[RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change

Rorie Reyes posted 4 patches 6 months, 1 week ago
Maintainers: Thomas Huth <thuth@redhat.com>, Richard Henderson <richard.henderson@linaro.org>, David Hildenbrand <david@redhat.com>, Ilya Leoshkevich <iii@linux.ibm.com>, Halil Pasic <pasic@linux.ibm.com>, Christian Borntraeger <borntraeger@linux.ibm.com>, Eric Farman <farman@linux.ibm.com>, Alex Williamson <alex.williamson@redhat.com>, "Cédric Le Goater" <clg@redhat.com>, Tony Krowiak <akrowiak@linux.ibm.com>, Jason Herne <jjherne@linux.ibm.com>
[RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Rorie Reyes 6 months, 1 week ago
Handle interception of the CHSC SEI instruction for requests
indicating the guest's AP configuration has changed.

If configuring --without-default-devices, hw/s390x/ap-stub.c
was created to handle such circumstance. Also added the
following to hw/s390x/meson.build if CONFIG_VFIO_AP is
false, it will use the stub file.

Signed-off-by: Rorie Reyes <rreyes@linux.ibm.com>
Reviewed-by: Anthony Krowiak <akrowiak@linux.ibm.com>
Reviewed-by: Cédric Le Goater <clg@redhat.com>
---
 MAINTAINERS           |  1 +
 hw/s390x/ap-stub.c    | 21 +++++++++++++++++++++
 hw/s390x/meson.build  |  1 +
 target/s390x/ioinst.c | 11 +++++++++--
 4 files changed, 32 insertions(+), 2 deletions(-)
 create mode 100644 hw/s390x/ap-stub.c

diff --git a/MAINTAINERS b/MAINTAINERS
index aa6763077e..1e84bfeaee 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -112,6 +112,7 @@ F: hw/intc/s390_flic.c
 F: hw/intc/s390_flic_kvm.c
 F: hw/s390x/
 F: hw/vfio/ap.c
+F: hw/s390x/ap-stub.c
 F: hw/vfio/ccw.c
 F: hw/watchdog/wdt_diag288.c
 F: include/hw/s390x/
diff --git a/hw/s390x/ap-stub.c b/hw/s390x/ap-stub.c
new file mode 100644
index 0000000000..001fe5f8b0
--- /dev/null
+++ b/hw/s390x/ap-stub.c
@@ -0,0 +1,21 @@
+/*
+ * VFIO based AP matrix device assignment
+ *
+ * Copyright 2025 IBM Corp.
+ * Author(s): Rorie Reyes <rreyes@linux.ibm.com>
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "qemu/osdep.h"
+#include "hw/s390x/ap-bridge.h"
+
+int ap_chsc_sei_nt0_get_event(void *res)
+{
+    return EVENT_INFORMATION_NOT_STORED;
+}
+
+bool ap_chsc_sei_nt0_have_event(void)
+{
+    return false;
+}
diff --git a/hw/s390x/meson.build b/hw/s390x/meson.build
index 3bbebfd817..99cbcbd7d6 100644
--- a/hw/s390x/meson.build
+++ b/hw/s390x/meson.build
@@ -33,6 +33,7 @@ s390x_ss.add(when: 'CONFIG_S390_CCW_VIRTIO', if_true: files(
 ))
 s390x_ss.add(when: 'CONFIG_TERMINAL3270', if_true: files('3270-ccw.c'))
 s390x_ss.add(when: 'CONFIG_VFIO', if_true: files('s390-pci-vfio.c'))
+s390x_ss.add(when: 'CONFIG_VFIO_AP', if_false: files('ap-stub.c'))
 
 virtio_ss = ss.source_set()
 virtio_ss.add(files('virtio-ccw.c'))
diff --git a/target/s390x/ioinst.c b/target/s390x/ioinst.c
index fe62ba5b06..2320dd4c12 100644
--- a/target/s390x/ioinst.c
+++ b/target/s390x/ioinst.c
@@ -18,6 +18,7 @@
 #include "trace.h"
 #include "hw/s390x/s390-pci-bus.h"
 #include "target/s390x/kvm/pv.h"
+#include "hw/s390x/ap-bridge.h"
 
