* [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG @ 2021-09-30 18:50 Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 1/3] NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker ` (3 more replies) 0 siblings, 4 replies; 12+ messages in thread From: Jean-Philippe Brucker @ 2021-09-30 18:50 UTC (permalink / raw) To: eric.auger; +Cc: Jean-Philippe Brucker, qemu-devel, mst Replace the VIRTIO_IOMMU_F_BYPASS feature with VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch global bypass on and off. Add a boot-bypass option, which defaults to 'on' to be in line with other vIOMMUs and to allow running firmware/bootloader that are unaware of the IOMMU. See the spec change for more rationale https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html Jean-Philippe Brucker (3): NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG virtio-iommu: Default to bypass during boot virtio-iommu: Support bypass domain include/hw/virtio/virtio-iommu.h | 1 + include/standard-headers/linux/virtio_iommu.h | 10 +++- hw/virtio/virtio-iommu.c | 60 ++++++++++++++++--- hw/virtio/trace-events | 4 +- 4 files changed, 64 insertions(+), 11 deletions(-) -- 2.33.0 ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH 1/3] NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG 2021-09-30 18:50 [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker @ 2021-09-30 18:50 ` Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 2/3] virtio-iommu: Default to bypass during boot Jean-Philippe Brucker ` (2 subsequent siblings) 3 siblings, 0 replies; 12+ messages in thread From: Jean-Philippe Brucker @ 2021-09-30 18:50 UTC (permalink / raw) To: eric.auger; +Cc: Jean-Philippe Brucker, qemu-devel, mst Pull VIRTIO_IOMMU_F_BYPASS_CONFIG changes from Linux (not upstream yet). Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> --- include/standard-headers/linux/virtio_iommu.h | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/include/standard-headers/linux/virtio_iommu.h b/include/standard-headers/linux/virtio_iommu.h index b9443b83a1..d14808e3fb 100644 --- a/include/standard-headers/linux/virtio_iommu.h +++ b/include/standard-headers/linux/virtio_iommu.h @@ -13,9 +13,10 @@ #define VIRTIO_IOMMU_F_INPUT_RANGE 0 #define VIRTIO_IOMMU_F_DOMAIN_RANGE 1 #define VIRTIO_IOMMU_F_MAP_UNMAP 2 -#define VIRTIO_IOMMU_F_BYPASS 3 +#define VIRTIO_IOMMU_F_BYPASS 3 /* Deprecated */ #define VIRTIO_IOMMU_F_PROBE 4 #define VIRTIO_IOMMU_F_MMIO 5 +#define VIRTIO_IOMMU_F_BYPASS_CONFIG 6 struct virtio_iommu_range_64 { uint64_t start; @@ -36,6 +37,8 @@ struct virtio_iommu_config { struct virtio_iommu_range_32 domain_range; /* Probe buffer size */ uint32_t probe_size; + uint8_t bypass; + uint8_t reserved[7]; }; /* Request types */ @@ -66,11 +69,14 @@ struct virtio_iommu_req_tail { uint8_t reserved[3]; }; +#define VIRTIO_IOMMU_ATTACH_F_BYPASS (1 << 0) + struct virtio_iommu_req_attach { struct virtio_iommu_req_head head; uint32_t domain; uint32_t endpoint; - uint8_t reserved[8]; + uint32_t flags; + uint8_t reserved[4]; struct virtio_iommu_req_tail tail; }; -- 2.33.0 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH 2/3] virtio-iommu: Default to bypass during boot 2021-09-30 18:50 [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 1/3] NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker @ 2021-09-30 18:50 ` Jean-Philippe Brucker 2021-10-06 13:04 ` Eric Auger 2021-09-30 18:50 ` [PATCH 3/3] virtio-iommu: Support bypass domain Jean-Philippe Brucker 2022-01-11 9:02 ` [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Eric Auger 3 siblings, 1 reply; 12+ messages in thread From: Jean-Philippe Brucker @ 2021-09-30 18:50 UTC (permalink / raw) To: eric.auger; +Cc: Jean-Philippe Brucker, qemu-devel, mst Currently the virtio-iommu device must be programmed before it allows DMA from any PCI device. This can make the VM entirely unusable when a virtio-iommu driver isn't present, for example in a bootloader that loads the OS from storage. Similarly to the other vIOMMU implementations, default to DMA bypassing the IOMMU during boot. Add a "boot-bypass" option that lets users change this behavior. Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> --- include/hw/virtio/virtio-iommu.h | 1 + hw/virtio/virtio-iommu.c | 28 +++++++++++++++++++++++----- hw/virtio/trace-events | 4 ++-- 3 files changed, 26 insertions(+), 7 deletions(-) diff --git a/include/hw/virtio/virtio-iommu.h b/include/hw/virtio/virtio-iommu.h index 273e35c04b..4c66989ca4 100644 --- a/include/hw/virtio/virtio-iommu.h +++ b/include/hw/virtio/virtio-iommu.h @@ -58,6 +58,7 @@ struct VirtIOIOMMU { GTree *domains; QemuMutex mutex; GTree *endpoints; + bool boot_bypass; }; #endif diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c index 1b23e8e18c..82edeaa101 100644 --- a/hw/virtio/virtio-iommu.c +++ b/hw/virtio/virtio-iommu.c @@ -728,8 +728,7 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr, .perm = IOMMU_NONE, }; - bypass_allowed = virtio_vdev_has_feature(&s->parent_obj, - VIRTIO_IOMMU_F_BYPASS); + bypass_allowed = s->config.bypass; sid = virtio_iommu_get_bdf(sdev); @@ -828,7 +827,8 @@ static void virtio_iommu_get_config(VirtIODevice *vdev, uint8_t *config_data) config->input_range.start, config->input_range.end, config->domain_range.end, - config->probe_size); + config->probe_size, + config->bypass); memcpy(config_data, &dev->config, sizeof(struct virtio_iommu_config)); } @@ -836,13 +836,29 @@ static void virtio_iommu_set_config(VirtIODevice *vdev, const uint8_t *config_data) { struct virtio_iommu_config config; + VirtIOIOMMU *dev = VIRTIO_IOMMU(vdev); memcpy(&config, config_data, sizeof(struct virtio_iommu_config)); + + if (config.bypass != dev->config.bypass) { + if (!virtio_vdev_has_feature(vdev, VIRTIO_IOMMU_F_BYPASS_CONFIG)) { + virtio_error(vdev, "cannot set config.bypass"); + return; + } + if (config.bypass != 0 && config.bypass != 1) { + warn_report("invalid config.bypass value '%d'", config.bypass); + dev->config.bypass = 0; + return; + } + dev->config.bypass = config.bypass; + } + trace_virtio_iommu_set_config(config.page_size_mask, config.input_range.start, config.input_range.end, config.domain_range.end, - config.probe_size); + config.probe_size, + config.bypass); } static uint64_t virtio_iommu_get_features(VirtIODevice *vdev, uint64_t f, @@ -986,6 +1002,7 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp) s->config.input_range.end = -1UL; s->config.domain_range.end = 32; s->config.probe_size = VIOMMU_PROBE_SIZE; + s->config.bypass = s->boot_bypass; virtio_add_feature(&s->features, VIRTIO_RING_F_EVENT_IDX); virtio_add_feature(&s->features, VIRTIO_RING_F_INDIRECT_DESC); @@ -993,9 +1010,9 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp) virtio_add_feature(&s->features, VIRTIO_IOMMU_F_INPUT_RANGE); virtio_add_feature(&s->features, VIRTIO_IOMMU_F_DOMAIN_RANGE); virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MAP_UNMAP); - virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS); virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MMIO); virtio_add_feature(&s->features, VIRTIO_IOMMU_F_PROBE); + virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS_CONFIG); qemu_mutex_init(&s->mutex); @@ -1169,6 +1186,7 @@ static const VMStateDescription vmstate_virtio_iommu = { static Property virtio_iommu_properties[] = { DEFINE_PROP_LINK("primary-bus", VirtIOIOMMU, primary_bus, "PCI", PCIBus *), + DEFINE_PROP_BOOL("boot-bypass", VirtIOIOMMU, boot_bypass, true), DEFINE_PROP_END_OF_LIST(), }; diff --git a/hw/virtio/trace-events b/hw/virtio/trace-events index 8ed19e9d0c..6bc3821ba3 100644 --- a/hw/virtio/trace-events +++ b/hw/virtio/trace-events @@ -90,8 +90,8 @@ virtio_mmio_setting_irq(int level) "virtio_mmio setting IRQ %d" virtio_iommu_device_reset(void) "reset!" virtio_iommu_get_features(uint64_t features) "device supports features=0x%"PRIx64 virtio_iommu_device_status(uint8_t status) "driver status = %d" -virtio_iommu_get_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_range=%d probe_size=0x%x" -virtio_iommu_set_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_bits=%d