From: "Du, Changbin" <[email protected]>
This is a reworked patch based on reverted commit d8f00cd685f5 ("usb:
hub: do not clear BOS field during reset device").
The privious one caused double mem-free if run to re_enumerate label.
New patch title changed to distinguish from old one. And I have tested
it with memory debugging options.
In function usb_reset_and_verify_device, the old BOS descriptor may
still be used before allocating a new one. (usb_disable_lpm function
uses it under the situation that it fails at usb_disable_link_state.)
So we cannot set the udev->bos to NULL before that, just keep what it
was. It will be overwrite when allocating a new one.
How to reproduce:
1. connect one usb3 hub to xhci port.
2. connect several lpm-capable super-speed usb disk to the hub.
3. copy big files to the usb disks.
4. disconnect the hub and repeat step 1-4.
Crash log:
BUG: unable to handle kernel NULL pointer dereference at
0000000000000010
IP: [<ffffffff8171f98d>] usb_enable_link_state+0x2d/0x2f0
Call Trace:
[<ffffffff8171ed5b>] ? usb_set_lpm_timeout+0x12b/0x140
[<ffffffff8171fcd1>] usb_enable_lpm+0x81/0xa0
[<ffffffff8171fdd8>] usb_disable_lpm+0xa8/0xc0
[<ffffffff8171fe1c>] usb_unlocked_disable_lpm+0x2c/0x50
[<ffffffff81723933>] usb_reset_and_verify_device+0xc3/0x710
[<ffffffff8172c4ed>] ? usb_sg_wait+0x13d/0x190
[<ffffffff81724743>] usb_reset_device+0x133/0x280
[<ffffffff8179ccd1>] usb_stor_port_reset+0x61/0x70
[<ffffffff8179cd68>] usb_stor_invoke_transport+0x88/0x520
Signed-off-by: Du, Changbin <[email protected]>
---
drivers/usb/core/hub.c | 14 +++++++++-----
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
index 51b43691..27444e5 100644
--- a/drivers/usb/core/hub.c
+++ b/drivers/usb/core/hub.c
@@ -5401,7 +5401,6 @@ static int usb_reset_and_verify_device(struct usb_device *udev)
}
bos = udev->bos;
- udev->bos = NULL;
for (i = 0; i < SET_CONFIG_TRIES; ++i) {
@@ -5494,13 +5493,18 @@ done:
usb_set_usb2_hardware_lpm(udev, 1);
usb_unlocked_enable_lpm(udev);
usb_enable_ltm(udev);
- usb_release_bos_descriptor(udev);
- udev->bos = bos;
+ /* release the new BOS descriptor allocated by hub_port_init() */
+ if (udev->bos != bos) {
+ usb_release_bos_descriptor(udev);
+ udev->bos = bos;
+ }
return 0;
re_enumerate:
- usb_release_bos_descriptor(udev);
- udev->bos = bos;
+ if (udev->bos != bos) {
+ usb_release_bos_descriptor(udev);
+ udev->bos = bos;
+ }
re_enumerate_no_bos:
/* LPM state doesn't matter when we're about to destroy the device. */
hub_port_logical_disconnect(parent_hub, port1);
--
2.5.0
On Tue, Mar 08, 2016 at 05:15:17PM +0800, [email protected] wrote:
> From: "Du, Changbin" <[email protected]>
>
> This is a reworked patch based on reverted commit d8f00cd685f5 ("usb:
> hub: do not clear BOS field during reset device").
>
> The privious one caused double mem-free if run to re_enumerate label.
> New patch title changed to distinguish from old one. And I have tested
> it with memory debugging options.
>
> In function usb_reset_and_verify_device, the old BOS descriptor may
> still be used before allocating a new one. (usb_disable_lpm function
> uses it under the situation that it fails at usb_disable_link_state.)
> So we cannot set the udev->bos to NULL before that, just keep what it
> was. It will be overwrite when allocating a new one.
>
> How to reproduce:
> 1. connect one usb3 hub to xhci port.
> 2. connect several lpm-capable super-speed usb disk to the hub.
> 3. copy big files to the usb disks.
> 4. disconnect the hub and repeat step 1-4.
