On Mon, Sep 23, 2013 at 03:17:57PM -0400, Tom Rini wrote: > On Mon, Sep 23, 2013 at 08:50:55PM +0200, Wolfgang Denk wrote: > > > Dear Fabio Estevam, > > > > In message > > <caomzo5aj56kvtfrqbd3wq7oip8q3wa0yv4evuzerkxfufvx...@mail.gmail.com> you > > wrote: > > > > > > > Then shouldn't the patch subject/summary be "print command name only > > > > if cmdtp is not NULL" rather than the quite uninformative "prevent a > > > > crash"? > > > > > > Yes, I agree that original subject is a bit misleading. > > > > > > When I read it I thought it was a Wandboard related problem. > > > > I don't know if it's only Wandboard, or if other boards are affected, > > too (which are these? under which exact test cases?). In any case. > > the problem is not here, but on the caller's side. It should not call > > a function which expects a command name with a NULL pointer passed as > > argument. > > I looked around at this a bit this morning. cmd_pxe.c would need a lot > of mangling to pass around cmdtp, just for the sake of an error message > that's then ignored as the caller logic is: > 1) Try bootm on the image > 2) If CONFIG_CMD_BOOTZ, if bootm returned, maybe we got a zImage? > do_bootz instead (also NULL cmdtp). > > The error message wouldn't exactly make sense here either, being invoked > via menu.
So, I went off an modified cmd_pxe.c to pass around cmdtp so that we get an error message out, and it's not too bad looking, but it highlights another problem, which is that we could really use a way to get at least the "is this a ... ?" code, and just get the error code, rather than printouts. The pxe (and really it's the syslinux.conf/extlinux.conf parsing) code shouldn't be doing bootm();bootz() but checking the image type and calling the right boot. -- Tom
signature.asc
Description: Digital signature
_______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
