+ } else {
+ mapped[2*slot+0] = 0x1bad0000;
+ mapped[2*slot+1] = 0x1bad0000;
+ }
}

job->cancelled = true;
--
2.36.1