Date: Mon, 23 May 2005 23:01:06 -0700
From: Ashok Raj <ashok.raj@intel.com>
To: Srivatsa Vaddagiri <vatsa@in.ibm.com>
Cc: Ashok Raj <ashok.raj@intel.com>, Andi Kleen <ak@muc.de>,
       zwane@arm.linux.org.uk, discuss@x86-64.org, shaohua.li@intel.com,
       linux-kernel@vger.kernel.org, rusty@rustycorp.com.au
Subject: Re: [discuss] Re: [patch 0/4] CPU hot-plug support for x86_64
Message-ID: <20050523230106.A13839@unix-os.sc.intel.com>
References: <20050520221622.124069000@csdlinux-2.jf.intel.com> <20050523164046.GB39821@muc.de> <20050523095450.A8193@unix-os.sc.intel.com> <20050523171212.GF39821@muc.de> <20050523104046.B8692@unix-os.sc.intel.com> <20050524054617.GA5510@in.ibm.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.2.5.1i
In-Reply-To: <20050524054617.GA5510@in.ibm.com>; from vatsa@in.ibm.com on Tue, May 24, 2005 at 11:16:17AM +0530
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1805
Lines: 43

On Tue, May 24, 2005 at 11:16:17AM +0530, Srivatsa Vaddagiri wrote:
> On Mon, May 23, 2005 at 10:40:46AM -0700, Ashok Raj wrote:
> > Iam not a 100% sure about above either, if the smp_call_function 
> > is started with 3 cpus initially, and 1 just came up, the counts in 
> > the smp_call data struct could be set to 3 as a result of the new cpu 
> > received this broadcast as well, and we might quit earlier in the wait.
> 
> True.

Thanks for confirming. 

I think the window would be a little bit more narrow, but not completely 
closed even if we bring cpu in cli state completely.

> 
> > sending to only relevant cpus removes that ambiquity. 
> 
> Or grab the 'call_lock' before setting the upcoming cpu in the online_map.
> This should also avoid the race when a CPU is coming online.
> 

We do this today in x86_64 case when we setup this upcomming cpu in 
cpu_online_map. But the issue is when we use ipi broadcast, its an ugly
case when we dont know if the new cpu is receiving this as well, and 
we dont have real control there.

i converted to use send_IPI_mask(cpu_online_map, CALL_FUNCTION_VECTOR)
instead of the send_IPI_allbutself(CALL_FUNCTION_VECTOR) in 
__smp_call_function(), apart from taking the call_lock before setting the
bit in online_map.

Since Andi is concerned about tlb flush intr performance in the 8cpu and less
case, iam planning temporarily use a startup cmd or choose this option 
automatically if CONFIG_HOTPLUG_CPU is set for the time being, until we can 
find a clean solution that satisfies everyone.

Cheers,
Ashok Raj
-
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/