Re: [Qemu-devel] [PATCH v2 2/3] tcg: reorganize tb_find_physical loop

[Richard Henderson <rth@twiddle.net>]

On 03/29/2016 12:48 PM, Sergey Fedorov wrote:
> +    while (tb) {
>          if (tb->pc == pc &&
>              tb->page_addr[0] == phys_page1 &&
>              tb->cs_base == cs_base &&
>              tb->flags == flags) {
> -            /* check next page if needed */
> -            if (tb->page_addr[1] != -1) {
> -                tb_page_addr_t phys_page2;
>  
> -                virt_page2 = (pc & TARGET_PAGE_MASK) +
> -                    TARGET_PAGE_SIZE;
> -                phys_page2 = get_page_addr_code(env, virt_page2);
> +            if (tb->page_addr[1] == -1) {
> +                /* done, we have a match */
> +                break;
> +            } else {
> +                /* check next page if needed */
> +                target_ulong virt_page2 = (pc & TARGET_PAGE_MASK) +
> +                                          TARGET_PAGE_SIZE;
> +                tb_page_addr_t phys_page2 = get_page_addr_code(env, virt_page2);
> +
>                  if (tb->page_addr[1] == phys_page2) {
>                      break;
>                  }
> -            } else {
> -                break;
>              }
>          }

FYI,

There's an issue here before and after this patch: calling get_page_addr_code
will cause an exception to be thrown if the page isn't mapped.

Except this is a search routine looking for matching TB's.  We shouldn't be
raising an exception within this loop.  We need a variant of get_page_addr_code
that reloads the TLB, if possible, but without generating a fault.

Something that I don't think we can actually do with the current tlb_fill
target hooks.  So not something for this patch, but something to put on
someone's radar, hopefully...


r~