linux-kernel - Re: [PATCH] usb: chipidea: move ci_ulpi

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <ecb8d3e8-d525-4a2d-a868-803202c16296@franken-peeters.be>
Date: Thu, 16 May 2024 21:11:46 +0000
From: Wouter Franken <wouter@...nken-peeters.be>
To: Marek Szyprowski <m.szyprowski@...sung.com>, Michael Grzeschik <m.grzeschik@...gutronix.de>, Peter Chen <peter.chen@...nel.org>, Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc: linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] usb: chipidea: move ci_ulpi_init after the phy initialization

Hello all,

On 4/25/24 21:40, Marek Szyprowski wrote:
> Dear All,
>
> On 02.04.2024 08:23, Michael Grzeschik wrote:
>> The function ci_usb_phy_init is already handling the
>> hw_phymode_configure path which is also only possible after we have
>> a valid phy. So we move the ci_ulpi_init after the phy initialization
>> to be really sure to be able to communicate with the ulpi phy.
>>
>> Signed-off-by: Michael Grzeschik <m.grzeschik@...gutronix.de>
> This patch landed in linux-next some time ago as commit 22ffd399e6e7
> ("usb: chipidea: move ci_ulpi_init after the phy initialization").
> Unfortunately it breaks host USB operation on DragonBoard410c
> (arch/arm64/boot/dts/qcom/apq8016-sbc.dts). There is no error nor
> warning in the kernel log besides the information about deferred probe
> on the chipidea controller:
>
> platform ci_hdrc.0: deferred probe pending: (reason unknown)
>
> Tomorrow I will try to investigate which operation during driver probe
> triggers it. If there is anything else to check that might help fixing
> this issue, let me know.

I have the same problem on sony yuga 
(arch/arm/boot/dts/qcom/qcom-apq8064-sony-xperia-lagan-yuga.dts):
it breaks USB operation and shows the same message in the kernel log.

I have analyzed this issue and came to the following conclusion:

>> ---
>>    drivers/usb/chipidea/core.c | 8 ++++----
>>    drivers/usb/chipidea/ulpi.c | 5 -----
>>    2 files changed, 4 insertions(+), 9 deletions(-)
>>
>> diff --git a/drivers/usb/chipidea/core.c b/drivers/usb/chipidea/core.c
>> index 835bf2428dc6e..bada13f704b62 100644
>> --- a/drivers/usb/chipidea/core.c
>> +++ b/drivers/usb/chipidea/core.c
>> @@ -1084,10 +1084,6 @@ static int ci_hdrc_probe(struct platform_device *pdev)
>>    		return -ENODEV;
>>    	}
>>    
>> -	ret = ci_ulpi_init(ci);
>> -	if (ret)
>> -		return ret;
>> -

This ci_ulpi_init(...) call eventually calls probe(..) on the USB phy 
when the ulpi driver
registers all its child nodes. A couple lines later devm_phy_get(..) is 
called
to get the initialized phy device. Due to the move of the 
ci_ulpi_init(..) the phy
is not probed and the devm_phy_get(..) call fails with EPROBE_DEFER. 
This results
in a deadlock.

I don't understand the complete train of thought behind this patch to 
propose a fix
but hopefully my investigation can help to solve the issue.

>>    	if (ci->platdata->phy) {
>>    		ci->phy = ci->platdata->phy;
>>    	} else if (ci->platdata->usb_phy) {
>> @@ -1142,6 +1138,10 @@ static int ci_hdrc_probe(struct platform_device *pdev)
>>    		goto ulpi_exit;
>>    	}
>>    
>> +	ret = ci_ulpi_init(ci);
>> +	if (ret)
>> +		return ret;
>> +
>>    	ci->hw_bank.phys = res->start;
>>    
>>    	ci->irq = platform_get_irq(pdev, 0);
>> diff --git a/drivers/usb/chipidea/ulpi.c b/drivers/usb/chipidea/ulpi.c
>> index dfec07e8ae1d2..89fb51e2c3ded 100644
>> --- a/drivers/usb/chipidea/ulpi.c
>> +++ b/drivers/usb/chipidea/ulpi.c
>> @@ -68,11 +68,6 @@ int ci_ulpi_init(struct ci_hdrc *ci)
>>    	if (ci->platdata->phy_mode != USBPHY_INTERFACE_MODE_ULPI)
>>    		return 0;
>>    
>> -	/*
>> -	 * Set PORTSC correctly so we can read/write ULPI registers for
>> -	 * identification purposes
>> -	 */
>> -	hw_phymode_configure(ci);
>>    
>>    	ci->ulpi_ops.read = ci_ulpi_read;
>>    	ci->ulpi_ops.write = ci_ulpi_write;
>>
>> ---
>> base-commit: 5bab5dc780c9ed0c69fc2f828015532acf4a7848
>> change-id: 20240328-chipidea-phy-misc-b3f2bc814784
>>
>> Best regards,
> Best regards

Best regards
--
Wouter Franken