Skip to contentSkip to conversation

irqchip/sg2042-msi: Fix broken affinity setting

PATCH Applied Inochi Amaoto 10 participants 24 messages Aug 13, 2025 – Aug 27, 2025

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v2 cover irqchip/sg2042-msi: Fix broken affinity setting

When using NVME on SG2044, the NVME always complains "I/O tag XXX (XXX) QID XX timeout, completion polled", which is caused by the broken handler of the sg2042-msi driver.

As PLIC driver can only setting affinity when enabling, the sg2042-msi does not properly handled affinity setting previously and enable irq in an unexpected executing path.

Add irq_startup/irq_shutdown support to the PCI template domain, then set irq_chip_[startup/shutdown]_parent for irq_startup/ irq_shutdown of the sg2042-msi driver. So the irq can be started properly.

Change from v1:
1. patch 1: Fix comment format problem, and structure the changelog.
2. patch 2: Improve the comment title and body, add describtion about the fact the PLIC is used as parent chip.
3. patch 2: Remove __always_inline for cond_[shutdown/startup]_parent().
4. patch 3: Update the align of the "XXX_MSI_FLAGS_XXX" macro.
5. patch 4: Claim the fact that the added flag is used by the negotiation of MSI controller driver and PCIe device driver, and can be only used when both of them support this flag.

Inochi Amaoto (4):
genirq: Add irq_chip_(startup/shutdown)_parent() PCI/MSI: Add startup/shutdown for per device domains irqchip/sg2042-msi: Fix broken affinity setting irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044

 drivers/irqchip/irq-sg2042-msi.c | 19 +++++++++---
 drivers/pci/msi/irqdomain.c      | 52 ++++++++++++++++++++++++++++++++
 include/linux/irq.h              |  2 ++
 include/linux/msi.h              |  2 ++
 kernel/irq/chip.c                | 37 +++++++++++++++++++++++
 5 files changed, 107 insertions(+), 5 deletions(-)

--
2.50.1

#1 lore raw

PATCH v2 1/4 genirq: Add irq_chip_(startup/shutdown)_parent()

As the MSI controller on SG2044 uses PLIC as the underlying interrupt controller, it needs to call the irq_enable() and irq_disable() to startup/shutdown irqs. Otherwise, the MSI interrupt can not be startup correctly and will not respond any incoming interrupt.

Introduce helper irq_chip_startup_parent() and irq_chip_shutdown_parent() to allow the interrupt controller to call the irq_startup() or irq_shutdown() of the parent interrupt chip. In case irq_startup() or irq_shutdown() is not implemented for the parent interrupt chip, which will fallback to irq_chip_enable_parent() or irq_chip_disable_parent().

---
 include/linux/irq.h |  2 ++
 kernel/irq/chip.c   | 37 +++++++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+)
include/linux/irq.h +2 -0
index 1d6b606a81ef..890e1371f5d4 100644
@@ -669,6 +669,8 @@ extern int irq_chip_set_parent_state(struct irq_data *data,
 extern int irq_chip_get_parent_state(struct irq_data *data,
 				     enum irqchip_irq_state which,
 				     bool *state);
+extern void irq_chip_shutdown_parent(struct irq_data *data);
+extern unsigned int irq_chip_startup_parent(struct irq_data *data);
 extern void irq_chip_enable_parent(struct irq_data *data);
 extern void irq_chip_disable_parent(struct irq_data *data);
 extern void irq_chip_ack_parent(struct irq_data *data);
kernel/irq/chip.c +37 -0
index 0d0276378c70..3ffa0d80ddd1 100644
@@ -1259,6 +1259,43 @@ int irq_chip_get_parent_state(struct irq_data *data,
 }
 EXPORT_SYMBOL_GPL(irq_chip_get_parent_state);
 
+/**
+ * irq_chip_shutdown_parent - Shutdown the parent interrupt
+ * @data:	Pointer to interrupt specific data
+ *
+ * Invokes the irq_shutdown() callback of the parent if available or falls
+ * back to irq_chip_disable_parent().
+ */
+void irq_chip_shutdown_parent(struct irq_data *data)
+{
+	struct irq_data *parent = data->parent_data;
+
+	if (parent->chip->irq_shutdown)
+		parent->chip->irq_shutdown(parent);
+	else
+		irq_chip_disable_parent(data);
+}
+EXPORT_SYMBOL_GPL(irq_chip_shutdown_parent);
+
+/**
+ * irq_chip_startup_parent - Startup the parent interrupt
+ * @data:	Pointer to interrupt specific data
+ *
+ * Invokes the irq_startup() callback of the parent if available or falls
+ * back to irq_chip_enable_parent().
+ */
+unsigned int irq_chip_startup_parent(struct irq_data *data)
+{
+	struct irq_data *parent = data->parent_data;
+
+	if (parent->chip->irq_startup)
+		return parent->chip->irq_startup(parent);
+
+	irq_chip_enable_parent(data);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(irq_chip_startup_parent);
+
 /**
  * irq_chip_enable_parent - Enable the parent interrupt (defaults to unmask if
  * NULL)
Signature · 1 line
2.50.1
merge bot Applied
#2 lore raw

[tip: irq/core] genirq: Add irq_chip_(startup/shutdown)_parent()

The following commit has been merged into the irq/core branch of tip:

The following commit has been merged into the irq/core branch of tip:

Commit-ID:     7a721a2fee2bce01af26699a87739db8ca8ea3c8
Gitweb:        https://git.kernel.org/tip/7a721a2fee2bce01af26699a87739db8ca8ea3c8
Author:        Inochi Amaoto <[email protected]>
AuthorDate:    Thu, 14 Aug 2025 07:28:31 +08:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Sat, 23 Aug 2025 21:20:25 +02:00

genirq: Add irq_chip_(startup/shutdown)_parent()

As the MSI controller on SG2044 uses PLIC as the underlying interrupt controller, it needs to call irq_enable() and irq_disable() to startup/shutdown interrupts. Otherwise, the MSI interrupt can not be startup correctly and will not respond any incoming interrupt.

Introduce irq_chip_startup_parent() and irq_chip_shutdown_parent() to allow the interrupt controller to call the irq_startup()/irq_shutdown() callbacks of the parent interrupt chip.

In case the irq_startup()/irq_shutdown() callbacks are not implemented for the parent interrupt chip, this will fallback to irq_chip_enable_parent() or irq_chip_disable_parent().

