Re: Subject:[ PATCH 001:003] OFFSCHED CORE

[ebiederm@xmission.com (Eric W. Biederman)]

raz ben yehuda <raziebe@gmail.com> writes:

> 		OFFSCHED support.
>
> . A callback invoked just before halting the processor.
> . SMP alternative work-around. we do not run smp alternatives if we are
>    left with a single processor.
> . A set of utilities used by the offsched driver.

Nacked-by: "Eric W. Biederman" <ebiederm@xmission.com>


Cpu hotunplug is mostly broken when it comes to irqs.
Building on and extending broken code is just asking for trouble,
especially for out of tree code.

> process.c |  109 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> smpboot.c |   13 +++++--
> 2 files changed, 118 insertions(+), 4 deletions(-)
>
> Signed-off-by: raziebe@gmail.com
> ---
> diff -X 2.6.30-offsched/Documentation/dontdiff -urN tmp/linux-2.6.30/arch/x86/kernel/process.c 2.6.30-offsched/arch/x86/kernel/process.c
> --- tmp/linux-2.6.30/arch/x86/kernel/process.c	2009-06-10 06:05:27.000000000 +0300
> +++ 2.6.30-offsched/arch/x86/kernel/process.c	2009-08-21 22:02:44.000000000 +0300
> @@ -613,3 +613,112 @@
>  }
>  early_param("idle", idle_setup);
>  
> +#ifdef CONFIG_HOTPLUG_CPU
> +struct hotplug_cpu{
> +	long flags;
> +	void (*hotplug_cpu_dead)(void);
> +};
> +
> +static long 	offsched_services;
> +
> +#define CPU_OFFSCHED		31
> +#define CPU_OFFSCHED_KMEM	30
> +#define SERVICE_TIMER		01
> +#define SERVICE_NAPI		02
> +
> +DEFINE_PER_CPU(struct hotplug_cpu, offschedcpu);
> +
> +void unregister_offsched(int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	cpu->hotplug_cpu_dead = NULL;
> +	clear_bit(CPU_OFFSCHED, &cpu->flags);
> +}
> +EXPORT_SYMBOL_GPL(unregister_offsched);
> +
> +int is_offsched(int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	return test_bit(CPU_OFFSCHED, &cpu->flags);
> +}
> +EXPORT_SYMBOL_GPL(is_offsched);
> +
> +int is_offsched_kmem(int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	return test_bit(CPU_OFFSCHED_KMEM, &cpu->flags);
> +}
> +EXPORT_SYMBOL_GPL(is_offsched_kmem);
> +
> +void set_offsched_kmem(int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	set_bit(CPU_OFFSCHED_KMEM, &cpu->flags);
> +}
> +EXPORT_SYMBOL_GPL(set_offsched_kmem);
> +
> +void unset_offsched_kmem(int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	clear_bit(CPU_OFFSCHED_KMEM, &cpu->flags);
> +}
> +EXPORT_SYMBOL_GPL(unset_offsched_kmem);
> +
> +int register_offsched(void (*offsched_callback)(void), int cpuid)
> +{
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	if (is_offsched(cpuid))
> +		return -1;
> +	cpu->hotplug_cpu_dead = offsched_callback;
> +	set_bit(CPU_OFFSCHED, &cpu->flags);
> +	return 0;
> +}
> +EXPORT_SYMBOL_GPL(register_offsched);
> +
> +void run_offsched(void)
> +{
> +	int cpuid = raw_smp_processor_id();
> +	struct hotplug_cpu *cpu = &per_cpu(offschedcpu, cpuid);
> +	cpu->hotplug_cpu_dead();
> +}
> +
> +int  offsched_is_napi_avail(void)
> +{
> +	return SERVICE_NAPI & offsched_services;
> +}
> +
> +int (*offsched_rule)(void *);
> +EXPORT_SYMBOL(offsched_rule);
> +/*
> +* if we a have filter than apply filter.
> +* return 0 if packet should go to kernel
> +* return 1 if packet should go to offsched.
> +*/
> +int offsched_pkt(void *s)
> +{
> +	if (offsched_rule)
> +		return offsched_rule(s);
> +	return 0;
> +}
> +
> +void offsched_register_napi(void)
> +{
> +	offsched_services |= SERVICE_NAPI;
> +}
> +EXPORT_SYMBOL_GPL(offsched_register_napi);
> +
> +void offsched_unregister_napi(void)
> +{
> +	offsched_services &= ~SERVICE_NAPI;
> +}
> +EXPORT_SYMBOL_GPL(offsched_unregister_napi);
> +
> +/*
> + * napi scheudule is called in offsched context.
> +*/
> +void offsched_napi_schedule(void *n)
> +{
> +
> +}
> +
> +#endif
> diff -X 2.6.30-offsched/Documentation/dontdiff -urN tmp/linux-2.6.30/arch/x86/kernel/smpboot.c 2.6.30-offsched/arch/x86/kernel/smpboot.c
> --- tmp/linux-2.6.30/arch/x86/kernel/smpboot.c	2009-06-10 06:05:27.000000000 +0300
> +++ 2.6.30-offsched/arch/x86/kernel/smpboot.c	2009-07-25 13:59:19.000000000 +0300
> @@ -39,6 +39,8 @@
>   *	Glauber Costa		:	i386 and x86_64 integration
>   */
>  
> +#define DEBUG
> +
>  #include <linux/init.h>
>  #include <linux/smp.h>
>  #include <linux/module.h>
> @@ -686,8 +688,8 @@
>  	};
>  
>  	INIT_WORK(&c_idle.work, do_fork_idle);
> -
> -	alternatives_smp_switch(1);
> +	if (!is_offsched(cpu))
> +		alternatives_smp_switch(1);
>  
>  	c_idle.idle = get_idle_for_cpu(cpu);
>  
> @@ -1283,8 +1285,9 @@
>  	for (i = 0; i < 10; i++) {
>  		/* They ack this in play_dead by setting CPU_DEAD */
>  		if (per_cpu(cpu_state, cpu) == CPU_DEAD) {
> -			printk(KERN_INFO "CPU %d is now offline\n", cpu);
> -			if (1 == num_online_cpus())
> +			printk(KERN_INFO "CPU %d is now offline %s\n", cpu,
> +				is_offsched(cpu) ? "and OFFSCHED" : "");
> +			if (1 == num_online_cpus() && !is_offsched(cpu))
>  				alternatives_smp_switch(0);
>  			return;
>  		}
> @@ -1313,6 +1316,8 @@
>  void native_play_dead(void)
>  {
>  	play_dead_common();
> +	if (is_offsched(raw_smp_processor_id()))
> +		run_offsched();
>  	wbinvd_halt();
>  }
>  
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/