Re: [PATCH v2 2/2] arm64: dts: sm8350: fix tlmm base address

[Katherine Perez]

On Tue, Dec 07, 2021 at 06:35:41PM -0800, Bjorn Andersson wrote:
> On Tue 07 Dec 18:21 PST 2021, Katherine Perez wrote:
> 
> > On Tue, Dec 07, 2021 at 05:16:14PM +0530, Vinod Koul wrote:
> > > On 30-11-21, 21:15, Bjorn Andersson wrote:
> > > > On Mon 22 Nov 22:03 CST 2021, Vinod Koul wrote:
> > > > 
> > > > > On 22-11-21, 11:05, Katherine Perez wrote:
> > > > > > TLMM controller base address is incorrect and will hang on some platforms.
> > > > > > Fix by giving the correct address.
> > > > > 
> > > > > Thanks, recheck the spec this looks correct. We should have tlmm reg
> > > > > space here and not tlmm base which also contains xpu region (thus hang)
> > > > > 
> > > > 
> > > > Aren't you reading the patch backwards?
> > > 
> > > I guess :(
> > > 
> > > > Afaict downstream the driver carries an offset of 0x100000, which we
> > > > dropped as we upstreamed the driver. As such changing reg to 0x0f000000
> > > > should cause most gpio register accesses to fall outside the actual
> > > > register window.
> > > > 
> > > > Or perhaps I'm missing something here?
> > > 
> > > I relooked and XPU is at 0xF000000 and Reg at 0xF100000
> > > So this patch should be dropped as such. The size mentioned in
> > > documentation is also correct
> > > 
> > > Katherine, can you elaborate more on the hang you have observed? Any
> > > specific pins you use which causes this?
> > 
> > Hi Vinod,
> > 
> > Yes, it seems to hang in msm_pinctrl_probe. Specifically, line 734 in
> > gpiolib.c: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/gpio/gpiolib.c#n734.
> > On i=4, it hangs on assign_bit and the system goes into a reboot loop.
> > When I set the TLMM address to f000000, I don't see this issue at all.
> > 
> 
> The cause for that is quite likely that gc->get_direction() will read
> the configuration from gpio<i>'s registers and gpio4 in your system is
> reserved for use by some trusted application.
> 
> When you change the TLMM address you avoid this problem by just reading
> random registers outside the region that contains protected registers.
> 
> 
> Adjust the gpio-reserved-ranges in your device's tlmm node to mark gpio4
> (probably 4 pins long) as "invalid", gpiolib will then not touch them.
> 
> Regards,
> Bjorn

Thanks, Bjorn. That makes sense. I'll resubmit with the changes to my
device's TLMM node and will drop this patch.

-Katherine
> 
> > > 
> > > 
> > > -- 
> > > ~Vinod