Hi Bart,
On 7/26/22 1:07 AM, Bart Van Assche wrote:
On 7/24/22 20:47, Peter Wang wrote:
Because kernel_restart is export, and mediatek may call kernel_restart
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Is this code upstream?
No, it not upstream.
And sorry that I correct mediatek usage, it is kernel_power_off, not
kernel_restart.
while shutdown is on going.
EXPORT_SYMBOL_GPL(kernel_restart);
kernel_restart -> kernel_restart_prepare -> device_shutdown
So, there may have two thread execute device_shutdown concurrently.
And the list_lock in device_shutdown function is not protect shutdown
callback function,
caused two callback function(ufshcd_shutdown/ufshcd_wl_shutdown)
could run concurrently.
Here is the error log that two thread in device_shutdown.
[37684.002227] [T1500641] platform +platform:112b0000.ufshci
kpoc_charger: ufshcd-mtk 112b0000.ufshci: [name:core&]shutdown
[37684.002264] [T1600339] scsi +scsi:0:0:0:49488 charger_thread:
ufs_device_wlun 0:0:0:49488: [name:core&]shutdown
Hi Peter,
I had not yet taken a look at the kernel_restart() function. Now that
I have taken a look, it seems to me that kernel_restart() calls must
be serialized via the system_transition_mutex. Please make sure that
the kernel_restart() calls are serialized.
Thanks,
Bart.
I think kernel_power_off could use system_transition_mutex to protect
shutdown racing.
We will try it, Thanks for the suggestion.
And if no need wait, it could more simple in this patch. I will upload
next version again.
Thanks.
Peter
