LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH] x86/tsc: Make cur->adjusted values in package#1 to be the same
@ 2021-08-31  7:08 Tony W Wang-oc
  0 siblings, 0 replies; only message in thread
From: Tony W Wang-oc @ 2021-08-31  7:08 UTC (permalink / raw)
  To: tglx, mingo, bp, x86, hpa, linux-kernel
  Cc: TimGuo-oc, CooperYan, QiyuanWang, HerryYang, CobeChen, FelixZhang

When resume from S4 on Zhaoxin 2 packages platform that support
X86_FEATURE_TSC_ADJUST, the following warning messages appear:

[  327.445302] [Firmware Bug]: TSC ADJUST differs: CPU15 45960750 --> 78394770. Restoring
[  329.209120] [Firmware Bug]: TSC ADJUST differs: CPU14 45960750 --> 78394770. Restoring
[  329.209128] [Firmware Bug]: TSC ADJUST differs: CPU13 45960750 --> 78394770. Restoring
[  329.209138] [Firmware Bug]: TSC ADJUST differs: CPU12 45960750 --> 78394770. Restoring
[  329.209151] [Firmware Bug]: TSC ADJUST differs: CPU11 45960750 --> 78394770. Restoring
[  329.209160] [Firmware Bug]: TSC ADJUST differs: CPU10 45960750 --> 78394770. Restoring
[  329.209169] [Firmware Bug]: TSC ADJUST differs: CPU9 45960750 --> 78394770. Restoring

The reason is:

Step 1: Bring up.
	TSC is sync after bring up with following settings:
				MSR 0x3b	cur->adjusted
	Package#0 CPU 0-7	  0		  0
	Package#1 first CPU	  value1	  value1
	Package#1 non-first CPU	  value1	  value1

Step 2: Suspend to S4.
	Settings in Step 1 are not changed in this Step.

Step 3: Bring up caused by S4 wake up event.
	TSC is sync when bring up with following settings:
				MSR 0x3b	cur->adjusted
	Package#0 CPU 0-7	  0		  0
	Package#1 first CPU	  value2	  value2
	Package#1 non-first CPU	  value2	  value2

Step 4: Resume from S4.
	When resuming from S4, Current TSC synchronous mechanism
	cause following settings:
				MSR 0x3b	cur->adjusted
	Package#0 CPU 0-7	  0		  0
	Package#1 first CPU	  value2	  value2
	Package#1 non-first CPU	  value2	  value1

In these Steps, value1 != 0 and value2 != value1.

In Step4, as function tsc_store_and_check_tsc_adjust() do, when the
value of MSR 0x3b on the non-first online CPU in package#1 is equal
to the value of cur->adjusted on the first online CPU in the same
package, the cur->adjusted value on this non-first online CPU will
hold the old value1. This cause function tsc_verify_tsc_adjust() set
the value of MSR 0x3b on the non-first online CPUs in the package#1
to the old value1 and print the beginning warning messages.

Fix it by setting cur->adjusted value on the non-first online CPU
in a package to the value of MSR 0x3b on the same CPU when they are
not equal.

Signed-off-by: Tony W Wang-oc <TonyWWang-oc@zhaoxin.com>
---
 arch/x86/kernel/tsc_sync.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kernel/tsc_sync.c b/arch/x86/kernel/tsc_sync.c
index 50a4515..b5886b4 100644
--- a/arch/x86/kernel/tsc_sync.c
+++ b/arch/x86/kernel/tsc_sync.c
@@ -190,7 +190,8 @@ bool tsc_store_and_check_tsc_adjust(bool bootcpu)
 	if (bootval != ref->adjusted) {
 		cur->adjusted = ref->adjusted;
 		wrmsrl(MSR_IA32_TSC_ADJUST, ref->adjusted);
-	}
+	} else if (cur->adjusted != bootval)
+		cur->adjusted = bootval;
 	/*
 	 * We have the TSCs forced to be in sync on this package. Skip sync
 	 * test:
-- 
2.7.4


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2021-08-31  7:08 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-31  7:08 [PATCH] x86/tsc: Make cur->adjusted values in package#1 to be the same Tony W Wang-oc

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).