>
> Crash log:
> BUG: unable to handle kernel NULL pointer dereference at
> 0000000000000010
> IP: [<ffffffff8171f98d>] usb_enable_link_state+0x2d/0x2f0
> Call Trace:
> [<ffffffff8171ed5b>] ? usb_set_lpm_timeout+0x12b/0x140
> [<ffffffff8171fcd1>] usb_enable_lpm+0x81/0xa0
> [<ffffffff8171fdd8>] usb_disable_lpm+0xa8/0xc0
> [<ffffffff8171fe1c>] usb_unlocked_disable_lpm+0x2c/0x50
> [<ffffffff81723933>] usb_reset_and_verify_device+0xc3/0x710
> [<ffffffff8172c4ed>] ? usb_sg_wait+0x13d/0x190
> [<ffffffff81724743>] usb_reset_device+0x133/0x280
> [<ffffffff8179ccd1>] usb_stor_port_reset+0x61/0x70
> [<ffffffff8179cd68>] usb_stor_invoke_transport+0x88/0x520
>
> Signed-off-by: Du, Changbin <[email protected]>
> ---
> drivers/usb/core/hub.c | 14 +++++++++-----
> 1 file changed, 9 insertions(+), 5 deletions(-)
Is this patch still needed? I thought we had some other fix in this
area...
confused,
greg k-h
>
> diff --git a/drivers/usb/core/hub.c b/drivers/usb/core/hub.c
> index 51b43691..27444e5 100644
> --- a/drivers/usb/core/hub.c
> +++ b/drivers/usb/core/hub.c
> @@ -5401,7 +5401,6 @@ static int usb_reset_and_verify_device(struct usb_device *udev)
> }
>
> bos = udev->bos;
> - udev->bos = NULL;
>
> for (i = 0; i < SET_CONFIG_TRIES; ++i) {
>
> @@ -5494,13 +5493,18 @@ done:
> usb_set_usb2_hardware_lpm(udev, 1);
> usb_unlocked_enable_lpm(udev);
> usb_enable_ltm(udev);
> - usb_release_bos_descriptor(udev);
> - udev->bos = bos;
> + /* release the new BOS descriptor allocated by hub_port_init() */
> + if (udev->bos != bos) {
> + usb_release_bos_descriptor(udev);
> + udev->bos = bos;
> + }
> return 0;
>
> re_enumerate:
> - usb_release_bos_descriptor(udev);
> - udev->bos = bos;
> + if (udev->bos != bos) {
> + usb_release_bos_descriptor(udev);
> + udev->bos = bos;
> + }
> re_enumerate_no_bos:
> /* LPM state doesn't matter when we're about to destroy the device. */
> hub_port_logical_disconnect(parent_hub, port1);
> --
> 2.5.0
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-usb" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> On Tue, Mar 08, 2016 at 05:15:17PM +0800, [email protected] wrote:
> > From: "Du, Changbin" <[email protected]>
> >
> > This is a reworked patch based on reverted commit d8f00cd685f5 ("usb:
> > hub: do not clear BOS field during reset device").
> >
> > The privious one caused double mem-free if run to re_enumerate label.
> > New patch title changed to distinguish from old one. And I have tested
> > it with memory debugging options.
> >
> > In function usb_reset_and_verify_device, the old BOS descriptor may
> > still be used before allocating a new one. (usb_disable_lpm function
> > uses it under the situation that it fails at usb_disable_link_state.)
> > So we cannot set the udev->bos to NULL before that, just keep what it
> > was. It will be overwrite when allocating a new one.
> >
> > How to reproduce:
> > 1. connect one usb3 hub to xhci port.
> > 2. connect several lpm-capable super-speed usb disk to the hub.
> > 3. copy big files to the usb disks.
> > 4. disconnect the hub and repeat step 1-4.
> >
> > Crash log:
> > BUG: unable to handle kernel NULL pointer dereference at
> > 0000000000000010
> > IP: [<ffffffff8171f98d>] usb_enable_link_state+0x2d/0x2f0
> > Call Trace:
> > [<ffffffff8171ed5b>] ? usb_set_lpm_timeout+0x12b/0x140
> > [<ffffffff8171fcd1>] usb_enable_lpm+0x81/0xa0
> > [<ffffffff8171fdd8>] usb_disable_lpm+0xa8/0xc0
> > [<ffffffff8171fe1c>] usb_unlocked_disable_lpm+0x2c/0x50
> > [<ffffffff81723933>] usb_reset_and_verify_device+0xc3/0x710
> > [<ffffffff8172c4ed>] ? usb_sg_wait+0x13d/0x190
> > [<ffffffff81724743>] usb_reset_device+0x133/0x280
> > [<ffffffff8179ccd1>] usb_stor_port_reset+0x61/0x70
> > [<ffffffff8179cd68>] usb_stor_invoke_transport+0x88/0x520
> >
> > Signed-off-by: Du, Changbin <[email protected]>
> > ---
> > drivers/usb/core/hub.c | 14 +++++++++-----
> > 1 file changed, 9 insertions(+), 5 deletions(-)
>
> Is this patch still needed? I thought we had some other fix in this
> area...
