Re: [patch 1/3] add basic accounting fields to taskstats

From: Balbir Singh
Date: Wed Aug 02 2006 - 13:15:06 EST

Shailabh Nagar wrote:
Jay Lan wrote:
Balbir Singh wrote:

Jay Lan wrote:

+#define TASK_COMM_LEN 16

We should find a way to keep this in sync with with the definition
in linux/sched.h (won't we a warning if both this header and
linux/sched.h are included together?)

I do not know how to sync it up. This header linux/taskstats.h is
meant to be included by userspace programs. If an application
happens to include linux/sched.h, which includes linux/time.h,
the application will very likely have compilation errors because
the "struct timespec" declaration in <linux/time.h> and <time.h>
are conflicting.

The <linux/acct.h> defines it to
#define ACCT_COMM 16

I can change our define to TS_COMM_LEN with remakes saying it
should be in sync with the TAKS_COMM_LEN defined in linux/sched.h.

This seems like a good enough way to do it. There's no real need for
the taskstats comm length to remain exactly in sync with the task struct's
comm length (by way of trying to include sched.h etc.) though avoiding the
compile error by renaming is desirable as Balbir pointed out.

Moreover, TASK_COMM_LEN in linux/sched.h isn't likely to change much -
if it increases and csa_acct users also really need the extra info provided,
taskstats can always be changed and version bumped up. If the size decreases
there's no harm done (strncpy should be sufficient protection).


I am not sure if there is a version of BUG_ON() for compile time
asserts. Basically, if we have an infrastructure of the form

* From C/C++ users journal November 2004
#define STATIC_BUG_ON(e) \
switch (0) { \
case 0: \
case (e): \
; \

Then the STATIC_BUG_ON() can catch as shown below.

#define TASK_COMM_LEN 16
#define T_COMM_LEN 20


STATIC_BUG_ON gives the following warning

bug_on_c.c: In function `main':
bug_on_c.c:19: duplicate case value
bug_on_c.c:19: previously used here

but with T_COMM_LEN set to 16

It compiles without any errors, the code generated also
looks like it has no overhead

8048310: 55 push %ebp
8048311: 89 e5 mov %esp,%ebp
8048313: 83 ec 08 sub $0x8,%esp
8048316: 83 e4 f0 and $0xfffffff0,%esp
8048319: c9 leave
804831a: c3 ret
804831b: 90 nop

Assuming such infrastructure is available, you could then

#ifdef __KERNEL__
#include <linux/sched.h>
#define TS_COMM_LEN 16



Balbir Singh,
Linux Technology Center,
IBM Software Labs
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at