Re: [RFC PATCH 3/6] kvm: arm64: Fix up RELR relocation in hyp code/data

[Marc Zyngier <maz@kernel.org>]

On 2020-11-19 16:25, David Brazdil wrote:
> The arm64 kernel also supports packing of relocation data using the 
> RELR
> format. Implement a parser of RELR data and fixup the relocations using
> the same infra as RELA relocs.
> 
> Signed-off-by: David Brazdil <dbrazdil@google.com>
> ---
>  arch/arm64/kvm/va_layout.c | 41 ++++++++++++++++++++++++++++++++++++++
>  1 file changed, 41 insertions(+)
> 
> diff --git a/arch/arm64/kvm/va_layout.c b/arch/arm64/kvm/va_layout.c
> index b80fab974896..7f45a98eacfd 100644
> --- a/arch/arm64/kvm/va_layout.c
> +++ b/arch/arm64/kvm/va_layout.c
> @@ -145,6 +145,43 @@ static void __fixup_hyp_rela(void)
>  		__fixup_hyp_rel(rel[i].r_offset);
>  }
> 
> +#ifdef CONFIG_RELR
> +static void __fixup_hyp_relr(void)
> +{
> +	u64 *rel, *end;
> +
> +	rel = (u64*)(kimage_vaddr + __load_elf_u64(__relr_offset));
> +	end = rel + (__load_elf_u64(__relr_size) / sizeof(*rel));
> +
> +	while (rel < end) {
> +		unsigned n;
> +		u64 addr = *(rel++);
> +
> +		/* Address must not have the LSB set. */
> +		BUG_ON(addr & BIT(0));
> +
> +		/* Fix up the first address of the chain. */
> +		__fixup_hyp_rel(addr);
> +
> +		/*
> +		 * Loop over bitmaps, i.e. as long as words' LSB is 1.
> +		 * Each bit (ordered from LSB to MSB) represents one word from
> +		 * the last full address (exclusive). If the corresponding bit
> +		 * is 1, there is a relative relocation on that word.
> +		 */

What is the endianness of this bitmap? Is it guaranteed to be in
CPU-endian format?

> +		for (n = 0; rel < end && (*rel & BIT(0)); n++) {
> +			unsigned i;
> +			u64 bitmap = *(rel++);

nit: if you change this u64 for an unsigned long...

> +
> +			for (i = 1; i < 64; ++i) {
> +				if ((bitmap & BIT(i)))
> +					__fixup_hyp_rel(addr + 8 * (63 * n + i));
> +			}

... this can be written as:

         i = 1;
         for_each_set_bit_from(i, &bitmap, 64)
                 __fixup_hyp_rel(addr + 8 * (63 * n + i));

> +		}
> +	}
> +}
> +#endif
> +
>  /*
>   * The kernel relocated pointers to kernel VA. Iterate over 
> relocations in
>   * the hypervisor ELF sections and convert them to hyp VA. This avoids 
> the
> @@ -156,6 +193,10 @@ __init void kvm_fixup_hyp_relocations(void)
>  		return;
> 
>  	__fixup_hyp_rela();
> +
> +#ifdef CONFIG_RELR
> +	__fixup_hyp_relr();
> +#endif
>  }
> 
>  static u32 compute_instruction(int n, u32 rd, u32 rn)

Thanks,

         M.
-- 
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@lists.cs.columbia.edu
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm