Return-path: Received: from userp1040.oracle.com ([156.151.31.81]:41219 "EHLO userp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751600AbaBRTXG (ORCPT ); Tue, 18 Feb 2014 14:23:06 -0500 Date: Tue, 18 Feb 2014 22:22:39 +0300 From: Dan Carpenter To: Andrea Merello Cc: John Linville , Linux Wireless List , Larry Finger , Bernhard Schiffner , Huqiu Liu Subject: Re: [PATCH 3/7] rtl818x: check for pci_map_single() success when initializing RX ring Message-ID: <20140218192239.GP26776@mwanda> (sfid-20140218_202310_574246_1A750995) References: <1392685846-10116-1-git-send-email-andrea.merello@gmail.com> <1392685846-10116-4-git-send-email-andrea.merello@gmail.com> <20140218093135.GJ26776@mwanda> <20140218182714.GO26776@mwanda> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: Sender: linux-wireless-owner@vger.kernel.org List-ID: On Tue, Feb 18, 2014 at 08:06:51PM +0100, Andrea Merello wrote: > Thank for pointing out this. Yes, I probably exaggerated trying hard > to allocate memory. > > If your point is about if pci_map_single may or not ever fail in real > life (that is was I thought you meant in your first mail), then I > think that it's worth to keep the check anyway (I think it could > happen). Please keep the check. > > If your point is that it can be not so much useful to try and retry > hard the allocation (I must admit I'm not even sure that after freeing > the skb the kernel will not likely re-allocate the same one at the > next try), then I will keep the error check, but I can remove code > that retries allocation (failing at the first error as you suggested). Yes. Please remove the retry code unless there is some real life justification for it where you have seen partial allocations before. > > BTW consider that the allocation is done in ieee80211 start callback, > that is called every time the interface is brought down/up, and not > once at the begining. > You are right. I didn't realize that. > On one hand I cared avoiding wasting memory when the interface is not up. > On the other hand I had thought also to move memory allocation in > initialization, exactly to increase success probability as you pointed > out.. > > I chosen the first option because after the interface is brought up, > the RX ant TX processes will start to perform a lot of other skb > allocations and dma maps. > If we assume, as you pointed out, they will likely ALL fail or > succeeds, not just few of them, then probably even if we allocated > some memory at init time, we have gained not advantage. > > Do you agree ? Yes. regards, dan carpenter