]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
irqchip/gic-v3: Switch high priority PPIs over to handle_percpu_devid_irq()
authorMarc Zyngier <maz@kernel.org>
Mon, 20 Oct 2025 12:29:27 +0000 (13:29 +0100)
committerThomas Gleixner <tglx@linutronix.de>
Mon, 27 Oct 2025 16:16:34 +0000 (17:16 +0100)
It so appears that handle_percpu_devid_irq() is extremely similar to
handle_percpu_devid_fasteoi_nmi(), and that the differences do no justify
the horrid machinery in the GICv3 driver to handle the flow handler switch.

Stick with the standard flow handler, even for NMIs.

Suggested-by: Will Deacon <will@kernel.org>
Signed-off-by: Marc Zyngier <maz@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Will Deacon <will@kernel.org>
Link: https://patch.msgid.link/20251020122944.3074811-11-maz@kernel.org
drivers/irqchip/irq-gic-v3.c

index cf0ba83c6eda61ca3bddb016029e0c966ac67cf7..dd2d6d722fc5e8d730f4e415d1f8204be5edaaee 100644 (file)
@@ -230,9 +230,6 @@ static void __init gic_prio_init(void)
                !cpus_have_group0);
 }
 
-/* rdist_nmi_refs[n] == number of cpus having the rdist interrupt n set as NMI */
-static refcount_t *rdist_nmi_refs;
-
 static struct gic_kvm_info gic_v3_kvm_info __initdata;
 static DEFINE_PER_CPU(bool, has_rss);
 
@@ -608,24 +605,6 @@ static u32 __gic_get_ppi_index(irq_hw_number_t hwirq)
        }
 }
 
-static u32 __gic_get_rdist_index(irq_hw_number_t hwirq)
-{
-       switch (__get_intid_range(hwirq)) {
-       case SGI_RANGE:
-       case PPI_RANGE:
-               return hwirq;
-       case EPPI_RANGE:
-               return hwirq - EPPI_BASE_INTID + 32;
-       default:
-               unreachable();
-       }
-}
-
-static u32 gic_get_rdist_index(struct irq_data *d)
-{
-       return __gic_get_rdist_index(d->hwirq);
-}
-
 static int gic_irq_nmi_setup(struct irq_data *d)
 {
        struct irq_desc *desc = irq_to_desc(d->irq);
@@ -646,20 +625,8 @@ static int gic_irq_nmi_setup(struct irq_data *d)
                return -EINVAL;
 
        /* desc lock should already be held */
-       if (gic_irq_in_rdist(d)) {
-               u32 idx = gic_get_rdist_index(d);
-
-               /*
-                * Setting up a percpu interrupt as NMI, only switch handler
-                * for first NMI
-                */
-               if (!refcount_inc_not_zero(&rdist_nmi_refs[idx])) {
-                       refcount_set(&rdist_nmi_refs[idx], 1);
-                       desc->handle_irq = handle_percpu_devid_fasteoi_nmi;
-               }
-       } else {
+       if (!gic_irq_in_rdist(d))
                desc->handle_irq = handle_fasteoi_nmi;
-       }
 
        gic_irq_set_prio(d, dist_prio_nmi);
 
@@ -686,15 +653,8 @@ static void gic_irq_nmi_teardown(struct irq_data *d)
                return;
 
        /* desc lock should already be held */
-       if (gic_irq_in_rdist(d)) {
-               u32 idx = gic_get_rdist_index(d);
-
-               /* Tearing down NMI, only switch handler for last NMI */
-               if (refcount_dec_and_test(&rdist_nmi_refs[idx]))
-                       desc->handle_irq = handle_percpu_devid_irq;
-       } else {
+       if (!gic_irq_in_rdist(d))
                desc->handle_irq = handle_fasteoi_irq;
-       }
 
        gic_irq_set_prio(d, dist_prio_irq);
 }
@@ -2080,19 +2040,9 @@ static const struct gic_quirk gic_quirks[] = {
 
 static void gic_enable_nmi_support(void)
 {
-       int i;
-
        if (!gic_prio_masking_enabled() || nmi_support_forbidden)
                return;
 
-       rdist_nmi_refs = kcalloc(gic_data.ppi_nr + SGI_NR,
-                                sizeof(*rdist_nmi_refs), GFP_KERNEL);
-       if (!rdist_nmi_refs)
-               return;
-
-       for (i = 0; i < gic_data.ppi_nr + SGI_NR; i++)
-               refcount_set(&rdist_nmi_refs[i], 0);
-
        pr_info("Pseudo-NMIs enabled using %s ICC_PMR_EL1 synchronisation\n",
                gic_has_relaxed_pmr_sync() ? "relaxed" : "forced");