probe_size=0x%x" +virtio_iommu_get_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size, uint8_t bypass) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_range=%d probe_size=0x%x bypass=0x%x" +virtio_iommu_set_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size, uint8_t bypass) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_bits=%d probe_size=0x%x bypass=0x%x" virtio_iommu_attach(uint32_t domain_id, uint32_t ep_id) "domain=%d endpoint=%d" virtio_iommu_detach(uint32_t domain_id, uint32_t ep_id) "domain=%d endpoint=%d" virtio_iommu_map(uint32_t domain_id, uint64_t virt_start, uint64_t virt_end, uint64_t phys_start, uint32_t flags) "domain=%d virt_start=0x%"PRIx64" virt_end=0x%"PRIx64 " phys_start=0x%"PRIx64" flags=%d" -- 2.33.0 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH 2/3] virtio-iommu: Default to bypass during boot 2021-09-30 18:50 ` [PATCH 2/3] virtio-iommu: Default to bypass during boot Jean-Philippe Brucker @ 2021-10-06 13:04 ` Eric Auger 2021-10-08 10:21 ` Jean-Philippe Brucker 0 siblings, 1 reply; 12+ messages in thread From: Eric Auger @ 2021-10-06 13:04 UTC (permalink / raw) To: Jean-Philippe Brucker; +Cc: qemu-devel, mst Hi jean, On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > Currently the virtio-iommu device must be programmed before it allows > DMA from any PCI device. This can make the VM entirely unusable when a > virtio-iommu driver isn't present, for example in a bootloader that > loads the OS from storage. > > Similarly to the other vIOMMU implementations, default to DMA bypassing > the IOMMU during boot. Add a "boot-bypass" option that lets users change > this behavior. s/option/property set to true by default. Also mention this deprecates the legacy VIRTIO_IOMMU_F_BYPASS feature > > Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> > --- > include/hw/virtio/virtio-iommu.h | 1 + > hw/virtio/virtio-iommu.c | 28 +++++++++++++++++++++++----- > hw/virtio/trace-events | 4 ++-- > 3 files changed, 26 insertions(+), 7 deletions(-) > > diff --git a/include/hw/virtio/virtio-iommu.h b/include/hw/virtio/virtio-iommu.h > index 273e35c04b..4c66989ca4 100644 > --- a/include/hw/virtio/virtio-iommu.h > +++ b/include/hw/virtio/virtio-iommu.h > @@ -58,6 +58,7 @@ struct VirtIOIOMMU { > GTree *domains; > QemuMutex mutex; > GTree *endpoints; > + bool boot_bypass; > }; > > #endif > diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c > index 1b23e8e18c..82edeaa101 100644 > --- a/hw/virtio/virtio-iommu.c > +++ b/hw/virtio/virtio-iommu.c > @@ -728,8 +728,7 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr, > .perm = IOMMU_NONE, > }; > > - bypass_allowed = virtio_vdev_has_feature(&s->parent_obj, > - VIRTIO_IOMMU_F_BYPASS); > + bypass_allowed = s->config.bypass; > > sid = virtio_iommu_get_bdf(sdev); > > @@ -828,7 +827,8 @@ static void virtio_iommu_get_config(VirtIODevice *vdev, uint8_t *config_data) > config->input_range.start, > config->input_range.end, > config->domain_range.end, > - config->probe_size); > + config->probe_size, > + config->bypass); > memcpy(config_data, &dev->config, sizeof(struct virtio_iommu_config)); > } > > @@ -836,13 +836,29 @@ static void virtio_iommu_set_config(VirtIODevice *vdev, > const uint8_t *config_data) > { > struct virtio_iommu_config config; > + VirtIOIOMMU *dev = VIRTIO_IOMMU(vdev); > > memcpy(&config, config_data, sizeof(struct virtio_iommu_config)); > + > + if (config.bypass != dev->config.bypass) { > + if (!virtio_vdev_has_feature(vdev, VIRTIO_IOMMU_F_BYPASS_CONFIG)) { > + virtio_error(vdev, "cannot set config.bypass"); > + return; > + } > + if (config.bypass != 0 && config.bypass != 1) { > + warn_report("invalid config.bypass value '%d'", config.bypass); > + dev->config.bypass = 0; Shouldn't we leave the old value instead. The spec proposal says it takes 0 though. > + return; > + } > + dev->config.bypass = config.bypass; > + } > + > trace_virtio_iommu_set_config(config.page_size_mask, > config.input_range.start, > config.input_range.end, > config.domain_range.end, > - config.probe_size); > + config.probe_size, > + config.bypass); > } > > static