From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:47002) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Zew1V-0005y5-KV for qemu-devel@nongnu.org; Wed, 23 Sep 2015 22:12:38 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Zew1S-0006Lp-D4 for qemu-devel@nongnu.org; Wed, 23 Sep 2015 22:12:37 -0400 Message-ID: <1443060750.23936.517.camel@redhat.com> From: Alex Williamson Date: Wed, 23 Sep 2015 20:12:30 -0600 In-Reply-To: <20150924011127.GK15944@voom.fritz.box> References: <1442647117-2726-1-git-send-email-david@gibson.dropbear.id.au> <1442647117-2726-2-git-send-email-david@gibson.dropbear.id.au> <1443029309.23936.483.camel@redhat.com> <20150924011127.GK15944@voom.fritz.box> Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [RFC PATCH 01/14] vfio: Start adding VFIO/EEH interface List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: David Gibson Cc: lvivier@redhat.com, thuth@redhat.com, mdroth@linux.vnet.ibm.com, aik@ozlabs.ru, qemu-devel@nongnu.org, gwshan@linux.vnet.ibm.com, qemu-ppc@nongnu.org On Thu, 2015-09-24 at 11:11 +1000, David Gibson wrote: > On Wed, Sep 23, 2015 at 11:28:29AM -0600, Alex Williamson wrote: > > On Sat, 2015-09-19 at 17:18 +1000, David Gibson wrote: > > > At present the code handling IBM's Enhanced Error Handling (EEH) interface > > > on VFIO devices operates by bypassing the usual VFIO logic with > > > vfio_container_ioctl(). That's a poorly designed interface with unclear > > > semantics about exactly what can be operated on. > > > > > > As a first step to cleaning that up, start creating a new VFIO interface > > > for EEH operations. Because EEH operates in units of a "Partitionable > > > Endpoint" (PE) - a group of devices that can't be mutually isolated), it > > > needs to expose host PEs (that is, IOMMU groups) to the guest. This means > > > EEH needs VFIO concepts exposed that other VFIO users don't. > > > > > > At present all EEH ioctl()s in use operate conceptually on a single PE and > > > take no parameters except the opcode itself. So, expose a vfio_eeh_op() > > > function to apply a single no-argument operation to a VFIO group. > > > > > > At present the kernel VFIO/EEH interface is broken, because it assumes > > > there is only one VFIO group per container, which is no longer always the > > > case. So, add logic to detect this case and warn. > > > > > > Signed-off-by: David Gibson > > > --- > > > hw/vfio/Makefile.objs | 1 + > > > hw/vfio/eeh.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++ > > > include/hw/vfio/vfio-eeh.h | 42 ++++++++++++++++++++++++++++++ > > > 3 files changed, 107 insertions(+) > > > create mode 100644 hw/vfio/eeh.c > > > create mode 100644 include/hw/vfio/vfio-eeh.h > > > > > > diff --git a/hw/vfio/Makefile.objs b/hw/vfio/Makefile.objs > > > index d540c9d..384c702 100644 > > > --- a/hw/vfio/Makefile.objs > > > +++ b/hw/vfio/Makefile.objs > > > @@ -3,4 +3,5 @@ obj-$(CONFIG_SOFTMMU) += common.o > > > obj-$(CONFIG_PCI) += pci.o > > > obj-$(CONFIG_SOFTMMU) += platform.o > > > obj-$(CONFIG_SOFTMMU) += calxeda-xgmac.o > > > +obj-$(CONFIG_PSERIES) += eeh.o > > > endif > > > diff --git a/hw/vfio/eeh.c b/hw/vfio/eeh.c > > > new file mode 100644 > > > index 0000000..35bd06c > > > --- /dev/null > > > +++ b/hw/vfio/eeh.c > > > @@ -0,0 +1,64 @@ > > > +/* > > > + * EEH (IBM Enhanced Error Handling) functions for VFIO devices > > > + * > > > + * Copyright Red Hat, Inc. 2015 > > > + * > > > + * Authors: > > > + * David Gibson > > > + * > > > + * This program is free software: you can redistribute it and/or > > > + * modify it under the terms of the GNU General Public License as > > > + * published by the Free Software Foundation, either version 2 of the > > > + * License, or (at your option) any later version. > > > + * > > > + * This program is distributed in the hope that it will be useful, > > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > > > + * GNU General Public License for more details. > > > + * > > > + * You should have received a copy of the GNU General Public License > > > + * along with this program. If not, see . > > > + * > > > + * Based on earlier EEH implementations: > > > + * Copyright (c) 2011-2014 Alexey Kardashevskiy, IBM Corporation. > > > + */ > > > +#include > > > +#include > > > + > > > +#include "qemu/error-report.h" > > > + > > > +#include "hw/vfio/vfio-common.h" > > > +#include "hw/vfio/vfio-eeh.h" > > > + > > > +int vfio_eeh_op(VFIOGroup *group, uint32_t op) > > > +{ > > > + VFIOContainer *container = group->container; > > > + struct vfio_eeh_pe_op pe_op = { > > > + .argsz = sizeof(pe_op), > > > + .op = op, > > > + }; > > > + int ret; > > > + > > > + if (!container) { > > > + error_report("vfio/eeh: EEH_PE_OP 0x%x called on unattached group %d", > > > + op, group->groupid); > > > + return -ENODEV; > > > + } > > > + > > > + /* A broken kernel interface means EEH operations can't work > > > + * correctly if there are multiple groups in a container */ > > > + if ((QLIST_FIRST(&container->group_list) != group) > > > + || QLIST_NEXT(group, container_next)) { > > > + error_report("vfio/eeh: EEH_PE_OP 0x%x called on container" > > > + " with multiple groups", op); > > > + return -ENOSPC; > > > > -EINVAL really seems more appropriate > > So, I agree that ENOSPC is a dubious choice, but EINVAL is definitely > wrong here. > > Broad as it is, EINVAL should always indicate that the caller has > supplied some sort of bad parameter. In this case the parameters are > just fine, it's just that the kernel is broken so we can't handle that > case. > > Perhaps EBUSY? Since there isn't an ESHOULDWORKBUTDOESNT or EDEVBROKEN. The caller has supplied a bad parameter, a group that doesn't match the existing group in the container. If you really object, EPERM? EACCES? > > > + } > > > + > > > + ret = ioctl(container->fd, VFIO_EEH_PE_OP, &pe_op); > > > + if (ret < 0) { > > > + error_report("vfio/eeh: EEH_PE_OP 0x%x failed on group %d: %m", > > > + op, group->groupid); > > > + } > > > + > > > + return ret; > > > > Would -errno be more interesting in the failure case? > > Oh, yes. Too much kernel work, I'm used to things returning the error > code. >