---
 include/linux/irq.h |  2 ++
 kernel/irq/chip.c   | 37 +++++++++++++++++++++++++++++++++++++
 2 files changed, 39 insertions(+)
include/linux/irq.h +2 -0
index c9bcdbf..c67e76f 100644
@@ -669,6 +669,8 @@ extern int irq_chip_set_parent_state(struct irq_data *data,
 extern int irq_chip_get_parent_state(struct irq_data *data,
 				     enum irqchip_irq_state which,
 				     bool *state);
+extern void irq_chip_shutdown_parent(struct irq_data *data);
+extern unsigned int irq_chip_startup_parent(struct irq_data *data);
 extern void irq_chip_enable_parent(struct irq_data *data);
 extern void irq_chip_disable_parent(struct irq_data *data);
 extern void irq_chip_ack_parent(struct irq_data *data);
kernel/irq/chip.c +37 -0
index 0d02763..3ffa0d8 100644
@@ -1260,6 +1260,43 @@ int irq_chip_get_parent_state(struct irq_data *data,
 EXPORT_SYMBOL_GPL(irq_chip_get_parent_state);
 
 /**
+ * irq_chip_shutdown_parent - Shutdown the parent interrupt
+ * @data:	Pointer to interrupt specific data
+ *
+ * Invokes the irq_shutdown() callback of the parent if available or falls
+ * back to irq_chip_disable_parent().
+ */
+void irq_chip_shutdown_parent(struct irq_data *data)
+{
+	struct irq_data *parent = data->parent_data;
+
+	if (parent->chip->irq_shutdown)
+		parent->chip->irq_shutdown(parent);
+	else
+		irq_chip_disable_parent(data);
+}
+EXPORT_SYMBOL_GPL(irq_chip_shutdown_parent);
+
+/**
+ * irq_chip_startup_parent - Startup the parent interrupt
+ * @data:	Pointer to interrupt specific data
+ *
+ * Invokes the irq_startup() callback of the parent if available or falls
+ * back to irq_chip_enable_parent().
+ */
+unsigned int irq_chip_startup_parent(struct irq_data *data)
+{
+	struct irq_data *parent = data->parent_data;
+
+	if (parent->chip->irq_startup)
+		return parent->chip->irq_startup(parent);
+
+	irq_chip_enable_parent(data);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(irq_chip_startup_parent);
+
+/**
  * irq_chip_enable_parent - Enable the parent interrupt (defaults to unmask if
  * NULL)
  * @data:	Pointer to interrupt specific data
#3 lore raw

PATCH v2 2/4 PCI/MSI: Add startup/shutdown for per device domains

In reply to Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

---
 drivers/pci/msi/irqdomain.c | 52 +++++++++++++++++++++++++++++++++++++
 include/linux/msi.h         |  2 ++
 2 files changed, 54 insertions(+)
drivers/pci/msi/irqdomain.c +52 -0
index 0938ef7ebabf..e0a800f918e8 100644
@@ -148,6 +148,23 @@ static void pci_device_domain_set_desc(msi_alloc_info_t *arg, struct msi_desc *d
 	arg->hwirq = desc->msi_index;
 }
 
+static void cond_shutdown_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		irq_chip_shutdown_parent(data);
+}
+
+static unsigned int cond_startup_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		return irq_chip_startup_parent(data);
+	return 0;
+}
+
 static __always_inline void cond_mask_parent(struct irq_data *data)
 {
 	struct msi_domain_info *info = data->domain->host_data;
@@ -164,6 +181,23 @@ static __always_inline void cond_unmask_parent(struct irq_data *data)
 		irq_chip_unmask_parent(data);
 }
 
+static void pci_irq_shutdown_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+
+	pci_msi_mask(desc, BIT(data->irq - desc->irq));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msi_unmask(desc, BIT(data->irq - desc->irq));
+	return ret;
+}
+
 static void pci_irq_mask_msi(struct irq_data *data)
 {
 	struct msi_desc *desc = irq_data_get_msi_desc(data);
@@ -194,6 +228,8 @@ static void pci_irq_unmask_msi(struct irq_data *data)
 static const struct msi_domain_template pci_msi_template = {
 	.chip = {
 		.name			= "PCI-MSI",
+		.irq_startup		= pci_irq_startup_msi,
+		.irq_shutdown		= pci_irq_shutdown_msi,
 		.irq_mask		= pci_irq_mask_msi,
 		.irq_unmask		= pci_irq_unmask_msi,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
@@ -210,6 +246,20 @@ static const struct msi_domain_template pci_msi_template = {
 	},
 };
 
+static void pci_irq_shutdown_msix(struct irq_data *data)
+{
+	pci_msix_mask(irq_data_get_msi_desc(data));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msix(struct irq_data *data)
+{
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msix_unmask(irq_data_get_msi_desc(data));
+	return ret;
+}
+
 static void pci_irq_mask_msix(struct irq_data *data)
 {
 	pci_msix_mask(irq_data_get_msi_desc(data));
@@ -234,6 +284,8 @@ EXPORT_SYMBOL_GPL(pci_msix_prepare_desc);
 static const struct msi_domain_template pci_msix_template = {
 	.chip = {
 		.name			= "PCI-MSIX",
+		.irq_startup		= pci_irq_startup_msix,
+		.irq_shutdown		= pci_irq_shutdown_msix,
 		.irq_mask		= pci_irq_mask_msix,
 		.irq_unmask		= pci_irq_unmask_msix,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
include/linux/msi.h +2 -0
index e5e86a8529fb..3111ba95fbde 100644
@@ -568,6 +568,8 @@ enum {
 	MSI_FLAG_PARENT_PM_DEV		= (1 << 8),
 	/* Support for parent mask/unmask */
 	MSI_FLAG_PCI_MSI_MASK_PARENT	= (1 << 9),
+	/* Support for parent startup/shutdown */
+	MSI_FLAG_PCI_MSI_STARTUP_PARENT	= (1 << 10),
 
 	/* Mask for the generic functionality */
 	MSI_GENERIC_FLAGS_MASK		= GENMASK(15, 0),
Signature · 1 line
2.50.1
Acked-by on v2 2/4
#4 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 3 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

s/unavailble/unavailable/ (mentioned previously)

Quoted 6 lines

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

s/irq/IRQ/ a couple times above

Quoted 2 lines

Thomas, I assume you'll merge this series; let me know if not.

Quoted 112 lines
---
 drivers/pci/msi/irqdomain.c | 52 +++++++++++++++++++++++++++++++++++++
 include/linux/msi.h         |  2 ++
 2 files changed, 54 insertions(+)
drivers/pci/msi/irqdomain.c +52 -0
index 0938ef7ebabf..e0a800f918e8 100644
@@ -148,6 +148,23 @@ static void pci_device_domain_set_desc(msi_alloc_info_t *arg, struct msi_desc *d
 	arg->hwirq = desc->msi_index;
 }
 
+static void cond_shutdown_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		irq_chip_shutdown_parent(data);
+}
+
+static unsigned int cond_startup_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		return irq_chip_startup_parent(data);
+	return 0;
+}
+
 static __always_inline void cond_mask_parent(struct irq_data *data)
 {
 	struct msi_domain_info *info = data->domain->host_data;
@@ -164,6 +181,23 @@ static __always_inline void cond_unmask_parent(struct irq_data *data)
 		irq_chip_unmask_parent(data);
 }
 
