On 2018-09-04 13:29, Ralf Ramsauer wrote:
On 02/09/2018 17:49, Jan Kiszka wrote:
From: Jan Kiszka <[email protected]>
This can provide hints why latencies are not as low as expected, even
without any hypervisor exits.
Signed-off-by: Jan Kiszka <[email protected]>
---
Changes in v2:
- hmpf, not all CPUs support this register...
like the Xeon E5 v4 (model 0x4f) in front of me...
According to Intel's SDM, there is no register MSR_SMI_COUNT for this
model (p. 248 f.), but register 0x34 isn't used for anything else.
Hmm, let's try to read it!
A first attempt returned 1825 across all cores, after a warm reboot it
increased to 182c, and after a cold reboot it increased again to 1835.
After the next reboot it dropped to 182f. Meh, doesn't look good. :-(
I don't know what the value reflects, probably nothing, but /maybe/ it's
a SMI count, though, I'll keep an eye on it.
A reboot (unless its kexec based) should reset that counter. It
typically increases during early boot-up, e.g. while the BIOS is
handling legacy devices accesses. However, your number is comparably high.
- fix typo "MSR_MSI_COUNT"
inmates/demos/x86/apic-demo.c | 45 ++++++++++++++++++++++++++++++++++++++++++-
1 file changed, 44 insertions(+), 1 deletion(-)
diff --git a/inmates/demos/x86/apic-demo.c b/inmates/demos/x86/apic-demo.c
index ce03872e..65f6f55e 100644
--- a/inmates/demos/x86/apic-demo.c
+++ b/inmates/demos/x86/apic-demo.c
@@ -12,24 +12,61 @@
#include <inmate.h>
+#define MSR_SMI_COUNT 0x34
+
#define POLLUTE_CACHE_SIZE (512 * 1024)
#define APIC_TIMER_VECTOR 32
static unsigned long expected_time;
static unsigned long min = -1, max;
+static bool has_smi_count;
+static u32 initial_smis;
+
+unsigned int smi_count_models[] = {
can become static
Actually, static const.
+ 0x37, 0x4a, 0x4d, 0x5a, 0x5d, 0x5c, 0x7a,
/* Silvermont */
Missing: 0x57, 0x85: Xeon Phi 3200/5200/7200 (ref. Intel SDM v4 p. 298)
I was copying from section 2.23 (MSR Index), but I missed the page
break. And Intel missed 06_85 in that index. Hopefully not more.
+ 0x1a, 0x1e, 0x1f, 0x2e,
/* Nehalem */
+ 0x2a, 0x2d,
/* Sandy Bridge */
Can add the comments, thanks for checking.
Jan
--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux
--
You received this message because you are subscribed to the Google Groups
"Jailhouse" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