>
> confused,
>
> greg k-h
>
Hi, Greg k-h,
Sorry for it confused you. This patch still need. This is same fix with
previous commit d8f00cd685f5 ("usb: hub: do not clear BOS field
during reset device"). But d8f00cd685f5 is buggy and reverted. This
new patch should be the final fix.
Best Regards,
Du, Changbin
On 04/26/2016 10:53 PM, Du, Changbin wrote:
>> On Tue, Mar 08, 2016 at 05:15:17PM +0800, [email protected] wrote:
>>> From: "Du, Changbin" <[email protected]>
>>>
>>> This is a reworked patch based on reverted commit d8f00cd685f5 ("usb:
>>> hub: do not clear BOS field during reset device").
>>>
>>> The privious one caused double mem-free if run to re_enumerate label.
>>> New patch title changed to distinguish from old one. And I have tested
>>> it with memory debugging options.
>>>
>>> In function usb_reset_and_verify_device, the old BOS descriptor may
>>> still be used before allocating a new one. (usb_disable_lpm function
>>> uses it under the situation that it fails at usb_disable_link_state.)
>>> So we cannot set the udev->bos to NULL before that, just keep what it
>>> was. It will be overwrite when allocating a new one.
>>>
>>> How to reproduce:
>>> 1. connect one usb3 hub to xhci port.
>>> 2. connect several lpm-capable super-speed usb disk to the hub.
>>> 3. copy big files to the usb disks.
>>> 4. disconnect the hub and repeat step 1-4.
>>>
>>> Crash log:
>>> BUG: unable to handle kernel NULL pointer dereference at
>>> 0000000000000010
>>> IP: [<ffffffff8171f98d>] usb_enable_link_state+0x2d/0x2f0
>>> Call Trace:
>>> [<ffffffff8171ed5b>] ? usb_set_lpm_timeout+0x12b/0x140
>>> [<ffffffff8171fcd1>] usb_enable_lpm+0x81/0xa0
>>> [<ffffffff8171fdd8>] usb_disable_lpm+0xa8/0xc0
>>> [<ffffffff8171fe1c>] usb_unlocked_disable_lpm+0x2c/0x50
>>> [<ffffffff81723933>] usb_reset_and_verify_device+0xc3/0x710
>>> [<ffffffff8172c4ed>] ? usb_sg_wait+0x13d/0x190
>>> [<ffffffff81724743>] usb_reset_device+0x133/0x280
>>> [<ffffffff8179ccd1>] usb_stor_port_reset+0x61/0x70
>>> [<ffffffff8179cd68>] usb_stor_invoke_transport+0x88/0x520
>>>
>>> Signed-off-by: Du, Changbin <[email protected]>
>>> ---
>>> drivers/usb/core/hub.c | 14 +++++++++-----
>>> 1 file changed, 9 insertions(+), 5 deletions(-)
>> Is this patch still needed? I thought we had some other fix in this
>> area...
>>
>> confused,
>>
>> greg k-h
>>
> Hi, Greg k-h,
> Sorry for it confused you. This patch still need. This is same fix with
> previous commit d8f00cd685f5 ("usb: hub: do not clear BOS field
> during reset device"). But d8f00cd685f5 is buggy and reverted. This
> new patch should be the final fix.
>
> Best Regards,
> Du, Changbin
>
I think Greg is referring to commit 464ad8c43a9e ("usb: core : hub: Fix
BOS 'NULL pointer' kernel panic"), which has already been applied
upstream. It looks to me like that patch might have fixed the same
problem in a different way, in which case Changbin's patch is not
needed. But I haven't been involved in developing or testing that
patch, so I can't say for sure. At the very least, 464ad8c43a9e
conflicts with Changbin's patch.
Changbin, can you take a look at 464ad8c43a9e and see if that fixes the
same problem that your patch did?
Thanks,
Tony Battersby