KVM: s390: fix gisa destroy operation might lead to cpu stalls
authorMichael Mueller <mimu@linux.ibm.com>
Fri, 1 Sep 2023 10:58:23 +0000 (12:58 +0200)
committerClaudio Imbrenda <imbrenda@linux.ibm.com>
Mon, 25 Sep 2023 06:31:47 +0000 (08:31 +0200)
A GISA cannot be destroyed as long it is linked in the GIB alert list
as this would break the alert list. Just waiting for its removal from
the list triggered by another vm is not sufficient as it might be the
only vm. The below shown cpu stall situation might occur when GIB alerts
are delayed and is fixed by calling process_gib_alert_list() instead of
waiting.

At this time the vcpus of the vm are already destroyed and thus
no vcpu can be kicked to enter the SIE again if for some reason an
interrupt is pending for that vm.

Additionally the IAM restore value is set to 0x00. That would be a bug
introduced by incomplete device de-registration, i.e. missing
kvm_s390_gisc_unregister() call.

Setting this value and the IAM in the GISA to 0x00 guarantees that late
interrupts don't bring the GISA back into the alert list.

CPU stall caused by kvm_s390_gisa_destroy():

 [ 4915.311372] rcu: INFO: rcu_sched detected expedited stalls on CPUs/tasks: { 14-.... } 24533 jiffies s: 5269 root: 0x1/.
 [ 4915.311390] rcu: blocking rcu_node structures (internal RCU debug): l=1:0-15:0x4000/.
 [ 4915.311394] Task dump for CPU 14:
 [ 4915.311395] task:qemu-system-s39 state:R  running task     stack:0     pid:217198 ppid:1      flags:0x00000045
 [ 4915.311399] Call Trace:
 [ 4915.311401]  [<0000038003a33a10>] 0x38003a33a10
 [ 4933.861321] rcu: INFO: rcu_sched self-detected stall on CPU
 [ 4933.861332] rcu:  14-....: (42008 ticks this GP) idle=53f4/1/0x4000000000000000 softirq=61530/61530 fqs=14031
 [ 4933.861353] rcu:  (t=42008 jiffies g=238109 q=100360 ncpus=18)
 [ 4933.861357] CPU: 14 PID: 217198 Comm: qemu-system-s39 Not tainted 6.5.0-20230816.rc6.git26.a9d17c5d8813.300.fc38.s390x #1
 [ 4933.861360] Hardware name: IBM 8561 T01 703 (LPAR)
 [ 4933.861361] Krnl PSW : 0704e00180000000 000003ff804bfc66 (kvm_s390_gisa_destroy+0x3e/0xe0 [kvm])
 [ 4933.861414]            R:0 T:1 IO:1 EX:1 Key:0 M:1 W:0 P:0 AS:3 CC:2 PM:0 RI:0 EA:3
 [ 4933.861416] Krnl GPRS: 0000000000000000 00000372000000fc 00000002134f8000 000000000d5f5900
 [ 4933.861419]            00000002f5ea1d18 00000002f5ea1d18 0000000000000000 0000000000000000
 [ 4933.861420]            00000002134fa890 00000002134f8958 000000000d5f5900 00000002134f8000
 [ 4933.861422]            000003ffa06acf98 000003ffa06858b0 0000038003a33c20 0000038003a33bc8
 [ 4933.861430] Krnl Code: 000003ff804bfc58ec66002b007e cij %r6,0,6,000003ff804bfcae
                           000003ff804bfc5eb904003a lgr %r3,%r10
                          #000003ff804bfc62a7f40005 brc 15,000003ff804bfc6c
                          >000003ff804bfc66e330b7300204 lg %r3,10032(%r11)
                           000003ff804bfc6c58003000 l %r0,0(%r3)
                           000003ff804bfc70ec03fffb6076 crj %r0,%r3,6,000003ff804bfc66
                           000003ff804bfc76e320b7600271 lay %r2,10080(%r11)
                           000003ff804bfc7cc0e5fffea339 brasl %r14,000003ff804942ee
 [ 4933.861444] Call Trace:
 [ 4933.861445]  [<000003ff804bfc66>] kvm_s390_gisa_destroy+0x3e/0xe0 [kvm]
 [ 4933.861460] ([<00000002623523de>] free_unref_page+0xee/0x148)
 [ 4933.861507]  [<000003ff804aea98>] kvm_arch_destroy_vm+0x50/0x120 [kvm]
 [ 4933.861521]  [<000003ff8049d374>] kvm_destroy_vm+0x174/0x288 [kvm]
 [ 4933.861532]  [<000003ff8049d4fe>] kvm_vm_release+0x36/0x48 [kvm]
 [ 4933.861542]  [<00000002623cd04a>] __fput+0xea/0x2a8
 [ 4933.861547]  [<00000002620d5bf8>] task_work_run+0x88/0xf0
 [ 4933.861551]  [<00000002620b0aa6>] do_exit+0x2c6/0x528
 [ 4933.861556]  [<00000002620b0f00>] do_group_exit+0x40/0xb8
 [ 4933.861557]  [<00000002620b0fa6>] __s390x_sys_exit_group+0x2e/0x30
 [ 4933.861559]  [<0000000262d481f4>] __do_syscall+0x1d4/0x200
 [ 4933.861563]  [<0000000262d59028>] system_call+0x70/0x98
 [ 4933.861565] Last Breaking-Event-Address:
 [ 4933.861566]  [<0000038003a33b60>] 0x38003a33b60

Fixes: 9f30f6216378 ("KVM: s390: add gib_alert_irq_handler()")
Signed-off-by: Michael Mueller <mimu@linux.ibm.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.ibm.com>
Reviewed-by: Nico Boehr <nrb@linux.ibm.com>
Link: https://lore.kernel.org/r/20230901105823.3973928-1-mimu@linux.ibm.com
Message-ID: <20230901105823.3973928-1-mimu@linux.ibm.com>
Signed-off-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
arch/s390/kvm/interrupt.c

index c1b47d6..efaebba 100644 (file)
@@ -303,11 +303,6 @@ static inline u8 gisa_get_ipm_or_restore_iam(struct kvm_s390_gisa_interrupt *gi)
        return 0;
 }
 
-static inline int gisa_in_alert_list(struct kvm_s390_gisa *gisa)
-{
-       return READ_ONCE(gisa->next_alert) != (u32)virt_to_phys(gisa);
-}
-
 static inline void gisa_set_ipm_gisc(struct kvm_s390_gisa *gisa, u32 gisc)
 {
        set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
@@ -3216,11 +3211,12 @@ void kvm_s390_gisa_destroy(struct kvm *kvm)
 
        if (!gi->origin)
                return;
-       if (gi->alert.mask)
-               KVM_EVENT(3, "vm 0x%pK has unexpected iam 0x%02x",
-                         kvm, gi->alert.mask);
-       while (gisa_in_alert_list(gi->origin))
-               cpu_relax();
+       WARN(gi->alert.mask != 0x00,
+            "unexpected non zero alert.mask 0x%02x",
+            gi->alert.mask);
+       gi->alert.mask = 0x00;
+       if (gisa_set_iam(gi->origin, gi->alert.mask))
+               process_gib_alert_list();
        hrtimer_cancel(&gi->timer);
        gi->origin = NULL;
        VM_EVENT(kvm, 3, "gisa 0x%pK destroyed", gisa);