From mboxrd@z Thu Jan 1 00:00:00 1970 From: Alan Ott Subject: Re: [PATCH beta 1] 0/3] Fix race conditions in mrf24j40 interrupts Date: Thu, 23 May 2013 15:33:50 -0400 Message-ID: <519E6F1E.8000200@signal11.us> References: <1369188080-8904-1-git-send-email-alan@signal11.us> <519C278F.5030809@signal11.us> <519DB8E6.4020709@signal11.us> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, linux-kernel@vger.kernel.org, linux-zigbee-devel@lists.sourceforge.net To: David Hauweele Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org On 05/23/2013 01:54 PM, David Hauweele wrote: > 2013/5/23 Alan Ott : >> On 5/22/13 4:32 PM, David Hauweele wrote: >>> >>> I cannot use level-triggered interrupts with GPIO on the RPi, so I >>> cannot test this specific patch. >> >> >> Is there another interrupt line you can tie into which does support >> level-trigger interrupts (INT0 or something)? > > According to the datasheet it should be possible but the bcm2708 port > does not support it. I've been told that we shouldn't use level-triggered > interrupts in the first place. Who is "we," and why shouldn't we use level-triggered interrupts? The CPU needs to detect whatever type of interrupt the attached hardware generates. > >> >> >>> However I agree with the idea of level-triggered interrupts, that >>> would fix all major problems related to missed interrupts. >>> >>> Beside this I'm running a ping -f since more than two hours now and it >>> seems to work well. >>> >> >> So that surprises me. I thought level-trigger interrupts were the thing that >> would fix this problem, and if you're not running with that patch, you just >> have the INIT_COMPLETION() fix (which you said didn't fix your issue) and >> the threaded interrupts patch, which I was fairly sure I had determined >> wasn't fixing any actual race-condition-related problems. > > I should have been more clear about this. I've tested [PATCH 1/3] > which fixes the race condition with tx_complete. That is the > INIT_COMPLETION() fix. But it is still possible to miss an interrupt, > perhaps it just took longer this time. I ran the test again today and > it failed after 30 minutes. > > I did not test [PATCH 2/3], that is the threaded IRQ. Instead I > removed interrupt enable/disable from the IRQ handler and the > workqueue. Without this the driver would fail within seconds of a ping > -f. Without what? What do you mean by "without this?" Without the enable/disable, or without the change that removes the enable/disable? > Have you observed this too ? Perhaps this problem is specific to > the bcm2708 port. > What I observe right now is that it seems to work great (ping -f for 6.5 hours) when using the three patches in this patch set on a BeagleBone. > >> >> I'm glad, but surprised that you're no longer seeing issues. >> >> Alan. >> >> >>> >>> 2013/5/22 Alan Ott : >>>> >>>> On 05/21/2013 10:01 PM, Alan Ott wrote: >>>>> >>>>> David Hauweele noticed that the mrf24j40 would hang arbitrarily after >>>>> some >>>>> period of heavy traffic. Two race conditions were discovered, and the >>>>> driver was changed to use threaded interrupts, since the enable/disable >>>>> of >>>>> interrupts in the driver has recently been a lighning rod whenever >>>>> issues >>>>> arise related to interrupts (costing engineering time), and since >>>>> threaded >>>>> interrupts are the right way to do it. >>>>> >>>>> Alan Ott (3): >>>>> mrf24j40: Move INIT_COMPLETION() to before packet transmission >>>>> mrf24j40: Use threaded IRQ handler >>>>> mrf24j40: Use level-triggered interrupts >>>>> >>>>> drivers/net/ieee802154/mrf24j40.c | 31 +++++++++---------------------- >>>>> 1 file changed, 9 insertions(+), 22 deletions(-) >>>> >>>> >>>> I forgot to add, I ran ping -f both ways all afternoon (6.5 hours), and >>>> it seems solid. >>>> >>