+static void pci_irq_shutdown_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+
+	pci_msi_mask(desc, BIT(data->irq - desc->irq));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msi_unmask(desc, BIT(data->irq - desc->irq));
+	return ret;
+}
+
 static void pci_irq_mask_msi(struct irq_data *data)
 {
 	struct msi_desc *desc = irq_data_get_msi_desc(data);
@@ -194,6 +228,8 @@ static void pci_irq_unmask_msi(struct irq_data *data)
 static const struct msi_domain_template pci_msi_template = {
 	.chip = {
 		.name			= "PCI-MSI",
+		.irq_startup		= pci_irq_startup_msi,
+		.irq_shutdown		= pci_irq_shutdown_msi,
 		.irq_mask		= pci_irq_mask_msi,
 		.irq_unmask		= pci_irq_unmask_msi,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
@@ -210,6 +246,20 @@ static const struct msi_domain_template pci_msi_template = {
 	},
 };
 
+static void pci_irq_shutdown_msix(struct irq_data *data)
+{
+	pci_msix_mask(irq_data_get_msi_desc(data));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msix(struct irq_data *data)
+{
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msix_unmask(irq_data_get_msi_desc(data));
+	return ret;
+}
+
 static void pci_irq_mask_msix(struct irq_data *data)
 {
 	pci_msix_mask(irq_data_get_msi_desc(data));
@@ -234,6 +284,8 @@ EXPORT_SYMBOL_GPL(pci_msix_prepare_desc);
 static const struct msi_domain_template pci_msix_template = {
 	.chip = {
 		.name			= "PCI-MSIX",
+		.irq_startup		= pci_irq_startup_msix,
+		.irq_shutdown		= pci_irq_shutdown_msix,
 		.irq_mask		= pci_irq_mask_msix,
 		.irq_unmask		= pci_irq_unmask_msix,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
include/linux/msi.h +2 -0
index e5e86a8529fb..3111ba95fbde 100644
@@ -568,6 +568,8 @@ enum {
 	MSI_FLAG_PARENT_PM_DEV		= (1 << 8),
 	/* Support for parent mask/unmask */
 	MSI_FLAG_PCI_MSI_MASK_PARENT	= (1 << 9),
+	/* Support for parent startup/shutdown */
+	MSI_FLAG_PCI_MSI_STARTUP_PARENT	= (1 << 10),
 
 	/* Mask for the generic functionality */
 	MSI_GENERIC_FLAGS_MASK		= GENMASK(15, 0),
Signature · 1 line
2.50.1
#5 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 14 lines · Bjorn Helgaas
Quoted 3 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

s/unavailble/unavailable/ (mentioned previously)

Quoted 6 lines

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

Quoted 8 lines

s/irq/IRQ/ a couple times above

Quoted 2 lines

Thomas, I assume you'll merge this series; let me know if not.

I'll pick it up and fixup the wording as I go.

merge bot Applied
#6 lore raw

[tip: irq/drivers] PCI/MSI: Add startup/shutdown for per device domains

In reply to Inochi Amaoto
The following commit has been merged into the irq/drivers branch of tip:

The following commit has been merged into the irq/drivers branch of tip:

Commit-ID:     54f45a30c0d0153d2be091ba2d683ab6db6d1d5b
Gitweb:        https://git.kernel.org/tip/54f45a30c0d0153d2be091ba2d683ab6db6d1d5b
Author:        Inochi Amaoto <[email protected]>
AuthorDate:    Thu, 14 Aug 2025 07:28:32 +08:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Sat, 23 Aug 2025 21:21:13 +02:00

PCI/MSI: Add startup/shutdown for per device domains

As the RISC-V PLIC cannot apply affinity settings without invoking irq_enable(), it will make the interrupt unavailble when used as an underlying interrupt chip for the MSI controller.

Implement the irq_startup() and irq_shutdown() callbacks for the PCI MSI and MSI-X templates.

For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, the parent startup and shutdown functions are invoked. That allows the interrupt on the parent chip to be enabled if the interrupt has not been enabled during allocation. This is necessary for MSI controllers which use PLIC as underlying parent interrupt chip.

---
 drivers/pci/msi/irqdomain.c | 52 ++++++++++++++++++++++++++++++++++++-
 include/linux/msi.h         |  2 +-
 2 files changed, 54 insertions(+)
drivers/pci/msi/irqdomain.c +52 -0
index 0938ef7..e0a800f 100644
@@ -148,6 +148,23 @@ static void pci_device_domain_set_desc(msi_alloc_info_t *arg, struct msi_desc *d
 	arg->hwirq = desc->msi_index;
 }
 
+static void cond_shutdown_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		irq_chip_shutdown_parent(data);
+}
+
+static unsigned int cond_startup_parent(struct irq_data *data)
+{
+	struct msi_domain_info *info = data->domain->host_data;
+
+	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
+		return irq_chip_startup_parent(data);
+	return 0;
+}
+
 static __always_inline void cond_mask_parent(struct irq_data *data)
 {
 	struct msi_domain_info *info = data->domain->host_data;
@@ -164,6 +181,23 @@ static __always_inline void cond_unmask_parent(struct irq_data *data)
 		irq_chip_unmask_parent(data);
 }
 
+static void pci_irq_shutdown_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+
+	pci_msi_mask(desc, BIT(data->irq - desc->irq));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msi(struct irq_data *data)
+{
+	struct msi_desc *desc = irq_data_get_msi_desc(data);
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msi_unmask(desc, BIT(data->irq - desc->irq));
+	return ret;
+}
+
 static void pci_irq_mask_msi(struct irq_data *data)
 {
 	struct msi_desc *desc = irq_data_get_msi_desc(data);
@@ -194,6 +228,8 @@ static void pci_irq_unmask_msi(struct irq_data *data)
 static const struct msi_domain_template pci_msi_template = {
 	.chip = {
 		.name			= "PCI-MSI",
+		.irq_startup		= pci_irq_startup_msi,
+		.irq_shutdown		= pci_irq_shutdown_msi,
 		.irq_mask		= pci_irq_mask_msi,
 		.irq_unmask		= pci_irq_unmask_msi,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
@@ -210,6 +246,20 @@ static const struct msi_domain_template pci_msi_template = {
 	},
 };
 
