Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 13 Jun 2018 23:02:16 +0100 (BST)
From:   James Simmons <jsimmons@infradead.org>
To:     NeilBrown <neilb@suse.com>
cc:     Andreas Dilger <andreas.dilger@intel.com>,
        Oleg Drokin <oleg.drokin@intel.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Lustre Development List <lustre-devel@lists.lustre.org>,
        Doug Oucharek <dougso@me.com>,
        Amir Shehata <amir.shehata@intel.com>
Subject: Re: [PATCH v2 01/25] staging: lustre: libcfs: restore UMP handling
In-Reply-To: <87muwiuiuz.fsf@notabene.neil.brown.name>
Message-ID: <alpine.LFD.2.21.1806132248500.12664@casper.infradead.org>
References: <1527603725-30560-1-git-send-email-jsimmons@infradead.org> <1527603725-30560-2-git-send-email-jsimmons@infradead.org> <87muwiuiuz.fsf@notabene.neil.brown.name>
User-Agent: Alpine 2.21 (LFD 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> > With the cleanup of the libcfs SMP handling all UMP handling
> > was removed. In the process now various NULL pointers and
> > empty fields are return in the UMP case which causes lustre
> > to crash hard. Restore the proper UMP handling so Lustre can
> > properly function.
> 
> Can't we just get lustre to handle the NULL pointer?
> Is most cases, the pointer is accessed through an accessor function, and
> on !CONFIG_SMP, that can be a static inline that doesn't even look at
> the pointer.

Lots of NULL pointer checks for a structure allocated at libcfs module 
start and only cleaned up at libcfs removal is not a clean approach.
So I have thought about it and I have to ask why allocate a global
struct cfs_cpu_table. It could be made static and fill it in which would
avoid the whole NULL pointer issue. Plus for the UMP case why allocate
a new cfs_cpu_table with cfs_cpt_table_alloc() which is exactly like
the default UMP cfs_cpu_table. Instead we could just return the pointer
to the static default cfs_cpt_tab every time. We still have the NULL
ctb_cpumask field to deal with. Does that sound like a better solution
to you? Doug what do you think?
 
> I really think this is a step backwards.  If you can identify specific
> problems caused by the current code, I'm sure we can fix them.
> 
> >
> > Signed-off-by: James Simmons <uja.ornl@yahoo.com>
> > Signed-off-by: Amir Shehata <amir.shehata@intel.com>
> > Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7734
> 
> This bug doesn't seem to mention this patch at all
> 
> > Reviewed-on: http://review.whamcloud.com/18916
> 
> Nor does this review.

Yeah its mutated so much from what is in the Intel tree.
I do believe it was the last patch to touch this.