On Fri, Nov 26, 2010 at 04:13:28PM +0100, Mark Kettenis wrote:
> > Date: Fri, 26 Nov 2010 16:35:28 +1000
> > From: David Gwynne <l...@animata.net>
> >
> > information about each cpu is stored in a statically sized array.
> > if we have more physical cpus than entries in that array, we currently
> > start storing cpu_info structures in unowned memory after that
> > array. this leads to Bad Things(tm).
> >
> > MAXCPUs should be bumped up on amd64, but we should also avoid
> > panicking on ridiculously overspecced boxes. i am sure i will regret
> > claiming that a 48core machine is overspecced in the very near
> > future though.
> >
> > Index: amd64/cpu.c
> > ===================================================================
> > RCS file: /cvs/src/sys/arch/amd64/amd64/cpu.c,v
> > retrieving revision 1.38
> > diff -u -p -r1.38 cpu.c
> > --- amd64/cpu.c 13 Nov 2010 04:16:42 -0000 1.38
> > +++ amd64/cpu.c 26 Nov 2010 06:31:53 -0000
> > @@ -167,10 +165,24 @@ cpu_match(struct device *parent, void *m
> > {
> > struct cfdata *cf = match;
> > struct cpu_attach_args *caa = aux;
> > + int cpunum;
> >
> > - if (strcmp(caa->caa_name, cf->cf_driver->cd_name) == 0)
> > + /*
> > + * we cannot attach more cpus than we are set up to store in the
> > + * cpu_info array, so fail to match if we have more cpus than slots
> > + * in the array.
> > + */
> > +
> > + for (cpunum = 0; cpunum < nitems(cpu_info); cpunum++) {
> > + if (cpu_info[cpunum] == NULL)
> > + break;
> > + }
> > +
> > + if (cpunum < nitems(cpu_info) &&
> > + strcmp(caa->caa_name, cf->cf_driver->cd_name) == 0)
> > return 1;
> > - return 0;
> > +
> > + return (0);
> > }
> >
> > static void
> >
>
> Sorry, but I think it is better to do this the same way it's done on
> hppa. The changes to i386 are a bit arger than strictly necessary to
> make the code identical to amd64. Only tested on i386, artificially
> limiting MAXCPUS to 1.
>
>
> Index: amd64/amd64/cpu.c
> ===================================================================
> RCS file: /cvs/src/sys/arch/amd64/amd64/cpu.c,v
> retrieving revision 1.38
> diff -u -p -r1.38 cpu.c
> --- amd64/amd64/cpu.c 13 Nov 2010 04:16:42 -0000 1.38
> +++ amd64/amd64/cpu.c 26 Nov 2010 15:00:49 -0000
> @@ -168,9 +168,13 @@ cpu_match(struct device *parent, void *m
> struct cfdata *cf = match;
> struct cpu_attach_args *caa = aux;
>
> - if (strcmp(caa->caa_name, cf->cf_driver->cd_name) == 0)
> - return 1;
> - return 0;
> + if (strcmp(caa->caa_name, cf->cf_driver->cd_name) != 0)
> + return 0;
> +
> + if (cf->cf_unit >= MAXCPUS)
> + return 0;
> +
> + return 1;
> }
>
> static void
> Index: i386/i386/cpu.c
> ===================================================================
> RCS file: /cvs/src/sys/arch/i386/i386/cpu.c,v
> retrieving revision 1.41
> diff -u -p -r1.41 cpu.c
> --- i386/i386/cpu.c 25 Jul 2010 21:43:35 -0000 1.41
> +++ i386/i386/cpu.c 26 Nov 2010 15:00:53 -0000
> @@ -157,14 +157,18 @@ struct cfdriver cpu_cd = {
> };
>
> int
> -cpu_match(struct device *parent, void *matchv, void *aux)
> +cpu_match(struct device *parent, void *match, void *aux)
> {
> - struct cfdata *match = (struct cfdata *)matchv;
> - struct cpu_attach_args *caa = (struct cpu_attach_args *)aux;
> + struct cfdata *cf = match;
> + struct cpu_attach_args *caa = aux;
>
> - if (strcmp(caa->caa_name, match->cf_driver->cd_name) == 0)
> - return (1);
> - return (0);
> + if (strcmp(caa->caa_name, cf->cf_driver->cd_name) != 0)
> + return 0;
> +
> + if (cf->cf_unit >= MAXCPUS)
> + return 0;
> +
> + return 1;
> }
>
> void
Seems to work on my 6-core amd64 box. I'm all for consistency, so this
is ok k...@.
.... Ken
