[v4,04/14] irqchip: mips-gic: Support multi-cluster in for_each_online_cpu_gic()

Commit Message

Aleksandar Rikalo May 11, 2024, 10:43 a.m. UTC

From: Paul Burton <[e-mail]>

Introduce support for multi-cluster GIC register access in
__gic_with_next_online_cpu(), and therefore in its user
for_each_online_cpu_gic(). We access registers in remote clusters
using the CM's GCR_CL_REDIRECT register, and so here we delegate
to mips_cm_lock_other() in order to configure this access.

With this done, users of gic_with_each_online_cpu() gain support
for multi-cluster with no further changes.

Signed-off-by: Paul Burton <[e-mail]>
Signed-off-by: Chao-ying Fu <[e-mail]>
Signed-off-by: Dragan Mladjenovic <[e-mail]>
Signed-off-by: Aleksandar Rikalo <[e-mail]>
---
 drivers/irqchip/irq-mips-gic.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

Comments

Thomas Gleixner June 21, 2024, 8:21 p.m. UTC | #1

On Sat, May 11 2024 at 12:43, Aleksandar Rikalo wrote:
> From: Paul Burton <[e-mail]>
>
> Introduce support for multi-cluster GIC register access in
> __gic_with_next_online_cpu(), and therefore in its user
> for_each_online_cpu_gic(). We access registers in remote clusters
> using the CM's GCR_CL_REDIRECT register, and so here we delegate
> to mips_cm_lock_other() in order to configure this access.

Again: We do nothing. See docs.

> @@ -70,6 +70,20 @@ static int __gic_with_next_online_cpu(int prev)
>  {
>  	unsigned int cpu;
>  
> +	/*
> +	 * Unlock access to the previous CPU's GIC local register block.
> +	 *
> +	 * Delegate to the CM locking code in the multi-cluster case, since
> +	 * other clusters can only be accessed using GCR_CL_REDIRECT.
> +	 *
> +	 * In the single cluster case we don't need to do anything; the caller
> +	 * is responsible for maintaining gic_lock & nothing should be
> +	 * expecting any particular value of GIC_VL_OTHER so we can leave it
> +	 * as-is.
> +	 */
> +	if ((prev != -1) && mips_cps_multicluster_cpus())
> +		mips_cm_unlock_other();

Eew.

static inline void gic_unlock_cluster(void)
{
	if (mips_cps_multicluster_cpus())
		mips_cm_unlock_other();
}

#define for_each_online_cpu_gic(cpu, gic_lock)		\
	guard(raw_spinlock_irqsave)(gic_lock);		\
	for ((cpu) = __gic_with_next_online_cpu(-1);	\
	     (cpu) < nr_cpu_ids;			\
             gic_unlock_cluster(),			\
	     (cpu) = __gic_with_next_online_cpu(cpu);)

No?

Thanks,

        tglx

Patch

diff --git a/drivers/irqchip/irq-mips-gic.c b/drivers/irqchip/irq-mips-gic.c
index 9e7182150b5c..317ccc2593d1 100644
--- a/drivers/irqchip/irq-mips-gic.c
+++ b/drivers/irqchip/irq-mips-gic.c
@@ -70,6 +70,20 @@  static int __gic_with_next_online_cpu(int prev)
 {
 	unsigned int cpu;
 
+	/*
+	 * Unlock access to the previous CPU's GIC local register block.
+	 *
+	 * Delegate to the CM locking code in the multi-cluster case, since
+	 * other clusters can only be accessed using GCR_CL_REDIRECT.
+	 *
+	 * In the single cluster case we don't need to do anything; the caller
+	 * is responsible for maintaining gic_lock & nothing should be
+	 * expecting any particular value of GIC_VL_OTHER so we can leave it
+	 * as-is.
+	 */
+	if ((prev != -1) && mips_cps_multicluster_cpus())
+		mips_cm_unlock_other();
+
 	/* Discover the next online CPU */
 	cpu = cpumask_next(prev, cpu_online_mask);
 
@@ -80,10 +94,16 @@  static int __gic_with_next_online_cpu(int prev)
 	/*
 	 * Lock access to the next CPU's GIC local register block.
 	 *
+	 * Delegate to the CM locking code in the multi-cluster case, since
+	 * other clusters can only be accessed using GCR_CL_REDIRECT.
+	 *
 	 * In the single cluster case we simply set GIC_VL_OTHER. The caller
 	 * holds gic_lock so nothing can clobber the value we write.
 	 */
-	write_gic_vl_other(mips_cm_vp_id(cpu));
+	if (mips_cps_multicluster_cpus())
+		mips_cm_lock_other_cpu(cpu, CM_GCR_Cx_OTHER_BLOCK_LOCAL);
+	else
+		write_gic_vl_other(mips_cm_vp_id(cpu));
 
 	return cpu;
 }
@@ -105,6 +125,9 @@  static inline void __lockdep_assert_held(raw_spinlock_t *gic_lock)
  *
  * The caller must hold gic_lock throughout the loop, such that GIC_VL_OTHER
  * cannot be clobbered.
+ *
+ * Please note that if you start a loop this way, it must be completed to the
+ * end, otherwise, GIC local register block will remain locked.
  */
 #define for_each_online_cpu_gic(cpu, gic_lock)		\
 	for (__lockdep_assert_held(gic_lock),		\