arm64: kexec: Check if IRQ is already masked before masking
Subsystems (5)
PATCH v2 1/1 arm64: kexec: Check if IRQ is already masked before masking
During machine kexec, the function machine_kexec_mask_interrupts() is responsible for disabling or masking all interrupts. While the irq_disable hook ensures that an already-disabled IRQ is not disabled again, the current implementation unconditionally invokes the irq_mask() function for every interrupt descriptor, even when the interrupt is already masked.
A specific issue was observed in the crash kernel flow after unbinding a device (prior to kexec) that used a GPIO as an IRQ source. The warning was triggered by the gpiochip_disable_irq() function, which attempted to clear the FLAG_IRQ_IS_ENABLED flag when FLAG_USED_AS_IRQ was not set:
```
void gpiochip_disable_irq(struct gpio_chip *gc, unsigned int offset)
{
struct gpio_desc *desc = gpiochip_get_desc(gc, offset);
if (!IS_ERR(desc) &&
!WARN_ON(!test_bit(FLAG_USED_AS_IRQ, &desc->flags)))
clear_bit(FLAG_IRQ_IS_ENABLED, &desc->flags);
}
```The issue emerged after commit a8173820f441 ("gpio: gpiolib: Allow GPIO IRQs to lazy disable"), which introduced lazy disablement for GPIO IRQs by replacing disable/enable hooks with mask/unmask hooks in some cases. While irq_disable guarded against redundant operations, the unguarded irq_mask in machine_kexec_mask_interrupts() led to warnings when invoked on already-masked IRQs.
When a GPIO-IRQ-using driver is unbound, the IRQ is released, invoking __irq_disable() and irq_state_set_masked(). A subsequent call to machine_kexec_mask_interrupts() reinvoked chip->irq_mask(), leading to a call chain that included gpiochip_irq_mask() and gpiochip_disable_irq(). Because FLAG_USED_AS_IRQ was cleared earlier, a warning was printed.
This patch replaces the direct invocation of the irq_mask() and irq_disable() hooks with simplified code that leverages the irq_disable() kernel infrastructure. This higher-level function checks the interrupt's state to prevent redundant operations. Additionally, the IRQ_DISABLE_UNLAZY status flag is set to ensure that, for interrupt chips lacking an irq_disable callback, the disable operation is handled using the lazy approach.
As part of this change, the irq_disable() declaration was moved from kernel/irq/internals.h to include/linux/irq.h to make it accessible outside the kernel/irq/ directory, as the former can only be included within that directory.
---
V1 -> V2:
- Implement an alternative solution by utilizing the kernel's irq_disable() infrastructure.
- Apply the fix to additional architectures, including ARM, PowerPC, and RISC-V.
--- arch/arm/kernel/machine_kexec.c | 7 ++----- arch/arm64/kernel/machine_kexec.c | 7 ++----- arch/powerpc/kexec/core.c | 7 ++----- arch/riscv/kernel/machine_kexec.c | 7 ++----- include/linux/irq.h | 2 ++ kernel/irq/internals.h | 1 - 6 files changed, 10 insertions(+), 21 deletions(-)
arch/arm/kernel/machine_kexec.c +2 -5
@@ -142,11 +142,8 @@ static void machine_kexec_mask_interrupts(void)if (chip->irq_eoi &&irqd_irq_inprogress (&desc->irq_data)) chip->irq_eoi (&desc->irq_data); -if (chip->irq_mask) - chip->irq_mask (&desc->irq_data); - -if (chip->irq_disable && !irqd_irq_disabled (&desc->irq_data)) - chip->irq_disable (&desc->irq_data); +irq_set_status_flags (i,IRQ_DISABLE_UNLAZY ); +irq_disable (desc); } }
arch/arm64/kernel/machine_kexec.c +2 -5
@@ -230,11 +230,8 @@ static void machine_kexec_mask_interrupts(void) chip->irq_eoi) chip->irq_eoi (&desc->irq_data); -if (chip->irq_mask) - chip->irq_mask (&desc->irq_data); - -if (chip->irq_disable && !irqd_irq_disabled (&desc->irq_data)) - chip->irq_disable (&desc->irq_data); +irq_set_status_flags (i,IRQ_DISABLE_UNLAZY ); +irq_disable (desc); } }
arch/powerpc/kexec/core.c +2 -5
