RE: [PATCH] irqchip/gic-v3: workaround for ASR8601 when reading mpidr

[Yan Zheng（严政）]

> -----Original Message-----
> From: Marc Zyngier [mailto:maz@xxxxxxxxxx]
> Sent: Wednesday, May 17, 2023 4:32 PM
> To: Yan Zheng（严政） <zhengyan@xxxxxxxxxxxx>
> Cc: tglx@xxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Gao Meitao（高玫涛）
> <meitaogao@xxxxxxxxxxxx>; Zhou Qiao(周侨) <qiaozhou@xxxxxxxxxxxx>;
> Zhang Zhizhou(张治洲) <zhizhouzhang@xxxxxxxxxxxx>
> Subject: Re: [PATCH] irqchip/gic-v3: workaround for ASR8601 when reading
> mpidr
> 
> On Wed, 17 May 2023 08:55:00 +0100,
> zhengyan <zhengyan@xxxxxxxxxxxx> wrote:
> >
> > This patch add workaround for ASR8601, which uses an armv8.2 processor
> > with a gic-500. ARMv8.2 uses Multiprocessor Affinity Register to
> > identify the logical address of the core by
> > | cluster | core | thread |.
> 
> Not quite. The ARMv8.2 architecture doesn't say *any* of that. It is ARM's
> *implementations* that follow this scheme.
> 

Really thank you for rapid response, 
Yes, as arm documents https://developer.arm.com/docuentation/ka002107/latest said
It comes from armv8.2 get 3 types for affinity (arm v8.0 cpus only get 2 types)
And it's an implementations issue.

> > However, gic-500 only supports topologies with affinity levels less
> > than 2 as
> > | cluster | core|.
> >
> > So it needs this patch to shift the MPIDR values to ensure proper
> > functionality
> >
> > Signed-off-by: zhengyan <zhengyan@xxxxxxxxxxxx>
> > ---
> >  drivers/irqchip/irq-gic-v3.c | 28 +++++++++++++++++++++++++++-
> >  1 file changed, 27 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/irqchip/irq-gic-v3.c
> > b/drivers/irqchip/irq-gic-v3.c index 6fcee221f201..435b98a8641e 100644
> > --- a/drivers/irqchip/irq-gic-v3.c
> > +++ b/drivers/irqchip/irq-gic-v3.c
> > @@ -39,6 +39,7 @@
> >
> >  #define FLAGS_WORKAROUND_GICR_WAKER_MSM8996	(1ULL << 0)
> >  #define FLAGS_WORKAROUND_CAVIUM_ERRATUM_38539	(1ULL << 1)
> > +#define FLAGS_WORKAROUND_MPIDR_ASR8601		(1ULL << 2)
> 
> What is ASR8601? Is it a system? Or an erratum number? For issues that are the
> result of a HW integration issue, please provide an official erratum number, and
> update Documentation/arm64/silicon-errata.rst.
> 

ASR8601 is our soc's name, and yes it’s a kind of HW integration issue
But maybe it’s not an erratum since our HW design is like that, although
Arm doesn't recommend this way.
And I would like to add more comments
Under the next part before *desc	= "GICv3: ASR 8601 MPIDR shift"*
Maybe this is a better way? Or add something under Documentation？