 /* All I/O instructions but chsc use the s format */
 static uint64_t get_address_from_regs(CPUS390XState *env, uint32_t ipb,
@@ -574,13 +575,19 @@ out:
 
 static int chsc_sei_nt0_get_event(void *res)
 {
-    /* no events yet */
+    if (s390_has_feat(S390_FEAT_AP)) {
+        return ap_chsc_sei_nt0_get_event(res);
+    }
+
     return 1;
 }
 
 static int chsc_sei_nt0_have_event(void)
 {
-    /* no events yet */
+    if (s390_has_feat(S390_FEAT_AP)) {
+        return ap_chsc_sei_nt0_have_event();
+    }
+
     return 0;
 }
 
-- 
2.48.1


Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Matthew Rosato 5 months, 3 weeks ago
On 6/9/25 12:44 PM, Rorie Reyes wrote:
> Handle interception of the CHSC SEI instruction for requests
> indicating the guest's AP configuration has changed.
> 
> If configuring --without-default-devices, hw/s390x/ap-stub.c
> was created to handle such circumstance. Also added the
> following to hw/s390x/meson.build if CONFIG_VFIO_AP is
> false, it will use the stub file.
> 
> Signed-off-by: Rorie Reyes <rreyes@linux.ibm.com>
> Reviewed-by: Anthony Krowiak <akrowiak@linux.ibm.com>
> Reviewed-by: Cédric Le Goater <clg@redhat.com>

FYI, this patch (or some part of this series) breaks hotplug for PCI devices on s390x.  I verified it breaks with both PCI passthrough and virtio-pci via virsh attach-device.

I get greeted with the following message:
qemu-system-s390x: ../util/qemu-thread-posix.c:92: qemu_mutex_lock_impl: Assertion `mutex->initialized' failed.
2025-06-26 14:12:23.031+0000: shutting down, reason=crashed

As a quick test, reverting this patch resolves the issue.  

Another test shows that if I have a vfio-ap device attached to my guest before I try the zPCI hotplug, everything works fine (because vfio_ap_realize was driven -- see more info below).

I note in the failing call chain that in response to the hotplug event for a zPCI device we end up calling into hw/vfio/ap.c and getting some locks which seems wrong:

