From: Boris Ostrovsky <boris.ostrovsky@oracle.com>
To: Krister Johansen <kjlx@templeofstupid.com>
Cc: Juergen Gross <jgross@suse.com>,
Thomas Gleixner <tglx@linutronix.de>,
Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
Dave Hansen <dave.hansen@linux.intel.com>,
x86@kernel.org, "H. Peter Anvin" <hpa@zytor.com>,
xen-devel@lists.xenproject.org, linux-kernel@vger.kernel.org,
Marcelo Tosatti <mtosatti@redhat.com>,
Anthony Liguori <aliguori@amazon.com>,
David Reaver <me@davidreaver.com>,
Brendan Gregg <brendan@intel.com>
Subject: Re: [PATCH linux-next v2] x86/xen/time: prefer tsc as clocksource when it is invariant
Date: Wed, 14 Dec 2022 16:46:10 -0500 [thread overview]
Message-ID: <0835453d-9617-48d5-b2dc-77a2ac298bad@oracle.com> (raw)
In-Reply-To: <20221214180147.GA1953@templeofstupid.com>
On 12/14/22 1:01 PM, Krister Johansen wrote:
> On Tue, Dec 13, 2022 at 04:25:32PM -0500, Boris Ostrovsky wrote:
>> On 12/12/22 5:09 PM, Krister Johansen wrote:
>>> On Mon, Dec 12, 2022 at 01:48:24PM -0500, Boris Ostrovsky wrote:
>>>> On 12/12/22 11:05 AM, Krister Johansen wrote:
>>>>> diff --git a/arch/x86/include/asm/xen/cpuid.h b/arch/x86/include/asm/xen/cpuid.h
>>>>> index 6daa9b0c8d11..d9d7432481e9 100644
>>>>> --- a/arch/x86/include/asm/xen/cpuid.h
>>>>> +++ b/arch/x86/include/asm/xen/cpuid.h
>>>>> @@ -88,6 +88,12 @@
>>>>> * EDX: shift amount for tsc->ns conversion
>>>>> * Sub-leaf 2: EAX: host tsc frequency in kHz
>>>>> */
>>>>> +#define XEN_CPUID_TSC_EMULATED (1u << 0)
>>>>> +#define XEN_CPUID_HOST_TSC_RELIABLE (1u << 1)
>>>>> +#define XEN_CPUID_RDTSCP_INSTR_AVAIL (1u << 2)
>>>>> +#define XEN_CPUID_TSC_MODE_DEFAULT (0)
>>>>> +#define XEN_CPUID_TSC_MODE_EMULATE (1u)
>>>>> +#define XEN_CPUID_TSC_MODE_NOEMULATE (2u)
>>>> This file is a copy of Xen public interface so this change should go to Xen first.
>>> Ok, should I split this into a separate patch on the linux side too?
>> Yes. Once the Xen patch has been accepted you will either submit the same patch for Linux or sync Linux file with Xen (if there are more differences).
> Thanks. Based upon the feedback I received from you and Jan, I may try
> to shrink the check in xen_tsc_safe_clocksource() down a bit. In that
> case, I may only need to refer to a single field in the leaf that
> provides this information. In that case, are you alright with dropping
> the change to the header and referring to the value directly, or would
> you prefer that I proceed with adding these to the public API?
It would certainly be appreciated if you updated the header files but it's up to maintainers to decide whether it's required.
>>>>> +static int __init xen_tsc_safe_clocksource(void)
>>>>> +{
>>>>> + u32 eax, ebx, ecx, edx;
>>>>> +
>>>>> + if (!(xen_hvm_domain() || xen_pvh_domain()))
>>>>> + return 0;
>>>>> +
>>>>> + if (!(boot_cpu_has(X86_FEATURE_CONSTANT_TSC)))
>>>>> + return 0;
>>>>> +
>>>>> + if (!(boot_cpu_has(X86_FEATURE_NONSTOP_TSC)))
>>>>> + return 0;
>>>>> +
>>>>> + if (check_tsc_unstable())
>>>>> + return 0;
>>>>> +
>>>>> + cpuid(xen_cpuid_base() + 3, &eax, &ebx, &ecx, &edx);
>>>>> +
>>>>> + if (eax & XEN_CPUID_TSC_EMULATED)
>>>>> + return 0;
>>>>> +
>>>>> + if (ebx != XEN_CPUID_TSC_MODE_NOEMULATE)
>>>>> + return 0;
>>>> Why is the last test needed?
>>> I was under the impression that if the mode was 0 (default) it would be
>>> possible for the tsc to become emulated in the future, perhaps after a
>>> migration. The presence of the tsc_mode noemulate meant that we could
>>> count on the falseneess of the XEN_CPUID_TSC_EMULATED check remaining
>>> constant.
>> This will filter out most modern processors with TSC scaling support where in default mode we don't intercept RDTCS after migration. But I don't think we have proper interface to determine this so we don't have much choice but to indeed make this check.
> Yes, if this remains a single boot-time check, I'm not sure that knowing
> whether the processor supports tsc scaling helps us. If tsc_mode is
> default, there's always a possibility of the tsc becoming emulated later
> on as part of migration, correct?
If the processor supports TSC scaling I don't think it's possible (it can happen in theory) but if it doesn't and you migrate to a CPU running at different frequency then yes, hypervisor will start emulating RDTSC.
>
> The other thing that might be possible here is to add a background
> timer that periodically checks if the tsc is still not emulated, and if
> it suddenly becomes so, change the rating again to prefer the xen
> clocksource. I had written this off initially as an impractical
> solution, since it seemed like a lot more mechanism and because it meant
> the performance characteristics of the system would change without user
> intervention. However, if this seems like a good idea, I'm not opposed
> to giving it a try.
I don't think we should do it. Having the kernel suddenly change clocksource will probably be somewhat of a surprise to users.
-boris
next prev parent reply other threads:[~2022-12-14 21:47 UTC|newest]
Thread overview: 16+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-12-08 16:36 [PATCH linux-next] x86/xen/time: prefer tsc as clocksource when it is invariant Krister Johansen
2022-12-09 19:32 ` Boris Ostrovsky
2022-12-12 15:57 ` Krister Johansen
2022-12-12 16:05 ` [PATCH linux-next v2] " Krister Johansen
2022-12-12 16:46 ` Jan Beulich
2022-12-12 22:05 ` Krister Johansen
2022-12-13 7:23 ` Jan Beulich
2022-12-13 18:58 ` Krister Johansen
2022-12-14 8:17 ` Jan Beulich
2022-12-14 18:01 ` Krister Johansen
2022-12-12 18:48 ` Boris Ostrovsky
2022-12-12 22:09 ` Krister Johansen
2022-12-13 21:25 ` Boris Ostrovsky
2022-12-14 18:01 ` Krister Johansen
2022-12-14 21:46 ` Boris Ostrovsky [this message]
2022-12-16 16:20 ` Krister Johansen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=0835453d-9617-48d5-b2dc-77a2ac298bad@oracle.com \
--to=boris.ostrovsky@oracle.com \
--cc=aliguori@amazon.com \
--cc=bp@alien8.de \
--cc=brendan@intel.com \
--cc=dave.hansen@linux.intel.com \
--cc=hpa@zytor.com \
--cc=jgross@suse.com \
--cc=kjlx@templeofstupid.com \
--cc=linux-kernel@vger.kernel.org \
--cc=me@davidreaver.com \
--cc=mingo@redhat.com \
--cc=mtosatti@redhat.com \
--cc=tglx@linutronix.de \
--cc=x86@kernel.org \
--cc=xen-devel@lists.xenproject.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).