> >
> >  #define GIC_IRQ_TYPE_PARTITION	(GIC_IRQ_TYPE_LPI + 1)
> >
> > @@ -659,6 +660,9 @@ static u64 gic_mpidr_to_affinity(unsigned long
> > mpidr)  {
> >  	u64 aff;
> >
> > +	if (gic_data.flags & FLAGS_WORKAROUND_MPIDR_ASR8601)
> > +		mpidr >>= 8;
> > +
> >  	aff = ((u64)MPIDR_AFFINITY_LEVEL(mpidr, 3) << 32 |
> >  	       MPIDR_AFFINITY_LEVEL(mpidr, 2) << 16 |
> >  	       MPIDR_AFFINITY_LEVEL(mpidr, 1) << 8  | @@ -970,6 +974,9
> @@
> > static int __gic_populate_rdist(struct redist_region *region, void __iomem
> *ptr)
> >  	 * Convert affinity to a 32bit value that can be matched to
> >  	 * GICR_TYPER bits [63:32].
> >  	 */
> > +	if (gic_data.flags & FLAGS_WORKAROUND_MPIDR_ASR8601)
> > +		mpidr >>= 8;
> > +
> >  	aff = (MPIDR_AFFINITY_LEVEL(mpidr, 3) << 24 |
> >  	       MPIDR_AFFINITY_LEVEL(mpidr, 2) << 16 |
> >  	       MPIDR_AFFINITY_LEVEL(mpidr, 1) << 8 | @@ -1265,6 +1272,8
> @@
> > static u16 gic_compute_target_list(int *base_cpu, const struct cpumask
> *mask,
> >  	unsigned long mpidr = cpu_logical_map(cpu);
> >  	u16 tlist = 0;
> >
> > +	if (gic_data.flags & FLAGS_WORKAROUND_MPIDR_ASR8601)
> > +		mpidr >>= 8;
> >  	while (cpu < nr_cpu_ids) {
> >  		tlist |= 1 << (mpidr & 0xf);
> >
> > @@ -1274,7 +1283,8 @@ static u16 gic_compute_target_list(int *base_cpu,
> const struct cpumask *mask,
> >  		cpu = next_cpu;
> >
> >  		mpidr = cpu_logical_map(cpu);
> > -
> > +		if (gic_data.flags & FLAGS_WORKAROUND_MPIDR_ASR8601)
> > +			mpidr >>= 8;
> >  		if (cluster_id != MPIDR_TO_SGI_CLUSTER_ID(mpidr)) {
> >  			cpu--;
> >  			goto out;
> > @@ -1321,6 +1331,8 @@ static void gic_ipi_send_mask(struct irq_data *d,
> const struct cpumask *mask)
> >  		u64 cluster_id = MPIDR_TO_SGI_CLUSTER_ID(cpu_logical_map(cpu));
> >  		u16 tlist;
> >
> > +		if (gic_data.flags & FLAGS_WORKAROUND_MPIDR_ASR8601)
> > +			cluster_id =
> MPIDR_TO_SGI_CLUSTER_ID(cpu_logical_map(cpu) >> 8);
> 
> You've written the same check 5 times. Maybe you could start by refactoring
> that code so that the hack can be in a single place?
> 

Okay, I'll try to refactor it

> >  		tlist = gic_compute_target_list(&cpu, mask, cluster_id);
> >  		gic_send_sgi(cluster_id, tlist, d->hwirq);
> >  	}
> > @@ -1729,6 +1741,15 @@ static bool gic_enable_quirk_cavium_38539(void
> *data)
> >  	return true;
> >  }
> >
> > +static bool gic_enable_quirk_asr8601(void *data) {
> > +	struct gic_chip_data *d = data;
> > +
> > +	d->flags |= FLAGS_WORKAROUND_MPIDR_ASR8601;
> > +
> > +	return true;
> > +}
> > +
> >  static bool gic_enable_quirk_hip06_07(void *data)  {
> >  	struct gic_chip_data *d = data;
> > @@ -1823,6 +1844,11 @@ static const struct gic_quirk gic_quirks[] = {
> >  		.mask	= 0xffffffff,
> >  		.init	= gic_enable_quirk_nvidia_t241,
> >  	},
> > +	{
> > +		.desc	= "GICv3: ASR 8601 MPIDR SHIFT",
> 
> s/SHIFT/shift/
> 

Okay
> > +		.compatible = "asr,asr8601-gic-v3",
> 
> So ASR8601 *is* a system... Is it DT only?
> 
> Thanks,
> 
> 	M.
> 
> --
> Without deviation from the norm, progress is not possible.

Yes, asr8601 is our soc, and we want to use compatible node in devices-tree to control it,
As I mentioned at previous part, it might works well under
armv8.2(3 types of affinity) with gic500, but these code get strongly order with HW integration

Thanks again,