Patchwork [U-Boot,v2,1/5] bootm: Handle errors consistently

login
register
mail settings
Submitter Simon Glass
Date July 4, 2013, 8:17 p.m.
Message ID <1372969032-20009-1-git-send-email-sjg@chromium.org>
Download mbox | patch
Permalink /patch/256992/
State Superseded
Delegated to: Tom Rini
Headers show

Comments

Simon Glass - July 4, 2013, 8:17 p.m.
A recent bootm fix left the error path incomplete. Reinstate this so that
failures in bootm stages are handled properly.

Signed-off-by: Simon Glass <sjg@chromium.org>
---
Changes in v2:
- Correct checking in the no-error case

 common/cmd_bootm.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)
Tom Rini - July 5, 2013, 12:59 p.m.
On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:

> A recent bootm fix left the error path incomplete. Reinstate this so that
> failures in bootm stages are handled properly.
> 
> Signed-off-by: Simon Glass <sjg@chromium.org>
> ---
> Changes in v2:
> - Correct checking in the no-error case


OK, this conflicts with the change I posted (and pushed later than I
thought I had).  Can you confirm the code is good in mainline now?
Thanks!
Simon Glass - July 5, 2013, 7:52 p.m.
Hi Tom,

On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:

> On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
>
> > A recent bootm fix left the error path incomplete. Reinstate this so that
> > failures in bootm stages are handled properly.
> >
> > Signed-off-by: Simon Glass <sjg@chromium.org>
> > ---
> > Changes in v2:
> > - Correct checking in the no-error case
>
>
> OK, this conflicts with the change I posted (and pushed later than I
> thought I had).  Can you confirm the code is good in mainline now?
> Thanks!
>

It's close, but I think it still needs this near the end
of do_bootm_states(), something like:

 else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv); + else
if (ret) + puts("subcommand not supported\n"); return ret;

If you agree, I can prepare a patch as part of the bootz update.

Regards,
Simon
Tom Rini - July 5, 2013, 8:15 p.m.
On Fri, Jul 05, 2013 at 12:52:03PM -0700, Simon Glass wrote:
> Hi Tom,
> 
> On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:
> 
> > On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
> >
> > > A recent bootm fix left the error path incomplete. Reinstate this so that
> > > failures in bootm stages are handled properly.
> > >
> > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > ---
> > > Changes in v2:
> > > - Correct checking in the no-error case
> >
> >
> > OK, this conflicts with the change I posted (and pushed later than I
> > thought I had).  Can you confirm the code is good in mainline now?
> > Thanks!
> >
> 
> It's close, but I think it still needs this near the end
> of do_bootm_states(), something like:
> 
>  else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv); + else
> if (ret) + puts("subcommand not supported\n"); return ret;
> 
> If you agree, I can prepare a patch as part of the bootz update.

How do we get there in the code?  When we do any subcalls is where we've
got that puts already.  Failures from that point on are either the OS
bootm part failed (and return is > 0) or one of the BOOTM_ERR codes.  Or
did I miss a case still?
Simon Glass - July 5, 2013, 8:21 p.m.
Hi Tom,

On Fri, Jul 5, 2013 at 1:15 PM, Tom Rini <trini@ti.com> wrote:

> On Fri, Jul 05, 2013 at 12:52:03PM -0700, Simon Glass wrote:
> > Hi Tom,
> >
> > On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:
> >
> > > On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
> > >
> > > > A recent bootm fix left the error path incomplete. Reinstate this so
> that
> > > > failures in bootm stages are handled properly.
> > > >
> > > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > > ---
> > > > Changes in v2:
> > > > - Correct checking in the no-error case
> > >
> > >
> > > OK, this conflicts with the change I posted (and pushed later than I
> > > thought I had).  Can you confirm the code is good in mainline now?
> > > Thanks!
> > >
> >
> > It's close, but I think it still needs this near the end
> > of do_bootm_states(), something like:
> >
> >  else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv); +
> else
> > if (ret) + puts("subcommand not supported\n"); return ret;
> >
> > If you agree, I can prepare a patch as part of the bootz update.
>
> How do we get there in the code?  When we do any subcalls is where we've
> got that puts already.  Failures from that point on are either the OS
> bootm part failed (and return is > 0) or one of the BOOTM_ERR codes.  Or
> did I miss a case still?
>

I think this is when the boot_os function returns an error. At least the
old code had quite a lot of printf()s for that case.

