| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <641c4fa2-5406-0254-42de-8fd821876ddd@gmail.com>
Date: Fri, 18 Nov 2022 23:29:29 +0100
From: Ferry Toth <fntoth@...il.com>
To: Thinh Nguyen <Thinh.Nguyen@...opsys.com>
Cc: "linux-usb@...r.kernel.org" <linux-usb@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Heikki Krogerus <heikki.krogerus@...ux.intel.com>,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Sean Anderson <sean.anderson@...o.com>,
Liu Shixin <liushixin2@...wei.com>,
Andrey Smirnov <andrew.smirnov@...il.com>,
Andy Shevchenko <andriy.shevchenko@...ux.intel.com>
Subject: Re: [PATCH v1 1/1] usb: ulpi: defer ulpi_register on ulpi_read_id
timeout
Hi,
Op 11-11-2022 om 02:31 schreef Thinh Nguyen:
> On Thu, Nov 10, 2022, Ferry Toth wrote:
>> Hi
>>
>> Op 10-11-2022 om 13:45 schreef Ferry Toth:
>>> (sorry sent html with previous attempt)
>>>
>>> On 10-11-2022 01:06, Thinh Nguyen wrote:
>>>> Hi Ferry,
>>>>
>>>> On Wed, Nov 09, 2022, Ferry Toth wrote:
>>>>> Since commit 0f010171
>>>>> Dual Role support on Intel Merrifield platform broke due to rearranging
>>>>> the call to dwc3_get_extcon().
>>>>>
>>>>> It appears to be caused by ulpi_read_id() on the first test
>>>>> write failing
>>>>> with -ETIMEDOUT. Currently ulpi_read_id() expects to discover
>>>>> the phy via
>>>>> DT when the test write fails and returns 0 in that case even if
>>>>> DT does not
>>>>> provide the phy. Due to the timeout being masked dwc3 probe continues by
>>>>> calling dwc3_core_soft_reset() followed by dwc3_get_extcon()
>>>>> which happens
>>>>> to return -EPROBE_DEFER. On deferred probe ulpi_read_id()
>>>>> finally succeeds.
>>>>>
>>>>> This patch changes ulpi_read_id() to return -ETIMEDOUT when it
>>>>> occurs and
>>>>> catches the error in dwc3_core_init(). It handles the error by calling
>>>>> dwc3_core_soft_reset() after which it requests -EPROBE_DEFER. On
>>>>> deferred
>>>>> probe ulpi_read_id() again succeeds.
>>>>>
>>>>> Signed-off-by: Ferry Toth<ftoth@...londelft.nl>
>>>>> ---
>>>>> drivers/usb/common/ulpi.c | 5 +++--
>>>>> drivers/usb/dwc3/core.c | 5 ++++-
>>>>> 2 files changed, 7 insertions(+), 3 deletions(-)
>>>>>
>>>> Can you split the dwc3 change and ulpi change to separate patches?
>>>
>>> Thanks for your comments.
>>>
>>> I will send v2
>>>
>>>>> diff --git a/drivers/usb/common/ulpi.c b/drivers/usb/common/ulpi.c
>>>>> index d7c8461976ce..d8f22bc2f9d0 100644
>>>>> --- a/drivers/usb/common/ulpi.c
>>>>> +++ b/drivers/usb/common/ulpi.c
>>>>> @@ -206,8 +206,9 @@ static int ulpi_read_id(struct ulpi *ulpi)
>>>>> /* Test the interface */
>>>>> ret = ulpi_write(ulpi, ULPI_SCRATCH, 0xaa);
>>>>> - if (ret < 0)
>>>>> - goto err;
>>>>> + if (ret < 0) {
>>>>> + return ret;
>>>>> + }
>>>>> ret = ulpi_read(ulpi, ULPI_SCRATCH);
>>>>> if (ret < 0)
>>>>> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
>>>>> index 648f1c570021..e293ef70039b 100644
>>>>> --- a/drivers/usb/dwc3/core.c
>>>>> +++ b/drivers/usb/dwc3/core.c
>>>>> @@ -1106,8 +1106,11 @@ static int dwc3_core_init(struct dwc3 *dwc)
>>>>> if (!dwc->ulpi_ready) {
>>>>> ret = dwc3_core_ulpi_init(dwc);
>>>>> - if (ret)
>>>>> + if (ret) {
>>>>> + dwc3_core_soft_reset(dwc);
>>>> We shouldn't need to do soft reset here. The controller shouldn't be at
>>>> a bad/incorrect state at this point to warrant a soft-reset. There will
>>>> be a soft-reset when it goes through the initialization again.
>>>
>>> It doesn't go through the initialization again unless we set
>>> -EPROBE_DEFER. And when we make ulpi_read_id() return -EPROBE_DEFER it
>>> will goto err0 here, so skips dwc3_core_soft_reset.
>>>
>>> Do you mean you prefer something like:
>>>
>>> if (ret) {
>>>
>>> if (ret == -ETIMEDOUT) ret = -EPROBE_DEFER;
>>>
>>> else goto err0;
>
> Why "else"? But I saw you remove that in the new patch.
>
>>>
>>> }
>>
>> I just tested, and calling dwc3_core_soft_reset() proves to be necessary as
>> we need to goto err0 directly after. Else ret is overwritten and
>> -EPROBE_DEFER lost.
>
> Looks like there's a strange dependency problem here.