uint64_t virtio_iommu_get_features(VirtIODevice *vdev, uint64_t f, > @@ -986,6 +1002,7 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp) > s->config.input_range.end = -1UL; > s->config.domain_range.end = 32; > s->config.probe_size = VIOMMU_PROBE_SIZE; > + s->config.bypass = s->boot_bypass; > > virtio_add_feature(&s->features, VIRTIO_RING_F_EVENT_IDX); > virtio_add_feature(&s->features, VIRTIO_RING_F_INDIRECT_DESC); > @@ -993,9 +1010,9 @@ static void virtio_iommu_device_realize(DeviceState *dev, Error **errp) > virtio_add_feature(&s->features, VIRTIO_IOMMU_F_INPUT_RANGE); > virtio_add_feature(&s->features, VIRTIO_IOMMU_F_DOMAIN_RANGE); > virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MAP_UNMAP); > - virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS); > virtio_add_feature(&s->features, VIRTIO_IOMMU_F_MMIO); > virtio_add_feature(&s->features, VIRTIO_IOMMU_F_PROBE); > + virtio_add_feature(&s->features, VIRTIO_IOMMU_F_BYPASS_CONFIG); > > qemu_mutex_init(&s->mutex); > > @@ -1169,6 +1186,7 @@ static const VMStateDescription vmstate_virtio_iommu = { > > static Property virtio_iommu_properties[] = { > DEFINE_PROP_LINK("primary-bus", VirtIOIOMMU, primary_bus, "PCI", PCIBus *), > + DEFINE_PROP_BOOL("boot-bypass", VirtIOIOMMU, boot_bypass, true), > DEFINE_PROP_END_OF_LIST(), > }; > > diff --git a/hw/virtio/trace-events b/hw/virtio/trace-events > index 8ed19e9d0c..6bc3821ba3 100644 > --- a/hw/virtio/trace-events > +++ b/hw/virtio/trace-events > @@ -90,8 +90,8 @@ virtio_mmio_setting_irq(int level) "virtio_mmio setting IRQ %d" > virtio_iommu_device_reset(void) "reset!" > virtio_iommu_get_features(uint64_t features) "device supports features=0x%"PRIx64 > virtio_iommu_device_status(uint8_t status) "driver status = %d" > -virtio_iommu_get_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_range=%d probe_size=0x%x" > -virtio_iommu_set_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_bits=%d probe_size=0x%x" > +virtio_iommu_get_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size, uint8_t bypass) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_range=%d probe_size=0x%x bypass=0x%x" > +virtio_iommu_set_config(uint64_t page_size_mask, uint64_t start, uint64_t end, uint32_t domain_range, uint32_t probe_size, uint8_t bypass) "page_size_mask=0x%"PRIx64" start=0x%"PRIx64" end=0x%"PRIx64" domain_bits=%d probe_size=0x%x bypass=0x%x" > virtio_iommu_attach(uint32_t domain_id, uint32_t ep_id) "domain=%d endpoint=%d" > virtio_iommu_detach(uint32_t domain_id, uint32_t ep_id) "domain=%d endpoint=%d" > virtio_iommu_map(uint32_t domain_id, uint64_t virt_start, uint64_t virt_end, uint64_t phys_start, uint32_t flags) "domain=%d virt_start=0x%"PRIx64" virt_end=0x%"PRIx64 " phys_start=0x%"PRIx64" flags=%d" Thanks Eric ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 2/3] virtio-iommu: Default to bypass during boot 2021-10-06 13:04 ` Eric Auger @ 2021-10-08 10:21 ` Jean-Philippe Brucker 0 siblings, 0 replies; 12+ messages in thread From: Jean-Philippe Brucker @ 2021-10-08 10:21 UTC (permalink / raw) To: Eric Auger; +Cc: qemu-devel, mst On Wed, Oct 06, 2021 at 03:04:15PM +0200, Eric Auger wrote: > > + if (config.bypass != dev->config.bypass) { > > + if (!virtio_vdev_has_feature(vdev, VIRTIO_IOMMU_F_BYPASS_CONFIG)) { > > + virtio_error(vdev, "cannot set config.bypass"); > > + return; > > + } > > + if (config.bypass != 0 && config.bypass != 1) { > > + warn_report("invalid config.bypass value '%d'", config.bypass); > > + dev->config.bypass = 0; > Shouldn't we leave the old value instead. The spec proposal says it > takes 0 though. Yes keeping the old value is better, I'll change the spec. Writing a different value is a driver bug since new values for the bypass field will be keyed off a new feature bit. Thanks, Jean ^ permalink raw reply [flat|nested] 12+ messages in thread
