Re: [PATCH v4 2/2] xen/riscv: introduce identity mapping

[Jan Beulich <jbeulich@xxxxxxxx>]

On 26.07.2023 13:23, Oleksii wrote:
> I would like to ask for advice on whether it would be easier, less bug-
> provoking ( during identity mapping to remove of whole Xen ) to have a
> separate identity section that won't be more than PAGE_SIZE.

I'm afraid you can't safely do this in C, or at least not without
further checking on what the compiler actually did.

> @@ -264,6 +268,19 @@ void __init enable_mmu(void)
>                RV_STAGE1_MODE << SATP_MODE_SHIFT);
>  }
>  
> +void __attribute__((naked)) __section(".ident") turn_on_mmu(unsigned
> long ra)

Did you read what gcc doc says about "naked"? Extended asm() isn't
supported there. Since ...

> +{
> +    /* Ensure page table writes precede loading the SATP */
> +    sfence_vma();
> +
> +    /* Enable the MMU and load the new pagetable for Xen */
> +    csr_write(CSR_SATP,
> +              PFN_DOWN((unsigned long)stage1_pgtbl_root) |
> +              RV_STAGE1_MODE << SATP_MODE_SHIFT);
> +
> +    asm volatile( "jr %0\n" : : "r"(ra) );
> +}

... none of this really requires C, I think we're at the point where
(iirc) Andrew's and my suggestion wants following, moving this to
assembly code (at which point it doesn't need to be a separate
function). You can still build page tables in C, of course. (Likely
you then also won't need a separate section; some minimal alignment
guarantees ought to suffice to make sure the critical code is
confined to a single page.)

Jan