* [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
@ 2016-12-19 16:04 Radim Krčmář
2016-12-22 10:02 ` Paolo Bonzini
2016-12-22 10:45 ` Paolo Bonzini
0 siblings, 2 replies; 6+ messages in thread
From: Radim Krčmář @ 2016-12-19 16:04 UTC (permalink / raw)
To: kvm; +Cc: Paolo Bonzini, David Hildenbrand
Broadcast to address 0xff is not exercised by seabios+linux so we better
test it.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
---
The bounded wait loop is bad ... time to introduce TSC magic?
---
x86/apic.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 46 insertions(+)
diff --git a/x86/apic.c b/x86/apic.c
index 39c7fd19913d..b25ba45e4b8f 100644
--- a/x86/apic.c
+++ b/x86/apic.c
@@ -5,6 +5,7 @@
#include "desc.h"
#include "isr.h"
#include "msr.h"
+#include "atomic.h"
static void test_lapic_existence(void)
{
@@ -16,6 +17,7 @@ static void test_lapic_existence(void)
}
#define TSC_DEADLINE_TIMER_VECTOR 0xef
+#define BROADCAST_VECTOR 0xcf
static int tdt_count;
@@ -411,6 +413,49 @@ static void test_apic_timer_one_shot(void)
(tsc2 - tsc1 >= interval));
}
+static atomic_t broadcast_counter;
+
+static void broadcast_handler(isr_regs_t *regs)
+{
+ atomic_inc(&broadcast_counter);
+ eoi();
+}
+
+static bool broadcast_received(unsigned ncpus)
+{
+ unsigned counter;
+
+ for (int i = 123456789; i; i--) {
+ counter = atomic_read(&broadcast_counter);
+ if (counter >= ncpus)
+ break;
+ pause();
+ }
+
+ atomic_set(&broadcast_counter, 0);
+
+ return counter == ncpus;
+}
+
+static void test_physical_broadcast(void)
+{
+ unsigned ncpus = cpu_count();
+ unsigned long cr3 = read_cr3();
+ u32 broadcast_address = enable_x2apic() ? 0xffffffff : 0xff;
+
+ handle_irq(BROADCAST_VECTOR, broadcast_handler);
+ for (int c = 1; c < ncpus; c++)
+ on_cpu(c, update_cr3, (void *)cr3);
+
+ apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_FIXED | APIC_INT_ASSERT |
+ BROADCAST_VECTOR, broadcast_address);
+ report("APIC physical broadcast address", broadcast_received(ncpus));
+
+ apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_FIXED | APIC_INT_ASSERT |
+ BROADCAST_VECTOR | APIC_DEST_ALLINC, 0);
+ report("APIC physical broadcast shorthand", broadcast_received(ncpus));
+}
+
int main()
{
setup_vm();
@@ -425,6 +470,7 @@ int main()
test_apicbase();
test_self_ipi();
+ test_physical_broadcast();
test_sti_nmi();
test_multiple_nmi();
--
2.11.0
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
2016-12-19 16:04 [kvm-unit-tests PATCH] x86: apic: add physical broadcast test Radim Krčmář
@ 2016-12-22 10:02 ` Paolo Bonzini
2016-12-22 18:23 ` Radim Krčmář
2016-12-22 10:45 ` Paolo Bonzini
1 sibling, 1 reply; 6+ messages in thread
From: Paolo Bonzini @ 2016-12-22 10:02 UTC (permalink / raw)
To: Radim Krčmář, kvm; +Cc: David Hildenbrand
On 19/12/2016 17:04, Radim Krčmář wrote:
> Broadcast to address 0xff is not exercised by seabios+linux so we better
> test it.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> ---
> The bounded wait loop is bad ... time to introduce TSC magic?
Yeah, the VMX tests already use the TSC (though in the other direction:
they check that _at least_ some time passed). Roughly speaking, 1
billion TSC units should be plenty, even on a loaded hosts.
Paolo
> x86/apic.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 46 insertions(+)
>
> diff --git a/x86/apic.c b/x86/apic.c
> index 39c7fd19913d..b25ba45e4b8f 100644
> --- a/x86/apic.c
> +++ b/x86/apic.c
> @@ -5,6 +5,7 @@
> #include "desc.h"
> #include "isr.h"
> #include "msr.h"
> +#include "atomic.h"
>
> static void test_lapic_existence(void)
> {
> @@ -16,6 +17,7 @@ static void test_lapic_existence(void)
> }
>
> #define TSC_DEADLINE_TIMER_VECTOR 0xef
> +#define BROADCAST_VECTOR 0xcf
>
> static int tdt_count;
>
> @@ -411,6 +413,49 @@ static void test_apic_timer_one_shot(void)
> (tsc2 - tsc1 >= interval));
> }
>
> +static atomic_t broadcast_counter;
> +
> +static void broadcast_handler(isr_regs_t *regs)
> +{
> + atomic_inc(&broadcast_counter);
> + eoi();
> +}
> +
> +static bool broadcast_received(unsigned ncpus)
> +{
> + unsigned counter;
> +
> + for (int i = 123456789; i; i--) {
> + counter = atomic_read(&broadcast_counter);
> + if (counter >= ncpus)
> + break;
> + pause();
> + }
> +
> + atomic_set(&broadcast_counter, 0);
> +
> + return counter == ncpus;
> +}
> +
> +static void test_physical_broadcast(void)
> +{
> + unsigned ncpus = cpu_count();
> + unsigned long cr3 = read_cr3();
> + u32 broadcast_address = enable_x2apic() ? 0xffffffff : 0xff;
> +
> + handle_irq(BROADCAST_VECTOR, broadcast_handler);
> + for (int c = 1; c < ncpus; c++)
> + on_cpu(c, update_cr3, (void *)cr3);
> +
> + apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_FIXED | APIC_INT_ASSERT |
> + BROADCAST_VECTOR, broadcast_address);
> + report("APIC physical broadcast address", broadcast_received(ncpus));
> +
> + apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_FIXED | APIC_INT_ASSERT |
> + BROADCAST_VECTOR | APIC_DEST_ALLINC, 0);
> + report("APIC physical broadcast shorthand", broadcast_received(ncpus));
> +}
> +
> int main()
> {
> setup_vm();
> @@ -425,6 +470,7 @@ int main()
> test_apicbase();
>
> test_self_ipi();
> + test_physical_broadcast();
>
> test_sti_nmi();
> test_multiple_nmi();
>
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
2016-12-19 16:04 [kvm-unit-tests PATCH] x86: apic: add physical broadcast test Radim Krčmář
2016-12-22 10:02 ` Paolo Bonzini
@ 2016-12-22 10:45 ` Paolo Bonzini
2016-12-22 18:15 ` Radim Krčmář
1 sibling, 1 reply; 6+ messages in thread
From: Paolo Bonzini @ 2016-12-22 10:45 UTC (permalink / raw)
To: Radim Krčmář, kvm; +Cc: David Hildenbrand
On 19/12/2016 17:04, Radim Krčmář wrote:
> Broadcast to address 0xff is not exercised by seabios+linux so we better
> test it.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> ---
> The bounded wait loop is bad ... time to introduce TSC magic?
> ---
> x86/apic.c | 46 ++++++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 46 insertions(+)
>
> diff --git a/x86/apic.c b/x86/apic.c
> index 39c7fd19913d..b25ba45e4b8f 100644
> --- a/x86/apic.c
> +++ b/x86/apic.c
> @@ -5,6 +5,7 @@
> #include "desc.h"
> #include "isr.h"
> #include "msr.h"
> +#include "atomic.h"
>
> static void test_lapic_existence(void)
> {
> @@ -16,6 +17,7 @@ static void test_lapic_existence(void)
> }
>
> #define TSC_DEADLINE_TIMER_VECTOR 0xef
> +#define BROADCAST_VECTOR 0xcf
>
> static int tdt_count;
>
> @@ -411,6 +413,49 @@ static void test_apic_timer_one_shot(void)
> (tsc2 - tsc1 >= interval));
> }
>
> +static atomic_t broadcast_counter;
> +
> +static void broadcast_handler(isr_regs_t *regs)
> +{
> + atomic_inc(&broadcast_counter);
> + eoi();
> +}
> +
> +static bool broadcast_received(unsigned ncpus)
> +{
> + unsigned counter;
> +
> + for (int i = 123456789; i; i--) {
> + counter = atomic_read(&broadcast_counter);
> + if (counter >= ncpus)
> + break;
> + pause();
> + }
> +
> + atomic_set(&broadcast_counter, 0);
> +
> + return counter == ncpus;
> +}
> +
> +static void test_physical_broadcast(void)
> +{
> + unsigned ncpus = cpu_count();
> + unsigned long cr3 = read_cr3();
> + u32 broadcast_address = enable_x2apic() ? 0xffffffff : 0xff;
> +
> + handle_irq(BROADCAST_VECTOR, broadcast_handler);
> + for (int c = 1; c < ncpus; c++)
> + on_cpu(c, update_cr3, (void *)cr3);
> +
> + apic_icr_write(APIC_DEST_PHYSICAL | APIC_DM_FIXED | APIC_INT_ASSERT |
> + BROADCAST_VECTOR, broadcast_address);
> + report("APIC physical broadcast address", broadcast_received(ncpus));
This test is broken by
commit f11e8b051ef613670f9db77a79726534dbc667ec
Author: Radim Krčmář <rkrcmar@redhat.com>
Date: Tue Dec 13 17:30:00 2016 +0100
KVM: x86: make interrupt delivery fast and slow path behave the same
Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
to xAPIC VCPUs and vice-versa. Make slow path behave like fast path,
which never distinguished that.
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
Fixes: 682f732ecf73 ("KVM: x86: bump MAX_VCPUS to 288")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
when run with "./x86/run x86/apic.flat -smp 2 -cpu Nehalem,-x2apic", so I'm removing
that series from kvm/queue.
