Skip to contentSkip to conversation

arch/arm/lib/board.c - uninitialized vars

U-Boot Wolfgang Denk 2 participants 3 messages Nov 6, 2014

lore
Share & tools
lore
#0 lore raw

Hi,

I'm trying to clean up some warnings/errors detected when running "cppcheck" on the U-Boot source tree. For arch/arm/lib/board.c I get this:

[arch/arm/lib/board.c:445]: (error) Uninitialized variable: id [arch/arm/lib/board.c:422]: (error) Uninitialized variable: addr_sp

The problem is not usually detected by GCC depending on which macros are active, here especially CONFIG_SPL_BUILD

The relevant code was last touched / introduced by commit f1d2b313:
"ARM: add relocation support" some two years ago...

I have some questions regarding the CONFIG_SPL_BUILD "else" case (i. e. when building with CONFIG_SPL_BUILD defined):

422 addr_sp += 128; /* leave 32 words for abort-stack */

Is this correct? The stack is growing downward, so should the '+' not be replaced by a '-', like we do a few lines above:

412 /* leave 3 words for abort-stack */
413 addr_sp -= 12;

Why do we need 128 words in the CONFIG_SPL_BUILD case, but only 3 otherwise?

Should we not move the "alignment for ABI compliance" part outside the CONFIG_SPL_BUILD if/else case, i. e. should this not always be done?

And of course, how should we correctly initialize the "id" and "addr_sp" variable in both cases?

Best regards,

Wolfgang Denk

Signature · 4 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Der Dativ ist dem Genitiv sein Tod.
#1 lore raw

Hello Wolfgang,

Am 06.11.2014 12:28, schrieb Wolfgang Denk:

Quoted 15 lines

Hi,

I'm trying to clean up some warnings/errors detected when running "cppcheck" on the U-Boot source tree. For arch/arm/lib/board.c I get this:

[arch/arm/lib/board.c:445]: (error) Uninitialized variable: id [arch/arm/lib/board.c:422]: (error) Uninitialized variable: addr_sp

The problem is not usually detected by GCC depending on which macros are active, here especially CONFIG_SPL_BUILD

The relevant code was last touched / introduced by commit f1d2b313:
"ARM: add relocation support" some two years ago...

Quoted 7 lines

I have some questions regarding the CONFIG_SPL_BUILD "else" case (i. e. when building with CONFIG_SPL_BUILD defined):

422 addr_sp += 128; /* leave 32 words for abort-stack */

Is this correct? The stack is growing downward, so should the '+' not be replaced by a '-', like we do a few lines above:

No, this is a typo ... it must be a "-"

Quoted 6 lines

412 /* leave 3 words for abort-stack */
413 addr_sp -= 12;

Why do we need 128 words in the CONFIG_SPL_BUILD case, but only 3 otherwise?

Good question ...

Quoted 2 lines

Should we not move the "alignment for ABI compliance" part outside the CONFIG_SPL_BUILD if/else case, i. e. should this not always be done?

Yes, I think thats correct.

But looking into common/board_f.c ... there it is also done only for the not SPL case ...

going back in history ...

http://git.denx.de/?p=u-boot.git;a=blob;f=arch/arm/cpu/arm926ejs/start.S;h=cf40ce12928359c5238d02a32ed361ccc153c101;hb=a59e27997637a2395ae2cc7f809127f24119a167
here I see:

leave 32 words for abort-stack
leave 3 words for abort-stack
8-byte alignment for ABI compliance

without SPL (former PRELOADER) define mess ...

http://git.denx.de/?p=u-boot.git;a=blob;f=arch/arm/cpu/arm1136/start.S;h=41eb82dae246b9509545f051daac605a02b7db05;hb=a59e27997637a2395ae2cc7f809127f24119a167#l179
here:

#ifdef CONFIG_PRELOADER
leave 32 words for abort-stack
#else
leave 3 words for abort-stack
#endif
8-byte alignment for ABI compliance

http://git.denx.de/?p=u-boot.git;a=blob;f=arch/arm/cpu/pxa/start.S;h=e07c8c2e0e70744c45152c1649cef65f87825708;hb=a59e27997637a2395ae2cc7f809127f24119a167
here
leave 3 words for abort-stack
8-byte alignment for ABI compliance

no "leave 32 words for abort-stack"

... :-(

Quoted 2 lines

And of course, how should we correctly initialize the "id" and "addr_sp" variable in both cases?

Hmm.. they are initialized for the !SPL case ... and in the SPL case, board__init_f is used from ./arch/arm/lib/spl.c or overwritten from SoC dependend code, like done in:

./arch/arm/cpu/arm926ejs/davinci/spl.c

-> so I think, we can drop the CONFIG_SPL_BUILD in u-boot:/arch/arm/lib/board.c board_init_f(), as it is not used, thats maybe the reason, why this issue never poped up!

bye,
Heiko

Signature · 2 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
#2 lore raw

Dear Heiko,

Quoted 4 lines · In message <[email protected]> you wrote:

-> so I think, we can drop the CONFIG_SPL_BUILD in u-boot:/arch/arm/lib/board.c board_init_f(), as it is not used, thats maybe the reason, why this issue never poped up!

I agree - this appears to be dead code.

Albert, what do you think?

And, most important: Who is going to provide a patch?

Best regards,

Wolfgang Denk

Signature · 5 lines
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
He'd been wrong, there _was_ a light at the end of the tunnel, and it
was a flamethrower.                         - Terry Pratchett, _Mort_