Re: [PATCH net-next 10/11] sfc: implement vdpa config_ops for dma operations

[Jason Wang <jasowang@redhat.com>]

On Wed, Dec 7, 2022 at 10:57 PM Gautam Dawar <gautam.dawar@amd.com> wrote:
>
> Although sfc uses the platform IOMMU but it still
> implements the DMA config operations to deal with
> possible IOVA overlap with the MCDI DMA buffer and
> relocates the latter if such overlap is detected.
>
> Signed-off-by: Gautam Dawar <gautam.dawar@amd.com>
> ---
>  drivers/net/ethernet/sfc/ef100_vdpa.c     | 140 ++++++++++++++++++++++
>  drivers/net/ethernet/sfc/ef100_vdpa.h     |   3 +
>  drivers/net/ethernet/sfc/ef100_vdpa_ops.c | 111 +++++++++++++++++
>  drivers/net/ethernet/sfc/net_driver.h     |  12 ++
>  4 files changed, 266 insertions(+)
>
> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa.c b/drivers/net/ethernet/sfc/ef100_vdpa.c
> index b9368eb1acd5..16681d164fd1 100644
> --- a/drivers/net/ethernet/sfc/ef100_vdpa.c
> +++ b/drivers/net/ethernet/sfc/ef100_vdpa.c
> @@ -309,6 +309,140 @@ static int vdpa_update_domain(struct ef100_vdpa_nic *vdpa_nic)
>                                           vdpa_nic->geo_aper_end + 1, 0);
>  }
>
> +static int ef100_vdpa_alloc_buffer(struct efx_nic *efx, struct efx_buffer *buf)
> +{
> +       struct ef100_vdpa_nic *vdpa_nic = efx->vdpa_nic;
> +       struct device *dev = &vdpa_nic->vdpa_dev.dev;
> +       int rc;
> +
> +       buf->addr = kzalloc(buf->len, GFP_KERNEL);
> +       if (!buf->addr)
> +               return -ENOMEM;
> +
> +       rc = iommu_map(vdpa_nic->domain, buf->dma_addr,
> +                      virt_to_phys(buf->addr), buf->len,
> +                      IOMMU_READ | IOMMU_WRITE | IOMMU_CACHE);
> +       if (rc)
> +               dev_err(dev, "iommu_map failed, rc: %d\n", rc);
> +
> +       return rc;
> +}
> +
> +static void ef100_vdpa_free_buffer(struct ef100_vdpa_nic *vdpa_nic,
> +                                  struct efx_buffer *buf)
> +{
> +       struct device *dev = &vdpa_nic->vdpa_dev.dev;
> +       int rc;
> +
> +       rc = iommu_unmap(vdpa_nic->domain, buf->dma_addr, buf->len);
> +       if (rc < 0)
> +               dev_err(dev, "iommu_unmap failed, rc: %d\n", rc);
> +
> +       kfree(buf->addr);
> +}
> +
> +int ef100_setup_ef100_mcdi_buffer(struct ef100_vdpa_nic *vdpa_nic)
> +{
> +       struct efx_nic *efx = vdpa_nic->efx;
> +       struct ef100_nic_data *nic_data;
> +       struct efx_mcdi_iface *mcdi;
> +       struct efx_buffer mcdi_buf;
> +       enum efx_mcdi_mode mode;
> +       struct device *dev;
> +       int rc;
> +
> +       /* Switch to poll mode MCDI mode */
> +       nic_data = efx->nic_data;
> +       dev = &vdpa_nic->vdpa_dev.dev;
> +       mcdi = efx_mcdi(efx);
> +       mode = mcdi->mode;
> +       efx_mcdi_mode_poll(efx);
> +       efx_mcdi_flush_async(efx);
> +
> +       /* First, allocate the MCDI buffer for EF100 mode */
> +       rc = efx_nic_alloc_buffer(efx, &mcdi_buf,
> +                                 MCDI_BUF_LEN, GFP_KERNEL);
> +       if (rc) {
> +               dev_err(dev, "nic alloc buf failed, rc: %d\n", rc);
