From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:59607) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1e1ZE0-00053l-MT for qemu-devel@nongnu.org; Mon, 09 Oct 2017 10:40:13 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1e1ZDu-0006df-He for qemu-devel@nongnu.org; Mon, 09 Oct 2017 10:40:08 -0400 Received: from mail-wm0-x234.google.com ([2a00:1450:400c:c09::234]:55307) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1e1ZDu-0006d6-7U for qemu-devel@nongnu.org; Mon, 09 Oct 2017 10:40:02 -0400 Received: by mail-wm0-x234.google.com with SMTP id u138so24208753wmu.4 for ; Mon, 09 Oct 2017 07:40:02 -0700 (PDT) MIME-Version: 1.0 In-Reply-To: <1504286483-23327-3-git-send-email-eric.auger@redhat.com> References: <1504286483-23327-1-git-send-email-eric.auger@redhat.com> <1504286483-23327-3-git-send-email-eric.auger@redhat.com> From: Peter Maydell Date: Mon, 9 Oct 2017 15:39:40 +0100 Message-ID: Content-Type: text/plain; charset="UTF-8" Subject: Re: [Qemu-devel] [PATCH v7 02/20] hw/arm/smmu-common: IOMMU memory region and address space setup List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Eric Auger Cc: eric.auger.pro@gmail.com, qemu-arm , QEMU Developers , Prem Mallappa , Alex Williamson , Andrew Jones , Christoffer Dall , Radha.Chintakuntla@cavium.com, Sunil.Goutham@cavium.com, Radha Mohan , Trey Cain , Bharat Bhushan , Tomasz Nowicki , "Michael S. Tsirkin" , Will Deacon , jean-philippe.brucker@arm.com, robin.murphy@arm.com, Peter Xu , "Edgar E. Iglesias" , wtownsen@redhat.com On 1 September 2017 at 18:21, Eric Auger wrote: > We enumerate all the PCI devices attached to the SMMU and > initialize an associated IOMMU memory region and address space. > This happens on SMMU base instance init. > > Those info are stored in SMMUDevice objects. The devices are > grouped according to the PCIBus they belong to. A hash table > indexed by the PCIBus poinet is used. Also an array indexed by > the bus number allows to find the list of SMMUDevices. > > Signed-off-by: Eric Auger > --- > hw/arm/smmu-common.c | 89 ++++++++++++++++++++++++++++++++++++++++++++ > include/hw/arm/smmu-common.h | 6 +++ > 2 files changed, 95 insertions(+) > > diff --git a/hw/arm/smmu-common.c b/hw/arm/smmu-common.c > index 56608f1..3e67992 100644 > --- a/hw/arm/smmu-common.c > +++ b/hw/arm/smmu-common.c > @@ -30,8 +30,97 @@ > #include "qemu/error-report.h" > #include "hw/arm/smmu-common.h" > > +/******************/ > +/* Infrastructure */ > +/******************/ Minor thing, but we don't really need this kind of fancy comment formatting. > +static inline gboolean smmu_uint64_equal(gconstpointer v1, gconstpointer v2) > +{ > + return *((const uint64_t *)v1) == *((const uint64_t *)v2); > +} > + > +static inline guint smmu_uint64_hash(gconstpointer v) > +{ > + return (guint)*(const uint64_t *)v; > +} > + > +SMMUPciBus *smmu_find_as_from_bus_num(SMMUState *s, uint8_t bus_num) > +{ > + SMMUPciBus *smmu_pci_bus = s->smmu_as_by_bus_num[bus_num]; > + > + if (!smmu_pci_bus) { > + GHashTableIter iter; > + > + g_hash_table_iter_init(&iter, s->smmu_as_by_busptr); > + while (g_hash_table_iter_next(&iter, NULL, (void **)&smmu_pci_bus)) { > + if (pci_bus_num(smmu_pci_bus->bus) == bus_num) { > + s->smmu_as_by_bus_num[bus_num] = smmu_pci_bus; > + return