Regards,
Simon
Tom Rini - July 5, 2013, 8:29 p.m.
On Fri, Jul 05, 2013 at 01:21:09PM -0700, Simon Glass wrote:
> Hi Tom,
> 
> On Fri, Jul 5, 2013 at 1:15 PM, Tom Rini <trini@ti.com> wrote:
> 
> > On Fri, Jul 05, 2013 at 12:52:03PM -0700, Simon Glass wrote:
> > > Hi Tom,
> > >
> > > On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:
> > >
> > > > On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
> > > >
> > > > > A recent bootm fix left the error path incomplete. Reinstate this so
> > that
> > > > > failures in bootm stages are handled properly.
> > > > >
> > > > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > > > ---
> > > > > Changes in v2:
> > > > > - Correct checking in the no-error case
> > > >
> > > >
> > > > OK, this conflicts with the change I posted (and pushed later than I
> > > > thought I had).  Can you confirm the code is good in mainline now?
> > > > Thanks!
> > > >
> > >
> > > It's close, but I think it still needs this near the end
> > > of do_bootm_states(), something like:
> > >
> > >  else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv); +
> > else
> > > if (ret) + puts("subcommand not supported\n"); return ret;
> > >
> > > If you agree, I can prepare a patch as part of the bootz update.
> >
> > How do we get there in the code?  When we do any subcalls is where we've
> > got that puts already.  Failures from that point on are either the OS
> > bootm part failed (and return is > 0) or one of the BOOTM_ERR codes.  Or
> > did I miss a case still?
> >
> 
> I think this is when the boot_os function returns an error. At least the
> old code had quite a lot of printf()s for that case.

We had a printf per subcommand before, and just a single one now.  We
didn't say the 'go' subcommand failed however, just what the
function happened to print out.
Simon Glass - July 5, 2013, 8:48 p.m.
Hi Tom,

On Fri, Jul 5, 2013 at 1:29 PM, Tom Rini <trini@ti.com> wrote:

> On Fri, Jul 05, 2013 at 01:21:09PM -0700, Simon Glass wrote:
> > Hi Tom,
> >
> > On Fri, Jul 5, 2013 at 1:15 PM, Tom Rini <trini@ti.com> wrote:
> >
> > > On Fri, Jul 05, 2013 at 12:52:03PM -0700, Simon Glass wrote:
> > > > Hi Tom,
> > > >
> > > > On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:
> > > >
> > > > > On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
> > > > >
> > > > > > A recent bootm fix left the error path incomplete. Reinstate
> this so
> > > that
> > > > > > failures in bootm stages are handled properly.
> > > > > >
> > > > > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > > > > ---
> > > > > > Changes in v2:
> > > > > > - Correct checking in the no-error case
> > > > >
> > > > >
> > > > > OK, this conflicts with the change I posted (and pushed later than
> I
> > > > > thought I had).  Can you confirm the code is good in mainline now?
> > > > > Thanks!
> > > > >
> > > >
> > > > It's close, but I think it still needs this near the end
> > > > of do_bootm_states(), something like:
> > > >
> > > >  else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv);
> +
> > > else
> > > > if (ret) + puts("subcommand not supported\n"); return ret;
> > > >
> > > > If you agree, I can prepare a patch as part of the bootz update.
> > >
> > > How do we get there in the code?  When we do any subcalls is where
> we've
> > > got that puts already.  Failures from that point on are either the OS
> > > bootm part failed (and return is > 0) or one of the BOOTM_ERR codes.
>  Or
> > > did I miss a case still?
> > >
> >
> > I think this is when the boot_os function returns an error. At least the
> > old code had quite a lot of printf()s for that case.
>
> We had a printf per subcommand before, and just a single one now.  We
> didn't say the 'go' subcommand failed however, just what the
> function happened to print out.
>

Yes that looks right to me. But I believe that the GO command is not
supposed to return, so it might be harmless to put the message there in all
cases. If not, we can add a special case for GO.

