[PATCH v2 0/3] x86/irq: Fix lost interrupts on CPU hot-unplug for Hyper-V PCI/MSI

[PATCH v2 0/3] x86/irq: Fix lost interrupts on CPU hot-unplug for Hyper-V PCI/MSI

[Naman Jain]

On CPU hot-unplug, fixup_irqs() re-injects interrupts that are still
pending in the outgoing CPU's IRR via the irqchip's irq_retrigger()
callback. The Hyper-V vPCI MSI irqchip never installs one, so such a
pending interrupt is silently dropped, leading to lost interrupts, stalls
and "No irq handler for vector" messages.

Patch 1 is the minimal, backportable fix (Fixes:/Cc: stable) that installs
irq_chip_retrigger_hierarchy() on the Hyper-V PCI/MSI chip. Patch 2 then
switches fixup_irqs() to call irq_chip_retrigger_hierarchy() directly, so
the retrigger works for any stacked irqchip without a per-chip callback.
Patch 3 drops the now-redundant irq_retrigger() initializations.
This way, patch 2 and 3 do not need to be backported.

Changes since v1:
https://lore.kernel.org/all/20260728135117.2888175-1-namjain@linux.microsoft.com/#t
* Added Review tags
* Adding minor clarification in commit msg of patch 2 (Shradha)

Original RFC: https://lore.kernel.org/all/87y0f3jhd6.ffs@fw13/

Naman Jain (3):
  PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip
  x86/irq: Use irq_chip_retrigger_hierarchy() in fixup_irqs()
  x86/irq, iommu/amd, PCI: Drop redundant irq_retrigger inits

 arch/x86/kernel/apic/io_apic.c |  2 --
 arch/x86/kernel/apic/msi.c     |  4 +---
 arch/x86/kernel/hpet.c         |  1 -
 arch/x86/kernel/irq.c          | 23 +++++++----------------
 drivers/iommu/amd/init.c       |  1 -
 5 files changed, 8 insertions(+), 23 deletions(-)

-- 
2.43.0

[PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[Naman Jain]

The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.

On CPU hot-unplug fixup_irqs() migrates the interrupts which are affine to
the outgoing CPU to a new target. If an interrupt still has its pending bit
set in the outgoing CPU's IRR at that point, fixup_irqs() resends it on the
new target through the irqchip's irq_retrigger() callback. As the Hyper-V
PCI/MSI chip does not provide that callback, the pending interrupt is
silently dropped, which can result in lost interrupts, stalls and "No irq
handler for vector" messages during CPU hotplug.

Install irq_chip_retrigger_hierarchy() as the irq_retrigger() callback for
the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on its
new target CPU via the parent x86 vector domain.

Fixes: 4daace0d8ce85 ("PCI: hv: Add paravirtual PCI front-end for Microsoft Hyper-V VMs")
Cc: stable@vger.kernel.org
Suggested-by: Long Li <longli@microsoft.com>
Suggested-by: Thomas Gleixner <tglx@kernel.org>
Reviewed-by: Aditya Garg <gargaditya@linux.microsoft.com>
Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Signed-off-by: Naman Jain <namjain@linux.microsoft.com>
---
 drivers/pci/controller/pci-hyperv.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index cfc8fa403dad6..89816a2bd7cd3 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c
@@ -2119,6 +2119,7 @@ static bool hv_pcie_init_dev_msi_info(struct device *dev, struct irq_domain *dom
 	info->ops->msi_prepare = hv_msi_prepare;
 
 	chip->irq_set_affinity = irq_chip_set_affinity_parent;
+	chip->irq_retrigger = irq_chip_retrigger_hierarchy;
 
 	if (IS_ENABLED(CONFIG_X86))
 		chip->flags |= IRQCHIP_MOVE_DEFERRED;
-- 
2.43.0

[PATCH v2 2/3] x86/irq: Use irq_chip_retrigger_hierarchy() in fixup_irqs()

[Naman Jain]

fixup_irqs() re-injects a pending interrupt on its new target CPU by
looking at the outermost domain chip and invoking its irq_retrigger()
callback directly. That only works when the outermost chip happens to
install an irq_retrigger() callback, which is not guaranteed for every
irqchip and could lead to lost interrupts on CPU hot-unplug.

Use irq_chip_retrigger_hierarchy() instead, which walks up the interrupt
hierarchy until it finds a chip that implements irq_retrigger().

