[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <d06bb33e-047f-c849-de6a-246bc361c7af@gmail.com>
Date: Tue, 27 Jun 2023 18:57:28 +0800
From: Tianyu Lan <ltykernel@...il.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: kys@...rosoft.com, haiyangz@...rosoft.com, wei.liu@...nel.org,
decui@...rosoft.com, tglx@...utronix.de, mingo@...hat.com,
bp@...en8.de, dave.hansen@...ux.intel.com, x86@...nel.org,
hpa@...or.com, daniel.lezcano@...aro.org, arnd@...db.de,
michael.h.kelley@...rosoft.com, Tianyu Lan <tiala@...rosoft.com>,
linux-arch@...r.kernel.org, linux-hyperv@...r.kernel.org,
linux-kernel@...r.kernel.org, vkuznets@...hat.com
Subject: Re: [EXTERNAL] Re: [PATCH 5/9] x86/hyperv: Use vmmcall to implement
Hyper-V hypercall in sev-snp enlightened guest
On 6/8/2023 11:15 PM, Tianyu Lan wrote:
> On 6/8/2023 9:21 PM, Peter Zijlstra wrote:
>> On Thu, Jun 01, 2023 at 11:16:18AM -0400, Tianyu Lan wrote:
>>> From: Tianyu Lan <tiala@...rosoft.com>
>>>
>>> In sev-snp enlightened guest, Hyper-V hypercall needs
>>> to use vmmcall to trigger vmexit and notify hypervisor
>>> to handle hypercall request.
>>>
>>> There is no x86 SEV SNP feature flag support so far and
>>> hardware provides MSR_AMD64_SEV register to check SEV-SNP
>>> capability with MSR_AMD64_SEV_ENABLED bit. ALTERNATIVE can't
>>> work without SEV-SNP x86 feature flag. May add later when
>>> the associated flag is introduced.
>>>
>>> Signed-off-by: Tianyu Lan <tiala@...rosoft.com>
>>> ---
>>> arch/x86/include/asm/mshyperv.h | 44 ++++++++++++++++++++++++---------
>>> 1 file changed, 33 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/arch/x86/include/asm/mshyperv.h
>>> b/arch/x86/include/asm/mshyperv.h
>>> index 31c476f4e656..d859d7c5f5e8 100644
>>> --- a/arch/x86/include/asm/mshyperv.h
>>> +++ b/arch/x86/include/asm/mshyperv.h
>>> @@ -61,16 +61,25 @@ static inline u64 hv_do_hypercall(u64 control,
>>> void *input, void *output)
>>> u64 hv_status;
>>> #ifdef CONFIG_X86_64
>>> - if (!hv_hypercall_pg)
>>> - return U64_MAX;
>>> + if (hv_isolation_type_en_snp()) {
>>> + __asm__ __volatile__("mov %4, %%r8\n"
>>> + "vmmcall"
>>> + : "=a" (hv_status), ASM_CALL_CONSTRAINT,
>>> + "+c" (control), "+d" (input_address)
>>> + : "r" (output_address)
>>> + : "cc", "memory", "r8", "r9", "r10", "r11");
>>> + } else {
>>> + if (!hv_hypercall_pg)
>>> + return U64_MAX;
>>> - __asm__ __volatile__("mov %4, %%r8\n"
>>> - CALL_NOSPEC
>>> - : "=a" (hv_status), ASM_CALL_CONSTRAINT,
>>> - "+c" (control), "+d" (input_address)
>>> - : "r" (output_address),
>>> - THUNK_TARGET(hv_hypercall_pg)
>>> - : "cc", "memory", "r8", "r9", "r10", "r11");
>>> + __asm__ __volatile__("mov %4, %%r8\n"
>>> + CALL_NOSPEC
>>> + : "=a" (hv_status), ASM_CALL_CONSTRAINT,
>>> + "+c" (control), "+d" (input_address)
>>> + : "r" (output_address),
>>> + THUNK_TARGET(hv_hypercall_pg)
>>> + : "cc", "memory", "r8", "r9", "r10", "r11");
>>> + }
>>> #else
>>
>> Remains unanswered:
>>
>> https://nam06.safelinks.protection.outlook.com/?url=https%3A%2F%2Flkml.kernel.org%2Fr%2F20230516102912.GG2587705%2540hirez.programming.kicks-ass.net&data=05%7C01%7CTianyu.Lan%40microsoft.com%7C60a576eb67634ffa27b108db68234d5a%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C638218273105649705%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=MFj67DON0K%2BUoUJbeaIA5oVTxyrzO3fb5DbxYgDWwX0%3D&reserved=0
>>
>> Would this not generate better code with an alternative?
>
>
> Hi Peter:
> Thanks to review. I put the explaination in the change log.
>
> "There is no x86 SEV SNP feature(X86_FEATURE_SEV_SNP) flag
> support so far and hardware provides MSR_AMD64_SEV register
> to check SEV-SNP capability with MSR_AMD64_SEV_ENABLED bit
> ALTERNATIVE can't work without SEV-SNP x86 feature flag."
> There is no cpuid leaf bit to check AMD SEV-SNP feature.
>
> After some Hyper-V doesn't provides SEV and SEV-ES guest before and so
> may reuse X86_FEATURE_SEV and X86_FEATURE_SEV_ES flag as alternative
> feature check for Hyper-V SEV-SNP guest. Will refresh patch.
>
Hi Peter:
I tried using alternative for "vmmcall" and CALL_NOSPEC in a single
Inline assembly. The output is different in the SEV-SNP mode. When SEV-
SNP is enabled, thunk_target is not required. While it's necessary in
the non SEV-SNP mode. Do you have any idea how to differentiate outputs
in the single Inline assembly which just like alternative works for
assembler template.
Powered by blists - more mailing lists