2020-09-17 23:20:22

by Shuah Khan

[permalink] [raw]
Subject: bus/mhi/core: Double lock in mhi_device_put() and dev_wake inc/dec

While looking at this file for an unrelated issue, I happen to notice
there is a double locking on mhi_cntrl->pm_lock in the mhi_device_put()
when it gets called from mhi_driver_remove()

The other two calls from mhi_driver_probe() don't hold the pm_lock.

In addition, lock holding while dev_wake updates is inconsistent.

dev_wake gets incremented and decremented without holding pm_lock in
mhi_device_get(), mhi_device_get_sync() and mhi_device_put().

Exception are when mhi_device_put() is called from mhi_driver_remove().

The following commit is where all this code is added.

bus: mhi: core: Add support for data transfer
https://github.com/torvalds/linux/commit/189ff97cca53e3fe2d8b38d64105040ce17fc62d

It appears to be real problem. I don't have a way to test this driver,
hence reaching out to let you know about my findings.

thanks,
-- Shuah


2020-09-19 02:38:40

by Bhaumik Bhatt

[permalink] [raw]
Subject: Re: bus/mhi/core: Double lock in mhi_device_put() and dev_wake inc/dec

On 2020-09-17 16:16, Shuah Khan wrote:
> While looking at this file for an unrelated issue, I happen to notice
> there is a double locking on mhi_cntrl->pm_lock in the mhi_device_put()
> when it gets called from mhi_driver_remove()
>
> The other two calls from mhi_driver_probe() don't hold the pm_lock.
>
> In addition, lock holding while dev_wake updates is inconsistent.
>
> dev_wake gets incremented and decremented without holding pm_lock in
> mhi_device_get(), mhi_device_get_sync() and mhi_device_put().
>
> Exception are when mhi_device_put() is called from mhi_driver_remove().
>
> The following commit is where all this code is added.
>
> bus: mhi: core: Add support for data transfer
> https://github.com/torvalds/linux/commit/189ff97cca53e3fe2d8b38d64105040ce17fc62d
>
> It appears to be real problem. I don't have a way to test this driver,
> hence reaching out to let you know about my findings.
>
> thanks,
> -- Shuah
Thank you for inputs.

Hemant and I discussed this and we agree that there are inconsistencies
we need to fix.

We will be uploading a patch to remove the read_lock_bh/read_unlock_bh
calls from the
mhi_driver_remove().

Thanks,
Bhaumik
'The Qualcomm Innovation Center, Inc. is a member of the Code Aurora
Forum, a Linux Foundation Collaborative Project'