* [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86
@ 2021-05-12 8:46 Vitaly Kuznetsov
2021-05-12 16:03 ` Michael Kelley
2021-05-12 20:27 ` Thomas Gleixner
0 siblings, 2 replies; 4+ messages in thread
From: Vitaly Kuznetsov @ 2021-05-12 8:46 UTC (permalink / raw)
To: linux-hyperv
Cc: Daniel Lezcano, Michael Kelley, Mohammed Gamal, Wei Liu,
K. Y. Srinivasan, Haiyang Zhang, Dexuan Cui, Stephen Hemminger,
linux-kernel
Mohammed reports (https://bugzilla.kernel.org/show_bug.cgi?id=213029)
the commit e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO
differences inline") broke vDSO on x86. The problem appears to be that
VDSO_CLOCKMODE_HVCLOCK is an enum value in 'enum vdso_clock_mode' and
'#ifdef VDSO_CLOCKMODE_HVCLOCK' branch evaluates to false (it is not
a define). Replace it with CONFIG_X86 as it is the only arch which
has this mode currently.
Reported-by: Mohammed Gamal <mgamal@redhat.com>
Fixes: e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
---
drivers/clocksource/hyperv_timer.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/clocksource/hyperv_timer.c b/drivers/clocksource/hyperv_timer.c
index 977fd05ac35f..e17421f5e47d 100644
--- a/drivers/clocksource/hyperv_timer.c
+++ b/drivers/clocksource/hyperv_timer.c
@@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct clocksource *arg)
hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
}
-#ifdef VDSO_CLOCKMODE_HVCLOCK
+#ifdef CONFIG_X86
static int hv_cs_enable(struct clocksource *cs)
{
vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
@@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc = {
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
.suspend= suspend_hv_clock_tsc,
.resume = resume_hv_clock_tsc,
-#ifdef VDSO_CLOCKMODE_HVCLOCK
+#ifdef CONFIG_X86
.enable = hv_cs_enable,
.vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
#else
--
2.31.1
^ permalink raw reply related [flat|nested] 4+ messages in thread
* RE: [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86
2021-05-12 8:46 [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86 Vitaly Kuznetsov
@ 2021-05-12 16:03 ` Michael Kelley
2021-05-12 20:27 ` Thomas Gleixner
1 sibling, 0 replies; 4+ messages in thread
From: Michael Kelley @ 2021-05-12 16:03 UTC (permalink / raw)
To: vkuznets, linux-hyperv
Cc: Daniel Lezcano, Mohammed Gamal, Wei Liu, KY Srinivasan,
Haiyang Zhang, Dexuan Cui, Stephen Hemminger, linux-kernel
From: Vitaly Kuznetsov <vkuznets@redhat.com> Sent: Wednesday, May 12, 2021 1:47 AM
>
> Mohammed reports (https://bugzilla.kernel.org/show_bug.cgi?id=213029)
> the commit e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO
> differences inline") broke vDSO on x86. The problem appears to be that
> VDSO_CLOCKMODE_HVCLOCK is an enum value in 'enum vdso_clock_mode' and
> '#ifdef VDSO_CLOCKMODE_HVCLOCK' branch evaluates to false (it is not
> a define). Replace it with CONFIG_X86 as it is the only arch which
> has this mode currently.
>
> Reported-by: Mohammed Gamal <mgamal@redhat.com>
> Fixes: e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
> ---
> drivers/clocksource/hyperv_timer.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/clocksource/hyperv_timer.c b/drivers/clocksource/hyperv_timer.c
> index 977fd05ac35f..e17421f5e47d 100644
> --- a/drivers/clocksource/hyperv_timer.c
> +++ b/drivers/clocksource/hyperv_timer.c
> @@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct clocksource *arg)
> hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
> }
>
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef CONFIG_X86
> static int hv_cs_enable(struct clocksource *cs)
> {
> vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
> @@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc = {
> .flags = CLOCK_SOURCE_IS_CONTINUOUS,
> .suspend= suspend_hv_clock_tsc,
> .resume = resume_hv_clock_tsc,
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef CONFIG_X86
> .enable = hv_cs_enable,
> .vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
> #else
> --
> 2.31.1
Well, bummer. I thought I was being so clever. I hate having to base the
behavior directly on the architecture instead of the actual VDSO feature,
but I don't see a way to detect whether VDSO_CLOCKMODE_HVCLOCK is
defined or not. The other alternative would be to add another #define
such as VDSO_CLOCKMODE_HVCLOCK_PRESENT in
arch/x86/include/asm/vdso/clocksource.h, and use it in the #ifdefs
here. But that's a bit messy too, and I'm not sure it's worth the
trouble. So,
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
Mohammed -- Thanks for finding this!
