[v4,01/15] genirq/msi: Silence set affinity failed warning

Commit Message

Marek Vasut July 23, 2024, 1:27 p.m. UTC

Various PCIe controllers that mux MSIs onto single IRQ line produce these
"IRQ%d: set affinity failed" warnings when entering suspend. This has been
discussed before [1] [2] and an example test case is included at the end
of this commit message.

Controller drivers which create MSI IRQ domain with MSI_FLAG_USE_DEF_CHIP_OPS
flag set and which do not override the .irq_set_affinity irqchip callback get
assigned default .irq_set_affinity = msi_domain_set_affinity() callback. That
is not desired on controllers where it is not possible to set affinity of each
MSI IRQ line to a specific CPU core due to hardware limitation.

Introduce dedicated flag MSI_FLAG_NO_AFFINITY, which keeps .irq_set_affinity
unset in case the controller driver did not assign the callback. This way, the
migrate_one_irq() code in cpuhotplug.c can exit right away, without printing
the aforementioned warning. The .irq_set_affinity implementations which only
return -EINVAL can be removed from multiple controller drivers.

```
$ grep 25 /proc/interrupts
 25:   0 0 0 0 0 0 0 0   PCIe MSI   0   Edge   PCIe PME

$ echo core > /sys/power/pm_test ; echo mem > /sys/power/state
...
Disabling non-boot CPUs ...
IRQ25: set affinity failed(-22). <---------- This is being silenced here
psci: CPU7 killed (polled 4 ms)
...
```

[1] https://lore.kernel.org/all/[e-mail]/
[2] https://lore.kernel.org/all/[e-mail]/

Signed-off-by: Marek Vasut <[e-mail]>
---
Cc: "Krzysztof Wilczyński" <[e-mail]>
Cc: "Pali Rohár" <[e-mail]>
Cc: "Uwe Kleine-König" <[e-mail]>
Cc: Aleksandr Mishin <[e-mail]>
Cc: Anna-Maria Behnsen <[e-mail]>
Cc: Anup Patel <[e-mail]>
Cc: Bjorn Helgaas <[e-mail]>
Cc: Broadcom internal kernel review list <[e-mail]>
Cc: Daire McNamara <[e-mail]>
Cc: Damien Le Moal <[e-mail]>
Cc: Florian Fainelli <[e-mail]>
Cc: Hou Zhiqiang <[e-mail]>
Cc: Jianjun Wang <[e-mail]>
Cc: Jim Quinlan <[e-mail]>
Cc: Jingoo Han <[e-mail]>
Cc: Jisheng Zhang <[e-mail]>
Cc: Jon Hunter <[e-mail]>
Cc: Jonathan Derrick <[e-mail]>
Cc: Jonathan Hunter <[e-mail]>
Cc: Joyce Ooi <[e-mail]>
Cc: Karthikeyan Mitran <[e-mail]>
Cc: Kishon Vijay Abraham I <[e-mail]>
Cc: Koichiro Den <[e-mail]>
Cc: Lorenzo Pieralisi <[e-mail]>
Cc: Manivannan Sadhasivam <[e-mail]>
Cc: Marc Zyngier <[e-mail]>
Cc: Michal Simek <[e-mail]>
Cc: Nicolas Saenz Julienne <[e-mail]>
Cc: Niklas Cassel <[e-mail]>
Cc: Nipun Gupta <[e-mail]>
Cc: Nirmal Patel <[e-mail]>
Cc: Rob Herring <[e-mail]>
Cc: Ryder Lee <[e-mail]>
Cc: Shivamurthy Shastri <[e-mail]>
Cc: Siddharth Vadapalli <[e-mail]>
Cc: Thierry Reding <[e-mail]>
Cc: Thomas Gleixner <[e-mail]>
Cc: Thomas Petazzoni <[e-mail]>
Cc: Yoshihiro Shimoda <[e-mail]>
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
---
V2: - Introduce MSI_FLAG_NO_AFFINITY to inhibit assignment
      of msi_domain_set_affinity()
V3: - Replace MSI_FLAG_USE_DEF_CHIP_OPS_NOAFF with MSI_FLAG_NO_AFFINITY
      and make MSI_FLAG_NO_AFFINITY into separate flag
    - Update commit message
    - Rebase on current linux-next
    - Use genirq/msi: subject prefix which is likely better fit now
    - Split off R-Car part of the patch
V4: No change
---
 include/linux/msi.h | 2 ++
 kernel/irq/msi.c    | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)

Comments

Damien Le Moal July 23, 2024, 10:54 p.m. UTC | #1

On 7/23/24 10:27 PM, Marek Vasut wrote:
> Various PCIe controllers that mux MSIs onto single IRQ line produce these
> "IRQ%d: set affinity failed" warnings when entering suspend. This has been
> discussed before [1] [2] and an example test case is included at the end
> of this commit message.
> 
> Controller drivers which create MSI IRQ domain with MSI_FLAG_USE_DEF_CHIP_OPS
> flag set and which do not override the .irq_set_affinity irqchip callback get
> assigned default .irq_set_affinity = msi_domain_set_affinity() callback. That
> is not desired on controllers where it is not possible to set affinity of each
> MSI IRQ line to a specific CPU core due to hardware limitation.
> 
> Introduce dedicated flag MSI_FLAG_NO_AFFINITY, which keeps .irq_set_affinity
> unset in case the controller driver did not assign the callback. This way, the
> migrate_one_irq() code in cpuhotplug.c can exit right away, without printing
> the aforementioned warning. The .irq_set_affinity implementations which only
> return -EINVAL can be removed from multiple controller drivers.
> 
> ```
> $ grep 25 /proc/interrupts
>  25:   0 0 0 0 0 0 0 0   PCIe MSI   0   Edge   PCIe PME
> 
> $ echo core > /sys/power/pm_test ; echo mem > /sys/power/state
> ...
> Disabling non-boot CPUs ...
> IRQ25: set affinity failed(-22). <---------- This is being silenced here
> psci: CPU7 killed (polled 4 ms)
> ...
> ```
> 
> [1] https://lore.kernel.org/all/[e-mail]/
> [2] https://lore.kernel.org/all/[e-mail]/
> 
> Signed-off-by: Marek Vasut <[e-mail]>

Looks good to me.

Reviewed-by: Damien Le Moal <[e-mail]>

Patch

diff --git a/include/linux/msi.h b/include/linux/msi.h
index 9449797638255..b10093c4d00ea 100644
--- a/include/linux/msi.h
+++ b/include/linux/msi.h
@@ -554,6 +554,8 @@  enum {
 	MSI_FLAG_MSIX_CONTIGUOUS	= (1 << 19),
 	/* PCI/MSI-X vectors can be dynamically allocated/freed post MSI-X enable */
 	MSI_FLAG_PCI_MSIX_ALLOC_DYN	= (1 << 20),
+	/* PCI MSIs cannot be steered separately to CPU cores */
+	MSI_FLAG_NO_AFFINITY		= (1 << 21),
 };
 
 /**
diff --git a/kernel/irq/msi.c b/kernel/irq/msi.c
index 5fa0547ece0c4..ca6e2ae6d6fc0 100644
--- a/kernel/irq/msi.c
+++ b/kernel/irq/msi.c
@@ -832,7 +832,7 @@  static void msi_domain_update_chip_ops(struct msi_domain_info *info)
 	struct irq_chip *chip = info->chip;
 
 	BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
-	if (!chip->irq_set_affinity)
+	if (!chip->irq_set_affinity && !(info->flags & MSI_FLAG_NO_AFFINITY))
 		chip->irq_set_affinity = msi_domain_set_affinity;
 }