+static void pci_irq_shutdown_msix(struct irq_data *data)
+{
+	pci_msix_mask(irq_data_get_msi_desc(data));
+	cond_shutdown_parent(data);
+}
+
+static unsigned int pci_irq_startup_msix(struct irq_data *data)
+{
+	unsigned int ret = cond_startup_parent(data);
+
+	pci_msix_unmask(irq_data_get_msi_desc(data));
+	return ret;
+}
+
 static void pci_irq_mask_msix(struct irq_data *data)
 {
 	pci_msix_mask(irq_data_get_msi_desc(data));
@@ -234,6 +284,8 @@ EXPORT_SYMBOL_GPL(pci_msix_prepare_desc);
 static const struct msi_domain_template pci_msix_template = {
 	.chip = {
 		.name			= "PCI-MSIX",
+		.irq_startup		= pci_irq_startup_msix,
+		.irq_shutdown		= pci_irq_shutdown_msix,
 		.irq_mask		= pci_irq_mask_msix,
 		.irq_unmask		= pci_irq_unmask_msix,
 		.irq_write_msi_msg	= pci_msi_domain_write_msg,
include/linux/msi.h +2 -0
index e5e86a8..3111ba9 100644
@@ -568,6 +568,8 @@ enum {
 	MSI_FLAG_PARENT_PM_DEV		= (1 << 8),
 	/* Support for parent mask/unmask */
 	MSI_FLAG_PCI_MSI_MASK_PARENT	= (1 << 9),
+	/* Support for parent startup/shutdown */
+	MSI_FLAG_PCI_MSI_STARTUP_PARENT	= (1 << 10),
 
 	/* Mask for the generic functionality */
 	MSI_GENERIC_FLAGS_MASK		= GENMASK(15, 0),
#7 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

In reply to Inochi Amaoto
Quoted 11 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

Quoted 2 lines

Regressions found while booting the Linux next-20250826 on the qemu-arm64, qemu-armv7 due to following kernel log.

Bisection identified this commit as the cause of the regression.

Regression Analysis:
- New regression? Yes
- Reproducible? Yes

First seen on the next-20250826
Good: next-20250825
Bad: next-20250826

Test regression: next-20250826 gcc-13 boot failed on qemu-arm64 and qemu-armv7.

Expected behavior: System should boot normally and virtio block devices should be detected and initialized immediately.

Actual behavior: System hangs for ~30 seconds during virtio block device initialization before showing scheduler deadline replenish errors and failing to complete boot.

[...]

<6>[    1.369038] virtio-pci 0000:00:01.0: enabling device (0000 ->
0003)
<6>[    1.420097] Serial: 8250/16550 driver, 4 ports, IRQ sharing
enabled
<6>[    1.450858] msm_serial: driver initialized
<6>[    1.454489] SuperH (H)SCI(F) driver initialized
<6>[    1.456056] STM32 USART driver initialized
<6>[    1.513325] loop: module loaded
<6>[    1.515744] virtio_blk virtio0: 2/0/0 default/read/poll queues
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much

[here it hangs]

Reverting this commit restores normal boot behavior.

qemu-arm64
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/log

qemu-armv7
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663615/suite/boot/test/gcc-13-lkftconfig/log

## Source
* Git tree:
* https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next.git
* Git sha: d0630b758e593506126e8eda6c3d56097d1847c5
* Git describe: next-20250826
* Project details: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826
* Architectures: arm64
* Toolchains: gcc-13
* Kconfigs: gcc-13-lkftconfig

## Build
* Test history: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/history/
* Test link: https://tuxapi.tuxsuite.com/v1/groups/linaro/projects/lkft/tests/31oo1cMOi0uSNKYApi80iQahbLi
* Build link: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/
* Kernel config: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/config

--
Linaro LKFT
https://lkft.linaro.org

#8 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 38 lines · Anders Roxell

Regressions found while booting the Linux next-20250826 on the qemu-arm64, qemu-armv7 due to following kernel log.

Bisection identified this commit as the cause of the regression.

Regression Analysis:
- New regression? Yes
- Reproducible? Yes

First seen on the next-20250826
Good: next-20250825
Bad: next-20250826

Test regression: next-20250826 gcc-13 boot failed on qemu-arm64 and qemu-armv7.

Expected behavior: System should boot normally and virtio block devices should be detected and initialized immediately.

Actual behavior: System hangs for ~30 seconds during virtio block device initialization before showing scheduler deadline replenish errors and failing to complete boot.

[...]

<6>[    1.369038] virtio-pci 0000:00:01.0: enabling device (0000 ->
0003)
<6>[    1.420097] Serial: 8250/16550 driver, 4 ports, IRQ sharing
enabled
<6>[    1.450858] msm_serial: driver initialized
<6>[    1.454489] SuperH (H)SCI(F) driver initialized
<6>[    1.456056] STM32 USART driver initialized
<6>[    1.513325] loop: module loaded
<6>[    1.515744] virtio_blk virtio0: 2/0/0 default/read/poll queues
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much
Quoted 1 line

[here it hangs]

FWIW, I am also seeing this on real arm64 hardware (an LX2160A board and an Ampere Altra one) but with my NVMe drives failing to be recognized. In somewhat ironic fashion, I am seeing the message from cover letter repeating.

nvme nvme0: I/O tag 8 (1008) QID 0 timeout, completion polled

  [  125.810062] dracut-initqueue[640]: Timed out while waiting for udev queue to empty.
  nvme nvme0: I/O tag 9 (1009) QID 0 timeout, completion polled

I am happy to test patches or provide information.

Cheers,
Nathan

# bad: [d0630b758e593506126e8eda6c3d56097d1847c5] Add linux-next specific files for 20250826
# good: [b6add54ba61890450fa54fd9327d10fdfd653439] Merge tag 'pinctrl-v6.17-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl
git bisect start 'd0630b758e593506126e8eda6c3d56097d1847c5' 'b6add54ba61890450fa54fd9327d10fdfd653439'
# good: [968d16786392f6e047329f5eff66acc131636019] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma.git
git bisect good 968d16786392f6e047329f5eff66acc131636019
# good: [042e9f528d5362c499b5d8e2716cf6f64ca53add] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394.git
git bisect good 042e9f528d5362c499b5d8e2716cf6f64ca53add
# bad: [beebb75399dc36e7c244db0a08426053b4581ecc] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git
git bisect bad beebb75399dc36e7c244db0a08426053b4581ecc
# good: [62df8fb299358a45a915381de09025cf5e6a4a8f] Merge branch 'for-next' of https://git.kernel.org/pub/scm/linux/kernel/git/robh/linux.git
git bisect good 62df8fb299358a45a915381de09025cf5e6a4a8f
# bad: [1e6d2dcb13c8d94b56de1eff60235ca90587046b] Merge branch 'master' of https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git
git bisect bad 1e6d2dcb13c8d94b56de1eff60235ca90587046b
# bad: [a0daa9e939dbcd7767090151771d94ade75a4fd5] Merge branch into tip/master: 'x86/build'
git bisect bad a0daa9e939dbcd7767090151771d94ade75a4fd5
# bad: [d147a3db0dfa15c8e460f007128bd0fe2e1b877f] Merge branch into tip/master: 'perf/core'
git bisect bad d147a3db0dfa15c8e460f007128bd0fe2e1b877f
# good: [be5697d7136525a91e7f30fdca2e7de737d9a8ed] Merge branch into tip/master: 'irq/core'
git bisect good be5697d7136525a91e7f30fdca2e7de737d9a8ed
# good: [5d299897f1e36025400ca84fd36c15925a383b03] perf: Split out the RB allocation
git bisect good 5d299897f1e36025400ca84fd36c15925a383b03
# bad: [7fb83eb664e9b3a0438dd28859e9f0fd49d4c165] irqchip/loongson-eiointc: Route interrupt parsed from bios table
git bisect bad 7fb83eb664e9b3a0438dd28859e9f0fd49d4c165
# bad: [7ee4a5a2ec3748facfb4ca96e4cce6cabbdecab2] irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044
git bisect bad 7ee4a5a2ec3748facfb4ca96e4cce6cabbdecab2
# bad: [9d8c41816bac518b4824f83b346ae30a1be83f68] irqchip/sg2042-msi: Fix broken affinity setting
git bisect bad 9d8c41816bac518b4824f83b346ae30a1be83f68
# bad: [54f45a30c0d0153d2be091ba2d683ab6db6d1d5b] PCI/MSI: Add startup/shutdown for per device domains
git bisect bad 54f45a30c0d0153d2be091ba2d683ab6db6d1d5b

