Message-Id: <4FBE1F600200007800085C75@nat28.tlf.novell.com>
Date: Thu, 24 May 2012 10:45:36 +0100
From: "Jan Beulich" <JBeulich@suse.com>
To: "Alex Shi" <alex.shi@intel.com>
Cc: <borislav.petkov@amd.com>, <arnd@arndb.de>,
        "Peter Zijlstra" <a.p.zijlstra@chello.nl>, <akinobu.mita@gmail.com>,
        <eric.dumazet@gmail.com>, <fweisbec@gmail.com>, <rostedt@goodmis.org>,
        <hughd@google.com>, <jeremy@goop.org>, <len.brown@intel.com>,
        <tony.luck@intel.com>, <yongjie.ren@intel.com>,
        <kamezawa.hiroyu@jp.fujitsu.com>, <seto.hidetoshi@jp.fujitsu.com>,
        <penberg@kernel.org>, <yinghai@kernel.org>, <tglx@linutronix.de>,
        <akpm@linux-foundation.org>, <ak@linux.intel.com>, <luto@mit.edu>,
        <avi@redhat.com>, <dhowells@redhat.com>, <mingo@redhat.com>,
        <riel@redhat.com>, <cpw@sgi.com>, <steiner@sgi.com>,
        <linux-kernel@vger.kernel.org>, <viro@zeniv.linux.org.uk>,
        <hpa@zytor.com>
Subject: Re: [PATCH v7 8/8] x86/tlb: just do tlb flush on one of
 siblings of SMT
References: <1337782555-8088-1-git-send-email-alex.shi@intel.com>
 <1337782555-8088-9-git-send-email-alex.shi@intel.com>
 <4FBD18D20200007800085951@nat28.tlf.novell.com>
 <1337792984.9783.37.camel@laptop> <4FBDF200.7060608@intel.com>
 <4FBE11EB0200007800085BD0@nat28.tlf.novell.com>
 <4FBDF91C.7050902@intel.com>
In-Reply-To: <4FBDF91C.7050902@intel.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 8BIT
Content-Disposition: inline
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1208
Lines: 34

>>> On 24.05.12 at 11:02, Alex Shi <alex.shi@intel.com> wrote:

>>>
>>> I just measured the cost of this function on my Romely EP(32 LCPUs) with
>>> cpumask_t and NR_CPUS = 32/256/512/4096, the cost are similar with
>>> 256/512/4096 and that increased about 20% time cost from 32.
>>>
>>> I also tried to use cpumask_var_t and alloc it in heap(use
>>> CPUMASK_OFFSTACK), actually, it cost same time with cpumask_t in stack.
>>> But, the allocation bring another big cost. So, I use cpumask_t in stack.
>>> The performance gain data in commit log is getting with NR_CPUS = 256.
>> 
>> Perhaps using a per-CPU cpumask would be the better choice here
> 
> 
> See.
> 
>> (I can't see how preemption could validly be enabled when this
>> code is utilized).
> 
> 
> Sorry, What's your meaning here?, the function is always in pre-empt
> safe mode.

That's exactly what I implied (I was merely pointing this out to
prove that using per-CPU data here is possible).

Jan

--
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/