While at it, move the loop-local variables into the loop scope and use a
scoped guard for desc->lock.

VECTOR_RETRIGGERED is now set only when the retrigger succeeds
(irq_chip_retrigger_hierarchy() returns non-zero) instead of
unconditionally. This is harmless today since apic_retrigger_irq()
always returns 1, and arguably more correct. No functional change
intended for chips which already provide an irq_retrigger() callback
on the outermost domain.

Suggested-by: Thomas Gleixner <tglx@kernel.org>
Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Signed-off-by: Naman Jain <namjain@linux.microsoft.com>
---
 arch/x86/kernel/irq.c | 23 +++++++----------------
 1 file changed, 7 insertions(+), 16 deletions(-)

diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c
index 30122f0b3af96..ef1bdd3c4659a 100644
--- a/arch/x86/kernel/irq.c
+++ b/arch/x86/kernel/irq.c
@@ -466,11 +466,6 @@ DEFINE_IDTENTRY_SYSVEC(sysvec_posted_msi_notification)
 /* A cpu has been removed from cpu_online_mask.  Reset irq affinities. */
 void fixup_irqs(void)
 {
-	unsigned int vector;
-	struct irq_desc *desc;
-	struct irq_data *data;
-	struct irq_chip *chip;
-
 	irq_migrate_all_off_this_cpu();
 
 	/*
@@ -489,22 +484,18 @@ void fixup_irqs(void)
 	 * vector_lock because the cpu is already marked !online, so
 	 * nothing else will touch it.
 	 */
-	for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
-		if (IS_ERR_OR_NULL(__this_cpu_read(vector_irq[vector])))
+	for (unsigned int vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
+		struct irq_desc *desc = __this_cpu_read(vector_irq[vector]);
+
+		if (IS_ERR_OR_NULL(desc))
 			continue;
 
 		if (is_vector_pending(vector)) {
-			desc = __this_cpu_read(vector_irq[vector]);
-
-			raw_spin_lock(&desc->lock);
-			data = irq_desc_get_irq_data(desc);
-			chip = irq_data_get_irq_chip(data);
-			if (chip->irq_retrigger) {
-				chip->irq_retrigger(data);
+			guard(raw_spinlock)(&desc->lock);
+			if (irq_chip_retrigger_hierarchy(&desc->irq_data))
 				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
-			}
-			raw_spin_unlock(&desc->lock);
 		}
+
 		if (__this_cpu_read(vector_irq[vector]) != VECTOR_RETRIGGERED)
 			__this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
 	}
-- 
2.43.0

[PATCH v2 3/3] x86/irq, iommu/amd, PCI: Drop redundant irq_retrigger inits

[Naman Jain]

With fixup_irqs() now using irq_chip_retrigger_hierarchy() and the generic
resend path (try_retrigger()) already falling back to it, installing
irq_chip_retrigger_hierarchy() as the irq_retrigger() callback on the
outermost domain chips is no longer required. The hierarchy walk reaches
the x86 vector domain's apic_retrigger_irq() for all of these stacked
domain chips.

Remove the now redundant irq_retrigger() initializations from the outermost
domain interrupt chips.
msi_set_affinity() is the only site that invoked the outermost chip's
irq_retrigger() directly. Convert it to irq_chip_retrigger_hierarchy() so
the x86_init_dev_msi_info() initialization can be dropped as well.

No functional change intended.

Suggested-by: Thomas Gleixner <tglx@kernel.org>
Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
Signed-off-by: Naman Jain <namjain@linux.microsoft.com>
---
 arch/x86/kernel/apic/io_apic.c      | 2 --
 arch/x86/kernel/apic/msi.c          | 4 +---
 arch/x86/kernel/hpet.c              | 1 -
 drivers/iommu/amd/init.c            | 1 -
 drivers/pci/controller/pci-hyperv.c | 1 -
 5 files changed, 1 insertion(+), 8 deletions(-)

