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;
Message-ID: <d5d0cc2e-79e1-3248-0f55-8f1afd21f926@gmail.com>
Date:   Wed, 8 Jun 2022 19:24:02 -0400
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.10.0
Subject: Re: [PATCH v3 29/33] PCI/P2PDMA: Convert to printbuf
Content-Language: en-US
To:     Bjorn Helgaas <helgaas@kernel.org>
Cc:     linux-kernel@vger.kernel.org, pmladek@suse.com,
        rostedt@goodmis.org, linux-pci@vger.kernel.org,
        Logan Gunthorpe <logang@deltatee.com>
References: <20220608211146.GA422296@bhelgaas>
From:   Kent Overstreet <kent.overstreet@gmail.com>
In-Reply-To: <20220608211146.GA422296@bhelgaas>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 6/8/22 17:11, Bjorn Helgaas wrote:
> [+cc Logan, maintainer of p2pdma.c]
> 
> On Sat, Jun 04, 2022 at 03:30:38PM -0400, Kent Overstreet wrote:
>> This converts from seq_buf to printbuf. We're using printbuf in external
>> buffer mode, so it's a direct conversion, aside from some trivial
>> refactoring in cpu_show_meltdown() to make the code more consistent.
> 
> cpu_show_meltdown() doesn't appear in p2pdma.c.  Leftover from another
> patch?  Maybe from 27/33 ("powerpc: Convert to printbuf")?
> 
> I'm not opposed to this, but it would be nice to say what the benefit
> is.  How is printbuf better than seq_buf?  It's not obvious from the
> patch how this is better/safer/shorter/etc.
> 
> Even the cover letter [1] is not very clear about the benefit.  Yes, I
> see it has something to do with improving buffer management, and I
> know from experience that's a pain.  Concrete examples of typical
> printbuf usage and bugs that printbufs avoid would be helpful.

Take a look at the vsprintf.c conversion if you want to see big 
improvements. Also, %pf() is another thing that's going to enable a lot 
more improvements.

> I guess "external buffer mode" means we use an existing buffer (on the
> stack in this case) instead of allocating a buffer from the heap [2]?
> And we do that for performance (i.e., we know the max size) and to
> avoid sleeping to alloc?

I did it that way because I didn't want to touch unrelated code more 
than was necessary - just doing a direct conversion. Heap allocation 
would probably make sense here, but it's not my code.

> Are there any other printf-type things in drivers/pci that
> could/should be converted?  Is this basically a seq_buf replacement,
> so we can find everything with "git grep seq_buf drivers/pci/"?

All seq_buf uses are fully converted to printbuf in this patch series, 
and seq_buf is removed.

There is a lot of non seq_buf code that still uses raw char * pointers 
and lengths that should be converted to printbuf, but this patch series 
already does a lot of that and I'm not trying to boil the oceans today... :)