Message ID | 1474655197-26919-1-git-send-email-jacob-chen@iotwrt.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
On 2016/9/24 2:26, Jacob Chen wrote: > From: Jacob Chen <jacob2.chen@rock-chips.com> > > 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? > > 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. > Change-Id: If19912c7658253e15531c04db6c70fdbffd5960a remove this. > Signed-off-by: Jacob Chen <jacob2.chen@rock-chips.com> > --- > drivers/pinctrl/pinctrl-rockchip.c | 4 ++++ > 1 file changed, 4 insertions(+) > > diff --git a/drivers/pinctrl/pinctrl-rockchip.c b/drivers/pinctrl/pinctrl-rockchip.c > index c6c04ac..9a8804a 100644 > --- a/drivers/pinctrl/pinctrl-rockchip.c > +++ b/drivers/pinctrl/pinctrl-rockchip.c > @@ -2334,8 +2334,12 @@ static void rockchip_irq_gc_mask_clr_bit(struct irq_data *d) > void rockchip_irq_gc_mask_set_bit(struct irq_data *d) > { > struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); > + struct irq_chip_type *ct = irq_data_get_chip_type(d); > struct rockchip_pin_bank *bank = gc->private; > > + if (*ct->mask_cache & d->mask) > + return; > + > irq_gc_mask_set_bit(d); > clk_disable(bank->clk); > } > -- > 2.7.4 > > > _______________________________________________ > Linux-rockchip mailing list > Linux-rockchip@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-rockchip >
2016-09-24 18:24 GMT+08:00 Shawn Lin <shawn.lin@rock-chips.com>: > On 2016/9/24 2:26, Jacob Chen wrote: >> >> From: Jacob Chen <jacob2.chen@rock-chips.com> >> >> 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.
Am Samstag, 24. September 2016, 18:24:11 CEST schrieb Shawn Lin: > On 2016/9/24 2:26, Jacob Chen wrote: > > From: Jacob Chen <jacob2.chen@rock-chips.com> > > > > 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? > > > 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. that may be true, but on first glance I tend to agree with Jacob. Look for example at (a driver using disable_irq I picked at random) drivertc/ rtcrs/-tps6586x.c . In its probe function it does devm_request_irq directly followed by disable_irq on its alarm interrupt. The interrupt only gets enabled again if the rtc alarm gets enabled. So you can easily run into the disable + free case there as well. Similar probably in a lot of other drivers. Heiko > > > Change-Id: If19912c7658253e15531c04db6c70fdbffd5960a > > remove this. > > > Signed-off-by: Jacob Chen <jacob2.chen@rock-chips.com> > > --- > > > > drivers/pinctrl/pinctrl-rockchip.c | 4 ++++ > > 1 file changed, 4 insertions(+) > > > > diff --git a/drivers/pinctrl/pinctrl-rockchip.c > > b/drivers/pinctrl/pinctrl-rockchip.c index c6c04ac..9a8804a 100644 > > --- a/drivers/pinctrl/pinctrl-rockchip.c > > +++ b/drivers/pinctrl/pinctrl-rockchip.c > > @@ -2334,8 +2334,12 @@ static void rockchip_irq_gc_mask_clr_bit(struct > > irq_data *d)> > > void rockchip_irq_gc_mask_set_bit(struct irq_data *d) > > { > > > > struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); > > > > + struct irq_chip_type *ct = irq_data_get_chip_type(d); > > > > struct rockchip_pin_bank *bank = gc->private; > > > > + if (*ct->mask_cache & d->mask) > > + return; > > + > > > > irq_gc_mask_set_bit(d); > > clk_disable(bank->clk); > > > > } > > > > -- > > 2.7.4 > > > > > > _______________________________________________ > > Linux-rockchip mailing list > > Linux-rockchip@lists.infradead.org > > http://lists.infradead.org/mailman/listinfo/linux-rockchip
在 2016/9/27 6:35, Heiko Stuebner 写道: > Am Samstag, 24. September 2016, 18:24:11 CEST schrieb Shawn Lin: >> On 2016/9/24 2:26, Jacob Chen wrote: >>> From: Jacob Chen <jacob2.chen@rock-chips.com> >>> >>> 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? >> >>> 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. > > that may be true, but on first glance I tend to agree with Jacob. > Look for example at (a driver using disable_irq I picked at random) drivertc/ > rtcrs/-tps6586x.c . yep, I took a quick look at the files you pointed to, and it seems there should be the problems for what this patch to deal with. Probably I didn't see this error since the drivers we was using happened to handle it properly but maybe not for other drivers. I am okay if this could solve problems for Jacob. > > In its probe function it does devm_request_irq directly followed by > disable_irq on its alarm interrupt. The interrupt only gets enabled again if > the rtc alarm gets enabled. So you can easily run into the disable + free case > there as well. Similar probably in a lot of other drivers. > > > Heiko > >> >>> Change-Id: If19912c7658253e15531c04db6c70fdbffd5960a >> >> remove this. >> >>> Signed-off-by: Jacob Chen <jacob2.chen@rock-chips.com> >>> --- >>> >>> drivers/pinctrl/pinctrl-rockchip.c | 4 ++++ >>> 1 file changed, 4 insertions(+) >>> >>> diff --git a/drivers/pinctrl/pinctrl-rockchip.c >>> b/drivers/pinctrl/pinctrl-rockchip.c index c6c04ac..9a8804a 100644 >>> --- a/drivers/pinctrl/pinctrl-rockchip.c >>> +++ b/drivers/pinctrl/pinctrl-rockchip.c >>> @@ -2334,8 +2334,12 @@ static void rockchip_irq_gc_mask_clr_bit(struct >>> irq_data *d)> >>> void rockchip_irq_gc_mask_set_bit(struct irq_data *d) >>> { >>> >>> struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); >>> >>> + struct irq_chip_type *ct = irq_data_get_chip_type(d); >>> >>> struct rockchip_pin_bank *bank = gc->private; >>> >>> + if (*ct->mask_cache & d->mask) >>> + return; >>> + >>> >>> irq_gc_mask_set_bit(d); >>> clk_disable(bank->clk); >>> >>> } >>> >>> -- >>> 2.7.4 >>> >>> >>> _______________________________________________ >>> Linux-rockchip mailing list >>> Linux-rockchip@lists.infradead.org >>> http://lists.infradead.org/mailman/listinfo/linux-rockchip > > > > _______________________________________________ > Linux-rockchip mailing list > Linux-rockchip@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-rockchip >
diff --git a/drivers/pinctrl/pinctrl-rockchip.c b/drivers/pinctrl/pinctrl-rockchip.c index c6c04ac..9a8804a 100644 --- a/drivers/pinctrl/pinctrl-rockchip.c +++ b/drivers/pinctrl/pinctrl-rockchip.c @@ -2334,8 +2334,12 @@ static void rockchip_irq_gc_mask_clr_bit(struct irq_data *d) void rockchip_irq_gc_mask_set_bit(struct irq_data *d) { struct irq_chip_generic *gc = irq_data_get_irq_chip_data(d); + struct irq_chip_type *ct = irq_data_get_chip_type(d); struct rockchip_pin_bank *bank = gc->private; + if (*ct->mask_cache & d->mask) + return; + irq_gc_mask_set_bit(d); clk_disable(bank->clk); }