Re: [PATCH] clocksource: sh_tmu: Set cpu_possible_mask to fix SMP broadcast

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Geert,

On Wednesday 17 December 2014 13:11:05 Geert Uytterhoeven wrote:
> On Wed, Dec 17, 2014 at 1:04 PM, Laurent Pinchart wrote:
> > On Wednesday 17 December 2014 10:42:52 Geert Uytterhoeven wrote:
> >> On Wed, Dec 17, 2014 at 9:30 AM, Geert Uytterhoeven wrote:
> >> > On Wed, Dec 17, 2014 at 3:08 AM, Laurent Pinchart wrote:
> >> >>>> Kzm9g-reference still hangs at "Calibrating local timer..." (it did
> >> >>>> work at some point in the past).
> >> >> 
> >> >> kzm9g-reference boots for me with kzm9g_defconfig on Simon's devel
> >> >> branch with
> >> > 
> >> > OK.
> >> > 
> >> > I had expected the breakage to be something in my tree, either an issue
> >> > in a -next branch I'm using, an interaction with the CCF patches or so,
> >> > or a config issue (e.g. CONFIG_CPU_IDLE became broken lately if the TWD
> >> > is not in DT).
> >> 
> >> Kzm9g-reference hangs at "Calibrating local timer..." because I added the
> >> TWD to sh73a0.dtsi
> >> (http://www.spinics.net/lists/arm-kernel/msg383413.html).
> >> 
> >> As twd_get_clock() does
> >> 
> >>         if (np)
> >>                 twd_clk = of_clk_get(np, 0);
> >>         else
> >>                 twd_clk = clk_get_sys("smp_twd", NULL);
> >> 
> >> it's not DT-without-CCF proof, and kzm9g-reference cannot get its clock
> >> :-(
> >> 
> >> So the TWD node should be in sh73a0-kzm9g-multiplatform.dts, and we need
> >> more dts files instead of less...
> >> 
> >> Note that I added the TWD node to DT to fix a hang on kzm9g-multiplatform
> >> with CONFIG_CPU_IDLE=y (recently introduced "regression" in core code)
> >> after:
> >> 
> >>      DMA: preallocated 256 KiB pool for atomic coherent allocations
> >> 
> >> On kzm9g-legacy the TWD is instantiated from C board code
> >> (machine_desc.init_time = sh73a0_earlytimer_init), so CONFIG_CPU_IDLE=y
> >> does not hang. Despite https://lkml.org/lkml/2014/12/2/339.
> >> 
> >> On kzm9g-reference, the TWD is not instantiated, causing the same hang.
> >> If I instantiate the TWD from C board code there, it fails with
> >> 
> >>     twd: can't register interrupt 29 (-22)
> >> 
> >> which looks like a symptom of https://lkml.org/lkml/2014/12/2/339
> >> 
> >> So no cookies for users of several -legacy and -reference platforms in
> >> v3.19-rc1 this Monday...
> > 
> > I'm not sure if fixing that would be worth it. Should we instead try to
> > replace kzm9g-legacy and kzm9g-reference by kzm9g-multiplatform in v3.20 ?
> 
> v3.20 is reasonable. Except that some breakage will arrive in v3.19-rc1.
> So there will be a brokenness window for the kzm9g platform between
> v3.18 and v3.20.

I know, and that's unfortunate, but I wonder if we should really spend time on 
fixing that if we can get multiplatform support merged in v3.20. We could 
consider multiplatform to be a regression fix and get it merged in v3.19, but 
that might be a bit far-fetched :-)

> > What are we missing ?
> > 
> > - The DIV6 multiparent series has been merged by Mike, and the sh73a0 CCF
> > is nearly ready.
> 
> Yep.
> 
> > - We need proper BSC support to get LAN working, although it could be
> > worked around temporarily by specifying the BSC clock in the LAN node if
> > I recall correctly.
> 
> Working on that. I hope to send out v2 later today...
> 
> > - Accelerometer and RTC have no DT bindings, but they should work with the
> > I2C core OF match support.
> 
> I have RTC in my DTS, and it works.

Great.

I can help with at least some of the remaining pieces if you share your 
working branch.

-- 
Regards,

Laurent Pinchart