[maemo-developers] N900 kernel 2.6.28 no compiles after changing cb[] array size

[From: Nathan Anderson]

 Sergio,

	Maybe the actual problem is that 62 is not a multiple of 4.   the
sizeof(void (*)) should be 4 on 32 bit platforms.   So if you read that; you
get:
> 62-8 (=54)
> 54 / 4 (=13.5)

On the old value you get
> 48-8 (=40)
> 40/4 (=10)


What happens if you change it to 
	char                    cb[64];

The compiler could be having a issue because of 13.5 and just masking the
issue until later.  (Gotta love C/C++ compilers, you can chase your tail
sometimes <G>)

Nathan





-----Original Message-----
From: maemo-developers-bounces at maemo.org
[mailto:maemo-developers-bounces at maemo.org] On Behalf Of Sergio Lembo
Sent: Thursday, June 17, 2010 1:59 PM
To: maemo-developers at maemo.org
Subject: RE: N900 kernel 2.6.28 no compiles after changing cb[] array size


Hi,

Thank you for your answer.

I observe that in include/net/mac80211.h the defines pasted below are
introducing a non-linear sizing (observe the hardcoded number 8) that may
cause problems for some ranges and thus validating (sizeof(struct
ieee80211_tx_info) > sizeof(skb->cb)).

#define IEEE80211_TX_INFO_DRIVER_DATA_SIZE \
	(sizeof(((struct sk_buff *)0)->cb) - 8)

#define IEEE80211_TX_INFO_DRIVER_DATA_PTRS \
	(IEEE80211_TX_INFO_DRIVER_DATA_SIZE / sizeof(void *))

Although this is not directly the problem that I have and

   BUILD_BUG_ON(sizeof(struct ieee80211_tx_info) > sizeof(skb->cb));

is failing for some other reason that I did not find yet..

Regards,

Sergio




Quoting Nathan Anderson <nathan at andersonsplace.net>:

> Sergio,
>
> 	If I had a guess, I would guess that the n900 probably has a 
> different set of compile time defines, that cause the ieee80211 
> structure to be defined differently on the n900 vs a stock machine.  
> The compile time failsafe is to make sure that the structure iee80211 is
larger than skb->cb.
> Run a grep on the code to find out where/how the ieee80211 structure 
> is being defined and you will probably find the issue.
>
> Nathan.
>
> -----Original Message-----
> From: maemo-developers-bounces at maemo.org
> [mailto:maemo-developers-bounces at maemo.org] On Behalf Of Sergio Lembo
> Sent: Thursday, June 17, 2010 3:08 AM
> To: maemo-developers at maemo.org
> Subject: N900 kernel 2.6.28 no compiles after changing cb[] array size
>
>
> Hi,
>
>    I am having an inconsistent state in the N900 kernel 2.6.28 when I 
> make a very simple
> change:
>
>    The problem is reproduced easily by just increasing the size of the 
> control buffer cb[] in struct sk_buff{} (include/linux/skbuff.h):
>
> -       char                    cb[48];
> -
> +       char                    cb[62];
>
>    Using cb[62] I get a compilation error :
>
> CC [M]  net/mac80211/main.o
> net/mac80211/main.c: In function 'ieee80211_init':
> net/mac80211/main.c:1026: error: size of array 'type name' is negative
> make[2]: *** [net/mac80211/main.o] Error 1
> make[1]: *** [net/mac80211] Error 2
> make: *** [net] Error 2
>
>    The line with the error is a failsafe at compilation time
> (net/mac80211/main.c):
>
>         BUILD_BUG_ON(sizeof(struct ieee80211_tx_info) > 
> sizeof(skb->cb));
>
>    Bypassing such failsafe line the kernel compiles Ok, but the binary 
> flashed to the N900 fails to boot the system.
>
>    I observe that such compilation error does not happen when 
> compiling a plain 2.6.28 kernel for a PC.
>
>    And note that this error is just caused by changing the size of 
> cb[] from
> 48 to 62; that is the only modification we made to the kernel sources.
>
>
> Regards,
>
> Sergio
>
>
>
> _______________________________________________
> maemo-developers mailing list
> maemo-developers at maemo.org
> https://lists.maemo.org/mailman/listinfo/maemo-developers
>
>



--
-----------------------
Sergio Lembo
sergio.lembo at tkk.fi
-----------------------

_______________________________________________
maemo-developers mailing list
maemo-developers at maemo.org
https://lists.maemo.org/mailman/listinfo/maemo-developers