(gdb) bt
#0  __pthread_kill_implementation (threadid=<optimized out>, signo=signo@entry=6, no_tid=no_tid@entry=0) at pthread_kill.c:44
#1  0x000003ffa85ad636 in __pthread_kill_internal (threadid=<optimized out>, signo=6) at pthread_kill.c:78
#2  0x000003ffa8553c20 in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26
#3  0x000003ffa8533e3c in __GI_abort () at abort.c:79
#4  0x000003ffa854ac9a in __assert_fail_base
    (fmt=<optimized out>, assertion=assertion@entry=0x2aa3fa76414 "mutex->initialized", file=file@entry=0x2aa3fa763f8 "../util/qemu-thread-posix.c", line=line@entry=92, function=function@entry=0x2aa3fa7647a <__PRETTY_FUNCTION__.23> "qemu_mutex_lock_impl") at assert.c:94
#5  0x000003ffa854acf4 in __assert_fail
    (assertion=0x2aa3fa76414 "mutex->initialized", file=0x2aa3fa763f8 "../util/qemu-thread-posix.c", line=92, function=0x2aa3fa7647a <__PRETTY_FUNCTION__.23> "qemu_mutex_lock_impl") at assert.c:103
#6  0x000002aa3f83ef02 in qemu_mutex_lock_impl (mutex=0x2aa3fed9e08 <cfg_chg_events_lock>, file=0x2aa3f9e653e "/usr/src/qemu/include/qemu/lockable.h", line=56) at ../util/qemu-thread-posix.c:92
#7  0x000002aa3f4c3a70 in qemu_lockable_mutex_lock (x=0x2aa3fed9e08 <cfg_chg_events_lock>) at /usr/src/qemu/include/qemu/lockable.h:56
#8  0x000002aa3f4c3b00 in qemu_lockable_lock (x=0x3fd076f7730) at /usr/src/qemu/include/qemu/lockable.h:110
#9  0x000002aa3f4c3b82 in qemu_lockable_auto_lock (x=0x3fd076f7730) at /usr/src/qemu/include/qemu/lockable.h:120
#10 0x000002aa3f4c3f90 in ap_chsc_sei_nt0_get_event (res=0x3fd076f79b8) at ../hw/vfio/ap.c:110
^^^^^^^^
#11 0x000002aa3f3a59ea in chsc_sei_nt0_get_event (res=0x3fd076f79b8) at ../target/s390x/ioinst.c:579
#12 0x000002aa3f3a5b76 in ioinst_handle_chsc_sei (req=0x3fd076f79a8, res=0x3fd076f79b8) at ../target/s390x/ioinst.c:620
#13 0x000002aa3f3a6062 in ioinst_handle_chsc (cpu=0x2aa40cc9430, ipb=2097152, ra=0) at ../target/s390x/ioinst.c:708
#14 0x000002aa3f4647d2 in handle_b2 (cpu=0x2aa40cc9430, run=0x3ffa4180000, ipa1=95 '_') at ../target/s390x/kvm/kvm.c:1218
#15 0x000002aa3f465db2 in handle_instruction (cpu=0x2aa40cc9430, run=0x3ffa4180000) at ../target/s390x/kvm/kvm.c:1627
#16 0x000002aa3f4661da in handle_intercept (cpu=0x2aa40cc9430) at ../target/s390x/kvm/kvm.c:1709
#17 0x000002aa3f466c10 in kvm_arch_handle_exit (cs=0x2aa40cc9430, run=0x3ffa4180000) at ../target/s390x/kvm/kvm.c:1914
#18 0x000002aa3f523c98 in kvm_cpu_exec (cpu=0x2aa40cc9430) at ../accel/kvm/kvm-all.c:3294
#19 0x000002aa3f527f52 in kvm_vcpu_thread_fn (arg=0x2aa40cc9430) at ../accel/kvm/kvm-accel-ops.c:51
#20 0x000002aa3f840002 in qemu_thread_start (args=0x2aa40ccd090) at ../util/qemu-thread-posix.c:393
#21 0x000003ffa85ab622 in start_thread (arg=0x3fd076fe8c0) at pthread_create.c:448
#22 0x000003ffa862b390 in thread_start () at ../sysdeps/unix/sysv/linux/s390/s390-64/clone3.S:71


Note that PCI devices were already using SEI before this series (e.g. see hw/s390x/s390-pci-bus.c:pci_chsc_sei_nt2_get_event()); I think the problem here is that nt0 events are looked for implicitly in the shared handler (ioinst_handle_chsc_sei()) which is fine -- but you cannot go acquiring locks from ap.c that might have never been initialized (cfg_chg_events_lock which is only initialized in vfio_ap_realize()); that's why we ultimately crash.

AFAICT this portion of the handler including mutex should be moved out of ap.c into chsc_sei_nt0_get_event().  When vfio_ap is not in use, we can't call into it.  Can't you also build without VFIO_AP?  I didn't try it but that sure seems like it would cause issues too.  

I guess the mutex and list would then need to be part of something common to s390x like S390CcwMachineState.

Alternatively, you could leave those in vfio_ap but would need some kind of indicator that vfio_ap is "armed" for these kinds of events and only bother calling into vfio_ap to look at cfg_chg_events when that indicator has been set (e.g. vfio_ap_realize has been called once.  And it actually looks like you already do something like this via lock_initialized in ap.c). 
Otherwise, we should always return EVENT_INFORMATION_NOT_STORED like the old code did (via return 1).

Or maybe someone else has a better idea.


