> > + spin_lock_irqsave(hba->host->host_lock, flags);
> > + hba->active_uic_cmd = NULL;
> > + if (ufshcd_is_link_broken(hba)) {
> > + spin_unlock_irqrestore(hba->host->host_lock, flags);
> > + ret = -ENOLINK;
> > + goto out;
> > + }
> > + hba->uic_async_done = cnf;
> > + cmd->argument2 = 0;
> > + cmd->argument3 = mode;
> > + ret = __ufshcd_send_uic_cmd(hba, cmd, true);
> > + spin_unlock_irqrestore(hba->host->host_lock, flags);
>
> Please consider adding lockdep_assert_held(&hba->uic_cmd_mutex) near
> the start of __ufshcd_poll_uic_pwr() to document that this function is
> serialized against other contexts that submit an UIC.
Got it.
>
> Additionally, I don't think that it is necessary to hold the host lock
> around the ufshcd_is_link_broken() or __ufshcd_send_uic_cmd() calls.
I agree with you on __ufshcd_send_uic_cmd because of the mutex. And I wonder why you make you think that on uic_link_state.
Is it because you think it's not possible to access uic_link_state simultaneously w/ the current code or it should be?
It seems most cases have no problems but anyway I think it seems that uic_link_state could be accessed from more than one context,
especially if auto hibern8 is enabled and only if some UIC errors occur.
Thanks.
Kiwoong Kim
>
> Thanks,
>
> Bart.
>
