From mboxrd@z Thu Jan 1 00:00:00 1970 From: w@1wt.eu (Willy Tarreau) Date: Thu, 23 Jan 2014 12:09:14 +0100 Subject: [BUG] FL1009: xHCI host not responding to stop endpoint command. In-Reply-To: <87sisfjeba.fsf@natisbad.org> References: <87fvon2khn.fsf@natisbad.org> <87zjmvau23.fsf@nemi.mork.no> <20140117205432.GB5618@xanatos> <87vbxhrmea.fsf@natisbad.org> <20140121211741.GA12657@xanatos> <87bnz364h0.fsf@natisbad.org> <20140122222641.GD29184@titan.lakedaemon.net> <87mwin4ozf.fsf@natisbad.org> <20140122235620.GC20359@xanatos> <87sisfjeba.fsf@natisbad.org> Message-ID: <20140123110914.GA22913@1wt.eu> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Thu, Jan 23, 2014 at 09:24:41AM +0100, Arnaud Ebalard wrote: > Hi Sarah, > > Sarah Sharp writes: > > >> > Can you pinpoint the commit that introduced the regression? > >> > >> f5182b4155b9d686c5540a6822486400e34ddd98 "xhci: Disable MSI for some Fresco Logic hosts." > >> > >> Technically, this is not per se the commit which introduced the > >> regression but the one that *partially* fixed it by introducing the XHCI > >> quirk to skip MSI enabling for Fresco Logic chips. The thing is it > >> should have included the FL1009 in the targets. Sarah, can you confirm > >> this? > > > > I don't know if it should have included FL1009, it was just a guess, > > based on the fact that the 0x1000 and 0x1400 devices did need MSI > > disabled. I can attempt to ask the Fresco Logic folks I know, but I'm > > not sure if/when I'll get a response back. > > > > That still doesn't necessarily rule out MSI issues in the Marvell PCI > > host controller code. Can you attach another PCI device with MSI > > support under the host and see if it works? > > The various Armada-based devices I have are NAS which do not have PCIe > slots to plug additional devices (everything is soldered). I don't know > which device Thomas used for its tests. Just in case, I also added Willy > in CC: who have various boards and may also have done more test with > additional PCIe devices and CONFIG_PCI_MSI enabled on 3.13 kernel. I've been running an intel i350 dual-port NIC (igb driver) supporting MSI on the mirabox, and it used to work in 3.10+many of the patches coming from the Free-electrons team. Some recent changes to the PCI code introduced a regression preventing this driver from correctly registering an MSI interrupt, and I did not have enough time to investigate it deep enough to fix it. That said, I know how to hack it to work again, so if it can be of any use, I can run a test on the mirabox (armada370) and on the XPGP board (armadaXP). Willy