* [PATCH 3/3] virtio-iommu: Support bypass domain 2021-09-30 18:50 [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 1/3] NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 2/3] virtio-iommu: Default to bypass during boot Jean-Philippe Brucker @ 2021-09-30 18:50 ` Jean-Philippe Brucker 2021-10-06 13:10 ` Eric Auger 2022-01-11 9:02 ` [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Eric Auger 3 siblings, 1 reply; 12+ messages in thread From: Jean-Philippe Brucker @ 2021-09-30 18:50 UTC (permalink / raw) To: eric.auger; +Cc: Jean-Philippe Brucker, qemu-devel, mst The driver can create a bypass domain by passing the VIRTIO_IOMMU_ATTACH_F_BYPASS flag on the ATTACH request. Bypass domains perform slightly better than domains with identity mappings since they skip translation. Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> --- hw/virtio/virtio-iommu.c | 32 ++++++++++++++++++++++++++++++-- 1 file changed, 30 insertions(+), 2 deletions(-) diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c index 82edeaa101..4f0207a3eb 100644 --- a/hw/virtio/virtio-iommu.c +++ b/hw/virtio/virtio-iommu.c @@ -42,6 +42,7 @@ typedef struct VirtIOIOMMUDomain { uint32_t id; + bool bypass; GTree *mappings; QLIST_HEAD(, VirtIOIOMMUEndpoint) endpoint_list; } VirtIOIOMMUDomain; @@ -257,12 +258,16 @@ static void virtio_iommu_put_endpoint(gpointer data) } static VirtIOIOMMUDomain *virtio_iommu_get_domain(VirtIOIOMMU *s, - uint32_t domain_id) + uint32_t domain_id, + bool bypass) { VirtIOIOMMUDomain *domain; domain = g_tree_lookup(s->domains, GUINT_TO_POINTER(domain_id)); if (domain) { + if (domain->bypass != bypass) { + return NULL; + } return domain; } domain = g_malloc0(sizeof(*domain)); @@ -270,6 +275,7 @@ static VirtIOIOMMUDomain *virtio_iommu_get_domain(VirtIOIOMMU *s, domain->mappings = g_tree_new_full((GCompareDataFunc)interval_cmp, NULL, (GDestroyNotify)g_free, (GDestroyNotify)g_free); + domain->bypass = bypass; g_tree_insert(s->domains, GUINT_TO_POINTER(domain_id), domain); QLIST_INIT(&domain->endpoint_list); trace_virtio_iommu_get_domain(domain_id); @@ -333,11 +339,16 @@ static int virtio_iommu_attach(VirtIOIOMMU *s, { uint32_t domain_id = le32_to_cpu(req->domain); uint32_t ep_id = le32_to_cpu(req->endpoint); + uint32_t flags = le32_to_cpu(req->flags); VirtIOIOMMUDomain *domain; VirtIOIOMMUEndpoint *ep; trace_virtio_iommu_attach(domain_id, ep_id); + if (flags & ~VIRTIO_IOMMU_ATTACH_F_BYPASS) { + return VIRTIO_IOMMU_S_INVAL; + } + ep = virtio_iommu_get_endpoint(s, ep_id); if (!ep) { return VIRTIO_IOMMU_S_NOENT; @@ -355,7 +366,12 @@ static int virtio_iommu_attach(VirtIOIOMMU *s, } } - domain = virtio_iommu_get_domain(s, domain_id); + domain = virtio_iommu_get_domain(s, domain_id, + flags & VIRTIO_IOMMU_ATTACH_F_BYPASS); + if (!domain) { + /* Incompatible flags */ + return VIRTIO_IOMMU_S_INVAL; + } QLIST_INSERT_HEAD(&domain->endpoint_list, ep, next); ep->domain = domain; @@ -418,6 +434,10 @@ static int virtio_iommu_map(VirtIOIOMMU *s, return VIRTIO_IOMMU_S_NOENT; } + if (domain->bypass) { + return VIRTIO_IOMMU_S_INVAL; + } + interval = g_malloc0(sizeof(*interval)); interval->low = virt_start; @@ -463,6 +483,11 @@ static int virtio_iommu_unmap(VirtIOIOMMU *s, if (!domain) { return VIRTIO_IOMMU_S_NOENT; } + + if (domain->bypass) { + return VIRTIO_IOMMU_S_INVAL; + } + interval.low = virt_start; interval.high = virt_end; @@ -779,6 +804,9 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr, entry.perm = flag; } goto unlock; + } else if (ep->domain->bypass) { + entry.perm = flag; + goto unlock; } found = g_tree_lookup_extended(ep->domain->mappings, (gpointer)(&interval), -- 2.33.0 ^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH 3/3] virtio-iommu: Support bypass domain 2021-09-30 18:50 ` [PATCH 3/3] virtio-iommu: Support bypass domain Jean-Philippe Brucker @ 2021-10-06 13:10 ` Eric Auger 0 siblings, 0 replies; 12+ messages in thread From: Eric Auger @ 2021-10-06 13:10 UTC (permalink / raw) To: Jean-Philippe Brucker; +Cc: qemu-devel, mst Hi jean, On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > The driver can create a bypass domain by passing the > VIRTIO_IOMMU_ATTACH_F_BYPASS flag