Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 19 Jun 2015 16:56:47 -0700
From:      Maxim Sobolev <sobomax@FreeBSD.org>
To:        Ian Lepore <ian@freebsd.org>
Cc:        src-committers@freebsd.org, svn-src-all@freebsd.org,  svn-src-head@freebsd.org
Subject:   Re: svn commit: r284614 - head/sys/boot/uboot/lib
Message-ID:  <CAH7qZfuj4ZqEVVWtROZXTsrhy7C2DFw06ijJxUwbLbGEDtRPbQ@mail.gmail.com>
In-Reply-To: <1434755385.1415.114.camel@freebsd.org>
References:  <201506192224.t5JMOxpC097306@svn.freebsd.org> <1434755385.1415.114.camel@freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help
Ian, that's cool and dandy, but I still suggest we put some sanity checking
and have certain workarounds in the loader (whenever it does not add
ambiguity or blows up a code too much of course), so that new folks in town
trying to port to new platforms like myself won't spend hours and hours
hunting known issues and bugs. And hitting those infinite loops is very
frustrating with no errors or anything. On top of that, in some cases you
may be stuck with vendor-provided u-boot with no way to patch and
re-compile. BTW, there is another stupid bug existing in the u-boot loader,
which basically sets fdtaddr in decimal not in hex. On my particular board
this makes ubldr to blow up with CPU exception, unfortunately no workaround
is possible since there is no 0x for hex values and majority of cases when
this variable is set is in hex.

-Maxim

On Fri, Jun 19, 2015 at 4:09 PM, Ian Lepore <ian@freebsd.org> wrote:

> On Fri, 2015-06-19 at 22:24 +0000, Maxim Sobolev wrote:
> > Author: sobomax
> > Date: Fri Jun 19 22:24:58 2015
> > New Revision: 284614
> > URL: https://svnweb.freebsd.org/changeset/base/284614
> >
> > Log:
> >   Provide bug4bug workaround for certain dumbiness of the u-boot's
> API_env_enum
> >   function, which is expected to set returned env to NULL upon reaching
> the end
> >   of the environment list but fails to do so in certain cases. The
> respective
> >   u-boot code looks like the following (HEAD at the time of this commit):
> >
> >   --- api.c ---
> >    496 static int API_env_enum(va_list ap)
> >    ...
> >    510                 *next = last;
> >    511
> >    512                 for (i = 0; env_get_char(i) != '\0'; i = n + 1) {
> >    513                         for (n = i; env_get_char(n) != '\0'; ++n)
> {
> >    514                                 if (n >= CONFIG_ENV_SIZE) {
> >    515                                         /* XXX shouldn't we set
> *next = NULL?? */
> >    516                                         return 0;
> >    517                                 }
> >    518                         }
> >   -------------
> >
> >   The net result is that any unfortunate user of the loader's
> ub_env_enum()
> >   function hitting this condition would be trapped in the infinite loop,
> as
> >   the main use pattern of ub_env_enum() is basically the following:
> >
> >   while ((env = ub_env_enum(env)) != NULL) { DO STUFF }
> >
> >   Which would stuck forever with the last element.
> >
> > Modified:
> >   head/sys/boot/uboot/lib/glue.c
> >
> > Modified: head/sys/boot/uboot/lib/glue.c
> >
> ==============================================================================
> > --- head/sys/boot/uboot/lib/glue.c    Fri Jun 19 21:55:12 2015
> (r284613)
> > +++ head/sys/boot/uboot/lib/glue.c    Fri Jun 19 22:24:58 2015
> (r284614)
> > @@ -513,7 +513,7 @@ ub_env_enum(const char *last)
> >       if (!syscall(API_ENV_ENUM, NULL, (uint32_t)last, (uint32_t)&env))
> >               return (NULL);
> >
> > -     if (env == NULL)
> > +     if (env == NULL || last == env)
> >               /* no more env. variables to enumerate */
> >               return (NULL);
> >
> >
>
> This is only a problem with an unpatched u-boot, which has a completely
> bogus and un-useful implementation of API_env_enum().  That's why every
> one of our u-boot ports has the same patch to put in an implementation
> that actually works.
>
> Your change works around the worst part of the bug (the infinite loop)
> but leaves the major problem of the implementation only returning values
> initially loaded from the saved environment, not anything set by the
> scripts that loaded ubldr().
>
> -- Ian
>
>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAH7qZfuj4ZqEVVWtROZXTsrhy7C2DFw06ijJxUwbLbGEDtRPbQ>