[PATCH] irqchip/ast2700-intc: Avoid allocating in the irq_domain activate callback
Ryan Chen
ryan_chen at aspeedtech.com
Tue Aug 18 18:09:19 PDT 2026
The IRQ core calls the irq_domain .activate callback from __setup_irq()
with desc->lock held and interrupts disabled. Both
aspeed_intc1_irq_domain_activate() and the aspeed_intc0_resolve_route()
it invokes test a compatible string with fwnode_device_is_compatible(),
which wraps fwnode_property_match_string() and allocates with GFP_KERNEL.
Besides sleeping in atomic context, that makes the irq_desc lock depend
on fs_reclaim, which lockdep reports as an inversion because
irq_desc_lock_class is also taken in hardirq context:
WARNING: possible irq lock inversion dependency detected
swapper/0/0 just changed the state of lock:
(&irq_desc_lock_class){-...}-{2:2}, at: handle_level_irq+0x28/0x168
but this lock took another, HARDIRQ-unsafe lock in the past:
(fs_reclaim){+.+.}-{0:0}
...
fs_reclaim_acquire
__kmalloc_noprof
fwnode_property_match_string
aspeed_intc0_resolve_route
aspeed_intc1_irq_domain_activate
Both call sites are only ever handed OF nodes, so use
of_device_is_compatible() instead: it walks the property in place and
does not allocate.
Fixes: 07825e41519a ("irqchip/ast2700-intc: Add AST2700-A2 support")
Reported-by: Andrew Jeffery <andrew at codeconstruct.com.au>
Signed-off-by: Ryan Chen <ryan_chen at aspeedtech.com>
Tested-by: Andrew Jeffery <andrew at codeconstruct.com.au>
---
Tested on an AST2700 EVB with CONFIG_PROVE_LOCKING=y: the irq lock
inversion splat quoted in the patch is gone, and the ast2700-intc0 KUnit
suite still passes 9/9.
---
drivers/irqchip/irq-ast2700-intc0.c | 2 +-
drivers/irqchip/irq-ast2700-intc1.c | 3 ++-
2 files changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/irqchip/irq-ast2700-intc0.c b/drivers/irqchip/irq-ast2700-intc0.c
index 14b8b88f1179..c85b98a19bd4 100644
--- a/drivers/irqchip/irq-ast2700-intc0.c
+++ b/drivers/irqchip/irq-ast2700-intc0.c
@@ -312,7 +312,7 @@ int aspeed_intc0_resolve_route(const struct irq_domain *c0domain, size_t nc1outs
return -ENOENT;
if (!IS_ENABLED(CONFIG_ASPEED_AST2700_INTC_TEST) &&
- !fwnode_device_is_compatible(c0domain->fwnode, "aspeed,ast2700-intc0"))
+ !of_device_is_compatible(to_of_node(c0domain->fwnode), "aspeed,ast2700-intc0"))
return -ENODEV;
intc0 = c0domain->host_data;
diff --git a/drivers/irqchip/irq-ast2700-intc1.c b/drivers/irqchip/irq-ast2700-intc1.c
index 59e8f0d5ddcd..df6acbafce2d 100644
--- a/drivers/irqchip/irq-ast2700-intc1.c
+++ b/drivers/irqchip/irq-ast2700-intc1.c
@@ -157,7 +157,8 @@ static int aspeed_intc1_irq_domain_activate(struct irq_domain *domain,
intc1->ranges.nranges,
intc1->ranges.ranges, &resolved);
if (rc < 0) {
- if (!fwnode_device_is_compatible(intc1->upstream->fwnode, "riscv,aplic")) {
+ if (!of_device_is_compatible(to_of_node(intc1->upstream->fwnode),
+ "riscv,aplic")) {
dev_warn(intc1->dev,
"Failed to resolve interrupt route for hwirq %lu in domain %s\n",
data->hwirq, domain->name);
---
base-commit: 4477a78374a57c3809b172ad30cceabda48c47c6
change-id: 20260819-fix-ast2700-intc-atomic-alloc-93248a29a1c0
Best regards,
--
Ryan Chen <ryan_chen at aspeedtech.com>
More information about the linux-riscv
mailing list