diff --git a/arch/x86/kernel/apic/io_apic.c b/arch/x86/kernel/apic/io_apic.c
index 7d7175d012288..18f49a251f23b 100644
--- a/arch/x86/kernel/apic/io_apic.c
+++ b/arch/x86/kernel/apic/io_apic.c
@@ -1857,7 +1857,6 @@ static struct irq_chip ioapic_chip __read_mostly = {
 	.irq_ack		= irq_chip_ack_parent,
 	.irq_eoi		= ioapic_ack_level,
 	.irq_set_affinity	= ioapic_set_affinity,
-	.irq_retrigger		= irq_chip_retrigger_hierarchy,
 	.irq_get_irqchip_state	= ioapic_irq_get_chip_state,
 	.flags			= IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MOVE_DEFERRED |
 				  IRQCHIP_AFFINITY_PRE_STARTUP,
@@ -1871,7 +1870,6 @@ static struct irq_chip ioapic_ir_chip __read_mostly = {
 	.irq_ack		= irq_chip_ack_parent,
 	.irq_eoi		= ioapic_ir_ack_level,
 	.irq_set_affinity	= ioapic_set_affinity,
-	.irq_retrigger		= irq_chip_retrigger_hierarchy,
 	.irq_get_irqchip_state	= ioapic_irq_get_chip_state,
 	.flags			= IRQCHIP_SKIP_SET_WAKE |
 				  IRQCHIP_AFFINITY_PRE_STARTUP,
diff --git a/arch/x86/kernel/apic/msi.c b/arch/x86/kernel/apic/msi.c
index 66bc5d3e79db3..aad2f15eb2176 100644
--- a/arch/x86/kernel/apic/msi.c
+++ b/arch/x86/kernel/apic/msi.c
@@ -137,7 +137,7 @@ msi_set_affinity(struct irq_data *irqd, const struct cpumask *mask, bool force)
 	 * IRR.
 	 */
 	if (lapic_vector_set_in_irr(cfg->vector))
-		irq_data_get_irq_chip(irqd)->irq_retrigger(irqd);
+		irq_chip_retrigger_hierarchy(irqd);
 
 	return ret;
 }
@@ -246,7 +246,6 @@ static bool x86_init_dev_msi_info(struct device *dev, struct irq_domain *domain,
 	info->ops->msi_prepare		= x86_msi_prepare;
 
 	info->chip->irq_ack		= irq_chip_ack_parent;
-	info->chip->irq_retrigger	= irq_chip_retrigger_hierarchy;
 	info->chip->flags		|= IRQCHIP_SKIP_SET_WAKE |
 					   IRQCHIP_AFFINITY_PRE_STARTUP;
 
@@ -313,7 +312,6 @@ static struct irq_chip dmar_msi_controller = {
 	.irq_mask		= dmar_msi_mask,
 	.irq_ack		= irq_chip_ack_parent,
 	.irq_set_affinity	= msi_domain_set_affinity,
-	.irq_retrigger		= irq_chip_retrigger_hierarchy,
 	.irq_compose_msi_msg	= dmar_msi_compose_msg,
 	.irq_write_msi_msg	= dmar_msi_write_msg,
 	.flags			= IRQCHIP_SKIP_SET_WAKE | IRQCHIP_MOVE_DEFERRED |
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index 8dc7b710e1250..2bb7a6f466af3 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -509,7 +509,6 @@ static struct irq_chip hpet_msi_controller __ro_after_init = {
 	.irq_mask = hpet_msi_mask,
 	.irq_ack = irq_chip_ack_parent,
 	.irq_set_affinity = msi_domain_set_affinity,
-	.irq_retrigger = irq_chip_retrigger_hierarchy,
 	.irq_write_msi_msg = hpet_msi_write_msg,
 	.flags = IRQCHIP_SKIP_SET_WAKE | IRQCHIP_AFFINITY_PRE_STARTUP,
 };
diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
index ea3377e61aa5c..0c362e86a09d9 100644
--- a/drivers/iommu/amd/init.c
+++ b/drivers/iommu/amd/init.c
@@ -2468,7 +2468,6 @@ static struct irq_chip intcapxt_controller = {
 	.irq_unmask		= intcapxt_unmask_irq,
 	.irq_mask		= intcapxt_mask_irq,
 	.irq_ack		= irq_chip_ack_parent,
-	.irq_retrigger		= irq_chip_retrigger_hierarchy,
 	.irq_set_affinity       = intcapxt_set_affinity,
 	.irq_set_wake		= intcapxt_set_wake,
 	.flags			= IRQCHIP_MASK_ON_SUSPEND | IRQCHIP_MOVE_DEFERRED,
diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
index 89816a2bd7cd3..cfc8fa403dad6 100644
--- a/drivers/pci/controller/pci-hyperv.c
+++ b/drivers/pci/controller/pci-hyperv.c
@@ -2119,7 +2119,6 @@ static bool hv_pcie_init_dev_msi_info(struct device *dev, struct irq_domain *dom
 	info->ops->msi_prepare = hv_msi_prepare;
 
 	chip->irq_set_affinity = irq_chip_set_affinity_parent;
-	chip->irq_retrigger = irq_chip_retrigger_hierarchy;
 
 	if (IS_ENABLED(CONFIG_X86))
 		chip->flags |= IRQCHIP_MOVE_DEFERRED;
-- 
2.43.0

Re: [PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[sashiko-bot]