# first bad commit: [54f45a30c0d0153d2be091ba2d683ab6db6d1d5b] PCI/MSI: Add startup/shutdown for per device domains
maintainer
#9 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 12 lines · Nathan Chancellor
Quoted 4 lines · Anders Roxell
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much

[here it hangs]

FWIW, I am also seeing this on real arm64 hardware (an LX2160A board and an Ampere Altra one) but with my NVMe drives failing to be recognized. In somewhat ironic fashion, I am seeing the message from cover letter repeating.

Quoted 5 lines

nvme nvme0: I/O tag 8 (1008) QID 0 timeout, completion polled

  [  125.810062] dracut-initqueue[640]: Timed out while waiting for udev queue to empty.
  nvme nvme0: I/O tag 9 (1009) QID 0 timeout, completion polled

I am happy to test patches or provide information.

Same here, it's breaking at least Orion O6.

#10 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

In reply to Anders Roxell
Quoted 76 lines · Anders Roxell
Quoted 11 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

Quoted 2 lines

Regressions found while booting the Linux next-20250826 on the qemu-arm64, qemu-armv7 due to following kernel log.

Bisection identified this commit as the cause of the regression.

Regression Analysis:
- New regression? Yes
- Reproducible? Yes

First seen on the next-20250826
Good: next-20250825
Bad: next-20250826

Test regression: next-20250826 gcc-13 boot failed on qemu-arm64 and qemu-armv7.

Expected behavior: System should boot normally and virtio block devices should be detected and initialized immediately.

Actual behavior: System hangs for ~30 seconds during virtio block device initialization before showing scheduler deadline replenish errors and failing to complete boot.

[...]

<6>[    1.369038] virtio-pci 0000:00:01.0: enabling device (0000 ->
0003)
<6>[    1.420097] Serial: 8250/16550 driver, 4 ports, IRQ sharing
enabled
<6>[    1.450858] msm_serial: driver initialized
<6>[    1.454489] SuperH (H)SCI(F) driver initialized
<6>[    1.456056] STM32 USART driver initialized
<6>[    1.513325] loop: module loaded
<6>[    1.515744] virtio_blk virtio0: 2/0/0 default/read/poll queues
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much

[here it hangs]

Reverting this commit restores normal boot behavior.

qemu-arm64
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/log

qemu-armv7
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663615/suite/boot/test/gcc-13-lkftconfig/log

## Source
* Git tree:
* https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next.git
* Git sha: d0630b758e593506126e8eda6c3d56097d1847c5
* Git describe: next-20250826
* Project details: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826
* Architectures: arm64
* Toolchains: gcc-13
* Kconfigs: gcc-13-lkftconfig

Quoted 6 lines

## Build
* Test history: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/history/
* Test link: https://tuxapi.tuxsuite.com/v1/groups/linaro/projects/lkft/tests/31oo1cMOi0uSNKYApi80iQahbLi
* Build link: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/
* Kernel config: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/config

Is there a link for me to get the command line args for qemu? So I can reproduce it locally.

Regards,
Inochi

#11 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 84 lines · Inochi Amaoto
Quoted 76 lines · Anders Roxell
Quoted 11 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

Quoted 2 lines

Regressions found while booting the Linux next-20250826 on the qemu-arm64, qemu-armv7 due to following kernel log.

Bisection identified this commit as the cause of the regression.

Regression Analysis:
- New regression? Yes
- Reproducible? Yes

First seen on the next-20250826
Good: next-20250825
Bad: next-20250826

Test regression: next-20250826 gcc-13 boot failed on qemu-arm64 and qemu-armv7.

Expected behavior: System should boot normally and virtio block devices should be detected and initialized immediately.

Actual behavior: System hangs for ~30 seconds during virtio block device initialization before showing scheduler deadline replenish errors and failing to complete boot.

[...]

<6>[    1.369038] virtio-pci 0000:00:01.0: enabling device (0000 ->
0003)
<6>[    1.420097] Serial: 8250/16550 driver, 4 ports, IRQ sharing
enabled
<6>[    1.450858] msm_serial: driver initialized
<6>[    1.454489] SuperH (H)SCI(F) driver initialized
<6>[    1.456056] STM32 USART driver initialized
<6>[    1.513325] loop: module loaded
<6>[    1.515744] virtio_blk virtio0: 2/0/0 default/read/poll queues
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much

[here it hangs]

Reverting this commit restores normal boot behavior.

qemu-arm64
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/log

qemu-armv7
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663615/suite/boot/test/gcc-13-lkftconfig/log

## Source
* Git tree:
* https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next.git
* Git sha: d0630b758e593506126e8eda6c3d56097d1847c5
* Git describe: next-20250826
* Project details: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826
* Architectures: arm64
* Toolchains: gcc-13
* Kconfigs: gcc-13-lkftconfig

Quoted 6 lines

## Build
* Test history: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/history/
* Test link: https://tuxapi.tuxsuite.com/v1/groups/linaro/projects/lkft/tests/31oo1cMOi0uSNKYApi80iQahbLi
* Build link: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/
* Kernel config: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/config

Quoted 3 lines

Is there a link for me to get the command line args for qemu? So I can reproduce it locally.

OK, I guess I know why: I have missed one condition for startup.

Could you test the following patch? If worked, I will send it as a fix.

---

drivers/pci/msi/irqdomain.c +5 -0
index e0a800f918e8..b11b7f63f0d6 100644
@@ -154,6 +154,8 @@ static void cond_shutdown_parent(struct irq_data *data)
 
 	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
 		irq_chip_shutdown_parent(data);
+	else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+		irq_chip_mask_parent(data);
 }
 
 static unsigned int cond_startup_parent(struct irq_data *data)
@@ -162,6 +164,9 @@ static unsigned int cond_startup_parent(struct irq_data *data)
 
 	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
 		return irq_chip_startup_parent(data);
+	else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+		irq_chip_unmask_parent(data);
+
 	return 0;
 }
 
#12 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 4 lines · Inochi Amaoto

OK, I guess I know why: I have missed one condition for startup.