> ---
>  MAINTAINERS           |  1 +
>  hw/s390x/ap-stub.c    | 21 +++++++++++++++++++++
>  hw/s390x/meson.build  |  1 +
>  target/s390x/ioinst.c | 11 +++++++++--
>  4 files changed, 32 insertions(+), 2 deletions(-)
>  create mode 100644 hw/s390x/ap-stub.c
> 
> diff --git a/MAINTAINERS b/MAINTAINERS
> index aa6763077e..1e84bfeaee 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -112,6 +112,7 @@ F: hw/intc/s390_flic.c
>  F: hw/intc/s390_flic_kvm.c
>  F: hw/s390x/
>  F: hw/vfio/ap.c
> +F: hw/s390x/ap-stub.c
>  F: hw/vfio/ccw.c
>  F: hw/watchdog/wdt_diag288.c
>  F: include/hw/s390x/
> diff --git a/hw/s390x/ap-stub.c b/hw/s390x/ap-stub.c
> new file mode 100644
> index 0000000000..001fe5f8b0
> --- /dev/null
> +++ b/hw/s390x/ap-stub.c
> @@ -0,0 +1,21 @@
> +/*
> + * VFIO based AP matrix device assignment
> + *
> + * Copyright 2025 IBM Corp.
> + * Author(s): Rorie Reyes <rreyes@linux.ibm.com>
> + *
> + * SPDX-License-Identifier: GPL-2.0-or-later
> + */
> +
> +#include "qemu/osdep.h"
> +#include "hw/s390x/ap-bridge.h"
> +
> +int ap_chsc_sei_nt0_get_event(void *res)
> +{
> +    return EVENT_INFORMATION_NOT_STORED;
> +}
> +
> +bool ap_chsc_sei_nt0_have_event(void)
> +{
> +    return false;
> +}
> diff --git a/hw/s390x/meson.build b/hw/s390x/meson.build
> index 3bbebfd817..99cbcbd7d6 100644
> --- a/hw/s390x/meson.build
> +++ b/hw/s390x/meson.build
> @@ -33,6 +33,7 @@ s390x_ss.add(when: 'CONFIG_S390_CCW_VIRTIO', if_true: files(
>  ))
>  s390x_ss.add(when: 'CONFIG_TERMINAL3270', if_true: files('3270-ccw.c'))
>  s390x_ss.add(when: 'CONFIG_VFIO', if_true: files('s390-pci-vfio.c'))
> +s390x_ss.add(when: 'CONFIG_VFIO_AP', if_false: files('ap-stub.c'))
>  
>  virtio_ss = ss.source_set()
>  virtio_ss.add(files('virtio-ccw.c'))
> diff --git a/target/s390x/ioinst.c b/target/s390x/ioinst.c
> index fe62ba5b06..2320dd4c12 100644
> --- a/target/s390x/ioinst.c
> +++ b/target/s390x/ioinst.c
> @@ -18,6 +18,7 @@
>  #include "trace.h"
>  #include "hw/s390x/s390-pci-bus.h"
>  #include "target/s390x/kvm/pv.h"
> +#include "hw/s390x/ap-bridge.h"
>  
>  /* All I/O instructions but chsc use the s format */
>  static uint64_t get_address_from_regs(CPUS390XState *env, uint32_t ipb,
> @@ -574,13 +575,19 @@ out:
>  
>  static int chsc_sei_nt0_get_event(void *res)
>  {
> -    /* no events yet */
> +    if (s390_has_feat(S390_FEAT_AP)) {
> +        return ap_chsc_sei_nt0_get_event(res);
> +    }
> +
>      return 1;
>  }
>  
>  static int chsc_sei_nt0_have_event(void)
>  {
> -    /* no events yet */
> +    if (s390_has_feat(S390_FEAT_AP)) {
> +        return ap_chsc_sei_nt0_have_event();
> +    }
> +
>      return 0;
>  }
>  


Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Matthew Rosato 5 months, 3 weeks ago
On 6/26/25 12:19 PM, Matthew Rosato wrote:
> On 6/9/25 12:44 PM, Rorie Reyes wrote:
>> Handle interception of the CHSC SEI instruction for requests
>> indicating the guest's AP configuration has changed.
>>
>> If configuring --without-default-devices, hw/s390x/ap-stub.c
>> was created to handle such circumstance. Also added the
>> following to hw/s390x/meson.build if CONFIG_VFIO_AP is
>> false, it will use the stub file.
>>
>> Signed-off-by: Rorie Reyes <rreyes@linux.ibm.com>
>> Reviewed-by: Anthony Krowiak <akrowiak@linux.ibm.com>
>> Reviewed-by: Cédric Le Goater <clg@redhat.com>
> 
> FYI, this patch (or some part of this series) breaks hotplug for PCI devices on s390x.  I verified it breaks with both PCI passthrough and virtio-pci via virsh attach-device.
> 

I was unaware of 639ff87a1a ("attribute constructor for cfg_chg_events_lock") which apparently was already sitting in vfio-next and merged to master the day after I reported this issue.

