Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Mon, 8 Aug 2022 21:07:34 -0500
From:   Bjorn Helgaas <helgaas@kernel.org>
To:     Kent Overstreet <kent.overstreet@gmail.com>
Cc:     "Matthew Wilcox (Oracle)" <willy@infradead.org>,
        linux-kernel@vger.kernel.org, pmladek@suse.com,
        linux-pci@vger.kernel.org
Subject: Re: [PATCH v5 27/32] PCI/P2PDMA: Convert to printbuf
Message-ID: <20220809020734.GA1260912@bhelgaas>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1b92761a-c166-b84f-8d77-a966d4c0ef9a@gmail.com>
Precedence: bulk

On Mon, Aug 08, 2022 at 02:42:03PM -0400, Kent Overstreet wrote:
> On 8/8/22 13:51, Bjorn Helgaas wrote:
> > I don't object to the patch, but it would be nice if the commit log
> > hinted at what the advantage is.  I assume it's faster/safer/better in
> > some way, but I have no idea what.
> 
> Printbufs have some additional features over seq_buf but they're not used
> here. The main one you might be interested in is heap allocation: that means
> no need to statically allocate buffers on the stack and no need to calculate
> the buffer size, printbufs will reallocate as necessary.

This doesn't tell me what the advantage of this patch is, since I
don't think *this* patch uses heap allocation.

As far as I can tell, this particular patch doesn't improve safety or
readability, so "convert X to Y even though we don't use any fancy Y
features" is a pointless message.

But if printbufs are better than seq_buf overall, and converting this
gets us closer to the goal of removing seq_buf completely, that's a
perfectly acceptable reason.  Just say that.

> I generally haven't been converting code to use that unless it's obvious
> that we're in a context where it's safe to allocate memory and can deal with
> allocation failures.
> 
> I notice that in calc_map_type_and_dist() you're using xa_store() which can
> fail, but you're not checking for that or returning errors properly :)
> perhaps a fix for that could also switch to using printbuf in
> heap-allocation mode.
> 
> > Also, cpu_show_meltdown() doesn't appear in this patch, so maybe
> > that's relevant to some other patch but not this one?
> 
> Whoops, was copying the commit message from another patch, yeah.n