Skip to contentSkip to conversation

usb: xhci: use list_for_each_entry

PATCH Julia Lawall 2 participants 2 messages Dec 18, 2015 – Dec 21, 2015 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 201512190233.xPC4M6l4%[email protected]

#1 lore raw

Geliang,

Please check whether it is acceptable that last_unlinked_td point to the dummy entry at th beginning of the list, in the case where the list_for_each_entry loop runs out normally.

It seems that you have sent a bunch of these patches. Please recheck them all to see if they really follow the semantics of list_for_each_entry properly. If particular, you should not normally use the index variable after leaving the loop, unless it is guaranteed that the exit from the loop was via a break.

julia

Quoted 86 lines · kbuild test robot

CC: [email protected]
In-Reply-To: <0fbea26fdbcb76e24188fc0800d425f927f45b6f.1450455485.git.geliangtang@163.com>
TO: Geliang Tang <[email protected]>
CC: Mathias Nyman <[email protected]>, Greg Kroah-Hartman <[email protected]>
CC: Geliang Tang <[email protected]>, [email protected], [email protected]

Hi Geliang,

[auto build test WARNING on usb/usb-testing] [also build test WARNING on v4.4-rc5 next-20151218]

url:    https://github.com/0day-ci/linux/commits/Geliang-Tang/usb-host-fotg210-use-list_for_each_entry_safe/20151219-003955
base:   https://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb.git usb-testing

:::::: branch date: 2 hours ago
:::::: commit date: 2 hours ago

Quoted 1 line
Quoted 1 line

drivers/usb/host/xhci-ring.c:714:20-26: ERROR: invalid reference to the index variable of the iterator on line 672

git remote add linux-review https://github.com/0day-ci/linux
git remote update linux-review
git checkout 0af06cc5bab47032f7fc8e2e6a3df0fb29513b52 vim +714 drivers/usb/host/xhci-ring.c