Regards,
Simon
Tom Rini - July 8, 2013, 1:24 p.m.
On Fri, Jul 05, 2013 at 01:48:30PM -0700, Simon Glass wrote:
> Hi Tom,
> 
> On Fri, Jul 5, 2013 at 1:29 PM, Tom Rini <trini@ti.com> wrote:
> 
> > On Fri, Jul 05, 2013 at 01:21:09PM -0700, Simon Glass wrote:
> > > Hi Tom,
> > >
> > > On Fri, Jul 5, 2013 at 1:15 PM, Tom Rini <trini@ti.com> wrote:
> > >
> > > > On Fri, Jul 05, 2013 at 12:52:03PM -0700, Simon Glass wrote:
> > > > > Hi Tom,
> > > > >
> > > > > On Fri, Jul 5, 2013 at 5:59 AM, Tom Rini <trini@ti.com> wrote:
> > > > >
> > > > > > On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:
> > > > > >
> > > > > > > A recent bootm fix left the error path incomplete. Reinstate
> > this so
> > > > that
> > > > > > > failures in bootm stages are handled properly.
> > > > > > >
> > > > > > > Signed-off-by: Simon Glass <sjg@chromium.org>
> > > > > > > ---
> > > > > > > Changes in v2:
> > > > > > > - Correct checking in the no-error case
> > > > > >
> > > > > >
> > > > > > OK, this conflicts with the change I posted (and pushed later than
> > I
> > > > > > thought I had).  Can you confirm the code is good in mainline now?
> > > > > > Thanks!
> > > > > >
> > > > >
> > > > > It's close, but I think it still needs this near the end
> > > > > of do_bootm_states(), something like:
> > > > >
> > > > >  else if (ret == BOOTM_ERR_RESET) do_reset(cmdtp, flag, argc, argv);
> > +
> > > > else
> > > > > if (ret) + puts("subcommand not supported\n"); return ret;
> > > > >
> > > > > If you agree, I can prepare a patch as part of the bootz update.
> > > >
> > > > How do we get there in the code?  When we do any subcalls is where
> > we've
> > > > got that puts already.  Failures from that point on are either the OS
> > > > bootm part failed (and return is > 0) or one of the BOOTM_ERR codes.
> >  Or
> > > > did I miss a case still?
> > > >
> > >
> > > I think this is when the boot_os function returns an error. At least the
> > > old code had quite a lot of printf()s for that case.
> >
> > We had a printf per subcommand before, and just a single one now.  We
> > didn't say the 'go' subcommand failed however, just what the
> > function happened to print out.
> 
> Yes that looks right to me. But I believe that the GO command is not
> supposed to return, so it might be harmless to put the message there in all
> cases. If not, we can add a special case for GO.

OK, I've been re-reading the before and after code, and at least wrt
error messages and such, we're OK now.  The 'GO' state is allowed to
return 1 and usually does both for detectable errors (for example, FIT
image for VxWorks, but a bad image) and "wth? we've been returned to
from the OS".  And I don't want to add a won't be reached string to
the build given how badly we see compilers dealing with optimizing
strings, along with the code bloat reason.
Tom Rini - July 10, 2013, 1:17 p.m.
On Thu, Jul 04, 2013 at 01:17:07PM -0700, Simon Glass wrote:

> A recent bootm fix left the error path incomplete. Reinstate this so that
> failures in bootm stages are handled properly.
> 
> Signed-off-by: Simon Glass <sjg@chromium.org>
> ---
> Changes in v2:
> - Correct checking in the no-error case
> 
>  common/cmd_bootm.c | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/common/cmd_bootm.c b/common/cmd_bootm.c
> index 02a5013..652513a 100644
> --- a/common/cmd_bootm.c
> +++ b/common/cmd_bootm.c
> @@ -684,12 +684,8 @@ static int do_bootm_states(cmd_tbl_t *cmdtp, int flag, int argc,
>  	if (!ret && (states & BOOTM_STATE_OS_GO)) {
>  		ret = boot_selected_os(argc, argv, BOOTM_STATE_OS_GO,
>  				images, boot_fn);
> -		if (ret)
> -			goto err;
>  	}
>  
> -	return ret;
> -
>  	/* Deal with any fallout */
>  err:
>  	if (iflag)
> @@ -699,7 +695,7 @@ err:
>  		bootstage_error(BOOTSTAGE_ID_DECOMP_UNIMPL);
>  	else if (ret == BOOTM_ERR_RESET)
>  		do_reset(cmdtp, flag, argc, argv);
> -	else
> +	else if (ret)
>  		puts("subcommand not supported\n");
>  
>  	return ret;

Looking this part over again, BOOTM_STATE_OS_GO handles its own prints
when returning 1, so we don't want to do the "subcommand not supported"
part here (the other case is a BOOTM_ERR_RESET).  But, trace does need
to be covered and isn't by the "subcommand not supported" print.  I'll
make a v3 of this shortly and post.

Patch

diff --git a/common/cmd_bootm.c b/common/cmd_bootm.c
index 02a5013..652513a 100644
--- a/common/cmd_bootm.c
+++ b/common/cmd_bootm.c
@@ -684,12 +684,8 @@  static int do_bootm_states(cmd_tbl_t *cmdtp, int flag, int argc,
 	if (!ret && (states & BOOTM_STATE_OS_GO)) {
 		ret = boot_selected_os(argc, argv, BOOTM_STATE_OS_GO,
 				images, boot_fn);
-		if (ret)
-			goto err;
 	}
 
-	return ret;
-
 	/* Deal with any fallout */
 err:
 	if (iflag)
@@ -699,7 +695,7 @@  err:
 		bootstage_error(BOOTSTAGE_ID_DECOMP_UNIMPL);
 	else if (ret == BOOTM_ERR_RESET)
 		do_reset(cmdtp, flag, argc, argv);
-	else
+	else if (ret)
 		puts("subcommand not supported\n");
 
 	return ret;