> * The setup needs a soft-reset before ulpi registration
> * The ulpi registration needs to go before the phy initialization
> * The soft-reset should be called after the phy initialization
>
> I can't explain the actual issue here, and we can't debug further
> because to look into it further would require looking at internal
> signals.
>
> This soft-reset and -EPROBE_DEFER seems to be a workaround to this
> dependency problem. Instead of using -EPROBE_DEFER, can you do this:
>
> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> index 2f0a9679686f..5a1aaf3741ec 100644
> --- a/drivers/usb/dwc3/core.c
> +++ b/drivers/usb/dwc3/core.c
> @@ -1097,6 +1097,8 @@ static int dwc3_core_init(struct dwc3 *dwc)
> goto err0;
>
> if (!dwc->ulpi_ready) {
> + /* Add comment */
> + dwc3_core_soft_reset(dwc);
> ret = dwc3_core_ulpi_init(dwc);
> if (ret)
> goto err0;
>
This indeed fixes the issue as well. Here is the trace:
# tracer: function_graph
#
# CPU DURATION FUNCTION CALLS
# | | | | | | |
0) | /* start_event: (dwc3_probe+0x0/0x1910) */
0) 7.070 us | dwc3_clk_enable.part.0();
0) 5.480 us | extcon_get_extcon_dev();
0) + 10.230 us | dwc3_runtime_idle();
0) | /* end_event: (platform_probe+0x3f/0xa0 <-
dwc3_probe) */
** multiple defers while waiting for extcon
0) | /* start_event: (dwc3_probe+0x0/0x1910) */
0) 7.320 us | dwc3_clk_enable.part.0();
0) 6.830 us | extcon_get_extcon_dev();
0) | dwc3_core_init() {
0) + 29.200 us | dwc3_core_soft_reset.part.0();
0) | dwc3_ulpi_init() {
0) | ulpi_register_interface() {
0) | dwc3_ulpi_write() {
0) 3.380 us | dwc3_ulpi_busyloop();
** without this patch this one times out after 10000us
0) 7.710 us | }
0) | dwc3_ulpi_read() {
0) 3.060 us | dwc3_ulpi_busyloop();
0) 7.210 us | }
0) | dwc3_ulpi_read() {
0) 2.830 us | dwc3_ulpi_busyloop();
0) 6.690 us | }
0) | dwc3_ulpi_read() {
0) 2.880 us | dwc3_ulpi_busyloop();
0) 6.670 us | }
0) | dwc3_ulpi_read() {
0) 2.940 us | dwc3_ulpi_busyloop();
0) 6.690 us | }
0) | dwc3_ulpi_read() {
0) 2.870 us | dwc3_ulpi_busyloop();
0) 6.620 us | }
0) + 18.150 us | ulpi_uevent();
0) 5.990 us | ulpi_match();
0) | ulpi_probe() {
0) | tusb1210_probe() {
0) | ulpi_read() {
0) | dwc3_ulpi_read() {
0) 4.440 us | dwc3_ulpi_busyloop();
0) 9.600 us | }
0) + 15.770 us | }
0) | ulpi_write() {
0) | dwc3_ulpi_write() {
0) 3.270 us | dwc3_ulpi_busyloop();
0) 6.820 us | }
0) + 11.020 us | }
0) ! 407.540 us | }
0) ! 416.980 us | }
0) 9.800 us | ulpi_uevent();
0) * 18604.00 us | }
0) * 18611.20 us | }
0) + 30.570 us | dwc3_core_soft_reset.part.0();
0) | tusb1210_power_on() {
1) | extcon_set_state_sync() {
1) 5.330 us | extcon_set_state.part.0();
1) + 90.550 us | extcon_sync.part.0();
1) ! 113.670 us | }
1) + 19.450 us | ulpi_uevent();
0) + 13.640 us | ulpi_uevent();
1) + 13.980 us | ulpi_uevent();
0) + 15.960 us | ulpi_uevent();
0) | ulpi_write() {
0) | dwc3_ulpi_write() {
0) * 10239.47 us | dwc3_ulpi_busyloop();
0) * 10250.57 us | }
0) * 10265.09 us | }
0) * 69518.95 us | }
0) 5.740 us | dwc3_event_buffers_setup();
0) * 88241.02 us | } /* dwc3_core_init */
0) ! 104.900 us | dwc3_debugfs_init();
0) | dwc3_drd_init() {
0) 4.720 us | extcon_register_notifier();
0) | extcon_get_state() {
0) 2.640 us | extcon_get_state.part.0();
0) 6.460 us | }
0) + 14.460 us | dwc3_set_mode();
0) + 43.300 us | }
0) | /* end_event: (platform_probe+0x3f/0xa0 <-
dwc3_probe) */
Maybe this is the preferred way to go if the dwc3_core_soft_reset()
doesn't hurt other users?
>>
>>>>> + ret = -EPROBE_DEFER;
>>>> We shouldn't automatically set every error status to correspond to
>>>> -EPROBE_DEFER. Check only the approapriate error codes (-ETIMEDOUT +
>>>> any other?).
>>> Other could be -ENOMEM. I think no need to do any new handling for that.
>
> There's also -EINVAL and some others too. Regardless, if it's -ENOMEM,
> then it should not return -EPROBE_DEFER.
>
>>>>> goto err0;
>>>>> + }
>>>>> dwc->ulpi_ready = true;
>>>>> }
>>>>>
>
> Thanks,
> Thinh
Powered by blists - more mailing lists