on the ATTACH request. Bypass domains > perform slightly better than domains with identity mappings since they > skip translation. > > Signed-off-by: Jean-Philippe Brucker <jean-philippe@linaro.org> > --- > hw/virtio/virtio-iommu.c | 32 ++++++++++++++++++++++++++++++-- > 1 file changed, 30 insertions(+), 2 deletions(-) > > diff --git a/hw/virtio/virtio-iommu.c b/hw/virtio/virtio-iommu.c > index 82edeaa101..4f0207a3eb 100644 > --- a/hw/virtio/virtio-iommu.c > +++ b/hw/virtio/virtio-iommu.c > @@ -42,6 +42,7 @@ > > typedef struct VirtIOIOMMUDomain { > uint32_t id; > + bool bypass; > GTree *mappings; > QLIST_HEAD(, VirtIOIOMMUEndpoint) endpoint_list; > } VirtIOIOMMUDomain; > @@ -257,12 +258,16 @@ static void virtio_iommu_put_endpoint(gpointer data) > } > > static VirtIOIOMMUDomain *virtio_iommu_get_domain(VirtIOIOMMU *s, > - uint32_t domain_id) > + uint32_t domain_id, > + bool bypass) > { > VirtIOIOMMUDomain *domain; > > domain = g_tree_lookup(s->domains, GUINT_TO_POINTER(domain_id)); > if (domain) { > + if (domain->bypass != bypass) { > + return NULL; > + } > return domain; > } > domain = g_malloc0(sizeof(*domain)); > @@ -270,6 +275,7 @@ static VirtIOIOMMUDomain *virtio_iommu_get_domain(VirtIOIOMMU *s, > domain->mappings = g_tree_new_full((GCompareDataFunc)interval_cmp, > NULL, (GDestroyNotify)g_free, > (GDestroyNotify)g_free); > + domain->bypass = bypass; > g_tree_insert(s->domains, GUINT_TO_POINTER(domain_id), domain); > QLIST_INIT(&domain->endpoint_list); > trace_virtio_iommu_get_domain(domain_id); > @@ -333,11 +339,16 @@ static int virtio_iommu_attach(VirtIOIOMMU *s, > { > uint32_t domain_id = le32_to_cpu(req->domain); > uint32_t ep_id = le32_to_cpu(req->endpoint); > + uint32_t flags = le32_to_cpu(req->flags); > VirtIOIOMMUDomain *domain; > VirtIOIOMMUEndpoint *ep; > > trace_virtio_iommu_attach(domain_id, ep_id); > > + if (flags & ~VIRTIO_IOMMU_ATTACH_F_BYPASS) { > + return VIRTIO_IOMMU_S_INVAL; > + } > + > ep = virtio_iommu_get_endpoint(s, ep_id); > if (!ep) { > return VIRTIO_IOMMU_S_NOENT; > @@ -355,7 +366,12 @@ static int virtio_iommu_attach(VirtIOIOMMU *s, > } > } > > - domain = virtio_iommu_get_domain(s, domain_id); > + domain = virtio_iommu_get_domain(s, domain_id, > + flags & VIRTIO_IOMMU_ATTACH_F_BYPASS); > + if (!domain) { > + /* Incompatible flags */ Incompatible bypass flag > + return VIRTIO_IOMMU_S_INVAL; > + } > QLIST_INSERT_HEAD(&domain->endpoint_list, ep, next); > > ep->domain = domain; > @@ -418,6 +434,10 @@ static int virtio_iommu_map(VirtIOIOMMU *s, > return VIRTIO_IOMMU_S_NOENT; > } > > + if (domain->bypass) { > + return VIRTIO_IOMMU_S_INVAL; > + } > + > interval = g_malloc0(sizeof(*interval)); > > interval->low = virt_start; > @@ -463,6 +483,11 @@ static int virtio_iommu_unmap(VirtIOIOMMU *s, > if (!domain) { > return VIRTIO_IOMMU_S_NOENT; > } > + > + if (domain->bypass) { > + return VIRTIO_IOMMU_S_INVAL; > + } > + > interval.low = virt_start; > interval.high = virt_end; > > @@ -779,6 +804,9 @@ static IOMMUTLBEntry virtio_iommu_translate(IOMMUMemoryRegion *mr, hwaddr addr, > entry.perm = flag; > } > goto unlock; > + } else if (ep->domain->bypass) { > + entry.perm = flag; > + goto unlock; > } > > found = g_tree_lookup_extended(ep->domain->mappings, (gpointer)(&interval), Otherwise looks good to me. Eric ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG 2021-09-30 18:50 [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker ` (2 preceding siblings ...) 