Could you test the following patch? If worked, I will send it as a fix.

Yes, that appears to resolve the issue on one system. I cannot test the other at the moment since it is under load.

Quoted 24 lines

---

drivers/pci/msi/irqdomain.c +5 -0
index e0a800f918e8..b11b7f63f0d6 100644
@@ -154,6 +154,8 @@ static void cond_shutdown_parent(struct irq_data *data)
 
 	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
 		irq_chip_shutdown_parent(data);
+	else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+		irq_chip_mask_parent(data);
 }
 
 static unsigned int cond_startup_parent(struct irq_data *data)
@@ -162,6 +164,9 @@ static unsigned int cond_startup_parent(struct irq_data *data)
 
 	if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
 		return irq_chip_startup_parent(data);
+	else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+		irq_chip_unmask_parent(data);
+
 	return 0;
 }
 
Tested-by Linux Kernel Functional Testing
#13 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 9 lines · Nathan Chancellor
Quoted 4 lines · Inochi Amaoto

OK, I guess I know why: I have missed one condition for startup.

Could you test the following patch? If worked, I will send it as a fix.

Yes, that appears to resolve the issue on one system. I cannot test the other at the moment since it is under load.

I have built on top of Linux next-20250826 tag and the qemu-arm64 boot test pass and LTP smoke test also pass.

Quoted 2 lines
Quoted 25 lines
Quoted 25 lines

---

drivers/pci/msi/irqdomain.c +5 -0
index e0a800f918e8..b11b7f63f0d6 100644
@@ -154,6 +154,8 @@ static void cond_shutdown_parent(struct irq_data *data)
 
      if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
              irq_chip_shutdown_parent(data);
+     else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+             irq_chip_mask_parent(data);
 }
 
 static unsigned int cond_startup_parent(struct irq_data *data)
@@ -162,6 +164,9 @@ static unsigned int cond_startup_parent(struct irq_data *data)
 
      if (unlikely(info->flags & MSI_FLAG_PCI_MSI_STARTUP_PARENT))
              return irq_chip_startup_parent(data);
+     else if (unlikely(info->flags & MSI_FLAG_PCI_MSI_MASK_PARENT))
+             irq_chip_unmask_parent(data);
+
      return 0;
 }
 
#14 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 11 lines · Naresh Kamboju
Quoted 9 lines · Nathan Chancellor
Quoted 4 lines · Inochi Amaoto

OK, I guess I know why: I have missed one condition for startup.

Could you test the following patch? If worked, I will send it as a fix.

Yes, that appears to resolve the issue on one system. I cannot test the other at the moment since it is under load.

Quoted 8 lines

I have built on top of Linux next-20250826 tag and the qemu-arm64 boot test pass and LTP smoke test also pass.

Quoted 2 lines

Thanks for your tag, can you resend you tag to the following url? I have sent a fix patch here. Thanks.

https://lore.kernel.org/all/[email protected]/

Regards,
Inochi

#15 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

In reply to Anders Roxell
Quoted 82 lines · Anders Roxell
Quoted 11 lines · Inochi Amaoto

As the RISC-V PLIC can not apply affinity setting without calling irq_enable(), it will make the interrupt unavailble when using as an underlying IRQ chip for MSI controller.

Implement .irq_startup() and .irq_shutdown() for the PCI MSI and MSI-X templates. For chips that specify MSI_FLAG_PCI_MSI_STARTUP_PARENT, these startup and shutdown the parent as well, which allows the irq on the parent chip to be enabled if the irq is not enabled when allocating. This is necessary for the MSI controllers which use PLIC as underlying IRQ chip.

Quoted 2 lines

Regressions found while booting the Linux next-20250826 on the qemu-arm64, qemu-armv7 due to following kernel log.

Bisection identified this commit as the cause of the regression.

Regression Analysis:
- New regression? Yes
- Reproducible? Yes

First seen on the next-20250826
Good: next-20250825
Bad: next-20250826

Test regression: next-20250826 gcc-13 boot failed on qemu-arm64 and qemu-armv7.

Expected behavior: System should boot normally and virtio block devices should be detected and initialized immediately.

Actual behavior: System hangs for ~30 seconds during virtio block device initialization before showing scheduler deadline replenish errors and failing to complete boot.

[...]

<6>[    1.369038] virtio-pci 0000:00:01.0: enabling device (0000 ->
0003)
<6>[    1.420097] Serial: 8250/16550 driver, 4 ports, IRQ sharing
enabled
<6>[    1.450858] msm_serial: driver initialized
<6>[    1.454489] SuperH (H)SCI(F) driver initialized
<6>[    1.456056] STM32 USART driver initialized
<6>[    1.513325] loop: module loaded
<6>[    1.515744] virtio_blk virtio0: 2/0/0 default/read/poll queues
<5>[    1.527859] virtio_blk virtio0: [vda] 5397504 512-byte logical
blocks (2.76 GB/2.57 GiB)
<4>[   29.761219] sched: DL replenish lagged too much

[here it hangs]

Reverting this commit restores normal boot behavior.

qemu-arm64
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/log

qemu-armv7
- https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663615/suite/boot/test/gcc-13-lkftconfig/log

## Source
* Git tree:
* https://kernel.googlesource.com/pub/scm/linux/kernel/git/next/linux-next.git
* Git sha: d0630b758e593506126e8eda6c3d56097d1847c5
* Git describe: next-20250826
* Project details: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826
* Architectures: arm64
* Toolchains: gcc-13
* Kconfigs: gcc-13-lkftconfig

## Build
* Test history: https://qa-reports.linaro.org/lkft/linux-next-master/build/next-20250826/testrun/29663822/suite/boot/test/gcc-13-lkftconfig/history/
* Test link: https://tuxapi.tuxsuite.com/v1/groups/linaro/projects/lkft/tests/31oo1cMOi0uSNKYApi80iQahbLi
* Build link: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/
* Kernel config: https://storage.tuxsuite.com/public/linaro/lkft/builds/31onzS5UmJVvvZucEhtB1veoJA1/config

Quoted 3 lines

--
Linaro LKFT
https://lkft.linaro.org

Fix patch is here:

https://lore.kernel.org/all/[email protected]/

Regards,
Inochi

#16 lore raw

PATCH v2 2/4 PCI/MSI: Add startup/shutdown for per device domains

In reply to Inochi Amaoto

We found an issue that the ENETC network port of our i.MX95 platform (arm64) does not work based the latest linux-next tree. According to my observation, the MSI-X interrupts statistics from
"cat /proc/interrupts" are all 0.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5

So I reverted this patch and then the MSI-X interrupts return to normal.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:       4365          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0        194          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0        227          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0        219          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0        176          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0        233 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5

It looks like that this patch causes this issue, but I don't know about the PCI MSI driver, so please help investigate this issue, thanks.

