Re: [PATCH] pinctrl: rockchip: don't disable clk when irq mask is already set

From: éè
Date: Sat Sep 24 2016 - 10:42:56 EST


2016-09-24 18:24 GMT+08:00 Shawn Lin <shawn.lin@xxxxxxxxxxxxxx>:
> On 2016/9/24 2:26, Jacob Chen wrote:
>>
>> From: Jacob Chen <jacob2.chen@xxxxxxxxxxxxxx>
>>
>> In some drivers, disable_irq() call don't be symmetric with enable_irq()
>> , disable_irq() will be called before call free_irq().
>
>
> Which upstream drivers you refer to?
>
> Shouldn't it be the unbalanced call for these drivers?
>


I met this in bcmdhd driver, it seems upstream brcmfmac driver don't
have this problem.

======================================================================
void bcmsdh_oob_intr_unregister(bcmsdh_info_t *bcmsdh)
{
int err = 0;
bcmsdh_os_info_t *bcmsdh_osinfo = bcmsdh->os_cxt;

SDLX_MSG(("%s: Enter\n", __FUNCTION__));
if (!bcmsdh_osinfo->oob_irq_registered) {
SDLX_MSG(("%s: irq is not registered\n", __FUNCTION__));
return;
}
if (bcmsdh_osinfo->oob_irq_wake_enabled) {
err = disable_irq_wake(bcmsdh_osinfo->oob_irq_num);
if (!err)
bcmsdh_osinfo->oob_irq_wake_enabled = FALSE;
}
if (bcmsdh_osinfo->oob_irq_enabled) {
disable_irq(bcmsdh_osinfo->oob_irq_num);
bcmsdh_osinfo->oob_irq_enabled = FALSE;
}
free_irq(bcmsdh_osinfo->oob_irq_num, bcmsdh);
bcmsdh_osinfo->oob_irq_registered = FALSE;
}
======================================================================
In this funciton, it will disable irq before free_irq.
At first, i think that commenting the line
"disable_irq(bcmsdh_osinfo->oob_irq_num);" can slove this problem,
but actually this driver have many hidden calls to disable_irq and
hardlly to correct....

Besides, I think that umask() and mask() aren't supposed to be strict symmetric.

>>
>> But both disable_irq() and free_irq() will call
>> rockchip_irq_gc_mask_set_bit,
>> and clk_disable() will be called more times than clk_enable(), which will
>> cause bugs.
>>
>> I think we can correct that by checking of mask.If mask is already set, do
>> nothing.
>>
>
> Looks like a little hacky to me.
>

Yes, it's pretty hacky, but i think it can work stable since this
condition only take effect when disbale_irq + free_irq was called.