[Qemu-devel] [PATCH v2 17&18/22] kvm: Unconditionally reenter kernel after IO exits

[Jan Kiszka]

On 2011-01-31 17:56, Gleb Natapov wrote:
>>>>>>> The only thing we miss by moving process_irqchip_events is a self-INIT
>>>>>>> of an AP - if such thing exists in real life. In that case, the AP would
>>>>>>> cause a reset of itself, followed by a transition to HALT state.
>>>>>>
>>>>>> I checked again with the Intel spec, and a self-INIT is invalid (at
>>>>>> least when specified via shorthand). So I'm under the impression now
>>>>>> that we can safely ignore this case and leave the patch as is.
>>>>>>
>>>>>> Any different views?
>>>>>>
>>>>> IIRC if you don't use shorthand you can send INIT to self.
>>>>
>>>> We didn't care so far (in qemu-kvm), do you think we should?
>>>>
>>> Doesn't kernel lapic emulation support this?
>>
>> See the my other mail: It supports it, but it apparently doesn't expects
>> this to happen.
>>
> I saw it, but I do not understand why do we print this message. May be
> it was used for debugging in early stages of KVM development.
> 

OK, lets' try to handle this in user space as well. The following patch
replaces both 17 & 18 from my original series as we can no longer split
things up.

Jan

--------8<--------

KVM requires to reenter the kernel after IO exits in order to complete
instruction emulation. Failing to do so will leave the kernel state
inconsistently behind. To ensure that we will get back ASAP, we issue a
self-signal that will cause KVM_RUN to return once the pending
operations are completed.

We can move kvm_arch_process_irqchip_events out of the inner VCPU loop.
The only state that mattered at its old place was a pending INIT
request. Catch it in kvm_arch_pre_run and also trigger a self-signal to
process the request on next kvm_cpu_exec.

This patch also fixes the missing exit_request check in kvm_cpu_exec in
the CONFIG_IOTHREAD case.

Signed-off-by: Jan Kiszka <address@hidden>
CC: Gleb Natapov <address@hidden>
---
 kvm-all.c         |   31 +++++++++++++++++--------------
 target-i386/kvm.c |    5 +++++
 2 files changed, 22 insertions(+), 14 deletions(-)

diff --git a/kvm-all.c b/kvm-all.c
index 5bfa8c0..d961697 100644
--- a/kvm-all.c
+++ b/kvm-all.c
@@ -199,7 +199,6 @@ int kvm_pit_in_kernel(void)
     return kvm_state->pit_in_kernel;
 }
 
-
 int kvm_init_vcpu(CPUState *env)
 {
     KVMState *s = kvm_state;
@@ -892,29 +891,33 @@ int kvm_cpu_exec(CPUState *env)
 
     DPRINTF("kvm_cpu_exec()\n");
 
-    do {
-#ifndef CONFIG_IOTHREAD
-        if (env->exit_request) {
-            DPRINTF("interrupt exit requested\n");
-            ret = 0;
-            break;
-        }
-#endif
-
-        if (kvm_arch_process_irqchip_events(env)) {
-            ret = 0;
-            break;
-        }
+    if (kvm_arch_process_irqchip_events(env)) {
+        env->exit_request = 0;
+        env->exception_index = EXCP_HLT;
+        return 0;
+    }
 
+    do {
         if (env->kvm_vcpu_dirty) {
             kvm_arch_put_registers(env, KVM_PUT_RUNTIME_STATE);
             env->kvm_vcpu_dirty = 0;
         }
 
         kvm_arch_pre_run(env, run);
+        if (env->exit_request) {
+            DPRINTF("interrupt exit requested\n");
+            /*
+             * KVM requires us to reenter the kernel after IO exits to complete
+             * instruction emulation. This self-signal will ensure that we
+             * leave ASAP again.
+             */
+            qemu_cpu_kick_self();
+        }
         cpu_single_env = NULL;
         qemu_mutex_unlock_iothread();
+
         ret = kvm_vcpu_ioctl(env, KVM_RUN, 0);
+
         qemu_mutex_lock_iothread();
         cpu_single_env = env;
         kvm_arch_post_run(env, run);
diff --git a/target-i386/kvm.c b/target-i386/kvm.c
index 9df8ff8..8a87244 100644
--- a/target-i386/kvm.c
+++ b/target-i386/kvm.c
@@ -1426,6 +1426,11 @@ int kvm_arch_get_registers(CPUState *env)
 
 int kvm_arch_pre_run(CPUState *env, struct kvm_run *run)
 {
+    /* Force the VCPU out of its inner loop to process the INIT request */
+    if (env->interrupt_request & CPU_INTERRUPT_INIT) {
+        env->exit_request = 1;
+    }
+
     /* Inject NMI */
     if (env->interrupt_request & CPU_INTERRUPT_NMI) {
         env->interrupt_request &= ~CPU_INTERRUPT_NMI;
-- 
1.7.1