FWIW, 639ff87a1a resolves the issue by ensuring the mutex_init happens even if there is no vfio-ap device attached so long as vfio_ap is configured/built (and I like the use of the constructor).

I also now see the stub functions in hw/s390x/ap-stub.c that will be called if vfio_ap is not configured.

So: my concerns are resolved.  Thanks!

Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Anthony Krowiak 5 months, 3 weeks ago

On 6/26/25 12:19 PM, Matthew Rosato wrote:
> On 6/9/25 12:44 PM, Rorie Reyes wrote:
>> Handle interception of the CHSC SEI instruction for requests
>> indicating the guest's AP configuration has changed.
>>
>> If configuring --without-default-devices, hw/s390x/ap-stub.c
>> was created to handle such circumstance. Also added the
>> following to hw/s390x/meson.build if CONFIG_VFIO_AP is
>> false, it will use the stub file.
>>
>> Signed-off-by: Rorie Reyes <rreyes@linux.ibm.com>
>> Reviewed-by: Anthony Krowiak <akrowiak@linux.ibm.com>
>> Reviewed-by: Cédric Le Goater <clg@redhat.com>
> FYI, this patch (or some part of this series) breaks hotplug for PCI devices on s390x.  I verified it breaks with both PCI passthrough and virtio-pci via virsh attach-device.
>
> I get greeted with the following message:
> qemu-system-s390x: ../util/qemu-thread-posix.c:92: qemu_mutex_lock_impl: Assertion `mutex->initialized' failed.
> 2025-06-26 14:12:23.031+0000: shutting down, reason=crashed
>
> As a quick test, reverting this patch resolves the issue.
>
> Another test shows that if I have a vfio-ap device attached to my guest before I try the zPCI hotplug, everything works fine (because vfio_ap_realize was driven -- see more info below).
>
> I note in the failing call chain that in response to the hotplug event for a zPCI device we end up calling into hw/vfio/ap.c and getting some locks which seems wrong:
>
> (gdb) bt
> #0  __pthread_kill_implementation (threadid=<optimized out>, signo=signo@entry=6, no_tid=no_tid@entry=0) at pthread_kill.c:44
> #1  0x000003ffa85ad636 in __pthread_kill_internal (threadid=<optimized out>, signo=6) at pthread_kill.c:78
> #2  0x000003ffa8553c20 in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26
> #3  0x000003ffa8533e3c in __GI_abort () at abort.c:79
> #4  0x000003ffa854ac9a in __assert_fail_base
>      (fmt=<optimized out>, assertion=assertion@entry=0x2aa3fa76414 "mutex->initialized", file=file@entry=0x2aa3fa763f8 "../util/qemu-thread-posix.c", line=line@entry=92, function=function@entry=0x2aa3fa7647a <__PRETTY_FUNCTION__.23> "qemu_mutex_lock_impl") at assert.c:94
> #5  0x000003ffa854acf4 in __assert_fail