> The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.
> 
> On CPU hot-unplug fixup_irqs() migrates the interrupts which are affine to
> the outgoing CPU to a new target. If an interrupt still has its pending bit
> set in the outgoing CPU's IRR at that point, fixup_irqs() resends it on the
> new target through the irqchip's irq_retrigger() callback. As the Hyper-V
> PCI/MSI chip does not provide that callback, the pending interrupt is
> silently dropped, which can result in lost interrupts, stalls and "No irq
> handler for vector" messages during CPU hotplug.
> 
> Install irq_chip_retrigger_hierarchy() as the irq_retrigger() callback for
> the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on its
> new target CPU via the parent x86 vector domain.
> 
> Fixes: 4daace0d8ce85 ("PCI: hv: Add paravirtual PCI front-end for Microsoft Hyper-V VMs")
> Cc: stable@vger.kernel.org
> Suggested-by: Long Li <longli@microsoft.com>
> Suggested-by: Thomas Gleixner <tglx@kernel.org>
> Reviewed-by: Aditya Garg <gargaditya@linux.microsoft.com>
> Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
> Signed-off-by: Naman Jain <namjain@linux.microsoft.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260810090716.2325295-1-namjain@linux.microsoft.com?part=1

Re: [PATCH v2 2/3] x86/irq: Use irq_chip_retrigger_hierarchy() in fixup_irqs()

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] `irq_chip_retrigger_hierarchy()` skips the outermost interrupt chip, breaking retriggering for interrupts where the top-level chip directly implements `.irq_retrigger` (e.g., single-level domains).
--

--- Patch [2]: [PATCH v2 2/3] x86/irq: Use irq_chip_retrigger_hierarchy() in fixup_irqs() ---
commit 51de6b45bc61b20c750e5760473230c207ce5194
Author: Naman Jain <namjain@linux.microsoft.com>

x86/irq: Use irq_chip_retrigger_hierarchy() in fixup_irqs()

This commit updates fixup_irqs() to use a scoped guard for the
irq_desc lock and replaces the direct invocation of the outermost
chip's irq_retrigger() callback with irq_chip_retrigger_hierarchy().

> diff --git a/arch/x86/kernel/irq.c b/arch/x86/kernel/irq.c
> --- a/arch/x86/kernel/irq.c
> +++ b/arch/x86/kernel/irq.c
[ ... ]
> @@ -489,22 +484,18 @@ void fixup_irqs(void)
[ ... ]
>  		if (is_vector_pending(vector)) {
> -			desc = __this_cpu_read(vector_irq[vector]);
> -
> -			raw_spin_lock(&desc->lock);
> -			data = irq_desc_get_irq_data(desc);
> -			chip = irq_data_get_irq_chip(data);
> -			if (chip->irq_retrigger) {
> -				chip->irq_retrigger(data);
> +			guard(raw_spinlock)(&desc->lock);
> +			if (irq_chip_retrigger_hierarchy(&desc->irq_data))

[Severity: High]
Does this skip the outermost interrupt chip for single-level domains?

Looking at irq_chip_retrigger_hierarchy() in kernel/irq/chip.c, the loop
initialization immediately advances to the parent data:

int irq_chip_retrigger_hierarchy(struct irq_data *data)
{
	for (data = data->parent_data; data; data = data->parent_data)
		if (data->chip && data->chip->irq_retrigger)
			return data->chip->irq_retrigger(data);
	return 0;
}

For domains where the outermost chip directly implements the retrigger 
callback (such as legacy interrupts in x86_vector_domain), data->parent_data 
will be NULL or lack the retrigger logic. Will this cause the hierarchy walk 
to just return 0 and silently drop the pending interrupt during CPU hot-unplug?

The generic try_retrigger() in kernel/irq/resend.c handles this by explicitly 
checking desc->irq_data.chip->irq_retrigger before falling back to the 
hierarchy walk. Should fixup_irqs() follow that same pattern to avoid skipping
the top-level chip?