Paolo
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
2016-12-22 10:45 ` Paolo Bonzini
@ 2016-12-22 18:15 ` Radim Krčmář
2016-12-22 18:24 ` Paolo Bonzini
0 siblings, 1 reply; 6+ messages in thread
From: Radim Krčmář @ 2016-12-22 18:15 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: kvm, David Hildenbrand
2016-12-22 11:45+0100, Paolo Bonzini:
> On 19/12/2016 17:04, Radim Krčmář wrote:
>> Broadcast to address 0xff is not exercised by seabios+linux so we better
>> test it.
>>
>> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
>> ---
>
> This test is broken by
>
> commit f11e8b051ef613670f9db77a79726534dbc667ec
> Author: Radim Krčmář <rkrcmar@redhat.com>
> Date: Tue Dec 13 17:30:00 2016 +0100
>
> KVM: x86: make interrupt delivery fast and slow path behave the same
>
> Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
> to xAPIC VCPUs and vice-versa. Make slow path behave like fast path,
> which never distinguished that.
>
> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
> Fixes: 682f732ecf73 ("KVM: x86: bump MAX_VCPUS to 288")
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>
> when run with "./x86/run x86/apic.flat -smp 2 -cpu Nehalem,-x2apic", so I'm removing
> that series from kvm/queue.
Yes, kvm/queue had v2 of the series and I have added the test because
David noticed the broadcast bug while reviewing v3 of that series.
(http://www.spinics.net/lists/kvm/msg142614.html)
The bug was fixed in v4. I have only posted v4 of this one patch as the
rest seemed ok in v3.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
2016-12-22 10:02 ` Paolo Bonzini
@ 2016-12-22 18:23 ` Radim Krčmář
0 siblings, 0 replies; 6+ messages in thread
From: Radim Krčmář @ 2016-12-22 18:23 UTC (permalink / raw)
To: Paolo Bonzini; +Cc: kvm, David Hildenbrand
2016-12-22 11:02+0100, Paolo Bonzini:
> On 19/12/2016 17:04, Radim Krčmář wrote:
>> Broadcast to address 0xff is not exercised by seabios+linux so we better
>> test it.
>>
>> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
>> ---
>> The bounded wait loop is bad ... time to introduce TSC magic?
>
> Yeah, the VMX tests already use the TSC (though in the other direction:
> they check that _at least_ some time passed). Roughly speaking, 1
> billion TSC units should be plenty, even on a loaded hosts.
I will convert it, thanks.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [kvm-unit-tests PATCH] x86: apic: add physical broadcast test
2016-12-22 18:15 ` Radim Krčmář
@ 2016-12-22 18:24 ` Paolo Bonzini
0 siblings, 0 replies; 6+ messages in thread
From: Paolo Bonzini @ 2016-12-22 18:24 UTC (permalink / raw)
To: Radim Krčmář; +Cc: kvm, David Hildenbrand
On 22/12/2016 19:15, Radim Krčmář wrote:
> 2016-12-22 11:45+0100, Paolo Bonzini:
>> On 19/12/2016 17:04, Radim Krčmář wrote:
>>> Broadcast to address 0xff is not exercised by seabios+linux so we better
>>> test it.
>>>
>>> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
>>> ---
>>
>> This test is broken by
>>
>> commit f11e8b051ef613670f9db77a79726534dbc667ec
>> Author: Radim Krčmář <rkrcmar@redhat.com>
>> Date: Tue Dec 13 17:30:00 2016 +0100
>>
>> KVM: x86: make interrupt delivery fast and slow path behave the same
>>
>> Slow path tried to prevent IPIs from x2APIC VCPUs from being delivered
>> to xAPIC VCPUs and vice-versa. Make slow path behave like fast path,
>> which never distinguished that.
>>
>> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
>> Fixes: 682f732ecf73 ("KVM: x86: bump MAX_VCPUS to 288")
>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>>
>> when run with "./x86/run x86/apic.flat -smp 2 -cpu Nehalem,-x2apic", so I'm removing
>> that series from kvm/queue.
>
> Yes, kvm/queue had v2 of the series and I have added the test because
> David noticed the broadcast bug while reviewing v3 of that series.
> (http://www.spinics.net/lists/kvm/msg142614.html)
>
> The bug was fixed in v4. I have only posted v4 of this one patch as the
> rest seemed ok in v3.
Oh, okay, then I'll recheck and fix it.
Currently I'm seeing WinXP/2003 failures with apicv=0 and Win2008r2
failures with apicv=0 ept=0. I guess I'll have to have some fun bisecting.
Paolo
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2016-12-22 18:24 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-12-19 16:04 [kvm-unit-tests PATCH] x86: apic: add physical broadcast test Radim Krčmář
2016-12-22 10:02 ` Paolo Bonzini
2016-12-22 18:23 ` Radim Krčmář
2016-12-22 10:45 ` Paolo Bonzini
2016-12-22 18:15 ` Radim Krčmář
2016-12-22 18:24 ` Paolo Bonzini
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.