[2/3] virtio-iommu: Default to bypass during boot

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(-)

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

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