On 27.03.2023 21:41, Andrew Cooper wrote:
> It is not valid to retain a bootstrap_map() across returning back to
> __start_xen(), but various pointers get stashed across calls.
It's error prone, yes, but "not valid" isn't really true imo: As long as
nothing calls bootstrap_map(NULL) all mappings will remain as they are.
> --- a/xen/arch/x86/cpu/microcode/core.c
> +++ b/xen/arch/x86/cpu/microcode/core.c
> @@ -755,47 +755,51 @@ int microcode_update_one(void)
> return microcode_update_cpu(NULL);
> }
>
> -static int __init early_update_cache(const void *data, size_t len)
> +int __init microcode_init_cache(unsigned long *module_map,
> + const struct multiboot_info *mbi)
> {
> int rc = 0;
> struct microcode_patch *patch;
> + struct ucode_mod_blob blob = {};
>
> - if ( !data )
> - return -ENOMEM;
This is lost afaict. To be in sync with earlier code ) think you want to ...
> + if ( ucode_scan )
> + /* Need to rescan the modules because they might have been relocated
> */
> + microcode_scan_module(module_map, mbi);
> +
> + if ( ucode_mod.mod_end )
> + {
> + blob.data = bootstrap_map(&ucode_mod);
... check here instead of ...
> + blob.size = ucode_mod.mod_end;
> + }
> + else if ( ucode_blob.size )
> + {
> + blob = ucode_blob;
> + }
(nit: unnecessary braces)
> - patch = parse_blob(data, len);
> + if ( !blob.data )
> + return 0;
... here, making the "return 0" the "else" to the earlier if/else-if.
Alternatively, if you think the -ENOMEM isn't sensible, I'm happy to
consider respective justification for its removal.
Jan
