GIC-v3 LPIs are impacted by an architectural design issue: they do not
have a global active state and as such the next LPI can be delivered to
a new CPU after an affinity change while the previous LPI handler has
not yet returned on the original CPU. If LPIs had an active state, then
an LPI would not be able to be retriggered until the first CPU had
issued a deactivation. As is, it is necessary to cater for this case in
software by enabling the IRQD_RESEND_WHEN_IN_PROGRESS flag for GIC-v3
interrupts. Setting this flag mitigates the issue by getting the
original CPU to resend the interrupt after the handler completed.
This RESEND_WHEN_INPROGRESS enablement is done for vanilla GIC-v3
interrupts allocated from its_irq_domain_alloc as well as directly
injected vLPIs for vPEs provided by GIC-v4 features, which still uses
the GIC-v3 code.
Suggested-by: Marc Zyngier <[email protected]>
Signed-off-by: James Gowans <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: KarimAllah Raslan <[email protected]>
Cc: Yipeng Zou <[email protected]>
Cc: Zhang Jianhua <[email protected]>
---
drivers/irqchip/irq-gic-v3-its.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 0ec2b1e1df75..1994541eaef8 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -3585,6 +3585,7 @@ static int its_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
irqd = irq_get_irq_data(virq + i);
irqd_set_single_target(irqd);
irqd_set_affinity_on_activate(irqd);
+ irqd_set_resend_when_in_progress(irqd);
pr_debug("ID:%d pID:%d vID:%d\n",
(int)(hwirq + i - its_dev->event_map.lpi_base),
(int)(hwirq + i), virq + i);
@@ -4523,6 +4524,7 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq
irq_domain_set_hwirq_and_chip(domain, virq + i, i,
irqchip, vm->vpes[i]);
set_bit(i, bitmap);
+ irqd_set_resend_when_in_progress(irq_get_irq_data(virq + i));
}
if (err) {
--
2.25.1
The following commit has been merged into the irq/irqchip-next branch of irqchip:
Commit-ID: 8f4b589595d01f882d63d21efe15af4a5ad7c59b
Gitweb: https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms/8f4b589595d01f882d63d21efe15af4a5ad7c59b
Author: James Gowans <[email protected]>
AuthorDate: Thu, 08 Jun 2023 14:00:21 +02:00
Committer: Marc Zyngier <[email protected]>
CommitterDate: Fri, 16 Jun 2023 12:23:40 +01:00
irqchip/gic-v3-its: Enable RESEND_WHEN_IN_PROGRESS for LPIs
GICv3 LPIs are impacted by an architectural design issue: they do not
have a global active state and as such a given LPI can be delivered to
a new CPU after an affinity change while the previous instance of the
same LPI handler has not yet completed on the original CPU.
If LPIs had an active state, this second LPI would not be delivered
until the first CPU deactivated the initial LPI, just like SPIs.
To solve this issue, use the newly introduced IRQD_RESEND_WHEN_IN_PROGRESS
flag, ensuring that we do not lose an LPI being delivered during that window
by getting the GIC to resend it.
This workaround gets enabled for all LPIs, including the VPE doorbells.
Suggested-by: Marc Zyngier <[email protected]>
Signed-off-by: James Gowans <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: KarimAllah Raslan <[email protected]>
Cc: Yipeng Zou <[email protected]>
Cc: Zhang Jianhua <[email protected]>
[maz: massaged commit message]
Signed-off-by: Marc Zyngier <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
drivers/irqchip/irq-gic-v3-its.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/irqchip/irq-gic-v3-its.c b/drivers/irqchip/irq-gic-v3-its.c
index 0ec2b1e..1994541 100644
--- a/drivers/irqchip/irq-gic-v3-its.c
+++ b/drivers/irqchip/irq-gic-v3-its.c
@@ -3585,6 +3585,7 @@ static int its_irq_domain_alloc(struct irq_domain *domain, unsigned int virq,
irqd = irq_get_irq_data(virq + i);
irqd_set_single_target(irqd);
irqd_set_affinity_on_activate(irqd);
+ irqd_set_resend_when_in_progress(irqd);
pr_debug("ID:%d pID:%d vID:%d\n",
(int)(hwirq + i - its_dev->event_map.lpi_base),
(int)(hwirq + i), virq + i);
@@ -4523,6 +4524,7 @@ static int its_vpe_irq_domain_alloc(struct irq_domain *domain, unsigned int virq
irq_domain_set_hwirq_and_chip(domain, virq + i, i,
irqchip, vm->vpes[i]);
set_bit(i, bitmap);
+ irqd_set_resend_when_in_progress(irq_get_irq_data(virq + i));
}
if (err) {