svn commit: r205683 - head/sys/compat/linprocfs

[John Baldwin]

On Friday 26 March 2010 7:43:15 am Alexander Leidinger wrote:
> Author: netchild
> Date: Fri Mar 26 11:43:15 2010
> New Revision: 205683
> URL: http://svn.freebsd.org/changeset/base/205683
> 
> Log:
>   Fix some bogus values in linprocfs.
>   
>   Submitted by:	Petr Salinger <Petr.Salinger at seznam.cz>
>   Verified on:	GNU/kFreeBSD debian 8.0-1-686 (by submitter)
>   PR:		144584
> 
> Modified:
>   head/sys/compat/linprocfs/linprocfs.c
> 
> Modified: head/sys/compat/linprocfs/linprocfs.c
> 
==============================================================================
> --- head/sys/compat/linprocfs/linprocfs.c	Fri Mar 26 11:33:12 2010	(r205682)
> +++ head/sys/compat/linprocfs/linprocfs.c	Fri Mar 26 11:43:15 2010	(r205683)
> @@ -110,13 +110,36 @@ __FBSDID("$FreeBSD$");
>  /*
>   * Various conversion macros
>   */
> +
> +/* The LINUX_USER_HZ is assumed 100 for now */
> +
> +#if defined(__i386__) && defined(__GNUCLIKE_ASM)
> +/* we need intermediate result as 64 bit, otherwise it overflows too early 
*/
> +#define DO64_MULDIV(v,m,d)       \
> +({                              \
> +   unsigned long rv0;           \
> +   unsigned long rv1;           \
> +   __asm__ __volatile__(        \
> +                "mull %1\n\t"   \
> +                "divl %2\n\t"   \
> +                :"=a" (rv0), "=d" (rv1) \
> +                :"r" (d), "0" (v), "1" (m) \
> +                :"cc" ); \
> +  rv0; \
> +})
> +
> +#define T2J(x) DO64_MULDIV((x), 100UL, (stathz ? stathz : hz)) /* ticks to 
jiffies */
> +#else
>  #define T2J(x) (((x) * 100UL) / (stathz ? stathz : hz))	/* ticks to jiffies 
*/
> +#endif

This is very bogus.  Just use a uint64_t cast or the like.

Extraneous ()'s around (x)->tv_sec and (x)->tv_usec in TV2J() as well.

> @@ -502,12 +525,24 @@ linprocfs_douptime(PFS_FILL_ARGS)
>  {
>  	long cp_time[CPUSTATES];
>  	struct timeval tv;
> +	int cnt, i;
>  
>  	getmicrouptime(&tv);
>  	read_cpu_time(cp_time);
> -	sbuf_printf(sb, "%lld.%02ld %ld.%02ld\n",
> +
> +	for (cnt = 0, i = 0; i <= mp_maxid; ++i)
> +		if (!(CPU_ABSENT(i)))
> +		    cnt++;
> +
> +	if (!cnt)
> +	    cnt = 1;
> +
> +	i = ((cp_time[CP_IDLE])/cnt) % (stathz ? stathz : hz);
> +	i = (i * 100) / (stathz ? stathz : hz);
> +
> +	sbuf_printf(sb, "%lld.%02ld %ld.%02d\n",
>  	    (long long)tv.tv_sec, tv.tv_usec / 10000,
> -	    T2S(cp_time[CP_IDLE]), T2J(cp_time[CP_IDLE]) % 100);
> +	    T2S((cp_time[CP_IDLE]/cnt)), i);
>  	return (0);

What's wrong with mp_ncpus?

Also, I don't understand how the machinations for i vs the original code 
(assuming you add in a divisor of the raw cp_time by mp_ncpus) give a 
different value.

> @@ -613,9 +648,17 @@ linprocfs_doprocstat(PFS_FILL_ARGS)
>  	struct kinfo_proc kp;
>  	char state;
>  	static int ratelimit = 0;
> +	unsigned long startcode, startdata;

Why not make these a vm_offset_t and then print them with %j?

-- 
John Baldwin