Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.34 as permitted sender) client-ip=23.128.96.34;
Sender: Ingo Molnar <mingo.kernel.org@gmail.com>
Date:   Mon, 25 Sep 2023 10:41:24 +0200
From:   Ingo Molnar <mingo@kernel.org>
To:     Alexey Kardashevskiy <aik@amd.com>
Cc:     x86@kernel.org, linux-kernel@vger.kernel.org,
        Tom Lendacky <thomas.lendacky@amd.com>
Subject: Re: [PATCH kernel] x86/compressed/64: reduce #VC nesting for
 intercepted CPUID for SEV-SNP guest
Message-ID: <ZRFHtCCsL4kKajKF@gmail.com>
References: <20230925042302.593317-1-aik@amd.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230925042302.593317-1-aik@amd.com>
Precedence: bulk


* Alexey Kardashevskiy <aik@amd.com> wrote:

> For certain intercepts an SNP guest uses the GHCB protocol to talk to
> the hypervisor from the #VC handler. The protocol requires a shared page so
> there is one per vCPU. In case NMI arrives in a middle of #VC or the NMI
> handler triggers a #VC, there is another "backup" GHCB page which stores
> the content of the first one while SVM_VMGEXIT_NMI_COMPLETE is sent.
> The vc_raw_handle_exception() handler manages main and backup GHCB pages
> via __sev_get_ghcb/__sev_put_ghcb.
> 
> This works fine for #VC and occasional NMIs. This does not work so fine if
> the #VC handler causes intercept + another #VC, if NMI arrives during
> the second #VC, there are no more pages for SVM_VMGEXIT_NMI_COMPLETE.
> The problem place is the #VC CPUID handler. Running perf in the SNP guest
> crashes with:
> 
> Kernel panic - not syncing: Unable to handle #VC exception! GHCB and Backup GHCB are already in use
> 
> vc_raw_handle_exception #1: exit_code 72 (CPUID) eax d ecx 1
> We lock the main GHCB and while it is locked we get to
> snp_cpuid_postprocess() which executes "rdmsr" of MSR_IA32_XSS==0xda0 which
> triggers:
> 
> vc_raw_handle_exception #2: exit_code 7c (MSR) ecx da0
> Here we lock the backup ghcb.
> 
> And then PMC NMI comes which cannot complete as there is no GHCB page left
> to use:
> 
> CPU: 5 PID: 566 Comm: touch Not tainted 6.5.0-rc2-aik-ad9c-g7413e71d3dcf-dirty #27
> Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS unknown unknown
> Call Trace:
>  <NMI>
>  dump_stack_lvl+0x44/0x60
>  panic+0x222/0x310
>  ____sev_get_ghcb+0x21e/0x220
>  __sev_es_nmi_complete+0x28/0xf0
>  exc_nmi+0x1ac/0x1c0
>  end_repeat_nmi+0x16/0x67
> ...
>  </NMI>
>  <TASK>
>  vc_raw_handle_exception+0x9e/0x2c0
>  kernel_exc_vmm_communication+0x4d/0xa0
>  asm_exc_vmm_communication+0x31/0x60
> RIP: 0010:snp_cpuid+0x2ad/0x420
> 
> Drop one #VC by replacing "rdmsr" with GHCB's VMGEXIT to read the value from
> the hypervisor.
> 
> Fixes: ee0bfa08a345 ("x86/compressed/64: Add support for SEV-SNP CPUID table in #VC handlers")
> Cc: x86@kernel.org
> Signed-off-by: Alexey Kardashevskiy <aik@amd.com>
> ---
> 
> This is made on top of (which has the "efi/unaccepted: Make sure unaccepted table is mapped"
> fix for booting SNP):
> b996cbe1203c (tip/master) 15 hours ago Ingo Molnar Merge branch into tip/master: 'x86/tdx'
> 
> plus:
> https://lore.kernel.org/lkml/a5856fa1ebe3879de91a8f6298b6bbd901c61881.1690578565.git.thomas.lendacky@amd.com/
> ---
>  arch/x86/kernel/sev-shared.c | 18 +++++++++++++-----
>  1 file changed, 13 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/x86/kernel/sev-shared.c b/arch/x86/kernel/sev-shared.c
> index e73c90c9cc5b..399219de5a9b 100644
> --- a/arch/x86/kernel/sev-shared.c
> +++ b/arch/x86/kernel/sev-shared.c
> @@ -477,11 +477,19 @@ static int snp_cpuid_postprocess(struct ghcb *ghcb, struct es_em_ctxt *ctxt,
>  		if (leaf->subfn == 1) {
>  			/* Get XSS value if XSAVES is enabled. */
>  			if (leaf->eax & BIT(3)) {
> -				unsigned long lo, hi;
> -
> -				asm volatile("rdmsr" : "=a" (lo), "=d" (hi)
> -						     : "c" (MSR_IA32_XSS));
> -				xss = (hi << 32) | lo;
> +				/*
> +				 * Since we're here, it is SNP and rdmsr will trigger
> +				 * another #VC and waste one of just two GHCB pages.
> +				 * Skip the intercept and do direct hypercall.
> +				 */
> +				ghcb_set_rcx(ghcb, MSR_IA32_XSS);
> +				if (sev_es_ghcb_hv_call(ghcb, ctxt, SVM_EXIT_MSR, 0, 0) != ES_OK)
> +					return -EINVAL;
> +
> +				xss = (ghcb->save.rdx << 32) | ghcb->save.rax;
> +
> +				/* Invalidate qwords for likely another following GHCB call */
> +				vc_ghcb_invalidate(ghcb);

Ok, so I agree with this fix, but could you please reduce the ugliness
of this open-coded RDMSR by factoring out this sequence into a new
helper, called rdmsr_GHCB() or so, with a similar signature as
rdmsr(), where rdmsr_GHCB() emulates RDMSR behavior via a hypercall?

That makes this workaround to reduce nesting a lot easier to read & maintain
in the longer run.

Thanks,

	Ingo