ae6367471 Sarah Sharp      2009-04-29  666
ae6367471 Sarah Sharp      2009-04-29  667  	/* Fix up the ep ring first, so HW stops executing cancelled TDs.
ae6367471 Sarah Sharp      2009-04-29  668  	 * We have the xHCI lock, so nothing can modify this list until we drop
ae6367471 Sarah Sharp      2009-04-29  669  	 * it.  We're also in the event handler, so we can't get re-interrupted
ae6367471 Sarah Sharp      2009-04-29  670  	 * if another Stop Endpoint command completes
ae6367471 Sarah Sharp      2009-04-29  671  	 */
0af06cc5b Geliang Tang     2015-12-19 @672  	list_for_each_entry(cur_td, &ep->cancelled_td_list, cancelled_td_list) {
aa50b2906 Xenia Ragiadakou 2013-08-14  673  		xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
aa50b2906 Xenia Ragiadakou 2013-08-14  674  				"Removing canceled TD starting at 0x%llx (dma).",
79688acfb Sarah Sharp      2011-12-19  675  				(unsigned long long)xhci_trb_virt_to_dma(
79688acfb Sarah Sharp      2011-12-19  676  					cur_td->start_seg, cur_td->first_trb));
e9df17eb1 Sarah Sharp      2010-04-02  677  		ep_ring = xhci_urb_to_transfer_ring(xhci, cur_td->urb);
e9df17eb1 Sarah Sharp      2010-04-02  678  		if (!ep_ring) {
e9df17eb1 Sarah Sharp      2010-04-02  679  			/* This shouldn't happen unless a driver is mucking
e9df17eb1 Sarah Sharp      2010-04-02  680  			 * with the stream ID after submission.  This will
e9df17eb1 Sarah Sharp      2010-04-02  681  			 * leave the TD on the hardware ring, and the hardware
e9df17eb1 Sarah Sharp      2010-04-02  682  			 * will try to execute it, and may access a buffer
e9df17eb1 Sarah Sharp      2010-04-02  683  			 * that has already been freed.  In the best case, the
e9df17eb1 Sarah Sharp      2010-04-02  684  			 * hardware will execute it, and the event handler will
e9df17eb1 Sarah Sharp      2010-04-02  685  			 * ignore the completion event for that TD, since it was
e9df17eb1 Sarah Sharp      2010-04-02  686  			 * removed from the td_list for that endpoint.  In
e9df17eb1 Sarah Sharp      2010-04-02  687  			 * short, don't muck with the stream ID after
e9df17eb1 Sarah Sharp      2010-04-02  688  			 * submission.
e9df17eb1 Sarah Sharp      2010-04-02  689  			 */
e9df17eb1 Sarah Sharp      2010-04-02  690  			xhci_warn(xhci, "WARN Cancelled URB %p "
e9df17eb1 Sarah Sharp      2010-04-02  691  					"has invalid stream ID %u.\n",
e9df17eb1 Sarah Sharp      2010-04-02  692  					cur_td->urb,
e9df17eb1 Sarah Sharp      2010-04-02  693  					cur_td->urb->stream_id);
e9df17eb1 Sarah Sharp      2010-04-02  694  			goto remove_finished_td;
e9df17eb1 Sarah Sharp      2010-04-02  695  		}
ae6367471 Sarah Sharp      2009-04-29  696  		/*
ae6367471 Sarah Sharp      2009-04-29  697  		 * If we stopped on the TD we need to cancel, then we have to
ae6367471 Sarah Sharp      2009-04-29  698  		 * move the xHC endpoint ring dequeue pointer past this TD.
ae6367471 Sarah Sharp      2009-04-29  699  		 */
63a0d9abd Sarah Sharp      2009-09-04  700  		if (cur_td == ep->stopped_td)
e9df17eb1 Sarah Sharp      2010-04-02  701  			xhci_find_new_dequeue_state(xhci, slot_id, ep_index,
e9df17eb1 Sarah Sharp      2010-04-02  702  					cur_td->urb->stream_id,
e9df17eb1 Sarah Sharp      2010-04-02  703  					cur_td, &deq_state);
ae6367471 Sarah Sharp      2009-04-29  704  		else
522989a27 Sarah Sharp      2011-07-29  705  			td_to_noop(xhci, ep_ring, cur_td, false);
e9df17eb1 Sarah Sharp      2010-04-02  706  remove_finished_td:
ae6367471 Sarah Sharp      2009-04-29  707  		/*
ae6367471 Sarah Sharp      2009-04-29  708  		 * The event handler won't see a completion for this TD anymore,
ae6367471 Sarah Sharp      2009-04-29  709  		 * so remove it from the endpoint ring's TD list.  Keep it in
ae6367471 Sarah Sharp      2009-04-29  710  		 * the cancelled TD list for URB completion later.
ae6367471 Sarah Sharp      2009-04-29  711  		 */
585df1d90 Sarah Sharp      2011-08-02  712  		list_del_init(&cur_td->td_list);
ae6367471 Sarah Sharp      2009-04-29  713  	}
ae6367471 Sarah Sharp      2009-04-29 @714  	last_unlinked_td = cur_td;
6f5165cf9 Sarah Sharp      2009-10-27  715  	xhci_stop_watchdog_timer_in_irq(xhci, ep);
ae6367471 Sarah Sharp      2009-04-29  716
ae6367471 Sarah Sharp      2009-04-29  717  	/* If necessary, queue a Set Transfer Ring Dequeue Pointer command */

:::::: The code at line 714 was first introduced by commit
:::::: ae636747146ea97efa18e04576acd3416e2514f5 USB: xhci: URB cancellation support.

:::::: TO: Sarah Sharp <[email protected]> :::::: CC: Greg Kroah-Hartman <[email protected]>

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#2 lore raw
Quoted 6 lines · Julia Lawall

Geliang,

Please check whether it is acceptable that last_unlinked_td point to the dummy entry at th beginning of the list, in the case where the list_for_each_entry loop runs out normally.

Quoted 6 lines

It seems that you have sent a bunch of these patches. Please recheck them all to see if they really follow the semantics of list_for_each_entry properly. If particular, you should not normally use the index variable after leaving the loop, unless it is guaranteed that the exit from the loop was via a break.

Julia is correct, we can't use list_for_each_entry() here as cur_td would end up pointing to list head, and we really need it to point to the last entry in the list at that point.

old:
- list_for_each(entry, &ep->cancelled_td_list) {
- cur_td = list_entry(entry, struct xhci_td, cancelled_td_list);
cur_td_will point to last element in list. "entry" will point to head, but is on longer used.

new:
+ list_for_each_entry(cur_td, &ep->cancelled_td_list, cancelled_td_list) {
cur_td will point to head of list.

This is important as newly canceled TDs can be added to the tail of the list while looping through and returning the canceled TDs. We want to make sure we only return and delete the TDs up to the point we have handled them on the ring.

(code continues with: last_unlinked_td = cur_td;)

Thanks Julia for spotting this

-Mathias