Skip to contentSkip to conversation

accel/kvm: return early from kvm_irqchip_create if kvm does not support irqchip

PATCH Superseded Ani Sinha 3 participants 6 messages Apr 13 KVM

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Ani Sinha

author
#0 lore raw

PATCH 1/1 accel/kvm: return early from kvm_irqchip_create if kvm does not support irqchip

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change")
---
 accel/kvm/kvm-all.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
accel/kvm/kvm-all.c +8 -3
index 774499d34f..92af42503b 100644
@@ -2575,7 +2575,7 @@ void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
     g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
 }
 
-static void do_kvm_irqchip_create(KVMState *s)
+static int do_kvm_irqchip_create(KVMState *s)
 {
     int ret;
     if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
@@ -2587,7 +2587,7 @@ static void do_kvm_irqchip_create(KVMState *s)
             exit(1);
         }
     } else {
-        return;
+        return -EOPNOTSUPP;
     }
 
     if (kvm_check_extension(s, KVM_CAP_IRQFD) <= 0) {
@@ -2610,13 +2610,17 @@ static void do_kvm_irqchip_create(KVMState *s)
         fprintf(stderr, "Create kernel irqchip failed: %s\n", strerror(-ret));
         exit(1);
     }
+
+    return 0;
 }
 
 static void kvm_irqchip_create(KVMState *s)
 {
     assert(s->kernel_irqchip_split != ON_OFF_AUTO_AUTO);
 
-    do_kvm_irqchip_create(s);
+    if (do_kvm_irqchip_create(s) < 0) {
+        return;
+    }
     kvm_kernel_irqchip = true;
     /* If we have an in-kernel IRQ chip then we must have asynchronous
      * interrupt delivery (though the reverse is not necessarily true)
@@ -2835,6 +2839,7 @@ static int kvm_reset_vmfd(MachineState *ms)
     }
 
     if (s->kernel_irqchip_allowed) {
+        /* ignore return from this function */
         do_kvm_irqchip_create(s);
     }
 
Signature · 1 line
2.49.0
#1 lore raw
Quoted 34 lines · Ani Sinha

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change")
---
 accel/kvm/kvm-all.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
accel/kvm/kvm-all.c +1 -1
index 774499d34f..92af42503b 100644
@@ -2575,7 +2575,7 @@ void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
     g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
 }
 
-static void do_kvm_irqchip_create(KVMState *s)
+static int do_kvm_irqchip_create(KVMState *s)
 {
     int ret;
     if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {

accel/kvm/kvm-all.c static void do_kvm_irqchip_create(KVMState *s) · line 2593

Quoted 8 lines
accel/kvm/kvm-all.c +1 -1
@@ -2587,7 +2587,7 @@ static void do_kvm_irqchip_create(KVMState *s)
             exit(1);
         }
     } else {
-        return;
+        return -EOPNOTSUPP;
     }
 
     if (kvm_check_extension(s, KVM_CAP_IRQFD) <= 0) {

This function confuses me. With this patch it can do three things:
* report an error and exit(1)
* return success
* return a failure code

What is the difference between the cases where we exit and where we return the error code? The commit message says we return an error "when required capabilities are absent", but we do the error-exit for e.g. "KVM_CAP_IRQFD not present", which sounds like "required capabilities are absent" to me...

thanks
-- PMM

author
#2 lore raw
Quoted 53 lines · Peter Maydell
Quoted 34 lines · Ani Sinha

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change")

---
accel/kvm/kvm-all.c | 11 ++++++++---

1 file changed, 8 insertions(+), 3 deletions(-)
accel/kvm/kvm-all.c +1 -1
index 774499d34f..92af42503b 100644
@@ -2575,7 +2575,7 @@ void kvm_irqchip_set_qemuirq_gsi(KVMState *s, qemu_irq irq, int gsi)
    g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
 }
 
-static void do_kvm_irqchip_create(KVMState *s)
+static int do_kvm_irqchip_create(KVMState *s)
 {
    int ret;
    if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {

accel/kvm/kvm-all.c static void do_kvm_irqchip_create(KVMState *s) · line 2593

Quoted 8 lines
accel/kvm/kvm-all.c +1 -1
@@ -2587,7 +2587,7 @@ static void do_kvm_irqchip_create(KVMState *s)
            exit(1);
        }
    } else {
-        return;
+        return -EOPNOTSUPP;
    }
 
    if (kvm_check_extension(s, KVM_CAP_IRQFD) <= 0) {

This function confuses me. With this patch it can do three things:
* report an error and exit(1)
* return success
* return a failure code

Quoted 5 lines

What is the difference between the cases where we exit and where we return the error code? The commit message says we return an error "when required capabilities are absent", but we do the error-exit for e.g. "KVM_CAP_IRQFD not present", which sounds like "required capabilities are absent" to me...

I think it should be an assertion there. That is instead of exit(1) we should have had

assert(kvm_check_extension(s, KVM_CAP_IRQFD));

That is you can’t have KVM_CAP_IRQCHIP cap and not have KVM_CAP_IRQFD . That is how I understand it.

Reviewed-by on v1 1/1
#3 lore raw
In reply to Ani Sinha
Quoted 9 lines · Ani Sinha

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

As a fix intended for 11.0 at this point in the release cycle, it would be helpful if the commit message stated what the user-visible consqeuences of this bug are. That helps in identifying whether this really does need to go into 11.0 (i.e. whether it is release-critical).

