From: Ariel Levkovich <lariel@xxxxxxxxxxxx>
This patch adds the mlx5_ib provider implementation for the device
memory allocatoin API.
It implements the ib_device callbacks for allocation and deallocation
operations as well as a new mmap command support which allows mapping
an allocated device memory to a VMA.
The change also adds reporting of device memory maximum size and
alignment parameters reported in device capabilities.
The allocation/deallocation operation are making use of the mlx5_core
memic callbacks to allocated and free the actual memory on the device.
Signed-off-by: Ariel Levkovich <lariel@xxxxxxxxxxxx>
Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
---
drivers/infiniband/hw/mlx5/main.c | 137 ++++++++++++++++++++++++++++++-
drivers/infiniband/hw/mlx5/mlx5_ib.h | 18 +++-
include/uapi/rdma/mlx5-abi.h | 1 +
include/uapi/rdma/mlx5_user_ioctl_cmds.h | 6 +-
4 files changed, 159 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 4a62f3ba9b91..c90a7805c9f3 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -896,6 +896,11 @@ static int mlx5_ib_query_device(struct ib_device *ibdev,
props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
}
+ if (MLX5_CAP_DEV_MEM(mdev, memic)) {
+ props->max_dm_size =
+ MLX5_CAP_DEV_MEM(mdev, max_memic_size);
+ }
+
if (mlx5_get_flow_namespace(dev->mdev, MLX5_FLOW_NAMESPACE_BYPASS))
props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
@@ -2019,6 +2024,8 @@ static inline char *mmap_cmd2str(enum mlx5_ib_mmap_cmd cmd)
return "best effort WC";
case MLX5_IB_MMAP_NC_PAGE:
return "NC";
+ case MLX5_IB_MMAP_DEVICE_MEM:
+ return "Device Memory";
default:
return NULL;
}
@@ -2177,6 +2184,34 @@ static int uar_mmap(struct mlx5_ib_dev *dev, enum mlx5_ib_mmap_cmd cmd,
return err;
}
+static int dm_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+ struct mlx5_ib_ucontext *mctx = to_mucontext(context);
+ struct mlx5_ib_dev *dev = to_mdev(context->device);
+ u16 page_idx = get_extended_index(vma->vm_pgoff);
+ size_t map_size = vma->vm_end - vma->vm_start;
+ u32 npages = map_size >> PAGE_SHIFT;
+ phys_addr_t pfn;
+ pgprot_t prot;
+
+ if (find_next_zero_bit(mctx->dm_pages, page_idx + npages, page_idx) !=
+ page_idx + npages)
+ return -EINVAL;
+
+ pfn = ((pci_resource_start(dev->mdev->pdev, 0) +
+ MLX5_CAP64_DEV_MEM(dev->mdev, memic_bar_start_addr)) >>
+ PAGE_SHIFT) +
+ page_idx;
+ prot = pgprot_writecombine(vma->vm_page_prot);
+ vma->vm_page_prot = prot;
+
+ if (io_remap_pfn_range(vma, vma->vm_start, pfn, map_size,
+ vma->vm_page_prot))
+ return -EAGAIN;
+
+ return mlx5_ib_set_vma_data(vma, mctx);
+}
+
static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
{
struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
@@ -2221,6 +2256,9 @@ static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vm
case MLX5_IB_MMAP_CLOCK_INFO:
return mlx5_ib_mmap_clock_info_page(dev, vma, context);
+ case MLX5_IB_MMAP_DEVICE_MEM:
+ return dm_mmap(ibcontext, vma);
+
default:
return -EINVAL;
}
@@ -2228,6 +2266,85 @@ static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vm
return 0;
}
+struct ib_dm *mlx5_ib_alloc_dm(struct ib_device *ibdev,
+ struct ib_ucontext *context,