#17 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

Quoted 24 lines · Wei Fang

We found an issue that the ENETC network port of our i.MX95 platform (arm64) does not work based the latest linux-next tree. According to my observation, the MSI-X interrupts statistics from
"cat /proc/interrupts" are all 0.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5

So I reverted this patch and then the MSI-X interrupts return to normal.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:       4365          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0        194          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0        227          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0        219          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0        176          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0        233 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5
Quoted 2 lines

It looks like that this patch causes this issue, but I don't know about the PCI MSI driver, so please help investigate this issue, thanks.

Some people reported a similar issue, check the whole mail thread and inochi has provided a fixing patch, you can try it out.

Thanks,

Chen

#18 lore raw

Re: [PATCH v2 2/4] PCI/MSI: Add startup/shutdown for per device domains

In reply to Wei Fang
Quoted 24 lines · Wei Fang

We found an issue that the ENETC network port of our i.MX95 platform (arm64) does not work based the latest linux-next tree. According to my observation, the MSI-X interrupts statistics from
"cat /proc/interrupts" are all 0.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5

So I reverted this patch and then the MSI-X interrupts return to normal.

root@imx95evk:~# cat /proc/interrupts | grep eth0
123:       4365          0          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   1 Edge      eth0-rxtx0
124:          0        194          0          0          0          0 ITS-PCI-MSIX-0002:00:00.0   2 Edge      eth0-rxtx1
125:          0          0        227          0          0          0 ITS-PCI-MSIX-0002:00:00.0   3 Edge      eth0-rxtx2
126:          0          0          0        219          0          0 ITS-PCI-MSIX-0002:00:00.0   4 Edge      eth0-rxtx3
127:          0          0          0          0        176          0 ITS-PCI-MSIX-0002:00:00.0   5 Edge      eth0-rxtx4
128:          0          0          0          0          0        233 ITS-PCI-MSIX-0002:00:00.0   6 Edge      eth0-rxtx5
Quoted 3 lines

It looks like that this patch causes this issue, but I don't know about the PCI MSI driver, so please help investigate this issue, thanks.

Can you try the following patch?

https://lore.kernel.org/all/[email protected]/

Regards,
Inochi

#19 lore raw

PATCH v2 3/4 irqchip/sg2042-msi: Fix broken affinity setting

In reply to Inochi Amaoto

When using NVME on SG2044, the NVME always complains "I/O tag XXX (XXX) QID XX timeout, completion polled", which is caused by the broken handler of the sg2042-msi driver.

As PLIC driver can only set affinity when enabling, the sg2042-msi does not properly handled affinity setting previously and enables irq in an unexpected executing path.

Since the PCI template domain supports irq_startup()/irq_shutdown(), set irq_chip_[startup/shutdown]_parent() for irq_startup() and irq_shutdown(). So the irq can be started properly.

Fixes e96b93a97c90 ("irqchip/sg2042-msi: Add the Sophgo SG2044 MSI interrupt controller")
---
 drivers/irqchip/irq-sg2042-msi.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
drivers/irqchip/irq-sg2042-msi.c +13 -5
index bcfddc51bc6a..2fd4d94f9bd7 100644
@@ -85,6 +85,8 @@ static void sg2042_msi_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *m
 
 static const struct irq_chip sg2042_msi_middle_irq_chip = {
 	.name			= "SG2042 MSI",
+	.irq_startup		= irq_chip_startup_parent,
+	.irq_shutdown		= irq_chip_shutdown_parent,
 	.irq_ack		= sg2042_msi_irq_ack,
 	.irq_mask		= irq_chip_mask_parent,
 	.irq_unmask		= irq_chip_unmask_parent,
@@ -114,6 +116,8 @@ static void sg2044_msi_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *m
 
 static struct irq_chip sg2044_msi_middle_irq_chip = {
 	.name			= "SG2044 MSI",
+	.irq_startup		= irq_chip_startup_parent,
+	.irq_shutdown		= irq_chip_shutdown_parent,
 	.irq_ack		= sg2044_msi_irq_ack,
 	.irq_mask		= irq_chip_mask_parent,
 	.irq_unmask		= irq_chip_unmask_parent,
@@ -185,8 +189,10 @@ static const struct irq_domain_ops sg204x_msi_middle_domain_ops = {
 	.select	= msi_lib_irq_domain_select,
 };
 
-#define SG2042_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |	\
-				   MSI_FLAG_USE_DEF_CHIP_OPS)
+#define SG2042_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |		\
+				   MSI_FLAG_USE_DEF_CHIP_OPS |		\
+				   MSI_FLAG_PCI_MSI_MASK_PARENT |	\
+				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
 #define SG2042_MSI_FLAGS_SUPPORTED MSI_GENERIC_FLAGS_MASK
 
@@ -200,10 +206,12 @@ static const struct msi_parent_ops sg2042_msi_parent_ops = {
 	.init_dev_msi_info	= msi_lib_init_dev_msi_info,
 };
 
-#define SG2044_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |	\
-				   MSI_FLAG_USE_DEF_CHIP_OPS)
+#define SG2044_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |		\
+				   MSI_FLAG_USE_DEF_CHIP_OPS |		\
+				   MSI_FLAG_PCI_MSI_MASK_PARENT |	\
+				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
-#define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |	\
+#define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |		\
 				    MSI_FLAG_PCI_MSIX)
 
 static const struct msi_parent_ops sg2044_msi_parent_ops = {
Signature · 1 line
2.50.1
merge bot Applied
#20 lore raw
The following commit has been merged into the irq/drivers branch of tip:

The following commit has been merged into the irq/drivers branch of tip:

Commit-ID:     9d8c41816bac518b4824f83b346ae30a1be83f68
Gitweb:        https://git.kernel.org/tip/9d8c41816bac518b4824f83b346ae30a1be83f68
Author:        Inochi Amaoto <[email protected]>
AuthorDate:    Thu, 14 Aug 2025 07:28:33 +08:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Sat, 23 Aug 2025 21:21:13 +02:00

irqchip/sg2042-msi: Fix broken affinity setting

When using NVME on SG2044, the NVME drvier always complains about "I/O tag XXX (XXX) QID XX timeout, completion polled", which is caused by the broken affinity setting mechanism of the sg2042-msi driver.

The PLIC driver can only the set the affinity when enabled, but the sg2042-msi driver invokes the affinity setter in disabled state, which causes the change to be lost.

Cure this by implementing the irq_startup()/shutdown() callbacks, which allow to startup (enabled) the underlying PLIC first.

