Re: [PATCH v2 1/2] x86/sev-es: Allow copy_from_kernel_nofault in earlier boot

From: Sean Christopherson
Date: Wed Sep 20 2023 - 16:37:47 EST


On Mon, Sep 11, 2023, Adam Dunlap wrote:
> Previously, if copy_from_kernel_nofault was called before
> boot_cpu_data.x86_virt_bits was set up, then it would trigger undefined
> behavior due to a shift by 64. This ended up causing boot failures in
> the latest version of ubuntu2204 in the gcp project when using SEV-SNP.
> Specifically, this function is called during an early #VC handler which
> is triggered by a cpuid to check if nx is implemented.

Why not stuff boot_cpu_data.x86_virt_bits to a "default" value that is guaranteed
to be accurate (or at least safe) for the purposes of the early boot code. I.e.
set it to 48 for 64-bit kernels.

That'd avoid the extra conditional, and would avoid laying whack-a-mole with
anything else that consumes x86_virt_bits.

> Fixes: 1aa9aa8ee517 ("x86/sev-es: Setup GHCB-based boot #VC handler")
> Suggested-by: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> Signed-off-by: Adam Dunlap <acdunlap@xxxxxxxxxx>
> ---
> arch/x86/mm/maccess.c | 19 ++++++++++++++-----
> 1 file changed, 14 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/mm/maccess.c b/arch/x86/mm/maccess.c
> index 5a53c2cc169c..6993f026adec 100644
> --- a/arch/x86/mm/maccess.c
> +++ b/arch/x86/mm/maccess.c
> @@ -9,12 +9,21 @@ bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
> unsigned long vaddr = (unsigned long)unsafe_src;
>
> /*
> - * Range covering the highest possible canonical userspace address
> - * as well as non-canonical address range. For the canonical range
> - * we also need to include the userspace guard page.
> + * Do not allow userspace addresses. This disallows
> + * normal userspace and the userspace guard page:
> */
> - return vaddr >= TASK_SIZE_MAX + PAGE_SIZE &&
> - __is_canonical_address(vaddr, boot_cpu_data.x86_virt_bits);
> + if (vaddr < TASK_SIZE_MAX + PAGE_SIZE)
> + return false;
> +
> + /*
> + * Allow everything during early boot before 'x86_virt_bits'
> + * is initialized. Needed for instruction decoding in early
> + * exception handlers.
> + */
> + if (!boot_cpu_data.x86_virt_bits)
> + return true;
> +
> + return __is_canonical_address(vaddr, boot_cpu_data.x86_virt_bits);
> }
> #else
> bool copy_from_kernel_nofault_allowed(const void *unsafe_src, size_t size)
> --
> 2.42.0.283.g2d96d420d3-goog
>