Hello Sean,
On 1/30/2025 7:48 PM, Sean Christopherson wrote:
> On Fri, Jan 31, 2025, Ashish Kalra wrote:
>> From: Sean Christopherson <seanjc@xxxxxxxxxx>
>>
>> This patch fixes the current SNP host enabling code and effectively SNP
> ^^^^^^^^^^
>> ---
>> drivers/iommu/amd/init.c | 18 ++++++++++++++----
>> 1 file changed, 14 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
>> index c5cd92edada0..ee887aa4442f 100644
>> --- a/drivers/iommu/amd/init.c
>> +++ b/drivers/iommu/amd/init.c
>> @@ -3194,7 +3194,7 @@ static bool __init detect_ivrs(void)
>> return true;
>> }
>>
>> -static void iommu_snp_enable(void)
>> +static __init void iommu_snp_enable(void)
>
> If you're feeling nitpicky, adding "__init" could be done in a separate patch.
>
>> {
>> #ifdef CONFIG_KVM_AMD_SEV
>> if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
>> @@ -3219,6 +3219,11 @@ static void iommu_snp_enable(void)
>> goto disable_snp;
>> }
>>
>> + if (snp_rmptable_init()) {
>> + pr_warn("SNP: RMP initialization failed, SNP cannot be supported.\n");
>> + goto disable_snp;
>> + }
>> +
>> pr_info("IOMMU SNP support enabled.\n");
>> return;
>>
>> @@ -3426,18 +3431,23 @@ void __init amd_iommu_detect(void)
>> int ret;
>>
>> if (no_iommu || (iommu_detected && !gart_iommu_aperture))
>> - return;
>> + goto disable_snp;
>>
>> if (!amd_iommu_sme_check())
>> - return;
>> + goto disable_snp;
>>
>> ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
>> if (ret)
>> - return;
>> + goto disable_snp;
>
> This handles initial failure, but it won't handle the case where amd_iommu_prepare()
> fails, as the iommu_go_to_state() call from amd_iommu_enable() will get
> short-circuited. I don't see any pleasant options. Maybe this?
>
> diff --git a/drivers/iommu/amd/init.c b/drivers/iommu/amd/init.c
> index c5cd92edada0..436e47f13f8f 100644
> --- a/drivers/iommu/amd/init.c
> +++ b/drivers/iommu/amd/init.c
> @@ -3318,6 +3318,8 @@ static int __init iommu_go_to_state(enum iommu_init_state state)
> ret = state_next();
> }
>
> + if (ret && !amd_iommu_snp_en && cc_platform_has(CC_ATTR_HOST_SEV_SNP))
> + cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
> return ret;
> }
>
This is surely hackish but it will work.
Though to add here, as mentioned in the patch commit logs, IOMMU needs to enabled for SNP_INIT to succeed.
If IOMMU initialization fails and iommu_snp_enable() is never reached, CC_ATTR_HOST_SEV_SNP will be
left set but that will cause PSP driver's SNP_INIT to fail with invalid configuration error as
IOMMU SNP sanity checks in SNP firmware will fail (if IOMMUs are not enabled) as below:
[ 9.723114] ccp 0000:23:00.1: sev enabled
[ 9.727602] ccp 0000:23:00.1: psp enabled
[ 9.732527] ccp 0000:a2:00.1: enabling device (0000 -> 0002)
[ 9.739098] ccp 0000:a2:00.1: no command queues available
[ 9.745167] ccp 0000:a2:00.1: psp enabled
[ 9.805337] ccp 0000:23:00.1: SEV-SNP: failed to INIT rc -5, error 0x3
[ 9.866426] ccp 0000:23:00.1: SEV API:1.53 build:5
...
and that will cause CC_ATTR_HOST_SEV_SNP flag to be cleared.
>
> Somewhat of a side topic, what happens if the RMP is fully configured and _then_
> IOMMU initialization fails? I.e. if amd_iommu_init_pci() or amd_iommu_enable_interrupts()
> fails? Is that even survivable?
Actually, SNP firmware initializes the IOMMU to perform RMP enforcement during SNP_INIT,
but as i mentioned above if IOMMU is not enabled then SNP_INIT will fail.
Thanks,
Ashish
>
>>
>> amd_iommu_detected = true;
>> iommu_detected = 1;
>> x86_init.iommu.iommu_init = amd_iommu_init;
>> + return;
>> +
>> +disable_snp:
>> + if (cc_platform_has(CC_ATTR_HOST_SEV_SNP))
>> + cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
>> }
>>
>> /****************************************************************************
>> --
>> 2.34.1
>>
