Warning message when compiling ioremap.c

Hi,

I'm not skilled with MM at all, so sorry if I'm saying something
stupid.

When compiling Linux (latest kernel from Linus' git) on ARM, I noticed the following warning:

CC arch/arm/mm/ioremap.o
arch/arm/mm/ioremap.c: In function '__arm_ioremap_pfn':
arch/arm/mm/ioremap.c:83: warning: control may reach end of non-void function 'remap_area_pte' being inlined

According to the message in the printk, we go to "bad" when the page already exists.

So, I'm wondering if we shouldn't return a -EEXIST (see the patch attached). This would remove that annoying warning message during compilation...

Is it a good/bad idea ?

Regards,

Claudio

From: Claudio Scordino <[email protected]>
Date: Mon, 1 Sep 2008 11:30:55 +0200
Subject: [PATCH 1/1] Fix compilation warning related to return value in remap_area_pte(...).

---
 arch/arm/mm/ioremap.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

index b81dbf9..1e59a03 100644
@@ -64,6 +64,7 @@ static int remap_area_pte(pmd_t *pmd, unsigned long addr, unsigned long end,
  bad:
 	printk(KERN_CRIT "remap_area_pte: page already exists\n");
 	BUG();
+	return -EEXIST;
 }
 
 static inline int remap_area_pmd(pgd_t *pgd, unsigned long addr,

1.5.4.3

Em Tue, 02 Sep 2008 09:37:14 +0200
Claudio Scordino <[email protected]> escreveu:

| Hi,
| 
|        I'm not skilled with MM at all, so sorry if I'm saying something
| stupid.
| 
| When compiling Linux (latest kernel from Linus' git) on ARM, I noticed 
| the following warning:
| 
| CC      arch/arm/mm/ioremap.o
| arch/arm/mm/ioremap.c: In function '__arm_ioremap_pfn':
| arch/arm/mm/ioremap.c:83: warning: control may reach end of non-void
| function 'remap_area_pte' being inlined
| 
| According to the message in the printk, we go to "bad" when the page
| already exists.

You see that right before the return you have added there is a
BUG() macro? That macro will call panic(), this means that this function will never return if it reaches that point.

If all you want is to silent gcc, you should remove the goto and
move the bad label contents there.

This is minor, but I see no need for the goto.

Luiz Fernando N. Capitulino

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Luiz Fernando N. Capitulino ha scritto:

Well, probably BUG() doesn't call panic() always. For instance, in arch/arm/include/asm/bug.h in case CONFIG_DEBUG_BUGVERBOSE is defined (and it might be), BUG just causes an oops (by dereferencing a NULL pointer). If I'm not wrong this doesn't always mean a panic...

However, in any case, the handler of pagefault eventually calls do_exit(), so you're right: what follows BUG() won't be executed.

Yes, it's obviously minor. But I don't like having meaningless warnings during compilation: they just confuse output, and people may miss some important warning message...

The need for the goto exists only if BUG() can return, and it doesn't, so we can safely remove it as you suggested.

Who's in charge of maintaining this piece of code? Should the patch in attachment be submitted to some specific person?

Many thanks,

Claudio

The structure of the original code (with the goto) is arranged so that the code path is a straight line for the normal, non-bad case. If you want to remove the goto, you should wrap the condition in unlikely() so as not to introduce another branch.

I guess you should send it to the linux-arm-kernel list.

p.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>