Re: Xorg indefinitely hangs in kernelspace

From: Gerd Hoffmann
Date: Mon Sep 09 2019 - 01:54:15 EST


Hi,

--verbose please. Do you see the same hang? Does the patch fix it?

> --- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
> +++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
> @@ -97,8 +97,9 @@ int ttm_eu_reserve_buffers(struct ww_acq
> struct list_head *dups, bool del_lru)
[ ... ]

> + if (locked)
> + ttm_eu_backoff_reservation_reverse(list, entry);

Hmm, I think the patch is wrong. As far I know it is the qxl drivers's
job to call ttm_eu_backoff_reservation(). Doing that automatically in
ttm will most likely break other ttm users.

So I guess the call is missing in the qxl driver somewhere, most likely
in some error handling code path given that this bug is a relatively
rare event.

There is only a single ttm_eu_reserve_buffers() call in qxl.
So how about this?

----------------------- cut here --------------------
diff --git a/drivers/gpu/drm/qxl/qxl_release.c b/drivers/gpu/drm/qxl/qxl_release.c
index 312216caeea2..2f9950fa0b8d 100644
--- a/drivers/gpu/drm/qxl/qxl_release.c
+++ b/drivers/gpu/drm/qxl/qxl_release.c
@@ -262,18 +262,20 @@ int qxl_release_reserve_list(struct qxl_release *release, bool no_intr)
ret = ttm_eu_reserve_buffers(&release->ticket, &release->bos,
!no_intr, NULL, true);
if (ret)
- return ret;
+ goto err_backoff;

list_for_each_entry(entry, &release->bos, tv.head) {
struct qxl_bo *bo = to_qxl_bo(entry->tv.bo);

ret = qxl_release_validate_bo(bo);
- if (ret) {
- ttm_eu_backoff_reservation(&release->ticket, &release->bos);
- return ret;
- }
+ if (ret)
+ goto err_backoff;
}
return 0;
+
+err_backoff:
+ ttm_eu_backoff_reservation(&release->ticket, &release->bos);
+ return ret;
}

void qxl_release_backoff_reserve_list(struct qxl_release *release)
----------------------- cut here --------------------

cheers,
Gerd