>      (assertion=0x2aa3fa76414 "mutex->initialized", file=0x2aa3fa763f8 "../util/qemu-thread-posix.c", line=92, function=0x2aa3fa7647a <__PRETTY_FUNCTION__.23> "qemu_mutex_lock_impl") at assert.c:103
> #6  0x000002aa3f83ef02 in qemu_mutex_lock_impl (mutex=0x2aa3fed9e08 <cfg_chg_events_lock>, file=0x2aa3f9e653e "/usr/src/qemu/include/qemu/lockable.h", line=56) at ../util/qemu-thread-posix.c:92
> #7  0x000002aa3f4c3a70 in qemu_lockable_mutex_lock (x=0x2aa3fed9e08 <cfg_chg_events_lock>) at /usr/src/qemu/include/qemu/lockable.h:56
> #8  0x000002aa3f4c3b00 in qemu_lockable_lock (x=0x3fd076f7730) at /usr/src/qemu/include/qemu/lockable.h:110
> #9  0x000002aa3f4c3b82 in qemu_lockable_auto_lock (x=0x3fd076f7730) at /usr/src/qemu/include/qemu/lockable.h:120
> #10 0x000002aa3f4c3f90 in ap_chsc_sei_nt0_get_event (res=0x3fd076f79b8) at ../hw/vfio/ap.c:110
> ^^^^^^^^
> #11 0x000002aa3f3a59ea in chsc_sei_nt0_get_event (res=0x3fd076f79b8) at ../target/s390x/ioinst.c:579
> #12 0x000002aa3f3a5b76 in ioinst_handle_chsc_sei (req=0x3fd076f79a8, res=0x3fd076f79b8) at ../target/s390x/ioinst.c:620
> #13 0x000002aa3f3a6062 in ioinst_handle_chsc (cpu=0x2aa40cc9430, ipb=2097152, ra=0) at ../target/s390x/ioinst.c:708
> #14 0x000002aa3f4647d2 in handle_b2 (cpu=0x2aa40cc9430, run=0x3ffa4180000, ipa1=95 '_') at ../target/s390x/kvm/kvm.c:1218
> #15 0x000002aa3f465db2 in handle_instruction (cpu=0x2aa40cc9430, run=0x3ffa4180000) at ../target/s390x/kvm/kvm.c:1627
> #16 0x000002aa3f4661da in handle_intercept (cpu=0x2aa40cc9430) at ../target/s390x/kvm/kvm.c:1709
> #17 0x000002aa3f466c10 in kvm_arch_handle_exit (cs=0x2aa40cc9430, run=0x3ffa4180000) at ../target/s390x/kvm/kvm.c:1914
> #18 0x000002aa3f523c98 in kvm_cpu_exec (cpu=0x2aa40cc9430) at ../accel/kvm/kvm-all.c:3294
> #19 0x000002aa3f527f52 in kvm_vcpu_thread_fn (arg=0x2aa40cc9430) at ../accel/kvm/kvm-accel-ops.c:51
> #20 0x000002aa3f840002 in qemu_thread_start (args=0x2aa40ccd090) at ../util/qemu-thread-posix.c:393
> #21 0x000003ffa85ab622 in start_thread (arg=0x3fd076fe8c0) at pthread_create.c:448
> #22 0x000003ffa862b390 in thread_start () at ../sysdeps/unix/sysv/linux/s390/s390-64/clone3.S:71
>
>
> Note that PCI devices were already using SEI before this series (e.g. see hw/s390x/s390-pci-bus.c:pci_chsc_sei_nt2_get_event()); I think the problem here is that nt0 events are looked for implicitly in the shared handler (ioinst_handle_chsc_sei()) which is fine -- but you cannot go acquiring locks from ap.c that might have never been initialized (cfg_chg_events_lock which is only initialized in vfio_ap_realize()); that's why we ultimately crash.