+ struct ib_dm_alloc_attr *attr,
+ struct uverbs_attr_bundle *attrs)
+{
+ u64 act_size = roundup(attr->length, MLX5_MEMIC_BASE_SIZE);
+ struct mlx5_core_dev *mdev = to_mdev(ibdev)->mdev;
+ phys_addr_t memic_addr;
+ struct mlx5_ib_dm *dm;
+ u64 start_offset;
+ u32 page_idx;
+ int err;
+
+ dm = kzalloc(sizeof(*dm), GFP_KERNEL);
+ if (!dm)
+ return ERR_PTR(-ENOMEM);
+
+ err = mlx5_core_alloc_memic(mdev, &memic_addr,
+ act_size, attr->alignment);
+ if (err)
+ goto err_free;
+
+ start_offset = memic_addr & ~PAGE_MASK;
+ page_idx = (memic_addr - pci_resource_start(mdev->pdev, 0) -
+ MLX5_CAP64_DEV_MEM(mdev, memic_bar_start_addr)) >>
+ PAGE_SHIFT;
+
+ err = uverbs_copy_to(attrs,
+ MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
+ &start_offset, sizeof(start_offset));
+ if (err)
+ goto err_dealloc;
+
+ err = uverbs_copy_to(attrs,
+ MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
+ &page_idx, sizeof(page_idx));
+ if (err)
+ goto err_dealloc;
+
+ bitmap_set(to_mucontext(context)->dm_pages, page_idx,
+ DIV_ROUND_UP(act_size, PAGE_SIZE));
+
+ dm->dev_addr = memic_addr;
+
+ return &dm->ibdm;
+
+err_dealloc:
+ mlx5_core_dealloc_memic(to_mdev(ibdev)->mdev, memic_addr,
+ act_size);
+err_free:
+ kfree(dm);
+ return ERR_PTR(err);
+}
+
+int mlx5_ib_dealloc_dm(struct ib_dm *ibdm)
+{
+ struct mlx5_ib_dev *dev = to_mdev(ibdm->device);
+ struct mlx5_ib_dm *dm = to_mdm(ibdm);
+ u64 act_size = roundup(dm->ibdm.length, MLX5_MEMIC_BASE_SIZE);
+ u32 page_idx;
+ int ret;
+
+ ret = mlx5_core_dealloc_memic(dev->mdev, dm->dev_addr, act_size);
+
+ if (ret)
+ return ret;
+
+ page_idx = (dm->dev_addr - pci_resource_start(dev->mdev->pdev, 0) -
+ MLX5_CAP64_DEV_MEM(dev->mdev, memic_bar_start_addr)) >>
+ PAGE_SHIFT;
+ bitmap_clear(to_mucontext(ibdm->uobject->context)->dm_pages,
+ page_idx,
+ DIV_ROUND_UP(act_size, PAGE_SIZE));
+
+ kfree(dm);
+
+ return 0;
+}
+
static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev,
struct ib_ucontext *context,
struct ib_udata *udata)
@@ -4844,13 +4961,22 @@ static void mlx5_ib_cleanup_multiport_master(struct mlx5_ib_dev *dev)
mlx5_nic_vport_disable_roce(dev->mdev);
}
+ADD_UVERBS_ATTRIBUTES_SIMPLE(mlx5_ib_dm, UVERBS_OBJECT_DM,
+ UVERBS_METHOD_DM_ALLOC,
+ &UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET,
+ UVERBS_ATTR_TYPE(u64),
+ UA_FLAGS(UVERBS_ATTR_SPEC_F_MANDATORY)),
+ &UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
+ UVERBS_ATTR_TYPE(u16),
+ UA_FLAGS(UVERBS_ATTR_SPEC_F_MANDATORY)));
+
ADD_UVERBS_ATTRIBUTES_SIMPLE(mlx5_ib_flow_action, UVERBS_OBJECT_FLOW_ACTION,
UVERBS_METHOD_FLOW_ACTION_ESP_CREATE,
&UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS,
UVERBS_ATTR_TYPE(u64),
UA_FLAGS(UVERBS_ATTR_SPEC_F_MANDATORY)));
-#define NUM_TREES 1
+#define NUM_TREES 2
static int populate_specs_root(struct mlx5_ib_dev *dev)
{
const struct uverbs_object_tree_def *default_root[NUM_TREES + 1] = {
@@ -4861,6 +4987,10 @@ static int populate_specs_root(struct mlx5_ib_dev *dev)
!WARN_ON(num_trees >= ARRAY_SIZE(default_root)))
default_root[num_trees++] = &mlx5_ib_flow_action;
+ if (MLX5_CAP_DEV_MEM(dev->mdev, memic) &&
+ !WARN_ON(num_trees >= ARRAY_SIZE(default_root)))