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86
2021-05-12 8:46 [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86 Vitaly Kuznetsov
2021-05-12 16:03 ` Michael Kelley
@ 2021-05-12 20:27 ` Thomas Gleixner
2021-05-13 7:18 ` Vitaly Kuznetsov
1 sibling, 1 reply; 4+ messages in thread
From: Thomas Gleixner @ 2021-05-12 20:27 UTC (permalink / raw)
To: Vitaly Kuznetsov, linux-hyperv
Cc: Daniel Lezcano, Michael Kelley, Mohammed Gamal, Wei Liu,
K. Y. Srinivasan, Haiyang Zhang, Dexuan Cui, Stephen Hemminger,
linux-kernel
On Wed, May 12 2021 at 10:46, Vitaly Kuznetsov wrote:
> Mohammed reports (https://bugzilla.kernel.org/show_bug.cgi?id=213029)
> the commit e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO
> differences inline") broke vDSO on x86. The problem appears to be that
> VDSO_CLOCKMODE_HVCLOCK is an enum value in 'enum vdso_clock_mode' and
> '#ifdef VDSO_CLOCKMODE_HVCLOCK' branch evaluates to false (it is not
> a define). Replace it with CONFIG_X86 as it is the only arch which
> has this mode currently.
>
> Reported-by: Mohammed Gamal <mgamal@redhat.com>
> Fixes: e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
> ---
> drivers/clocksource/hyperv_timer.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/clocksource/hyperv_timer.c b/drivers/clocksource/hyperv_timer.c
> index 977fd05ac35f..e17421f5e47d 100644
> --- a/drivers/clocksource/hyperv_timer.c
> +++ b/drivers/clocksource/hyperv_timer.c
> @@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct clocksource *arg)
> hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
> }
>
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef CONFIG_X86
> static int hv_cs_enable(struct clocksource *cs)
> {
> vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
> @@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc = {
> .flags = CLOCK_SOURCE_IS_CONTINUOUS,
> .suspend= suspend_hv_clock_tsc,
> .resume = resume_hv_clock_tsc,
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef CONFIG_X86
> .enable = hv_cs_enable,
> .vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
> #else
That's lame as it needs to be patched differently once ARM64 gains
support. What about the below?
Thanks,
tglx
---
--- a/arch/x86/include/asm/vdso/clocksource.h
+++ b/arch/x86/include/asm/vdso/clocksource.h
@@ -7,4 +7,6 @@
VDSO_CLOCKMODE_PVCLOCK, \
VDSO_CLOCKMODE_HVCLOCK
+#define HAVE_VDSO_CLOCKMODE_HVCLOCK
+
#endif /* __ASM_VDSO_CLOCKSOURCE_H */
--- a/drivers/clocksource/hyperv_timer.c
+++ b/drivers/clocksource/hyperv_timer.c
@@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct c
hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
}
-#ifdef VDSO_CLOCKMODE_HVCLOCK
+#ifdef HAVE_VDSO_CLOCKMODE_HVCLOCK
static int hv_cs_enable(struct clocksource *cs)
{
vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
@@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc
.flags = CLOCK_SOURCE_IS_CONTINUOUS,
.suspend= suspend_hv_clock_tsc,
.resume = resume_hv_clock_tsc,
-#ifdef VDSO_CLOCKMODE_HVCLOCK
+#ifdef HAVE_VDSO_CLOCKMODE_HVCLOCK
.enable = hv_cs_enable,
.vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
#else
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86
2021-05-12 20:27 ` Thomas Gleixner
@ 2021-05-13 7:18 ` Vitaly Kuznetsov
0 siblings, 0 replies; 4+ messages in thread
From: Vitaly Kuznetsov @ 2021-05-13 7:18 UTC (permalink / raw)
To: Thomas Gleixner, linux-hyperv
Cc: Daniel Lezcano, Michael Kelley, Mohammed Gamal, Wei Liu,
K. Y. Srinivasan, Haiyang Zhang, Dexuan Cui, Stephen Hemminger,
linux-kernel
Thomas Gleixner <tglx@linutronix.de> writes:
> On Wed, May 12 2021 at 10:46, Vitaly Kuznetsov wrote:
>> Mohammed reports (https://bugzilla.kernel.org/show_bug.cgi?id=213029)
>> the commit e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO
>> differences inline") broke vDSO on x86. The problem appears to be that
>> VDSO_CLOCKMODE_HVCLOCK is an enum value in 'enum vdso_clock_mode' and
>> '#ifdef VDSO_CLOCKMODE_HVCLOCK' branch evaluates to false (it is not
>> a define). Replace it with CONFIG_X86 as it is the only arch which
>> has this mode currently.