2021-09-30 18:50 ` [PATCH 3/3] virtio-iommu: Support bypass domain Jean-Philippe Brucker @ 2022-01-11 9:02 ` Eric Auger 2022-01-11 10:13 ` Jean-Philippe Brucker 2022-01-11 15:40 ` Michael S. Tsirkin 3 siblings, 2 replies; 12+ messages in thread From: Eric Auger @ 2022-01-11 9:02 UTC (permalink / raw) To: Jean-Philippe Brucker; +Cc: qemu-devel, mst Hi Jean, Michael, On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > Replace the VIRTIO_IOMMU_F_BYPASS feature with > VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch > global bypass on and off. > > Add a boot-bypass option, which defaults to 'on' to be in line with > other vIOMMUs and to allow running firmware/bootloader that are unaware > of the IOMMU. > > See the spec change for more rationale > https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html I guess the kernel bits should be merged in 5.17? Thanks Eric > > Jean-Philippe Brucker (3): > NOMERGE: virtio-iommu: Add definitions for > VIRTIO_IOMMU_F_BYPASS_CONFIG > virtio-iommu: Default to bypass during boot > virtio-iommu: Support bypass domain > > include/hw/virtio/virtio-iommu.h | 1 + > include/standard-headers/linux/virtio_iommu.h | 10 +++- > hw/virtio/virtio-iommu.c | 60 ++++++++++++++++--- > hw/virtio/trace-events | 4 +- > 4 files changed, 64 insertions(+), 11 deletions(-) > ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG 2022-01-11 9:02 ` [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Eric Auger @ 2022-01-11 10:13 ` Jean-Philippe Brucker 2022-01-11 10:33 ` Eric Auger 2022-01-11 15:40 ` Michael S. Tsirkin 1 sibling, 1 reply; 12+ messages in thread From: Jean-Philippe Brucker @ 2022-01-11 10:13 UTC (permalink / raw) To: Eric Auger; +Cc: qemu-devel, mst Hi Eric, On Tue, Jan 11, 2022 at 10:02:12AM +0100, Eric Auger wrote: > Hi Jean, Michael, > > On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > > Replace the VIRTIO_IOMMU_F_BYPASS feature with > > VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch > > global bypass on and off. > > > > Add a boot-bypass option, which defaults to 'on' to be in line with > > other vIOMMUs and to allow running firmware/bootloader that are unaware > > of the IOMMU. > > > > See the spec change for more rationale > > https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html > > I guess the kernel bits should be merged in 5.17? Yes, they should. I can resend at 5.17-rc1 along with a commit updating the linux headers. Or is there a specific process for synchronizing the headers? Looking at git log it looks like scripts/update-linux-headers.sh is run by whomever needs new UAPI features. Thanks, Jean ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG 2022-01-11 10:13 ` Jean-Philippe Brucker @ 2022-01-11 10:33 ` Eric Auger 0 siblings, 0 replies; 12+ messages in thread From: Eric Auger @ 2022-01-11 10:33 UTC (permalink / raw) To: Jean-Philippe Brucker; +Cc: qemu-devel, mst Hi Jean, On 1/11/22 11:13 AM, Jean-Philippe Brucker wrote: > Hi Eric, > > On Tue, Jan 11, 2022 at 10:02:12AM +0100, Eric Auger wrote: >> Hi Jean, Michael, >> >> On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: >>> Replace the VIRTIO_IOMMU_F_BYPASS feature with >>> VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch >>> global bypass on and off. >>> >>> Add a boot-bypass option, which defaults to 'on' to be in line with >>> other vIOMMUs and to allow running firmware/bootloader that are unaware >>> of the IOMMU. >>> >>> See the spec change for more rationale >>> https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html >> I guess the kernel bits should be merged in 5.17? > Yes, they should. I can resend at 5.17-rc1 along with a commit updating > the linux headers. Or is there a specific process for synchronizing the > headers? Looking at git log it looks like scripts/update-linux-headers.sh > is run by whomever needs new UAPI features. yes you need to generate the header update patch using scripts/update-linux-headers.sh once the kernel pieces have landed Thanks Eric > > Thanks, > Jean > ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG 2022-01-11 9:02 ` [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Eric Auger 2022-01-11 10:13 ` Jean-Philippe Brucker @ 2022-01-11 15:40 ` Michael S. Tsirkin 2022-01-11 16:24 ` Jean-Philippe Brucker 1 sibling, 1 reply; 12+ messages in thread From: Michael S. Tsirkin @ 2022-01-11 15:40 UTC (permalink / raw) To: Eric Auger; +Cc: Jean-Philippe Brucker, qemu-devel On Tue, Jan 11, 2022 at 10:02:12AM +0100, Eric Auger wrote: > Hi Jean, Michael, > > On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > > Replace the VIRTIO_IOMMU_F_BYPASS feature with > > VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch > > global bypass on and off. > > > > Add a boot-bypass option, which defaults to 'on' to be in line with > > other vIOMMUs and to allow running firmware/bootloader that are unaware > > of the IOMMU. > > > > See the spec change for more rationale > > https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html > > I guess the kernel bits should be merged in 5.17? > > Thanks > > Eric They are in fact in my tree and set to go into 5.16. They've been in linux-next for a whole cycle now. But if you feel I'm rushing things, pls let me know. Also, pls let me know whether my tree actually works well for you! > > > > Jean-Philippe Brucker (3): > > NOMERGE: virtio-iommu: Add definitions for > > VIRTIO_IOMMU_F_BYPASS_CONFIG > > virtio-iommu: Default to bypass during boot > > virtio-iommu: Support bypass domain > > > > include/hw/virtio/virtio-iommu.h | 1 + > > include/standard-headers/linux/virtio_iommu.h | 10 +++- > > hw/virtio/virtio-iommu.c | 60 ++++++++++++++++--- > > hw/virtio/trace-events | 4 +- > > 4 files changed, 64 insertions(+), 11 deletions(-) > > ^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG 2022-01-11 15:40 ` Michael S. Tsirkin @ 2022-01-11 16:24 ` Jean-Philippe Brucker 0 siblings, 0 replies; 12+ messages in thread From: Jean-Philippe Brucker @ 2022-01-11 16:24 UTC (permalink / raw) To: Michael S. Tsirkin; +Cc: Eric Auger, qemu-devel Hi Michael, On Tue, Jan 11, 2022 at 10:40:28AM -0500, Michael S. Tsirkin wrote: > On Tue, Jan 11, 2022 at 10:02:12AM +0100, Eric Auger wrote: > > Hi Jean, Michael, > > > > On 9/30/21 8:50 PM, Jean-Philippe Brucker wrote: > > > Replace the VIRTIO_IOMMU_F_BYPASS feature with > > > VIRTIO_IOMMU_F_BYPASS_CONFIG, which enables a config space bit to switch > > > global bypass on and off. > > > > > > Add a boot-bypass option, which defaults to 'on' to be in line with > > > other vIOMMUs and to allow running firmware/bootloader that are unaware > > > of the IOMMU. > > > > > > See the spec change for more rationale > > > https://lists.oasis-open.org/archives/virtio-dev/202109/msg00137.html > > > > I guess the kernel bits should be merged in 5.17? > > > > Thanks > > > > Eric > > They are in fact in my tree and set to go into 5.16. > They've been in linux-next for a whole cycle now. > But if you feel I'm rushing things, pls let me know. > Also, pls let me know whether my tree actually works well for you! Thanks, unfortunately I just noticed that those patches are from an older version of the series, the latest being v3 https://lore.kernel.org/linux-iommu/20211201173323.1045819-1-jean-philippe@linaro.org/ Since Joerg picked the latest one in his tree [1], I was wondering if you could drop those patches? 51a4c54f35ae iommu/virtio: Support identity-mapped domains 523e55a406aa iommu/virtio: Pass end address to viommu_add_mapping() 97301219dfb2 iommu/virtio: Sort reserved regions 720552613526 iommu/virtio: Support bypass domains 01444b9c772f iommu/virtio: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG Thanks, Jean [1] https://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu.git/ ^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2022-01-11 16:26 UTC | newest] Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2021-09-30 18:50 [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 1/3] NOMERGE: virtio-iommu: Add definitions for VIRTIO_IOMMU_F_BYPASS_CONFIG Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 2/3] virtio-iommu: Default to bypass during boot Jean-Philippe Brucker 2021-10-06 13:04 ` Eric Auger 2021-10-08 10:21 ` Jean-Philippe Brucker 2021-09-30 18:50 ` [PATCH 3/3] virtio-iommu: Support bypass domain Jean-Philippe Brucker 2021-10-06 13:10 ` Eric Auger 2022-01-11 9:02 ` [PATCH 0/3] virtio-iommu: Support VIRTIO_IOMMU_F_BYPASS_CONFIG Eric Auger 2022-01-11 10:13 ` Jean-Philippe Brucker 2022-01-11 10:33 ` Eric Auger 2022-01-11 15:40 ` Michael S. Tsirkin 2022-01-11 16:24 ` Jean-Philippe Brucker
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).