[LinuxBIOS] r2890 - in trunk/LinuxBIOSv2/src: arch/i386/boot boot cpu/x86/mtrr cpu/x86/tsc devices

Uwe Hermann uwe at hermann-uwe.de
Wed Oct 24 18:54:09 CEST 2007


On Wed, Oct 24, 2007 at 05:51:17PM +0200, Stefan Reinauer wrote:
> Uwe Hermann wrote:
> >> Modified: trunk/LinuxBIOSv2/src/boot/elfboot.c
> >> ===================================================================
> >> --- trunk/LinuxBIOSv2/src/boot/elfboot.c	2007-10-23 19:23:52 UTC (rev 2889)
> >> +++ trunk/LinuxBIOSv2/src/boot/elfboot.c	2007-10-23 22:17:45 UTC (rev 2890)
> >> @@ -144,7 +144,7 @@
> >>  {
> >>  	struct verify_callback *cb_chain;
> >>  	unsigned char *note, *end;
> >> -	char *program, *version;
> >> +	unsigned char *program, *version;
> >>     
> >
> > This doesn't look like the correct fix. Strings should always be 'char *'
> > or 'const char *' IMO, so the code where program/version is used should
> > be fixed instead.
> >   
> 
> Oh, but the code expects an unsigned char, so it should get one.

Depends, in this case the code may also be "wrong" in that it uses
'unsigned char' where it should rather use 'uint8_t' (didn't check though).


> This is processing ELF header data, 8bit, so, according to the rule you
> set up below, this fix should be correct.

uint8_t would be correct here, if it's 8bit data.


> The fact that in this case the code is parsing a string out of that data
> is secondary I think.

Yes.


> >>  	if (resource->flags & IORESOURCE_STORED) {
> >> -		unsigned char buf[10];
> >> +		char buf[10];
> >>     
> >
> > If buf contains a string, yes. If it should contain 8bit data, it should
> > be uint8_t or u8.
> >   
> Ok, then I assume you are saying everything is ok.

I think so (but I didn't really check the code).


> 
> >>  #if OPT_HT_LINK == 1
> >> @@ -123,15 +126,17 @@
> >>  
> >>  static int ht_setup_link(struct ht_link *prev, device_t dev, unsigned pos)
> >>  {
> >> +#if OPT_HT_LINK == 1
> >>  	static const uint8_t link_width_to_pow2[]= { 3, 4, 0, 5, 1, 2, 0, 0 };
> >>  	static const uint8_t pow2_to_link_width[] = { 0x7, 4, 5, 0, 1, 3 };
> >> -	struct ht_link cur[1];
> >>  	unsigned present_width_cap,    upstream_width_cap;
> >>  	unsigned present_freq_cap,     upstream_freq_cap;
> >>  	unsigned ln_present_width_in,  ln_upstream_width_in; 
> >>  	unsigned ln_present_width_out, ln_upstream_width_out;
> >>  	unsigned freq, old_freq;
> >>  	unsigned present_width, upstream_width, old_width;
> >> +#endif
> >> +	struct ht_link cur[1];
> >>     
> >
> > This change is not trivial, as it might very well have effects on the
> > run-time behaviour of LinuxBIOS / the hardware.
> >
> > Can we say for sure that there's no reason to (re-)initialize here?
> > Is this tested on hardware?
> >   
> Oh yes, this is highly trivial. Look at the code, all these are unused
> variables. The produced code does not change.
> The OPT_HT_LINK define is tested already below, where the real magic
> happens. But when this was introduced, all the variables were left unused.

Ah, yes, I see. There are further '#if OPT_HT_LINK == 1' down below
which actually do stuff with the variables (but only in the "== 1" case).


Uwe.
-- 
http://www.hermann-uwe.de  | http://www.holsham-traders.de
http://www.crazy-hacks.org | http://www.unmaintained-free-software.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://www.coreboot.org/pipermail/coreboot/attachments/20071024/07df388d/attachment.sig>


More information about the coreboot mailing list