>  				__this_cpu_write(vector_irq[vector], VECTOR_RETRIGGERED);
> -			}
> -			raw_spin_unlock(&desc->lock);
>  		}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260810090716.2325295-1-namjain@linux.microsoft.com?part=2

Re: [PATCH v2 3/3] x86/irq, iommu/amd, PCI: Drop redundant irq_retrigger inits

[sashiko-bot]

> With fixup_irqs() now using irq_chip_retrigger_hierarchy() and the generic
> resend path (try_retrigger()) already falling back to it, installing
> irq_chip_retrigger_hierarchy() as the irq_retrigger() callback on the
> outermost domain chips is no longer required. The hierarchy walk reaches
> the x86 vector domain's apic_retrigger_irq() for all of these stacked
> domain chips.
> 
> Remove the now redundant irq_retrigger() initializations from the outermost
> domain interrupt chips.
> msi_set_affinity() is the only site that invoked the outermost chip's
> irq_retrigger() directly. Convert it to irq_chip_retrigger_hierarchy() so
> the x86_init_dev_msi_info() initialization can be dropped as well.
> 
> No functional change intended.
> 
> Suggested-by: Thomas Gleixner <tglx@kernel.org>
> Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
> Signed-off-by: Naman Jain <namjain@linux.microsoft.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260810090716.2325295-1-namjain@linux.microsoft.com?part=3

RE: [PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[Michael Kelley]

From: Naman Jain <namjain@linux.microsoft.com> Sent: Monday, August 10, 2026 2:07 AM
> 
> The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.
> 
> On CPU hot-unplug fixup_irqs() migrates the interrupts which are affine to
> the outgoing CPU to a new target. If an interrupt still has its pending bit
> set in the outgoing CPU's IRR at that point, fixup_irqs() resends it on the
> new target through the irqchip's irq_retrigger() callback. As the Hyper-V
> PCI/MSI chip does not provide that callback, the pending interrupt is
> silently dropped, which can result in lost interrupts, stalls and "No irq
> handler for vector" messages during CPU hotplug.
> 
> Install irq_chip_retrigger_hierarchy() as the irq_retrigger() callback for
> the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on its
> new target CPU via the parent x86 vector domain.

The RFC version of this patch set listed the asynchronous nature
of the HVCALL_RETARGET_INTERRUPT hypercall as a cause of the
problem. Is that hypercall still believed to be asynchronous?
My understanding of "asynchronous" would be that Hyper-V
does not guarantee that the retargeting of the interrupt to a new
vCPU has occurred when the hypercall returns. There is an
unspecified time period during which the interrupt could still be
directed to the vCPU that is going offline. Furthermore, Hyper-V
provides no rendezvous mechanism for the guest to know when
the interrupt is guaranteed to go to the new vCPU.

If the async behavior is indeed present, isn't it a problem, even
after this patch? irq_migrate_all_off_this_cpu() causes the
RETARGET_INTERRUPT hypercall to be made, but a new interrupt
could still be generated to the going-offline vCPU after fixup_irqs()
runs its loop searching for any pending interrupts. A new such
interrupt would be lost. The 1 millisecond delay in fixup_irqs()
probably mitigates this problem by giving Hyper-V time to complete
the retargeting. But it's not a 100% solution.

Assuming my thinking is correct, there isn't a 100% solution. But
the problem's existence, and the likely benefit of the 1 ms delay,
should be called out in the commit message and/or code
comments for future readers.

I also wondered about arm64 where Hyper-V provides an
emulated GICv3. The RETARGET_INTERRUPT hypercall is not
used since the retargeting is done by writing to a GICD register.
Presumably on real GICv3 hardware, after the write the new
CPU is guaranteed to receive any subsequent interrupts. And if
an interrupt was pending at the time of the write, the appropriate
trigger happens in the new CPU, so everything works. As long
as Hyper-V is doing the emulation the same way (i.e., fully
synchronously), the problem would not occur on arm64
guests.

I'm reminded that a similar problem occurred with changing
the target CPU for a VMBus channel interrupt. The initial
Hyper-V implementation of the VMBus MODIFYCHANNEL
message was asynchronous, which caused the same problem
as seen here if a vCPU were to be taken offline. A later version
of the VMBus protocol added a notification that the change to
the target vCPU is complete, which solves the problem. See
vmbus_channel_set_cpu() and vmbus_send_modifychannel()
and the associated code comments.