>>
>> Reported-by: Mohammed Gamal <mgamal@redhat.com>
>> Fixes: e4ab4658f1cf ("clocksource/drivers/hyper-v: Handle vDSO differences inline")
>> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
>> ---
>> drivers/clocksource/hyperv_timer.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/clocksource/hyperv_timer.c b/drivers/clocksource/hyperv_timer.c
>> index 977fd05ac35f..e17421f5e47d 100644
>> --- a/drivers/clocksource/hyperv_timer.c
>> +++ b/drivers/clocksource/hyperv_timer.c
>> @@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct clocksource *arg)
>> hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
>> }
>>
>> -#ifdef VDSO_CLOCKMODE_HVCLOCK
>> +#ifdef CONFIG_X86
>> static int hv_cs_enable(struct clocksource *cs)
>> {
>> vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
>> @@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc = {
>> .flags = CLOCK_SOURCE_IS_CONTINUOUS,
>> .suspend= suspend_hv_clock_tsc,
>> .resume = resume_hv_clock_tsc,
>> -#ifdef VDSO_CLOCKMODE_HVCLOCK
>> +#ifdef CONFIG_X86
>> .enable = hv_cs_enable,
>> .vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
>> #else
>
> That's lame as it needs to be patched differently once ARM64 gains
> support. What about the below?
>
The solution I liked the most was Michael's: no need to do anything
except for adding VDSO_CLOCKMODE_HVCLOCK to the enum. Too bad it didn't
work)
You proposal seems to be slightly better than mine: when adding
VDSO_CLOCKMODE_HVCLOCK to ARM the change will be limited to
drivers/clocksource/hyperv_timer.c will stay intact.
I'll send v2 shortly, thanks!
> Thanks,
>
> tglx
> ---
> --- a/arch/x86/include/asm/vdso/clocksource.h
> +++ b/arch/x86/include/asm/vdso/clocksource.h
> @@ -7,4 +7,6 @@
> VDSO_CLOCKMODE_PVCLOCK, \
> VDSO_CLOCKMODE_HVCLOCK
>
> +#define HAVE_VDSO_CLOCKMODE_HVCLOCK
> +
> #endif /* __ASM_VDSO_CLOCKSOURCE_H */
> --- a/drivers/clocksource/hyperv_timer.c
> +++ b/drivers/clocksource/hyperv_timer.c
> @@ -419,7 +419,7 @@ static void resume_hv_clock_tsc(struct c
> hv_set_register(HV_REGISTER_REFERENCE_TSC, tsc_msr);
> }
>
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef HAVE_VDSO_CLOCKMODE_HVCLOCK
> static int hv_cs_enable(struct clocksource *cs)
> {
> vclocks_set_used(VDSO_CLOCKMODE_HVCLOCK);
> @@ -435,7 +435,7 @@ static struct clocksource hyperv_cs_tsc
> .flags = CLOCK_SOURCE_IS_CONTINUOUS,
> .suspend= suspend_hv_clock_tsc,
> .resume = resume_hv_clock_tsc,
> -#ifdef VDSO_CLOCKMODE_HVCLOCK
> +#ifdef HAVE_VDSO_CLOCKMODE_HVCLOCK
> .enable = hv_cs_enable,
> .vdso_clock_mode = VDSO_CLOCKMODE_HVCLOCK,
> #else
>
--
Vitaly
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2021-05-13 7:18 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-05-12 8:46 [PATCH] clocksource/drivers/hyper-v: Re-enable VDSO_CLOCKMODE_HVCLOCK on X86 Vitaly Kuznetsov
2021-05-12 16:03 ` Michael Kelley
2021-05-12 20:27 ` Thomas Gleixner
2021-05-13 7:18 ` Vitaly Kuznetsov
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).