[v4,3/7] arm/imx: add gic_handle_irq function

This is a plain translation of assembly gic irq handler to C function
for CONFIG_MULTI_IRQ_HANDLER support on imx family.

Signed-off-by: Shawn Guo <shawn.guo@linaro.org>
---
 arch/arm/plat-mxc/Makefile                   |    2 +-
 arch/arm/plat-mxc/gic.c                      |   47 ++++++++++++++++++++++++++
 arch/arm/plat-mxc/include/mach/common.h      |    2 +
 arch/arm/plat-mxc/include/mach/entry-macro.S |    6 +++
 4 files changed, 56 insertions(+), 1 deletions(-)
 create mode 100644 arch/arm/plat-mxc/gic.c

On Wed, Sep 28, 2011 at 05:06:44PM +0800, Shawn Guo wrote:
> +#ifdef CONFIG_SMP
> +		else if (irqnr < 16) {
> +			writel_relaxed(irqstat, gic_cpu_base_addr +
> +						GIC_CPU_EOI);
> +			do_IPI(irqnr, regs);
> +		}
> +#endif
> +#ifdef CONFIG_LOCAL_TIMERS
> +		else if (irqnr == 29) {
> +			writel_relaxed(irqstat, gic_cpu_base_addr +
> +						GIC_CPU_EOI);
> +			do_local_timer(regs);
> +		}
> +#endif

As I've said for similar patches. neither of these two functions are
designed to be called from another C function (because they're marked
__exception or __exception_irq_entry).  Both of these markers tell the
unwinder that a struct pt_regs is located directly above the functions
stack frame.

On Thu, Sep 29, 2011 at 10:34:09AM +0100, Russell King - ARM Linux wrote:
> On Wed, Sep 28, 2011 at 05:06:44PM +0800, Shawn Guo wrote:
> > +#ifdef CONFIG_SMP
> > +		else if (irqnr < 16) {
> > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > +						GIC_CPU_EOI);
> > +			do_IPI(irqnr, regs);
> > +		}
> > +#endif
> > +#ifdef CONFIG_LOCAL_TIMERS
> > +		else if (irqnr == 29) {
> > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > +						GIC_CPU_EOI);
> > +			do_local_timer(regs);
> > +		}
> > +#endif
> 
> As I've said for similar patches. neither of these two functions are
> designed to be called from another C function (because they're marked
> __exception or __exception_irq_entry).  Both of these markers tell the
> unwinder that a struct pt_regs is located directly above the functions
> stack frame.
> 
Can you please share your position on Marc's PPI and GIC
MULTI_IRQ_HANDLER series?  Are you possibly going to merge them in the
coming merge window?  If you are, I would directly move my imx6q onto
those series and save the local gic_handle_irq().  Otherwise, I may
have to add a wrapper for do_IPI() and do_local_timer() to work around
the pt_regs issue you pointed out here.

On Thu, Sep 29, 2011 at 10:34:09AM +0100, Russell King - ARM Linux wrote:
> On Wed, Sep 28, 2011 at 05:06:44PM +0800, Shawn Guo wrote:
> > +#ifdef CONFIG_SMP
> > +		else if (irqnr < 16) {
> > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > +						GIC_CPU_EOI);
> > +			do_IPI(irqnr, regs);
> > +		}
> > +#endif
> > +#ifdef CONFIG_LOCAL_TIMERS
> > +		else if (irqnr == 29) {
> > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > +						GIC_CPU_EOI);
> > +			do_local_timer(regs);
> > +		}
> > +#endif
> 
> As I've said for similar patches. neither of these two functions are
> designed to be called from another C function (because they're marked
> __exception or __exception_irq_entry).  Both of these markers tell the
> unwinder that a struct pt_regs is located directly above the functions
> stack frame.
> 
I'm also wondering the consequence of not doing so, because we are
seeing icip_handle_irq() and ichp_handle_irq() (arch/arm/mach-pxa/irq.c)
are doing the same thing, and we are not running into any problem with
above code.

On Thu, Sep 29, 2011 at 10:27:30PM +0800, Shawn Guo wrote:
> On Thu, Sep 29, 2011 at 10:34:09AM +0100, Russell King - ARM Linux wrote:
> > On Wed, Sep 28, 2011 at 05:06:44PM +0800, Shawn Guo wrote:
> > > +#ifdef CONFIG_SMP
> > > +		else if (irqnr < 16) {
> > > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > > +						GIC_CPU_EOI);
> > > +			do_IPI(irqnr, regs);
> > > +		}
> > > +#endif
> > > +#ifdef CONFIG_LOCAL_TIMERS
> > > +		else if (irqnr == 29) {
> > > +			writel_relaxed(irqstat, gic_cpu_base_addr +
> > > +						GIC_CPU_EOI);
> > > +			do_local_timer(regs);
> > > +		}
> > > +#endif
> > 
> > As I've said for similar patches. neither of these two functions are
> > designed to be called from another C function (because they're marked
> > __exception or __exception_irq_entry).  Both of these markers tell the
> > unwinder that a struct pt_regs is located directly above the functions
> > stack frame.
> > 
> I'm also wondering the consequence of not doing so, because we are
> seeing icip_handle_irq() and ichp_handle_irq() (arch/arm/mach-pxa/irq.c)
> are doing the same thing, and we are not running into any problem with
> above code.
> 
Sorry.  Please ignore the above.

On Thu, Sep 29, 2011 at 10:08:05PM +0800, Shawn Guo wrote:
> Can you please share your position on Marc's PPI and GIC
> MULTI_IRQ_HANDLER series?  Are you possibly going to merge them in the
> coming merge window?  If you are, I would directly move my imx6q onto
> those series and save the local gic_handle_irq().  Otherwise, I may
> have to add a wrapper for do_IPI() and do_local_timer() to work around
> the pt_regs issue you pointed out here.

My personal opinion of Marc's PPI patches hasn't changed: I think
integrating them into genirq just makes the whole thing a lot more
complicated (and error-prone) than it otherwise needs to be.

For example, it is completely invalid to call any of the existing
genirq APIs for a PPI interrupt, because there is no sane way to ensure
that the intended CPU is targetted - unless we're operating from a thread
which is bound specifically to a single CPU.

However, Marc has been working with Thomas to produce something more
generic (and not ARM specific) and I don't have an opinion on that.
This addresses the issues (like the one I mention above) but I've not
been following it any further than that.