Michael

> 
> Fixes: 4daace0d8ce85 ("PCI: hv: Add paravirtual PCI front-end for Microsoft Hyper-V VMs")
> Cc: stable@vger.kernel.org
> Suggested-by: Long Li <longli@microsoft.com>
> Suggested-by: Thomas Gleixner <tglx@kernel.org>
> Reviewed-by: Aditya Garg <gargaditya@linux.microsoft.com>
> Reviewed-by: Shradha Gupta <shradhagupta@linux.microsoft.com>
> Signed-off-by: Naman Jain <namjain@linux.microsoft.com>
> ---
>  drivers/pci/controller/pci-hyperv.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/pci/controller/pci-hyperv.c b/drivers/pci/controller/pci-hyperv.c
> index cfc8fa403dad6..89816a2bd7cd3 100644
> --- a/drivers/pci/controller/pci-hyperv.c
> +++ b/drivers/pci/controller/pci-hyperv.c
> @@ -2119,6 +2119,7 @@ static bool hv_pcie_init_dev_msi_info(struct device *dev,
> struct irq_domain *dom
>  	info->ops->msi_prepare = hv_msi_prepare;
> 
>  	chip->irq_set_affinity = irq_chip_set_affinity_parent;
> +	chip->irq_retrigger = irq_chip_retrigger_hierarchy;
> 
>  	if (IS_ENABLED(CONFIG_X86))
>  		chip->flags |= IRQCHIP_MOVE_DEFERRED;
> --
> 2.43.0
> 

Re: [PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[Naman Jain]

On 8/17/2026 9:08 AM, Michael Kelley wrote:
> From: Naman Jain <namjain@linux.microsoft.com> Sent: Monday, August 10, 2026 2:07 AM
>>
>> The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.
>>
>> On CPU hot-unplug fixup_irqs() migrates the interrupts which are affine to
>> the outgoing CPU to a new target. If an interrupt still has its pending bit
>> set in the outgoing CPU's IRR at that point, fixup_irqs() resends it on the
>> new target through the irqchip's irq_retrigger() callback. As the Hyper-V
>> PCI/MSI chip does not provide that callback, the pending interrupt is
>> silently dropped, which can result in lost interrupts, stalls and "No irq
>> handler for vector" messages during CPU hotplug.
>>
>> Install irq_chip_retrigger_hierarchy() as the irq_retrigger() callback for
>> the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on its
>> new target CPU via the parent x86 vector domain.
> 
> The RFC version of this patch set listed the asynchronous nature
> of the HVCALL_RETARGET_INTERRUPT hypercall as a cause of the
> problem. Is that hypercall still believed to be asynchronous?
> My understanding of "asynchronous" would be that Hyper-V
> does not guarantee that the retargeting of the interrupt to a new
> vCPU has occurred when the hypercall returns. There is an
> unspecified time period during which the interrupt could still be
> directed to the vCPU that is going offline. Furthermore, Hyper-V
> provides no rendezvous mechanism for the guest to know when
> the interrupt is guaranteed to go to the new vCPU.
> 
> If the async behavior is indeed present, isn't it a problem, even
> after this patch? irq_migrate_all_off_this_cpu() causes the
> RETARGET_INTERRUPT hypercall to be made, but a new interrupt
> could still be generated to the going-offline vCPU after fixup_irqs()
> runs its loop searching for any pending interrupts. A new such
> interrupt would be lost. The 1 millisecond delay in fixup_irqs()
> probably mitigates this problem by giving Hyper-V time to complete
> the retargeting. But it's not a 100% solution.
 > > Assuming my thinking is correct, there isn't a 100% solution. But
> the problem's existence, and the likely benefit of the 1 ms delay,
> should be called out in the commit message and/or code
> comments for future readers.

Please allow me to confirm with the Hyper-V team internally once. I may 
have assumed asynchronous behavior because of the pending interrupt 
after retarget. It could simply have been committed to the old route 
before the retarget operation.