> +               goto restore_mode;
> +       }
> +
> +       /* unmap and free the vDPA MCDI buffer now */
> +       ef100_vdpa_free_buffer(vdpa_nic, &nic_data->mcdi_buf);
> +       memcpy(&nic_data->mcdi_buf, &mcdi_buf, sizeof(struct efx_buffer));
> +       efx->mcdi_buf_mode = EFX_BUF_MODE_EF100;
> +
> +restore_mode:
> +       if (mode == MCDI_MODE_EVENTS)
> +               efx_mcdi_mode_event(efx);
> +
> +       return rc;
> +}
> +
> +int ef100_setup_vdpa_mcdi_buffer(struct efx_nic *efx, u64 mcdi_iova)
> +{
> +       struct ef100_nic_data *nic_data = efx->nic_data;
> +       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
> +       enum efx_mcdi_mode mode = mcdi->mode;
> +       struct efx_buffer mcdi_buf;
> +       int rc;
> +
> +       efx_mcdi_mode_poll(efx);
> +       efx_mcdi_flush_async(efx);
> +
> +       /* First, prepare the MCDI buffer for vDPA mode */
> +       mcdi_buf.dma_addr = mcdi_iova;
> +       /* iommu_map requires page aligned memory */
> +       mcdi_buf.len = PAGE_ALIGN(MCDI_BUF_LEN);
> +       rc = ef100_vdpa_alloc_buffer(efx, &mcdi_buf);
> +       if (rc) {
> +               pci_err(efx->pci_dev, "alloc vdpa buf failed, rc: %d\n", rc);
> +               goto restore_mode;
> +       }
> +
> +       /* All set-up, free the EF100 MCDI buffer now */
> +       efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
> +       memcpy(&nic_data->mcdi_buf, &mcdi_buf, sizeof(struct efx_buffer));
> +       efx->mcdi_buf_mode = EFX_BUF_MODE_VDPA;
> +
> +restore_mode:
> +       if (mode == MCDI_MODE_EVENTS)
> +               efx_mcdi_mode_event(efx);
> +       return rc;
> +}
> +
> +int ef100_remap_vdpa_mcdi_buffer(struct efx_nic *efx, u64 mcdi_iova)
> +{
> +       struct ef100_nic_data *nic_data = efx->nic_data;
> +       struct ef100_vdpa_nic *vdpa_nic = efx->vdpa_nic;
> +       struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
> +       struct efx_buffer *mcdi_buf;
> +       int rc;
> +
> +       mcdi_buf = &nic_data->mcdi_buf;
> +       spin_lock_bh(&mcdi->iface_lock);
> +
> +       rc = iommu_unmap(vdpa_nic->domain, mcdi_buf->dma_addr, mcdi_buf->len);
> +       if (rc < 0) {
> +               pci_err(efx->pci_dev, "iommu_unmap failed, rc: %d\n", rc);
> +               goto out;
> +       }
> +
> +       rc = iommu_map(vdpa_nic->domain, mcdi_iova,
> +                      virt_to_phys(mcdi_buf->addr),
> +                      mcdi_buf->len,
> +                      IOMMU_READ | IOMMU_WRITE | IOMMU_CACHE);
> +       if (rc) {
> +               pci_err(efx->pci_dev, "iommu_map failed, rc: %d\n", rc);
> +               goto out;
> +       }
> +
> +       mcdi_buf->dma_addr = mcdi_iova;
> +out:
> +       spin_unlock_bh(&mcdi->iface_lock);
> +       return rc;
> +}
> +
>  static struct ef100_vdpa_nic *ef100_vdpa_create(struct efx_nic *efx,
>                                                 const char *dev_name,
>                                                 enum ef100_vdpa_class dev_type,
> @@ -391,6 +525,12 @@ static struct ef100_vdpa_nic *ef100_vdpa_create(struct efx_nic *efx,
