Dropping thellst...@vmware.com from Cc from now on because of some
microsloth mail rule not delivering my mails.
On Tue, Apr 28, 2020 at 05:16:54PM +0200, Joerg Roedel wrote:
> From: Joerg Roedel <jroe...@suse.de>
>
> Allocate and map enough stacks for the #VC handler to support sufficient
> levels of nesting and the NMI-in-#VC scenario.
>
> Also setup the IST entrys for the #VC handler on all CPUs because #VC
"entries"
> needs to work before cpu_init() has set up the per-cpu TSS.
Add a sentence to the commit message pointing to that "VC Handler IST
Stacks" comment in the code explaining the justification for the need
for IST stacks and the nesting.
>
> Signed-off-by: Joerg Roedel <jroe...@suse.de>
> ---
> arch/x86/include/asm/cpu_entry_area.h | 61 +++++++++++++++++++++++++++
> arch/x86/include/asm/page_64_types.h | 1 +
> arch/x86/kernel/cpu/common.c | 1 +
> arch/x86/kernel/sev-es.c | 40 ++++++++++++++++++
> 4 files changed, 103 insertions(+)
>
> diff --git a/arch/x86/include/asm/cpu_entry_area.h
> b/arch/x86/include/asm/cpu_entry_area.h
> index 02c0078d3787..85aac6c63653 100644
> --- a/arch/x86/include/asm/cpu_entry_area.h
> +++ b/arch/x86/include/asm/cpu_entry_area.h
> @@ -64,6 +64,61 @@ enum exception_stack_ordering {
> #define CEA_ESTACK_PAGES \
> (sizeof(struct cea_exception_stacks) / PAGE_SIZE)
>
> +/*
> + * VC Handler IST Stacks
> + *
> + * The IST stacks for the #VC handler are only allocated when SEV-ES is
> active,
> + * so they are not part of 'struct exception_stacks'.
> + *
> + * The VC handler uses shift_ist so that #VC can be nested. Nesting happens
> for
> + * example when the #VC handler has to call printk in the case of and error
> or
"an"
> + * when emulating 'movs' instructions.
> + *
> + * NMIs are another special case which can cause nesting of #VC handlers. The
> + * do_nmi() code path can cause #VC, e.g. for RDPMC. An NMI can also hit in
> + * the time window when the #VC handler is raised but before it has shifted
> its
> + * IST entry. To make sure any #VC raised from the NMI code path uses a new
> + * stack, the NMI handler unconditionally shifts the #VC handlers IST entry.
> + * This can cause one IST stack for #VC to be omitted.
> + *
> + * To support sufficient levels of nesting for the #VC handler, make the
> number
> + * of nesting levels configurable. It is currently set to 5 to support this
> + * scenario:
> + *
> + * #VC - IST stack 4, IST entry already shifted to 3
> + *
> + * -> NMI - shifts #VC IST entry to 2
> + *
> + * -> #VC(RDPMC) - shifts #VC IST to 1, something goes wrong, print
> + * an error message
> + *
> + * -> #VC(printk) - shifts #VC IST entry to 0, output driver
> + * uses 'movs'
> + *
> + * -> #VC(movs) - shifts IST to unmapped stack, further #VCs will
> + * cause #DF
> + *
> + */
> +#define N_VC_STACKS 5
> +
> +#define VC_STACK_MEMBERS(guardsize, holesize) \
> + char hole[holesize]; \
> + struct { \
> + char guard[guardsize]; \
> + char stack[EXCEPTION_STKSZ]; \
> + } stacks[N_VC_STACKS]; \
> + char top_guard[guardsize]; \
> +
> +/* Physical storage */
> +struct vmm_exception_stacks {
> + VC_STACK_MEMBERS(0, 0)
> +};
> +
> +/* Mapping in cpu_entry_area */
> +struct cea_vmm_exception_stacks {
> + VC_STACK_MEMBERS(PAGE_SIZE, EXCEPTION_STKSZ)
> +};
All those things should be under an CONFIG_AMD_MEM_ENCRYPT ifdeffery.
> +
> #endif
>
> #ifdef CONFIG_X86_32
> @@ -110,6 +165,12 @@ struct cpu_entry_area {
> * Exception stacks used for IST entries with guard pages.
> */
> struct cea_exception_stacks estacks;
> +
> + /*
> + * IST Exception stacks for VC handler - Only allocated and mapped when
> + * SEV-ES is active.
> + */
> + struct cea_vmm_exception_stacks vc_stacks;
Ditto.
--
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