+ default_root[num_trees++] = &mlx5_ib_dm;
+
dev->ib_dev.specs_root =
uverbs_alloc_spec_tree(num_trees, default_root);
@@ -5097,6 +5227,11 @@ int mlx5_ib_stage_caps_init(struct mlx5_ib_dev *dev)
(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
}
+ if (MLX5_CAP_DEV_MEM(mdev, memic)) {
+ dev->ib_dev.alloc_dm = mlx5_ib_alloc_dm;
+ dev->ib_dev.dealloc_dm = mlx5_ib_dealloc_dm;
+ }
+
dev->ib_dev.create_flow = mlx5_ib_create_flow;
dev->ib_dev.destroy_flow = mlx5_ib_destroy_flow;
dev->ib_dev.uverbs_ex_cmd_mask |=
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 2b27ddafc354..a8ac6ce6ebd8 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -45,6 +45,7 @@
#include <linux/mlx5/transobj.h>
#include <rdma/ib_user_verbs.h>
#include <rdma/mlx5-abi.h>
+#include <rdma/uverbs_ioctl.h>
#define mlx5_ib_dbg(dev, format, arg...) \
pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__, \
@@ -131,6 +132,7 @@ struct mlx5_ib_ucontext {
struct mutex vma_private_list_mutex;
u64 lib_caps;
+ DECLARE_BITMAP(dm_pages, MLX5_MAX_MEMIC_PAGES);
};
static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
@@ -521,6 +523,11 @@ enum mlx5_ib_mtt_access_flags {
MLX5_IB_MTT_WRITE = (1 << 1),
};
+struct mlx5_ib_dm {
+ struct ib_dm ibdm;
+ phys_addr_t dev_addr;
+};
+
#define MLX5_IB_MTT_PRESENT (MLX5_IB_MTT_READ | MLX5_IB_MTT_WRITE)
struct mlx5_ib_mr {
@@ -897,6 +904,11 @@ static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
return container_of(msrq, struct mlx5_ib_srq, msrq);
}
+static inline struct mlx5_ib_dm *to_mdm(struct ib_dm *ibdm)
+{
+ return container_of(ibdm, struct mlx5_ib_dm, ibdm);
+}
+
static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
{
return container_of(ibmr, struct mlx5_ib_mr, ibmr);
@@ -1041,7 +1053,11 @@ struct ib_rwq_ind_table *mlx5_ib_create_rwq_ind_table(struct ib_device *device,
struct ib_udata *udata);
int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
bool mlx5_ib_dc_atomic_is_supported(struct mlx5_ib_dev *dev);
-
+struct ib_dm *mlx5_ib_alloc_dm(struct ib_device *ibdev,
+ struct ib_ucontext *context,
+ struct ib_dm_alloc_attr *attr,
+ struct uverbs_attr_bundle *attrs);
+int mlx5_ib_dealloc_dm(struct ib_dm *ibdm);
#ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
void mlx5_ib_internal_fill_odp_caps(struct mlx5_ib_dev *dev);
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index ad5c46eaedaa..16c56b02463b 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -429,6 +429,7 @@ enum mlx5_ib_mmap_cmd {
MLX5_IB_MMAP_CORE_CLOCK = 5,
MLX5_IB_MMAP_ALLOC_WC = 6,
MLX5_IB_MMAP_CLOCK_INFO = 7,
+ MLX5_IB_MMAP_DEVICE_MEM = 8,
};
enum {
diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
index 521813d5348c..f7d685ef2d1f 100644
--- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h
+++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h
@@ -40,5 +40,9 @@ enum mlx5_ib_create_flow_action_attrs {
MLX5_IB_ATTR_CREATE_FLOW_ACTION_FLAGS = (1U << UVERBS_ID_NS_SHIFT),
};
-#endif
+enum mlx5_ib_alloc_dm_attrs {
+ MLX5_IB_ATTR_ALLOC_DM_RESP_START_OFFSET = (1U << UVERBS_ID_NS_SHIFT),
+ MLX5_IB_ATTR_ALLOC_DM_RESP_PAGE_INDEX,
+};
+#endif
--
2.14.3
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