@@ -36,11 +36,8 @@ void machine_kexec_mask_interrupts(void) {if (chip->irq_eoi &&irqd_irq_inprogress (&desc->irq_data)) chip->irq_eoi (&desc->irq_data); -if (chip->irq_mask) - chip->irq_mask (&desc->irq_data); - -if (chip->irq_disable && !irqd_irq_disabled (&desc->irq_data)) - chip->irq_disable (&desc->irq_data); +irq_set_status_flags (i,IRQ_DISABLE_UNLAZY ); +irq_disable (desc); } }
arch/riscv/kernel/machine_kexec.c +2 -5
@@ -129,11 +129,8 @@ static void machine_kexec_mask_interrupts(void)if (chip->irq_eoi &&irqd_irq_inprogress (&desc->irq_data)) chip->irq_eoi (&desc->irq_data); -if (chip->irq_mask) - chip->irq_mask (&desc->irq_data); - -if (chip->irq_disable && !irqd_irq_disabled (&desc->irq_data)) - chip->irq_disable (&desc->irq_data); +irq_set_status_flags (i,IRQ_DISABLE_UNLAZY ); +irq_disable (desc); } }
include/linux/irq.h +2 -0
@@ -600,6 +600,8 @@ enum {#define IRQ_DEFAULT_INIT_FLAGS ARCH_IRQ_INIT_FLAGS +extern void irq_disable (struct irq_desc *desc); +struct irqaction ;extern int setup_percpu_irq (unsigned int irq,struct irqaction *new);extern void remove_percpu_irq (unsigned int irq,struct irqaction *act);
kernel/irq/internals.h +0 -1
@@ -91,7 +91,6 @@ extern int irq_startup(struct irq_desc *desc, bool resend, bool force);extern void irq_shutdown (struct irq_desc *desc);extern void irq_shutdown_and_deactivate (struct irq_desc *desc);extern void irq_enable (struct irq_desc *desc); -extern void irq_disable (struct irq_desc *desc);extern void irq_percpu_enable (struct irq_desc *desc,unsigned int cpu);extern void irq_percpu_disable (struct irq_desc *desc,unsigned int cpu);extern void mask_irq (struct irq_desc *desc);
Signature · 1 line
2.40.1
Quoted 5 lines · Eliav Farber
arch/arm/kernel/machine_kexec.c +0
-0
arch/arm/kernel/machine_kexec.c static void machine_kexec_mask_interrupts(void) · line 146
Quoted 10 lines
arch/arm/kernel/machine_kexec.c +2
-5@@ -142,11 +142,8 @@ static void machine_kexec_mask_interrupts(void)if (chip->irq_eoi &&irqd_irq_inprogress (&desc->irq_data)) chip->irq_eoi (&desc->irq_data); -if (chip->irq_mask) - chip->irq_mask (&desc->irq_data); - -if (chip->irq_disable && !irqd_irq_disabled (&desc->irq_data)) - chip->irq_disable (&desc->irq_data); +irq_set_status_flags (i,IRQ_DISABLE_UNLAZY ); +irq_disable (desc);
This is just wrong. If the interrupt was torn down, then its state is deactivated and it was masked already. So the EOI handling and the mask/disable dance are neither required nor make sense.
So this whole thing should be:
chip = irq_desc_get_chip(desc);
- if (!chip)
+ if (!chip || !irqd_is_started(&desc->irq_data))
continue;But what's worse is that we have 4 almost identical variants of the same code.
So instead of exposing core functionality and "fixing" up four variants, can we please have a consolidated version of this function in the core code:
struct irq_chip *chip;
int check_eoi = 1;
chip = irq_desc_get_chip(desc);
if (!chip || !irqd_is_started(&desc->irq_data))
continue;
if (IS_ENABLED(CONFIG_.....)) {
/*
* Add a sensible comment which explains this.
*/
check_eoi = irq_set_irqchip_state(....);
}
if (check_eoi && ....)
chip->irq_eoi(&desc->irq_data);
irq_shutdown(desc);No?
Thanks,
tglx
On Wed, Nov 27 2024 at 15:22, Eliav Farber wrote:
As a related note. The subject line is not really matching what the patch does. It want's to be split into a core change and one patch per architecture.
Quoted 1 line
This patch replaces the direct invocation of the
irq_mask()and
git grep 'This patch' Documentation/process/
Quoted 6 lines
irq_disable()hooks with simplified code that leverages theirq_disable()kernel infrastructure. This higher-level function checks the interrupt's state to prevent redundant operations. Additionally, theIRQ_DISABLE_UNLAZYstatus flag is set to ensure that, for interrupt chips lacking anirq_disablecallback, the disable operation is handled using the lazy approach.
Not that it matters much anymore, but the last sentence does not make sense:
Set the UNLAZY flag so disable is handled using the LAZY approach ...
Thanks,
tglx