ib mad: fix an incorrect use of list_for_each_entry

[markj]

Summary

In tf_dequeue(), if we reach the end of the fifo list without finding a
non-cancelled element, tmp will be a pointer to the list head, so the
tmp->canceled check is bogus. Use a flag instead to avoid this.

Submitted by: Tao Liu <[e-mail]>

[• hselasky]

sys/ofed/drivers/infiniband/core/mad.c

307 ↗

(On Diff #7491)

I think the correct fix is:

if (tmp == NULL)
....

No need for an extra variable?

You can assume that at the end of list_for_each_entry() "tmp" is NULL.

--HPS

[markj]

sys/ofed/drivers/infiniband/core/mad.c

307 ↗

(On Diff #7491)

That's true with queue(9) but not with Linux's list macros. Otherwise we'd always crash here.

[• hselasky]

sys/ofed/drivers/infiniband/core/mad.c

307 ↗

(On Diff #7491)

I was mixing list_for_each() with list_for_each_safe() ... Would it be an idea to switch to list_for_each_safe() ? And add a list entry pointer instead of a bool. It will consume a register anyway ...

[• hselasky]

sys/ofed/drivers/infiniband/core/mad.c

307 ↗

(On Diff #7491)

Nevermind. You're right Linux lists are not queue(9) compatible.

[• hselasky]

sys/ofed/drivers/infiniband/core/mad.c

307 ↗

(On Diff #7491)

Don't forget to MFC when you commit it.