From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:54872) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1dp9ZL-0005xB-N1 for qemu-devel@nongnu.org; Tue, 05 Sep 2017 04:50:56 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1dp9ZG-000546-Kb for qemu-devel@nongnu.org; Tue, 05 Sep 2017 04:50:51 -0400 Received: from mx1.redhat.com ([209.132.183.28]:53902) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1dp9ZG-00053g-BQ for qemu-devel@nongnu.org; Tue, 05 Sep 2017 04:50:46 -0400 Date: Tue, 5 Sep 2017 10:50:41 +0200 From: Cornelia Huck Message-ID: <20170905105041.3409c8cd.cohuck@redhat.com> In-Reply-To: <0923f97e-4f69-0bf3-b9ab-8e5f69a56bbb@linux.vnet.ibm.com> References: <1504239778-29893-1-git-send-email-zyimin@linux.vnet.ibm.com> <1504239778-29893-2-git-send-email-zyimin@linux.vnet.ibm.com> <20170905102928.6b23a28a.cohuck@redhat.com> <0923f97e-4f69-0bf3-b9ab-8e5f69a56bbb@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH v2 1/3] s390x/pci: remove idx from msix msg data List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Yi Min Zhao Cc: qemu-devel@nongnu.org, borntraeger@de.ibm.com, pasic@linux.vnet.ibm.com, pmorel@linux.vnet.ibm.com, agraf@suse.de, richard.henderson@linaro.org On Tue, 5 Sep 2017 16:44:37 +0800 Yi Min Zhao wrote: > =E5=9C=A8 2017/9/5 =E4=B8=8B=E5=8D=884:29, Cornelia Huck =E5=86=99=E9=81= =93: > > On Fri, 1 Sep 2017 06:22:56 +0200 > > Yi Min Zhao wrote: > > =20 > >> PCIDevice pointer has been a parameter of kvm_arch_fixup_msi_route(). > >> So we don't need to store zpci idx in msix message data to find out the > >> specific zpci device. Instead, we could use pci device id to find its > >> corresponding zpci device. > >> > >> Signed-off-by: Yi Min Zhao > >> --- > >> hw/s390x/s390-pci-bus.c | 16 +++++----------- > >> hw/s390x/s390-pci-bus.h | 2 ++ > >> hw/s390x/s390-pci-inst.c | 24 ------------------------ > >> hw/s390x/s390-pci-stub.c | 6 ++++++ > >> target/s390x/kvm.c | 7 +++++-- > >> 5 files changed, 18 insertions(+), 37 deletions(-) > >> > >> diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c > >> index 0a31a4ae88..bd8a3e1e1c 100644 > >> --- a/hw/s390x/s390-pci-bus.c > >> +++ b/hw/s390x/s390-pci-bus.c > >> @@ -199,8 +199,8 @@ static S390PCIBusDevice *s390_pci_find_dev_by_uid(= S390pciState *s, uint16_t uid) > >> return NULL; > >> } > >> =20 > >> -static S390PCIBusDevice *s390_pci_find_dev_by_target(S390pciState *s, > >> - const char *targ= et) > >> +S390PCIBusDevice *s390_pci_find_dev_by_target(S390pciState *s, > >> + const char *target) > >> { > >> S390PCIBusDevice *pbdev; > >> =20 > >> @@ -465,19 +465,13 @@ static void s390_msi_ctrl_write(void *opaque, hw= addr addr, uint64_t data, > >> unsigned int size) > >> { > >> S390PCIBusDevice *pbdev =3D opaque; > >> - uint32_t idx =3D data >> ZPCI_MSI_VEC_BITS; > >> uint32_t vec =3D data & ZPCI_MSI_VEC_MASK; > >> uint64_t ind_bit; > >> uint32_t sum_bit; > >> - uint32_t e =3D 0; > >> =20 > >> - DPRINTF("write_msix data 0x%" PRIx64 " idx %d vec 0x%x\n", data, = idx, vec); > >> - > >> - if (!pbdev) { > >> - e |=3D (vec << ERR_EVENT_MVN_OFFSET); > >> - s390_pci_generate_error_event(ERR_EVENT_NOMSI, idx, 0, addr, = e); > >> - return; > >> - } > >> + assert(pbdev); =20 > > I'm wondering whether you could/should generate an error event here. > > The one above probably won't work (as it seems to take idx as a > > parameter), but is this really 'this must not happen, we messed up in > > our code'? (Probably yes, but I want to be sure.) =20 > I think this must not happen. One a pci device is plugged into zPCI bus. > We would assign a new memory region with zpci device as opaque > for its msix. So if s390_msi_ctrl_write() is called, there must be a write > operation to a pci device's msix ctrl memory region which must has zpci > device as a opaque. The construct is one-msi-mr-per-pci-device. This makes sense. > > =20 > >> + DPRINTF("write_msix data 0x%" PRIx64 " idx %d vec 0x%x\n", data, > >> + pbdev->idx, vec); > >> =20 > >> if (pbdev->state !=3D ZPCI_FS_ENABLED) { > >> return; > >> diff --git a/hw/s390x/s390-pci-stub.c b/hw/s390x/s390-pci-stub.c > >> index 7a642d376c..e501e1b9ea 100644 > >> --- a/hw/s390x/s390-pci-stub.c > >> +++ b/hw/s390x/s390-pci-stub.c > >> @@ -74,3 +74,9 @@ S390PCIBusDevice *s390_pci_find_dev_by_idx(S390pciSt= ate *s, uint32_t idx) > >> { > >> return NULL; > >> } =20 > > Please remove s390_pci_find_dev_by_idx() from the stubs file, as it is > > not used outside of the conditionally-built pci code anymore. =20 > I'm confused. s390_pci_find_dev_by_idx() can be called in=20 > kvm_arch_fixup_msi_route(). > And kvm_arch_fixup_msi_route() can be called by kvm_irqchip_add_msi_route= (). > As the code, I think s390_pci_find_dev_by_idx() might be called. Could=20 > you please > explain more? But this patch replaces this with s390_pci_find_dev_by_target(), no? > > =20 > >> + > >> +S390PCIBusDevice *s390_pci_find_dev_by_target(S390pciState *s, > >> + const char *target) > >> +{ > >> + return NULL; > >> +} > >> diff --git a/target/s390x/kvm.c b/target/s390x/kvm.c > >> index 1338c29528..3d490c5e4b 100644 > >> --- a/target/s390x/kvm.c > >> +++ b/target/s390x/kvm.c > >> @@ -2533,10 +2533,13 @@ int kvm_arch_fixup_msi_route(struct kvm_irq_ro= uting_entry *route, > >> uint64_t address, uint32_t data, PCIDev= ice *dev) > >> { > >> S390PCIBusDevice *pbdev; > >> - uint32_t idx =3D data >> ZPCI_MSI_VEC_BITS; > >> uint32_t vec =3D data & ZPCI_MSI_VEC_MASK; > >> =20 > >> - pbdev =3D s390_pci_find_dev_by_idx(s390_get_phb(), idx); > >> + if (!dev) { > >> + return -ENODEV; =20 > > Can this actually happen? =20 > I think this cannot happen. But I'm afraid that I miss something. > So I added this to avoid NULL pointer. But from the code and > my test, there has not been NULL pointer happened. I'm wondering if that is in the same category as the instance I commented on above. Do you want to log something? > > =20 > >> + } > >> + > >> + pbdev =3D s390_pci_find_dev_by_target(s390_get_phb(), DEVICE(dev)= ->id); > >> if (!pbdev) { > >> DPRINTF("add_msi_route no dev\n"); > >> return -ENODEV; =20 > > =20 >=20