Re: [PATCH] performance counters 3.1 for 2.5.45 [4/4]: kernel

[Mikael Pettersson]

Stephen Hemminger writes:
 > On Thu, 2002-10-31 at 15:10, Mikael Pettersson wrote:
 > > This is part 4 of 4 of perfctr-3.1 for the 2.5.45 kernel:
 > > kernel changes to integrate the low and high-level drivers.
...
 > 
 > Rather than adding yet another system call, shouldn't this be done by
 > extending the cpu part of sysfs (or /proc)?

Why?
ptrace() is a system call
fork() is a system call
write() is a system call
other arch's performance counter interfaces are system calls
...

 > It looks like the operations in sys_vperfctr could be easily mapped to a
 > RAM based file system.  As it stands it reminds me of one of the old DEC
 > graphic libraries with one API entry for all the operations, and 65
 > different flag values.

There are about half a dozen operations, not 65. I could use 6 system calls,
no problem, but as long as Linus is silent about this I have no reason to
change it speculatively in random directions.

/Mikael