Quoted 6 lines
Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change")

as the minimal fix for the logic-bug introduced by 98884e0cc1.

thanks
-- PMM

author
#4 lore raw
Quoted 15 lines · Peter Maydell
Quoted 9 lines · Ani Sinha

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

Quoted 3 lines

As a fix intended for 11.0 at this point in the release cycle, it would be helpful if the commit message stated what the user-visible consqeuences of this bug are.

Ok I would appreciate if you can add this line in the commit log before putting it in CI:

“Due to this regression, all KVM guests on PPC 64 hang immediately during startup"

Or if you prefer, I can spin up a new version with the updated commit message.

Quoted 18 lines

That helps in
identifying whether this really does need to go into 11.0 (i.e. whether it is release-critical).

Quoted 6 lines
Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change")

as the minimal fix for the logic-bug introduced by 98884e0cc1.

thanks
-- PMM

Tested-by on v1 1/1
#5 lore raw
In reply to Ani Sinha

Hi Ani,
I've tested the v1 patch: [PATCH for-11.0] accel/kvm: return early from kvm_irqchip_create if kvm does not support irqchip, on PowerPC machine and it resolves the boot hang issue seen when booting KVM guest with >1 smp value.

Regression Details:
Commit 98884e0cc1 ("accel/kvm: add changes required to support KVM VM file descriptor change") led to a regression which caused KVM guests on ppc64le to hang indefinitely during boot when SMP was configured. Single vCPU guests booted successfully, but any configuration with more than one vCPU would hang. The hang occurred at bql_lock() in qemu_default_main() (system/main.c:49), where the system would become completely unresponsive after successfully completing qemu_system_reset().
Reference: https://lists.nongnu.org/archive/html/qemu-ppc/2026-03/msg00035.html

Test Environment:
- Host Arch: ppc64le
- Host and Guest OS: Fedora 42
- Machine Type: pseries with KVM acceleration
- QEMU: Latest master with this patch applied

Test Results:
All the following SMP topologies now boot successfully:

Single and simple multi-CPU:
- -smp 1
- -smp 2
- -smp 4
- -smp 32

Various socket/core/thread combinations (8 vCPUs):
- -smp 8,sockets=8,cores=1,threads=1
- -smp 8,sockets=1,cores=8,threads=1
- -smp 8,sockets=1,cores=1,threads=8
- -smp 8,sockets=2,cores=4,threads=1
- -smp 8,sockets=1,cores=4,threads=2
- -smp 8,sockets=2,cores=1,threads=4
- -smp 8,sockets=2,cores=2,threads=2

Higher vCPU count:
- -smp 16,sockets=2,cores=4,threads=2
- -smp 32,sockets=1,cores=8,threads=4

Thanks,
Misbah Anjum N <[email protected]>

Quoted 70 lines · Ani Sinha

During refactoring of kvm_irqchip_create(), the refactored code was returning early from do_kvm_irqchip_create() function if the required capabilities were not present in KVM. This was not translating to an early return from kvm_irqchip_create() as was the case before refactoring. This is because, do_kvm_irqchip_create() did not have a means to notify the caller of the lack of kvm capabilities. Fix this by making do_notify_irqchip_create() return EOPNOTSUPP error when required capabilities are absent and then the caller can check the return code and return early.

Fixes 98884e0cc1 ("accel/kvm: add changes required to support KVM VM

file descriptor change")

---
 accel/kvm/kvm-all.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
accel/kvm/kvm-all.c +1 -1
index 774499d34f..92af42503b 100644
@@ -2575,7 +2575,7 @@ void kvm_irqchip_set_qemuirq_gsi(KVMState *s,
 qemu_irq irq, int gsi)
     g_hash_table_insert(s->gsimap, irq, GINT_TO_POINTER(gsi));
 }
 
-static void do_kvm_irqchip_create(KVMState *s)
+static int do_kvm_irqchip_create(KVMState *s)
 {
     int ret;
     if (kvm_check_extension(s, KVM_CAP_IRQCHIP)) {
accel/kvm/kvm-all.c +6 -2
@@ -2587,7 +2587,7 @@ static void do_kvm_irqchip_create(KVMState *s)
             exit(1);
         }
     } else {
-        return;
+        return -EOPNOTSUPP;
     }
 
     if (kvm_check_extension(s, KVM_CAP_IRQFD) <= 0) {
@@ -2610,13 +2610,17 @@ static void do_kvm_irqchip_create(KVMState *s)
         fprintf(stderr, "Create kernel irqchip failed: %s\n", 
 strerror(-ret));
         exit(1);
     }
+
+    return 0;
 }
 
 static void kvm_irqchip_create(KVMState *s)
 {
     assert(s->kernel_irqchip_split != ON_OFF_AUTO_AUTO);
 
-    do_kvm_irqchip_create(s);
+    if (do_kvm_irqchip_create(s) < 0) {
+        return;
+    }
     kvm_kernel_irqchip = true;
     /* If we have an in-kernel IRQ chip then we must have asynchronous

* interrupt delivery (though the reverse is not necessarily true)

accel/kvm/kvm-all.c +1 -0
@@ -2835,6 +2839,7 @@ static int kvm_reset_vmfd(MachineState *ms)
     }
 
     if (s->kernel_irqchip_allowed) {
+        /* ignore return from this function */
         do_kvm_irqchip_create(s);
     }