Date: Sat, 15 Feb 2003 22:33:42 -0800 (PST) From: Nate Lawson <nate@root.org> To: "Matthew N. Dodd" <winter@jurai.net> Cc: current@FreeBSD.ORG, wpaul@FreeBSD.ORG Subject: Re: sys/pci/if* fixes Message-ID: <Pine.BSF.4.21.0302152224320.43778-100000@root.org> In-Reply-To: <20030216011126.D73971@sasami.jurai.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 16 Feb 2003, Matthew N. Dodd wrote: > On Sat, 15 Feb 2003, Nate Lawson wrote: > > The 'if' is a matter of coding style. Which is preferred: multiple > > returns from the same function (one which is only reachable from a goto) > > or one linear path with an 'if' to see if this is an error exit? > > In this case a single return is harder to read. I'm fine with going back to that style if someone else will corroborate this. One other advantage to the single return is that if a routine returns an error but this error is not checked, it will still be caught with my code whereas attach will return 0 unless code explicitly requests a failure. This is not a theoretical bug, many of the calls to bus_dma_tag_create() did not check for errors returned and would still return 0. However, this was never triggered due to that only happening on a very low memory condition. > > See the previous thread on this regarding my original dc(4) changes. > > It is bogus to create a lock with mtx_init and then immediately lock it > > for the entire attach routine just for the intention of ensuring > > exclusive access to the device. > > The problem is that the drivers aren't properly turning off interrrupts > before setting up the interrupt handler. This is on -current, turning off interrupts should be a very rare event. On -stable, I plan to only make the bugfixes in place since there is no locking to deal with. -Nate To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.21.0302152224320.43778-100000>