Re: [PATCH] Revert "staging:r8188eu: use lib80211 CCMP decrypt"

[Michael Straube <straube.linux@gmail.com>]

On 1/1/19 10:02 AM, Ivan Safonov wrote:
> I suggested a patch for loading modules from interruptible mode, but this patch remained unclaimed ( http://driverdev.linuxdriverproject.org/pipermail/driverdev-devel/2018-August/124851.html ).
> 

So with these changes try_then_request_module() would work properly?

> For some reason I thought that this patch had been removed and did not track the fate of this code ( http://driverdev.linuxdriverproject.org/pipermail/driverdev-devel/2018-August/124573.html ).
> 

I reverted that patch (there are conflicts meanwhile) and removed
try_then_request_module() in rtw_aes_encrypt() and it looks good.

Perhaps the same applies for the reverted TKIP changes?

Michael

> On 1/1/19 5:17 AM, Larry Finger wrote:
>> On 12/30/18 12:39 PM, Michael Straube wrote:
>>> Commit 6bd082af7e36 ("staging:r8188eu: use lib80211 CCMP decrypt")
>>> is causing hardfreeze whenever the driver tries to connect to my wifi
>>> network. That makes the driver unusable on my system. Reverting the
>>> commit fixes the issue and the driver works properly.
>>>
>>> Dec 29 19:21:17 gentoo kernel: BUG: scheduling while atomic: swapper/6/0/0x00000100
>>
>> Michael,
>>
>> I have verified the freezes that you see. Although I have not been able to capture the console dump, I think we are likely seeing the same problem.
>>
>> I do have a work-around in that I have not gotten any freezes when I force module lib80211_crypt_ccmp to be loaded before I load module r8188eu. This clue was used in finding what seems to be a good fix.
>>
>> I do not know anything about demand loading of modules using try_then_request_module(); however, I noticed that the macro actually calls __request_module(), which has the following comment:
>>
>>   * Load a module using the user mode module loader. The function returns
>>   * zero on success or a negative errno code or positive exit code from
>>   * "modprobe" on failure. Note that a successful module load does not mean
>>   * the module did not then unload and exit on an error of its own. Callers
>>   * must check that the service they requested is now available not blindly
>>   * invoke it.
>>
>> I note that it says "user mode module loader". Routine rtw_aes_decrypt() is likely inside some sort of locking, which leads to the "scheduling while atomic" bug that you see. As a result, I suspect that the module is not loaded, and that leads to the NULL dereference when the module is accessed. Please try the one-line patch attached, which forces lib80211 to load when r8188eu is loaded. With this patch, I have been connected to an AES-encrypted AP for nearly 3 hours with no problems.
>>
>> Larry
>>
>>