Re: [PATCH v2 3/3] efi: Add support for using efivars as a pstore backend

[Matthew Garrett <mjg59@srcf.ucam.org>]

On Tue, Jun 07, 2011 at 01:16:55PM -0700, Luck, Tony wrote:
> +	efi_pstore_info.buf = kmalloc(4096, GFP_KERNEL);
> +	if (efi_pstore_info.buf) {
> +		efi_pstore_info.bufsize = 1024;
> +		efi_pstore_info.data = efivars;
> +		mutex_init(&efi_pstore_info.buf_mutex);
> +		pstore_register(&efi_pstore_info);
> +	}
> 
> I'd imagined #ifdef CONFIG_PSTORE around this (and the
> efi_pstore_info definition) rather than providing stubs
> for the !PSTORE case.

We usually seem to frown on #ifdefs in the middle of functions. We could 
probably make this easier by adding a pstore_enabled that's #defined to 
0 in the !PSTORE case, then the compiler ought to just get rid of it 
all.

> You should avoid a memory leak with:
> 
> 		if (!pstore_register(&efi_pstore_info))
> 			kfree(efi_pstore_info.buf);

Ah, true.

> I've also been thinking some more about how to handle a
> system that has more than one pstore back-end available.
> I still don't have any good ideas how to make use of more
> than one backend - but it occurs to me that we may need
> a way to let the user choose which to use (e.g. in the
> unlikely event that a BIOS bug made one of ERST or EFI
> unusable by pstore). The current "whoever registers first
> gets to use it" now seems inadequate.

Mm. Given that the name of the source is in the file, there's no 
namespacing issues. In an ideal world I think we'd register all of them 
in order to provide a better chance of at least one of them ending up in 
actual persistent storage.

-- 
Matthew Garrett | mjg59@srcf.ucam.org