net/mlx5: Add support for release all pages event

If FW sets release_all_pages bit in MLX5_EVENT_TYPE_PAGE_REQUEST,
driver shall release all pages of a given function id, with no further
pages reclaim negotiation with FW nor MANAGE_PAGES commands from driver
towards FW.

Upon receiving this bit as part of pages reclaim event, driver will
initiate release all flow, in which it will iterate and release all
function's pages.

As part of driver <-> FW capabilities handshake, FW will report
release_all_pages max HCA cap bit, and driver will set the
release_all_pages bit in HCA cap.

NIC: ConnectX-4 Lx
CPU: Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz
Test case: Simulataniously FLR 4 VFs, and measure FW release pages by
driver.
Before: 3.18 Sec
After:  0.31 Sec

Signed-off-by: Eran Ben Elisha <eranbe@mellanox.com>
Reviewed-by: Moshe Shemesh <moshe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c
index fbbf51026b52ba9d0b6728b4de18adfb5b42d9e3..742ba012c2346151e456e15f5fb586a47d17c9bc 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c
@@ -549,6 +549,9 @@ static int handle_hca_cap(struct mlx5_core_dev *dev, void *set_ctx)
                         num_vhca_ports,
                         MLX5_CAP_GEN_MAX(dev, num_vhca_ports));
 
+       if (MLX5_CAP_GEN_MAX(dev, release_all_pages))
+               MLX5_SET(cmd_hca_cap, set_hca_cap, release_all_pages, 1);
+
        return set_caps(dev, set_ctx, MLX5_SET_HCA_CAP_OP_MOD_GENERAL_DEVICE);
 }
 

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
index c790d6e3d204c34268bb39006c6dc4d34786f8ff..8ce78f42dfc0bfb89880f15b9f14f5b891931085 100644 (file)
--- a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c
@@ -50,6 +50,7 @@ struct mlx5_pages_req {
        u8      ec_function;
        s32     npages;
        struct work_struct work;
+       u8      release_all;
 };
 
 struct fw_page {
@@ -341,6 +342,33 @@ out_free:
        return err;
 }
 
+static void release_all_pages(struct mlx5_core_dev *dev, u32 func_id,
+                             bool ec_function)
+{
+       struct rb_node *p;
+       int npages = 0;
+
+       p = rb_first(&dev->priv.page_root);
+       while (p) {
+               struct fw_page *fwp = rb_entry(p, struct fw_page, rb_node);
+
+               p = rb_next(p);
+               if (fwp->func_id != func_id)
+                       continue;
+               free_fwp(dev, fwp);
+               npages++;
+       }
+
+       dev->priv.fw_pages -= npages;
+       if (func_id)
+               dev->priv.vfs_pages -= npages;
+       else if (mlx5_core_is_ecpf(dev) && !ec_function)
+               dev->priv.peer_pf_pages -= npages;
+
+       mlx5_core_dbg(dev, "npages %d, ec_function %d, func_id 0x%x\n",
+                     npages, ec_function, func_id);
+}
+
 static int reclaim_pages_cmd(struct mlx5_core_dev *dev,
                             u32 *in, int in_size, u32 *out, int out_size)
 {
@@ -434,7 +462,9 @@ static void pages_work_handler(struct work_struct *work)
        struct mlx5_core_dev *dev = req->dev;
        int err = 0;
 
-       if (req->npages < 0)
+       if (req->release_all)
+               release_all_pages(dev, req->func_id, req->ec_function);
+       else if (req->npages < 0)
                err = reclaim_pages(dev, req->func_id, -1 * req->npages, NULL,
                                    req->ec_function);
        else if (req->npages > 0)
@@ -449,6 +479,7 @@ static void pages_work_handler(struct work_struct *work)
 
 enum {
        EC_FUNCTION_MASK = 0x8000,
+       RELEASE_ALL_PAGES_MASK = 0x4000,
 };
 
 static int req_pages_handler(struct notifier_block *nb,
@@ -459,6 +490,7 @@ static int req_pages_handler(struct notifier_block *nb,
        struct mlx5_priv *priv;
        struct mlx5_eqe *eqe;
        bool ec_function;
+       bool release_all;
        u16 func_id;
        s32 npages;
 
@@ -469,8 +501,10 @@ static int req_pages_handler(struct notifier_block *nb,
        func_id = be16_to_cpu(eqe->data.req_pages.func_id);
        npages  = be32_to_cpu(eqe->data.req_pages.num_pages);
        ec_function = be16_to_cpu(eqe->data.req_pages.ec_function) & EC_FUNCTION_MASK;
-       mlx5_core_dbg(dev, "page request for func 0x%x, npages %d\n",
-                     func_id, npages);
+       release_all = be16_to_cpu(eqe->data.req_pages.ec_function) &
+                     RELEASE_ALL_PAGES_MASK;
+       mlx5_core_dbg(dev, "page request for func 0x%x, npages %d, release_all %d\n",
+                     func_id, npages, release_all);
        req = kzalloc(sizeof(*req), GFP_ATOMIC);
        if (!req) {
                mlx5_core_warn(dev, "failed to allocate pages request\n");
@@ -481,6 +515,7 @@ static int req_pages_handler(struct notifier_block *nb,
        req->func_id = func_id;
        req->npages = npages;
        req->ec_function = ec_function;
+       req->release_all = release_all;
        INIT_WORK(&req->work, pages_work_handler);
        queue_work(dev->priv.pg_wq, &req->work);
        return NOTIFY_OK;