[prev in list] [next in list] [prev in thread] [next in thread] 

List:       lhcs-devel
Subject:    [lhcs-devel] Re: fix for physical cpu add
From:       Rusty Russell <rusty () rustcorp ! com ! au>
Date:       2004-03-18 4:31:55
Message-ID: 1079584314.19029.200.camel () bach
[Download RAW message or body]

On Thu, 2004-03-18 at 10:10, Ashok Raj wrote: 
> Hi Rusty
> 
> We are experimenting with physical cpu add, and as a first step we did this
> 
> - boot with maxcpus=2 on a 4 cpu system.
> - for the time being, phys_cpu_present_map in IA64 is set (i.e cpu_possible)
> will be true, and hence the control files for sysfs will also be created
> during topology_init()
> - now we try to bringup the cpu via echo 1> online
> 
> I hit the BUG check in arch/ia64/kernel/setup.c function cpu_init() line
> 626, when the check for idle task is done.
> 
> If (current->mm)
>    BUG()
> 
> The issue is when we do echo, its not called from a kernel thread, hence
> current->mm is not null. I did the following and that seems to get over the
> problem, and now we are able to do cpu_up on a new cpu that was not started
> during the boot time. I have only tested this on IA64, but seems like it
> would be a generic problem.

Hi Ashok,

This is one reasons why the PPC64 code generates all the threads at
boot.  But your problem is a little worse: the idle thread would inherit
other things from the echo command.  I recommend you use schedule_work()
to create a thread from keventd, which will be clean.

struct create_idle
{
	struct task_struct *idle;
	struct completion done;
}

static int do_idle(void *_ci)
{
	struct create_idle *ci = _ci;

	ci->idle = current;
	complete(&ci->done);
	for (;;)
		idle();
}

static void create_idle(void *_ci)
{
	int ret;

	ret = kernel_thread(do_idle, _ci, CLONE_KERNEL|CLONE_IDLETASK);
	if (ret < 0) {
		struct create_idle *ci = _ci;
		ci->idle = ERR_PTR(ret);
		complete(&ci->done);
	}
}


	struct create_idle ci;
	init_completion(&ci.done);
	DECLARE_WORK(work, create_idle, &ci);
	schedule_work(work);
	wait_for_completion(&ci.done);

Code is, of course, untested...
Hope that helps,
Rusty.

-- 
Anyone who quotes me in their signature is an idiot -- Rusty Russell



-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
lhcs-devel mailing list
lhcs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/lhcs-devel
[prev in list] [next in list] [prev in thread] [next in thread]