---
 drivers/irqchip/irq-sg2042-msi.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
drivers/irqchip/irq-sg2042-msi.c +13 -5
index bcfddc5..2fd4d94 100644
@@ -85,6 +85,8 @@ static void sg2042_msi_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *m
 
 static const struct irq_chip sg2042_msi_middle_irq_chip = {
 	.name			= "SG2042 MSI",
+	.irq_startup		= irq_chip_startup_parent,
+	.irq_shutdown		= irq_chip_shutdown_parent,
 	.irq_ack		= sg2042_msi_irq_ack,
 	.irq_mask		= irq_chip_mask_parent,
 	.irq_unmask		= irq_chip_unmask_parent,
@@ -114,6 +116,8 @@ static void sg2044_msi_irq_compose_msi_msg(struct irq_data *d, struct msi_msg *m
 
 static struct irq_chip sg2044_msi_middle_irq_chip = {
 	.name			= "SG2044 MSI",
+	.irq_startup		= irq_chip_startup_parent,
+	.irq_shutdown		= irq_chip_shutdown_parent,
 	.irq_ack		= sg2044_msi_irq_ack,
 	.irq_mask		= irq_chip_mask_parent,
 	.irq_unmask		= irq_chip_unmask_parent,
@@ -185,8 +189,10 @@ static const struct irq_domain_ops sg204x_msi_middle_domain_ops = {
 	.select	= msi_lib_irq_domain_select,
 };
 
-#define SG2042_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |	\
-				   MSI_FLAG_USE_DEF_CHIP_OPS)
+#define SG2042_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |		\
+				   MSI_FLAG_USE_DEF_CHIP_OPS |		\
+				   MSI_FLAG_PCI_MSI_MASK_PARENT |	\
+				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
 #define SG2042_MSI_FLAGS_SUPPORTED MSI_GENERIC_FLAGS_MASK
 
@@ -200,10 +206,12 @@ static const struct msi_parent_ops sg2042_msi_parent_ops = {
 	.init_dev_msi_info	= msi_lib_init_dev_msi_info,
 };
 
-#define SG2044_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |	\
-				   MSI_FLAG_USE_DEF_CHIP_OPS)
+#define SG2044_MSI_FLAGS_REQUIRED (MSI_FLAG_USE_DEF_DOM_OPS |		\
+				   MSI_FLAG_USE_DEF_CHIP_OPS |		\
+				   MSI_FLAG_PCI_MSI_MASK_PARENT |	\
+				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
-#define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |	\
+#define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |		\
 				    MSI_FLAG_PCI_MSIX)
 
 static const struct msi_parent_ops sg2044_msi_parent_ops = {
#21 lore raw

PATCH v2 4/4 irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044

In reply to Inochi Amaoto

The MSI controller on SG2044 has the ability to allocate multiple PCI MSI interrupts. So the PCIe controller driver can use this feature if it also supports multiple PCI MSI interrupts.

Add MSI_FLAG_MULTI_PCI_MSI flag for the supported_flags of SG2044 msi_parent_ops so the PCIe controller driver can use this feature if it also supports this feature.

---
 drivers/irqchip/irq-sg2042-msi.c | 1 +
 1 file changed, 1 insertion(+)
drivers/irqchip/irq-sg2042-msi.c +1 -0
index 2fd4d94f9bd7..3b13dbbfdb51 100644
@@ -212,6 +212,7 @@ static const struct msi_parent_ops sg2042_msi_parent_ops = {
 				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
 #define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |		\
+				    MSI_FLAG_MULTI_PCI_MSI |		\
 				    MSI_FLAG_PCI_MSIX)
 
 static const struct msi_parent_ops sg2044_msi_parent_ops = {
Signature · 1 line
2.50.1
merge bot Applied
#22 lore raw

[tip: irq/drivers] irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044

The following commit has been merged into the irq/drivers branch of tip:

The following commit has been merged into the irq/drivers branch of tip:

Commit-ID:     7ee4a5a2ec3748facfb4ca96e4cce6cabbdecab2
Gitweb:        https://git.kernel.org/tip/7ee4a5a2ec3748facfb4ca96e4cce6cabbdecab2
Author:        Inochi Amaoto <[email protected]>
AuthorDate:    Thu, 14 Aug 2025 07:28:34 +08:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Sat, 23 Aug 2025 21:21:13 +02:00

irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044

The MSI controller on SG2044 has the ability to allocate multiple PCI MSI interrupts. So the PCIe controller driver can use this feature if the hardware supports multiple PCI MSI interrupts.

Add the MSI_FLAG_MULTI_PCI_MSI flag to the supported_flags of SG2044 msi_parent_ops to enable this functionality.

---
 drivers/irqchip/irq-sg2042-msi.c | 1 +
 1 file changed, 1 insertion(+)
drivers/irqchip/irq-sg2042-msi.c +1 -0
index 2fd4d94..3b13dbb 100644
@@ -212,6 +212,7 @@ static const struct msi_parent_ops sg2042_msi_parent_ops = {
 				   MSI_FLAG_PCI_MSI_STARTUP_PARENT)
 
 #define SG2044_MSI_FLAGS_SUPPORTED (MSI_GENERIC_FLAGS_MASK |		\
+				    MSI_FLAG_MULTI_PCI_MSI |		\
 				    MSI_FLAG_PCI_MSIX)
 
 static const struct msi_parent_ops sg2044_msi_parent_ops = {
Reviewed-by Tested-by on the v2 series
#23 lore raw
In reply to Inochi Amaoto
Quoted 36 lines · Inochi Amaoto

When using NVME on SG2044, the NVME always complains "I/O tag XXX (XXX) QID XX timeout, completion polled", which is caused by the broken handler of the sg2042-msi driver.

As PLIC driver can only setting affinity when enabling, the sg2042-msi does not properly handled affinity setting previously and enable irq in an unexpected executing path.

Add irq_startup/irq_shutdown support to the PCI template domain, then set irq_chip_[startup/shutdown]_parent for irq_startup/ irq_shutdown of the sg2042-msi driver. So the irq can be started properly.

Change from v1:
1. patch 1: Fix comment format problem, and structure the changelog.
2. patch 2: Improve the comment title and body, add describtion about the fact the PLIC is used as parent chip.
3. patch 2: Remove __always_inline for cond_[shutdown/startup]_parent().
4. patch 3: Update the align of the "XXX_MSI_FLAGS_XXX" macro.
5. patch 4: Claim the fact that the added flag is used by the negotiation of MSI controller driver and PCIe device driver, and can be only used when both of them support this flag.

Inochi Amaoto (4):
genirq: Add irq_chip_(startup/shutdown)_parent() PCI/MSI: Add startup/shutdown for per device domains irqchip/sg2042-msi: Fix broken affinity setting irqchip/sg2042-msi: Set MSI_FLAG_MULTI_PCI_MSI flags for SG2044

  drivers/irqchip/irq-sg2042-msi.c | 19 +++++++++---
  drivers/pci/msi/irqdomain.c      | 52 ++++++++++++++++++++++++++++++++
  include/linux/irq.h              |  2 ++
  include/linux/msi.h              |  2 ++
  kernel/irq/chip.c                | 37 +++++++++++++++++++++++
  5 files changed, 107 insertions(+), 5 deletions(-)
Quoted 2 lines

--
2.50.1

It is recommended to add the following email link about the relevant discussion in the commit message for quich reference when there is a revision or merge:

https://lore.kernel.org/lkml/[email protected]/