Sarah Sharp wrote:
> On Wed, Jan 16, 2013 at 12:22:59PM -0500, Alan Stern wrote:
>> On Wed, 16 Jan 2013, Martin Mokrejs wrote:
>>
>>> A corresponding diff of dmesg output is attached. Note that the first kmemleak in there
>>> happened just without any prior fiddling with a USB drive. For about two days I haven't
>>> connected a drive. However, usb-storage might be in a wrong shape since several days
>>> when it happened for the first time. Did you want me to reboot? ;-) I did not.
>>
>> Sarah, I looked through the xhci-hcd driver. There does appear to be a
>> leak in xhci-ring.c:handle_tx_event().
>
> Thanks for catching that.
>
>> The routine looks like this:
>>
>> /* Leave the TD around for the reset endpoint function
>> * to use(but only if it's not a control endpoint,
>> * since we already queued the Set TR dequeue pointer
>> * command for stalled control endpoints).
>> */
>> if (usb_endpoint_xfer_control(&urb->ep->desc) ||
>> (trb_comp_code != COMP_STALL &&
>> trb_comp_code != COMP_BABBLE))
>> xhci_urb_free_priv(xhci, urb_priv);
>>
>> ...
>>
>> /* EHCI, UHCI, and OHCI always unconditionally set the
>> * urb->status of an isochronous endpoint to 0.
>> */
>> if (usb_pipetype(urb->pipe) == PIPE_ISOCHRONOUS)
>> status = 0;
>> usb_hcd_giveback_urb(bus_to_hcd(urb->dev->bus), urb, status);
>>
>> If the condition on the first "if" statement fails, urb_priv won't be
>> deallocated. It needs something like
>>
>> if (...)
>> xhci_urb_free_priv(xhci, urb_priv);
>> + else
>> + kfree(urb_priv);
>
> Ok, so you're proposing freeing the urb_priv, but leaving the TD
> allocated for the Set TR dequeue functions to use? Yes, that looks like
> the right solution, feel free to submit a patch.
>
>> Martin, can you tell if adding these two lines fixes the problem?
Hi Alan,
yes, the following change has helped. I managed to get twice the "error -71"
message (one case is shown here)
Jan 16 23:07:56 vostro kernel: usb 3-1.2: Device not responding to set address.
Jan 16 23:07:56 vostro kernel: usb 3-1.2: Device not responding to set address.
Jan 16 23:07:56 vostro kernel: usb 3-1.2: device not accepting address 17, error -71
Jan 16 23:07:56 vostro kernel: hub 3-1:1.0: unable to enumerate USB device on port 2
but the kmemleak did not report anything after next unplug of the disk anymore.
--- linux-3.7.1/drivers/usb/host/xhci-ring.c.ori 2013-01-16 22:51:25.000000000 +0100
+++ linux-3.7.1/drivers/usb/host/xhci-ring.c 2013-01-16 22:53:03.000000000 +0100
@@ -2580,6 +2580,8 @@
(trb_comp_code != COMP_STALL &&
trb_comp_code != COMP_BABBLE))
xhci_urb_free_priv(xhci, urb_priv);
+ else
+ kfree(urb_priv);
usb_hcd_unlink_urb_from_ep(bus_to_hcd(urb->dev->bus), urb);
if ((urb->actual_length != urb->transfer_buffer_length &&
Thanks,
Martin
P.S.: The other kmemleak is still in there but is probably not your business: ;-)
# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff88040b60a578 (size 256):
comm "swapper/0", pid 1, jiffies 4294937575 (age 688.030s)
hex dump (first 32 bytes):
00 00 00 00 ad 4e ad de ff ff ff ff 00 00 00 00 .....N..........
ff ff ff ff ff ff ff ff 38 3f 5d 82 ff ff ff ff ........8?].....
backtrace:
[<ffffffff815b1dbd>] kmemleak_alloc+0x21/0x3e
[<ffffffff81110536>] slab_post_alloc_hook+0x28/0x2a
[<ffffffff81112a6e>] __kmalloc+0xf2/0x104
[<ffffffff81302bd5>] kzalloc.constprop.14+0xe/0x10
[<ffffffff81303036>] device_private_init+0x14/0x63
[<ffffffff81305110>] dev_set_drvdata+0x19/0x2f
[<ffffffff815c1ed4>] i801_probe+0x5e/0x451
[<ffffffff81280fb3>] local_pci_probe+0x5b/0xa2
[<ffffffff81282074>] pci_device_probe+0xc8/0xf7
[<ffffffff813056cd>] driver_probe_device+0xa9/0x1c1
[<ffffffff8130583f>] __driver_attach+0x5a/0x7e
[<ffffffff81303f7a>] bus_for_each_dev+0x57/0x83
[<ffffffff81305276>] driver_attach+0x19/0x1b
[<ffffffff81304e48>] bus_add_driver+0xa8/0x1fa
[<ffffffff81305cb1>] driver_register+0x8c/0x106
[<ffffffff81281c6d>] __pci_register_driver+0x5a/0x5e
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