>                 goto err_put_device;
>         }
>
> +       rc = ef100_setup_vdpa_mcdi_buffer(efx, EF100_VDPA_IOVA_BASE_ADDR);
> +       if (rc) {
> +               pci_err(efx->pci_dev, "realloc mcdi failed, err: %d\n", rc);
> +               goto err_put_device;
> +       }
> +
>         rc = get_net_config(vdpa_nic);
>         if (rc)
>                 goto err_put_device;
> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa.h b/drivers/net/ethernet/sfc/ef100_vdpa.h
> index c3c77029973d..f15d8739dcde 100644
> --- a/drivers/net/ethernet/sfc/ef100_vdpa.h
> +++ b/drivers/net/ethernet/sfc/ef100_vdpa.h
> @@ -202,6 +202,9 @@ int ef100_vdpa_add_filter(struct ef100_vdpa_nic *vdpa_nic,
>  int ef100_vdpa_irq_vectors_alloc(struct pci_dev *pci_dev, u16 nvqs);
>  void ef100_vdpa_irq_vectors_free(void *data);
>  int ef100_vdpa_reset(struct vdpa_device *vdev);
> +int ef100_setup_ef100_mcdi_buffer(struct ef100_vdpa_nic *vdpa_nic);
> +int ef100_setup_vdpa_mcdi_buffer(struct efx_nic *efx, u64 mcdi_iova);
> +int ef100_remap_vdpa_mcdi_buffer(struct efx_nic *efx, u64 mcdi_iova);
>
>  static inline bool efx_vdpa_is_little_endian(struct ef100_vdpa_nic *vdpa_nic)
>  {
> diff --git a/drivers/net/ethernet/sfc/ef100_vdpa_ops.c b/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
> index 8c198d949fdb..7c632f179bcf 100644
> --- a/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
> +++ b/drivers/net/ethernet/sfc/ef100_vdpa_ops.c
> @@ -12,6 +12,7 @@
>  #include "ef100_vdpa.h"
>  #include "ef100_iova.h"
>  #include "io.h"
> +#include "ef100_iova.h"
>  #include "mcdi_vdpa.h"
>
>  /* Get the queue's function-local index of the associated VI
> @@ -739,14 +740,121 @@ static void ef100_vdpa_set_config(struct vdpa_device *vdev, unsigned int offset,
>         }
>  }
>
> +static bool is_iova_overlap(u64 iova1, u64 size1, u64 iova2, u64 size2)
> +{
> +       return max(iova1, iova2) < min(iova1 + size1, iova2 + size2);
> +}
> +
> +static int ef100_vdpa_dma_map(struct vdpa_device *vdev,
> +                             unsigned int asid,
> +                             u64 iova, u64 size,
> +                             u64 pa, u32 perm, void *opaque)
> +{
> +       struct ef100_vdpa_nic *vdpa_nic;
> +       struct ef100_nic_data *nic_data;
> +       unsigned int mcdi_buf_len;
> +       dma_addr_t mcdi_buf_addr;
> +       u64 mcdi_iova = 0;
> +       int rc;
> +
> +       vdpa_nic = get_vdpa_nic(vdev);
> +       nic_data = vdpa_nic->efx->nic_data;
> +       mcdi_buf_addr = nic_data->mcdi_buf.dma_addr;
> +       mcdi_buf_len = nic_data->mcdi_buf.len;
> +
> +       /* Validate the iova range against geo aperture */
> +       if (iova < vdpa_nic->geo_aper_start ||
> +           ((iova + size - 1) > vdpa_nic->geo_aper_end)) {
> +               dev_err(&vdpa_nic->vdpa_dev.dev,
> +                       "%s: iova range (%llx, %llx) not within geo aperture\n",
> +                       __func__, iova, (iova + size));
> +               return -EINVAL;

It might be helpful to advertise this geo via get_iova_range().

Thanks