On 5/22/25 20:55, Anthony Krowiak wrote:
On 5/22/25 9:30 AM, Cédric Le Goater wrote:
On 5/12/25 20:02, Rorie Reyes wrote:
These functions can be invoked by the function that handles interception
of the CHSC SEI instruction for requests indicating the accessibility of
one or more adjunct processors has changed.
Signed-off-by: Rorie Reyes <rre...@linux.ibm.com>
---
hw/vfio/ap.c | 39 ++++++++++++++++++++++++++++++++++++
include/hw/s390x/ap-bridge.h | 22 ++++++++++++++++++++
2 files changed, 61 insertions(+)
diff --git a/hw/vfio/ap.c b/hw/vfio/ap.c
index 5ea5dd9cca..4f88f80c54 100644
--- a/hw/vfio/ap.c
+++ b/hw/vfio/ap.c
@@ -96,6 +96,45 @@ static void vfio_ap_cfg_chg_notifier_handler(void *opaque)
}
+int ap_chsc_sei_nt0_get_event(void *res)
+{
+ ChscSeiNt0Res *nt0_res = (ChscSeiNt0Res *)res;
+ APConfigChgEvent *cfg_chg_event;
+
+ if (!ap_chsc_sei_nt0_have_event()) {
+ return 1;
+ }
+
+ cfg_chg_event = QTAILQ_FIRST(&cfg_chg_events);
+ memset(nt0_res, 0, sizeof(*nt0_res));
+
+ QTAILQ_REMOVE(&cfg_chg_events, cfg_chg_event, next);
btw, I don't know if this was discussed. Are we OK to manipulate the
'cfg_chg_events' construct withou locking ?
This has never been discussed, but it's an interesting question. The
ap_chsc_sei_nt0_get_event and ap_chsc_sei_nt0_have_event functions
are called as a result of a SIE exit to handle interception of a
CHSC SEI instruction. Handling of the intercepted instructions is
done under the Big QEMU Lock (see kvm_arch_handle_exit in
target/s390x/kvm/kvm.c),
so presumably no other processes will get access to these functions
until the instruction is handled.
On the other hand, the vfio_cfg_chg_notifier_handler function that handles the
eventfd
indicating the guest's AP configuration has been changed by the host device
driver
adds APConfigChgEvent objects to this queue. If, however, you think about the
flow,
when the notifier handler gets called to handle an AP config changed event, it
queues a channel request word (CRW) indicating there is an SEI pending.
Consequently,
the ap_chsc_sei_nt0_get_event and ap_chsc_sei_nt0_have_event functions will get
called
only after the guest receives the CRW event and executes the CHSC SEI
instruction. Since
the Big QEMU Lock is taken when the CHSC SE instruction is intercepted, it can
not proceed
until whatever the holding process releases it; so for that flow, it seems
highly likely if not
impossible for conflict given the event will always be added to the queue
before an attempt
can be made to retrieve it.
Having gone through this dissertation, I don't see how it can hurt to lock the
queue when
it is being accessed and would certainly make things bullet proof. What is your
opinion?
In that case, let's keep it simple (no mutex) and add a assert(bql_locked())
statement where we think the bql should be protecting access to shared
resources.
Thanks,
C.