> 
> I also wondered about arm64 where Hyper-V provides an
> emulated GICv3. The RETARGET_INTERRUPT hypercall is not
> used since the retargeting is done by writing to a GICD register.
> Presumably on real GICv3 hardware, after the write the new
> CPU is guaranteed to receive any subsequent interrupts. And if
> an interrupt was pending at the time of the write, the appropriate
> trigger happens in the new CPU, so everything works. As long
> as Hyper-V is doing the emulation the same way (i.e., fully
> synchronously), the problem would not occur on arm64
> guests.
> 
> I'm reminded that a similar problem occurred with changing
> the target CPU for a VMBus channel interrupt. The initial
> Hyper-V implementation of the VMBus MODIFYCHANNEL
> message was asynchronous, which caused the same problem
> as seen here if a vCPU were to be taken offline. A later version
> of the VMBus protocol added a notification that the change to
> the target vCPU is complete, which solves the problem. See
> vmbus_channel_set_cpu() and vmbus_send_modifychannel()
> and the associated code comments.
> 

Thanks for sharing this info. I'll check and get back.

Regards,
Naman

Re: [PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[Naman Jain]

On 8/17/2026 5:45 PM, Naman Jain wrote:
> 
> 
> On 8/17/2026 9:08 AM, Michael Kelley wrote:
>> From: Naman Jain <namjain@linux.microsoft.com> Sent: Monday, August 
>> 10, 2026 2:07 AM
>>>
>>> The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.
>>>
>>> On CPU hot-unplug fixup_irqs() migrates the interrupts which are 
>>> affine to
>>> the outgoing CPU to a new target. If an interrupt still has its 
>>> pending bit
>>> set in the outgoing CPU's IRR at that point, fixup_irqs() resends it 
>>> on the
>>> new target through the irqchip's irq_retrigger() callback. As the 
>>> Hyper-V
>>> PCI/MSI chip does not provide that callback, the pending interrupt is
>>> silently dropped, which can result in lost interrupts, stalls and "No 
>>> irq
>>> handler for vector" messages during CPU hotplug.
>>>
>>> Install irq_chip_retrigger_hierarchy() as the irq_retrigger() 
>>> callback for
>>> the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on 
>>> its
>>> new target CPU via the parent x86 vector domain.
>>
>> The RFC version of this patch set listed the asynchronous nature
>> of the HVCALL_RETARGET_INTERRUPT hypercall as a cause of the
>> problem. Is that hypercall still believed to be asynchronous?
>> My understanding of "asynchronous" would be that Hyper-V
>> does not guarantee that the retargeting of the interrupt to a new
>> vCPU has occurred when the hypercall returns. There is an
>> unspecified time period during which the interrupt could still be
>> directed to the vCPU that is going offline. Furthermore, Hyper-V
>> provides no rendezvous mechanism for the guest to know when
>> the interrupt is guaranteed to go to the new vCPU.
>>
>> If the async behavior is indeed present, isn't it a problem, even
>> after this patch? irq_migrate_all_off_this_cpu() causes the
>> RETARGET_INTERRUPT hypercall to be made, but a new interrupt
>> could still be generated to the going-offline vCPU after fixup_irqs()
>> runs its loop searching for any pending interrupts. A new such
>> interrupt would be lost. The 1 millisecond delay in fixup_irqs()
>> probably mitigates this problem by giving Hyper-V time to complete
>> the retargeting. But it's not a 100% solution.
>  > > Assuming my thinking is correct, there isn't a 100% solution. But
>> the problem's existence, and the likely benefit of the 1 ms delay,
>> should be called out in the commit message and/or code
>> comments for future readers.
> 
> Please allow me to confirm with the Hyper-V team internally once. I may 
> have assumed asynchronous behavior because of the pending interrupt 
> after retarget. It could simply have been committed to the old route 
> before the retarget operation.

This is actually synchronous. So the fix in this series should suffice.
Sorry for the confusion.
I don't see any mentions of asynchronous behavior of that hypercall 
anywhere in these 3 patches, so I think we should be good.

Regards,
Naman

> 
>>
>> I also wondered about arm64 where Hyper-V provides an
>> emulated GICv3. The RETARGET_INTERRUPT hypercall is not
>> used since the retargeting is done by writing to a GICD register.
>> Presumably on real GICv3 hardware, after the write the new
>> CPU is guaranteed to receive any subsequent interrupts. And if
>> an interrupt was pending at the time of the write, the appropriate
>> trigger happens in the new CPU, so everything works. As long
>> as Hyper-V is doing the emulation the same way (i.e., fully
>> synchronously), the problem would not occur on arm64
>> guests.
>>
>> I'm reminded that a similar problem occurred with changing
>> the target CPU for a VMBus channel interrupt. The initial
>> Hyper-V implementation of the VMBus MODIFYCHANNEL
>> message was asynchronous, which caused the same problem
>> as seen here if a vCPU were to be taken offline. A later version
>> of the VMBus protocol added a notification that the change to
>> the target vCPU is complete, which solves the problem. See
>> vmbus_channel_set_cpu() and vmbus_send_modifychannel()
>> and the associated code comments.
>>
> 
> Thanks for sharing this info. I'll check and get back.
> 
> Regards,
> Naman
> 

