Re: sched_clock - microblaze

[Michal Simek <michal.simek@petalogix.com>]

Wu Zhangjin wrote:
> On Thu, 2010-04-15 at 18:32 +0200, Michal Simek wrote:
>> Steven Rostedt wrote:
>>> On Thu, 2010-04-15 at 16:55 +0200, Michal Simek wrote:
>>>> Hi Thomas and Steven,
>>>>
>>>> I would like to improve time measuring for ftrace (Log below)
>>>>
>>>> I looked at http://lkml.org/lkml/2009/11/16/181
>>>> where Thomas suggested to create sched_clock() function.
>>>> I used the same solution as Wu proposed but it is not nice.
>>>>
>>>> Is unimplemented sched_clock the reason why ftrace not show fine grain time?
>>> Yeah, sched_clock is used by ftrace for timings, so if it only returns
>>> jiffies, then that will, unfortunately, be the resolution of the tracer.
>>>
>>> I've been told that if you make a higher resolution timer for sched
>>> clock, it will improve the scheduling in CFS.
>> We have two timers (in one IP).
>> timer0 - in IRQ mode - clock event
>> timer1 - free running up counter without IRQ - clock source
>>
>> I hope that I can use timer1 in sched_clock too.
>>
>> Anyway. I played with it a little bit and I also implemented microblaze 
>> specific sched_clock function (inspired by arch/arm/mach-ixp4xx/common.c)
>>
>> unsigned long long notrace sched_clock(void)
>> {
>> 	printk("%s\n", __func__);
>> 	cycle_t cyc = microblaze_read(NULL);
>> 	struct clocksource *cs = &clocksource_microblaze;
>> 	return clocksource_cyc2ns(cyc, cs->mult, cs->shift);
>> }
> 
> Hey, I guess you got:
> 
> sched_clock() -> printk() -> sched_clock() ...
> 
> since printk() is not annotated with notrace, it will call mcount() and
> then sched_clock() and ... so, you can not call printk() in
> sched_clock().
> 
> but you may try this if you want to print something in sched_clock() for
> debugging.
> 
> #include <>
> 
> sched_clock() {
> 	ftrace_stop();
> 	printk(...);
> 	ftrace_start();
> 	...
> }

But I am describing different problem. sched_clock is called before 
kernel initialized timers. Microblaze takes all information from DTS 
that's means that I am not able to read timer because I even don't know 
where it is. :-(

Thanks,
Michal




> 
> Regards,
> 	Wu Zhangjin
> 
>> (And comment sched_clock in kernel/sched_clock.c because our toolchain 
>> has the problem with weak symbols)
>>
>> Recompile the kernel and then show the log_buf
>>
>> <4>Ramdisk addr 0x00000003, Compiled-in FDT at 0xc0242a28
>> <5>Linux version 2.6.34-rc4-00053-gbc6ce8a-dirty (monstr@monstr.eu) (gcc 
>> version 4.1.2) #27 Thu Apr 15 16:38:16 CEST 2010
>> <6>setup_cpuinfo: initialising
>> <6>setup_cpuinfo: Using full CPU PVR support
>> <6>cache: wb_msr
>> <6>setup_memory: Main mem: 0x48000000-0x50000000, size 0x08000000, 
>> klimit 0xc09c7000
>> <6>setup_memory: max_mapnr: 0x8000
>> <6>setup_memory: min_low_pfn: 0x48000
>> <6>setup_memory: max_low_pfn: 0x50000
>> <4>reserved 0 - 0x48000000-0x009c8000
>> <4>reserved 1 - 0x48fe0000-0x00020000
>> <4>reserved 2 - 0x4fffd8c8-0x00002738
>> <7>On node 0 totalpages: 32768
>> <7>free_area_init_node: node 0, pgdat c02bc70c, node_mem_map c09c8000
>> <7>  Normal zone: 256 pages used for memmap
>> <7>  Normal zone: 32512 pages, LIFO batch:7
>> <4>Built 1 zonelists in Zone order, mobility grouping on.  Total pages: 
>> 32512
>> <5>Kernel command line:  console=ttyS0,115200
>> <6>PID hash table entries: 512 (order: -1, 2048 bytes)
>> <6>Dentry cache hash table entries: 16384 (order: 4, 65536 bytes)
>> <6>Inode-cache hash table entries: 8192 (order: 3, 32768 bytes)
>> <6>Memory: 119796k/131072k available
>> <4>sched_clock
>>
>> The problem I see is that scheck_clock is used before our timer 
>> subsystem is initialized.
>> What am I missing?
>>
>> Thanks,
>> Michal
>>
> 
> 


-- 
Michal Simek, Ing. (M.Eng)
PetaLogix - Linux Solutions for a Reconfigurable World
w: www.petalogix.com p: +61-7-30090663,+42-0-721842854 f: +61-7-30090663