smmu_pci_bus; > + } > + } > + } > + return smmu_pci_bus; > +} > + > +static AddressSpace *smmu_find_add_as(PCIBus *bus, void *opaque, int devfn) > +{ > + SMMUState *s = opaque; > + uintptr_t key = (uintptr_t)bus; > + SMMUPciBus *sbus = g_hash_table_lookup(s->smmu_as_by_busptr, &key); > + SMMUDevice *sdev; > + > + if (!sbus) { > + uintptr_t *new_key = g_malloc(sizeof(*new_key)); > + > + *new_key = (uintptr_t)bus; > + sbus = g_malloc0(sizeof(SMMUPciBus) + > + sizeof(SMMUDevice *) * SMMU_PCI_DEVFN_MAX); > + sbus->bus = bus; > + g_hash_table_insert(s->smmu_as_by_busptr, new_key, sbus); Why do we allocate memory containing a uintptr_t which we set to be the (integer value of the) pointer to the bus, and then use the pointer to that uintptr_t as the key, when we could just use the pointer to the bus as the key ? That would save you having a specialist equal function, hash function and having to free the keys. > + } > + > + sdev = sbus->pbdev[devfn]; > + if (!sdev) { > + char *name = g_strdup_printf("%s-%d-%d", > + s->mrtypename, > + pci_bus_num(bus), devfn); > + sdev = sbus->pbdev[devfn] = g_malloc0(sizeof(SMMUDevice)); g_new0() is slightly stylistically preferable for this kind of thing. > + > + sdev->smmu = s; > + sdev->bus = bus; > + sdev->devfn = devfn; > + > + memory_region_init_iommu(&sdev->iommu, sizeof(sdev->iommu), > + s->mrtypename, > + OBJECT(s), name, 1ULL << 48); What is this 1ULL << 48 ? Is it intended to be the input address size, intermediate address size or output address size? It's not clear to me that hardcoded 1 << 48 is right in any of those cases... > + address_space_init(&sdev->as, > + MEMORY_REGION(&sdev->iommu), name); > + } > + > + return &sdev->as; > +} > + > +static void smmu_init_iommu_as(SMMUState *s) > +{ > + PCIBus *pcibus = pci_find_primary_bus(); This looks odd. I would expect the board model to be instantiating and wiring up the SMMU somehow so that it is in the path of whatever PCI bus it is sitting in front of. It shouldn't need to look for the PCI bus like this, which prevents modelling a system where there are two PCI buses each of which has its own SMMU. > + > + if (pcibus) { > + pci_setup_iommu(pcibus, smmu_find_add_as, s); > + } else { > + error_report("No PCI bus, SMMU is not registered"); > + } > +} > + > static void smmu_base_instance_init(Object *obj) > { > + SMMUState *s = SMMU_SYS_DEV(obj); > + > + memset(s->smmu_as_by_bus_num, 0, sizeof(s->smmu_as_by_bus_num)); Instance init doesn't need to clear the data structure. > + > + s->smmu_as_by_busptr = g_hash_table_new_full(smmu_uint64_hash, > + smmu_uint64_equal, > + g_free, g_free); > + smmu_init_iommu_as(s); > } > > static void smmu_base_class_init(ObjectClass *klass, void *data) > diff --git a/include/hw/arm/smmu-common.h b/include/hw/arm/smmu-common.h > index 38cd18f..20f3fe6 100644 > --- a/include/hw/arm/smmu-common.h > +++ b/include/hw/arm/smmu-common.h > @@ -105,4 +105,10 @@ typedef struct { > #define SMMU_DEVICE_CLASS(klass) \ > OBJECT_CLASS_CHECK(SMMUBaseClass, (klass), TYPE_SMMU_DEV_BASE) > > +SMMUPciBus *smmu_find_as_from_bus_num(SMMUState *s, uint8_t bus_num); > + > +static inline uint16_t smmu_get_sid(SMMUDevice *sdev) > +{ > + return ((pci_bus_num(sdev->bus) & 0xff) << 8) | sdev->devfn; > +} > #endif /* HW_ARM_SMMU_COMMON */ > -- > 2.5.5 thanks -- PMM