RE: [PATCH v2 1/3] PCI: hv: Set irq_retrigger callback for the Hyper-V PCI MSI irqchip

[Michael Kelley]

From: Naman Jain <namjain@linux.microsoft.com> Sent: Monday, August 17, 2026 8:31 AM
> 
> On 8/17/2026 5:45 PM, Naman Jain wrote:
> >
> > On 8/17/2026 9:08 AM, Michael Kelley wrote:
> >> From: Naman Jain <namjain@linux.microsoft.com> Sent: Monday, August
> >> 10, 2026 2:07 AM
> >>>
> >>> The Hyper-V vPCI MSI irqchip never installs an irq_retrigger() callback.
> >>>
> >>> On CPU hot-unplug fixup_irqs() migrates the interrupts which are affine to
> >>> the outgoing CPU to a new target. If an interrupt still has its pending bit
> >>> set in the outgoing CPU's IRR at that point, fixup_irqs() resends it on the
> >>> new target through the irqchip's irq_retrigger() callback. As the Hyper-V
> >>> PCI/MSI chip does not provide that callback, the pending interrupt is
> >>> silently dropped, which can result in lost interrupts, stalls and "No irq
> >>> handler for vector" messages during CPU hotplug.
> >>>
> >>> Install irq_chip_retrigger_hierarchy() as the irq_retrigger() callback for
> >>> the Hyper-V PCI/MSI irqchip, so that a pending interrupt is resent on its
> >>> new target CPU via the parent x86 vector domain.
> >>
> >> The RFC version of this patch set listed the asynchronous nature
> >> of the HVCALL_RETARGET_INTERRUPT hypercall as a cause of the
> >> problem. Is that hypercall still believed to be asynchronous?
> >> My understanding of "asynchronous" would be that Hyper-V
> >> does not guarantee that the retargeting of the interrupt to a new
> >> vCPU has occurred when the hypercall returns. There is an
> >> unspecified time period during which the interrupt could still be
> >> directed to the vCPU that is going offline. Furthermore, Hyper-V
> >> provides no rendezvous mechanism for the guest to know when
> >> the interrupt is guaranteed to go to the new vCPU.
> >>
> >> If the async behavior is indeed present, isn't it a problem, even
> >> after this patch? irq_migrate_all_off_this_cpu() causes the
> >> RETARGET_INTERRUPT hypercall to be made, but a new interrupt
> >> could still be generated to the going-offline vCPU after fixup_irqs()
> >> runs its loop searching for any pending interrupts. A new such
> >> interrupt would be lost. The 1 millisecond delay in fixup_irqs()
> >> probably mitigates this problem by giving Hyper-V time to complete
> >> the retargeting. But it's not a 100% solution.
> >  > > Assuming my thinking is correct, there isn't a 100% solution. But
> >> the problem's existence, and the likely benefit of the 1 ms delay,
> >> should be called out in the commit message and/or code
> >> comments for future readers.
> >
> > Please allow me to confirm with the Hyper-V team internally once. I may
> > have assumed asynchronous behavior because of the pending interrupt
> > after retarget. It could simply have been committed to the old route
> > before the retarget operation.
> 
> This is actually synchronous. So the fix in this series should suffice.
> Sorry for the confusion.
> I don't see any mentions of asynchronous behavior of that hypercall
> anywhere in these 3 patches, so I think we should be good.
> 

Thanks for getting the definitive answer. Having it be async was
initially plausible to me given the experience we had with the
VMBus MODIFYCHANNEL message. But upon further reflection,
the VMBus case was really more because it is a messaging protocol
instead of a "call".  As a "call", the hypercall really should be
synchronous. And that's consistent with the arm64 behavior of
setting the GICD register.

With this top-level question now resolved, I'll review the patch
series at a more detailed level to see if anything else comes up,
but it looks pretty good in my initial read-through.

Michael