The following check is done in both the chsc_sei_nt0_get_event and 
chsc_sei_nt0_have_event
functions before calling into the AP versions of those functions:

     if (s390_has_feat(S390_FEAT_AP)) {
         return ap_chsc_sei_nt0_get_event(res);
     }

Given that the call to AP is made, it looks like AP is installed on the 
guest. So it seems the reason
that the vfio_ap_realize() function does not get called is because a 
mediated device is not
passed through to the guest. This check alone, therefore, does not 
protect us against making
a call into the ap_chsc_sei_nt0_get/have_event() functions which results 
in the problem you
are seeing.

>
> AFAICT this portion of the handler including mutex should be moved out of ap.c into chsc_sei_nt0_get_event().  When vfio_ap is not in use, we can't call into it.  Can't you also build without VFIO_AP?  I didn't try it but that sure seems like it would cause issues too.

If that were the case, I think the checks I mentioned above would be 
sufficient because
the S390_FEAT_AP feature could not be set.

>    
>
> I guess the mutex and list would then need to be part of something common to s390x like S390CcwMachineState.
>
> Alternatively, you could leave those in vfio_ap but would need some kind of indicator that vfio_ap is "armed" for these kinds of events and only bother calling into vfio_ap to look at cfg_chg_events when that indicator has been set (e.g. vfio_ap_realize has been called once.  And it actually looks like you already do something like this via lock_initialized in ap.c).
> Otherwise, we should always return EVENT_INFORMATION_NOT_STORED like the old code did (via return 1).
>
> Or maybe someone else has a better idea.

What about moving lock initialization to the vfio_ap_class_init() function?

>
>
>> ---
>>   MAINTAINERS           |  1 +
>>   hw/s390x/ap-stub.c    | 21 +++++++++++++++++++++
>>   hw/s390x/meson.build  |  1 +
>>   target/s390x/ioinst.c | 11 +++++++++--
>>   4 files changed, 32 insertions(+), 2 deletions(-)
>>   create mode 100644 hw/s390x/ap-stub.c
>>
>> diff --git a/MAINTAINERS b/MAINTAINERS
>> index aa6763077e..1e84bfeaee 100644
>> --- a/MAINTAINERS
>> +++ b/MAINTAINERS
>> @@ -112,6 +112,7 @@ F: hw/intc/s390_flic.c
>>   F: hw/intc/s390_flic_kvm.c
>>   F: hw/s390x/
>>   F: hw/vfio/ap.c
>> +F: hw/s390x/ap-stub.c
>>   F: hw/vfio/ccw.c
>>   F: hw/watchdog/wdt_diag288.c
>>   F: include/hw/s390x/
>> diff --git a/hw/s390x/ap-stub.c b/hw/s390x/ap-stub.c
>> new file mode 100644
>> index 0000000000..001fe5f8b0
>> --- /dev/null
>> +++ b/hw/s390x/ap-stub.c
>> @@ -0,0 +1,21 @@
>> +/*
>> + * VFIO based AP matrix device assignment
>> + *
>> + * Copyright 2025 IBM Corp.
>> + * Author(s): Rorie Reyes <rreyes@linux.ibm.com>
>> + *
>> + * SPDX-License-Identifier: GPL-2.0-or-later
>> + */
>> +
>> +#include "qemu/osdep.h"
>> +#include "hw/s390x/ap-bridge.h"
>> +
>> +int ap_chsc_sei_nt0_get_event(void *res)
>> +{
>> +    return EVENT_INFORMATION_NOT_STORED;
>> +}
>> +
>> +bool ap_chsc_sei_nt0_have_event(void)
>> +{
>> +    return false;
>> +}
>> diff --git a/hw/s390x/meson.build b/hw/s390x/meson.build
>> index 3bbebfd817..99cbcbd7d6 100644
>> --- a/hw/s390x/meson.build
>> +++ b/hw/s390x/meson.build
>> @@ -33,6 +33,7 @@ s390x_ss.add(when: 'CONFIG_S390_CCW_VIRTIO', if_true: files(
>>   ))
>>   s390x_ss.add(when: 'CONFIG_TERMINAL3270', if_true: files('3270-ccw.c'))
>>   s390x_ss.add(when: 'CONFIG_VFIO', if_true: files('s390-pci-vfio.c'))
>> +s390x_ss.add(when: 'CONFIG_VFIO_AP', if_false: files('ap-stub.c'))
>>   
>>   virtio_ss = ss.source_set()
>>   virtio_ss.add(files('virtio-ccw.c'))
>> diff --git a/target/s390x/ioinst.c b/target/s390x/ioinst.c
>> index fe62ba5b06..2320dd4c12 100644
>> --- a/target/s390x/ioinst.c
>> +++ b/target/s390x/ioinst.c
>> @@ -18,6 +18,7 @@
>>   #include "trace.h"
>>   #include "hw/s390x/s390-pci-bus.h"
>>   #include "target/s390x/kvm/pv.h"
>> +#include "hw/s390x/ap-bridge.h"
>>   
>>   /* All I/O instructions but chsc use the s format */
>>   static uint64_t get_address_from_regs(CPUS390XState *env, uint32_t ipb,
>> @@ -574,13 +575,19 @@ out:
>>   
>>   static int chsc_sei_nt0_get_event(void *res)
>>   {
>> -    /* no events yet */
>> +    if (s390_has_feat(S390_FEAT_AP)) {
>> +        return ap_chsc_sei_nt0_get_event(res);
>> +    }
>> +
>>       return 1;
>>   }
>>   
>>   static int chsc_sei_nt0_have_event(void)
>>   {
>> -    /* no events yet */
>> +    if (s390_has_feat(S390_FEAT_AP)) {
>> +        return ap_chsc_sei_nt0_have_event();
>> +    }
>> +
>>       return 0;
>>   }
>>   


Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Matthew Rosato 5 months, 3 weeks ago
>>
>> AFAICT this portion of the handler including mutex should be moved out of ap.c into chsc_sei_nt0_get_event().  When vfio_ap is not in use, we can't call into it.  Can't you also build without VFIO_AP?  I didn't try it but that sure seems like it would cause issues too.
> 
> If that were the case, I think the checks I mentioned above would be sufficient because
> the S390_FEAT_AP feature could not be set.
> 

OK, I tested it quick (remove VFIO_AP from config) and that appears to be the case; this also restores zPCI hotplug capability since we won't call info hw/vfio/ap.c/.

> 
> What about moving lock initialization to the vfio_ap_class_init() function?
> 

I tried a quick hack of that too and it appears to work, at least for the zPCI regression (note: I did not test the AP portion).

This approach is probably not the way I would have designed it initially...  But it does seem to be the simplest thing to do from a fix perspective.  

Can you or Rorie prepare a proper patch w/ a fixes tag and send (please add me to CC). Also please test the change against the vfio-ap environment; you can also test zPCI but I would also like to run some tests.

Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Anthony Krowiak 5 months, 3 weeks ago
Sure

On 6/27/25 11:12 AM, Matthew Rosato wrote:
>>> AFAICT this portion of the handler including mutex should be moved out of ap.c into chsc_sei_nt0_get_event().  When vfio_ap is not in use, we can't call into it.  Can't you also build without VFIO_AP?  I didn't try it but that sure seems like it would cause issues too.
>> If that were the case, I think the checks I mentioned above would be sufficient because
>> the S390_FEAT_AP feature could not be set.
>>
> OK, I tested it quick (remove VFIO_AP from config) and that appears to be the case; this also restores zPCI hotplug capability since we won't call info hw/vfio/ap.c/.
>
>> What about moving lock initialization to the vfio_ap_class_init() function?
>>
> I tried a quick hack of that too and it appears to work, at least for the zPCI regression (note: I did not test the AP portion).
>
> This approach is probably not the way I would have designed it initially...  But it does seem to be the simplest thing to do from a fix perspective.
>
> Can you or Rorie prepare a proper patch w/ a fixes tag and send (please add me to CC). Also please test the change against the vfio-ap environment; you can also test zPCI but I would also like to run some tests.


Re: [RFC PATCH v13 4/4] s390: implementing CHSC SEI for AP config change
Posted by Rorie Reyes 5 months, 3 weeks ago
On 6/27/25 11:12 AM, Matthew Rosato wrote:
>>> AFAICT this portion of the handler including mutex should be moved out of ap.c into chsc_sei_nt0_get_event().  When vfio_ap is not in use, we can't call into it.  Can't you also build without VFIO_AP?  I didn't try it but that sure seems like it would cause issues too.
>> If that were the case, I think the checks I mentioned above would be sufficient because
>> the S390_FEAT_AP feature could not be set.
>>
> OK, I tested it quick (remove VFIO_AP from config) and that appears to be the case; this also restores zPCI hotplug capability since we won't call info hw/vfio/ap.c/.
>
>> What about moving lock initialization to the vfio_ap_class_init() function?
>>
> I tried a quick hack of that too and it appears to work, at least for the zPCI regression (note: I did not test the AP portion).
>
> This approach is probably not the way I would have designed it initially...  But it does seem to be the simplest thing to do from a fix perspective.
>
> Can you or Rorie prepare a proper patch w/ a fixes tag and send (please add me to CC). Also please test the change against the vfio-ap environment; you can also test zPCI but I would also like to run some tests.
I can make those changes