#include <linux/export.h>
#include <linux/delay.h>
#include <rdma/ib_umem.h>
+#include <rdma/ib_verbs.h>
#include "mlx5_ib.h"
enum {
MLX5_UMR_ALIGN = 2048
};
-static __be64 *mr_align(__be64 *ptr, int align)
-{
- unsigned long mask = align - 1;
-
- return (__be64 *)(((unsigned long)ptr + mask) & ~mask);
-}
-
static int order2idx(struct mlx5_ib_dev *dev, int order)
{
struct mlx5_mr_cache *cache = &dev->cache;
mr->order = ent->order;
mr->umred = 1;
mr->dev = dev;
- in->seg.status = 1 << 6;
+ in->seg.status = MLX5_MKEY_STATUS_FREE;
in->seg.xlt_oct_size = cpu_to_be32((npages + 1) / 2);
in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
in->seg.flags = MLX5_ACCESS_MODE_MTT | MLX5_PERM_UMR_EN;
sizeof(*in), reg_mr_callback,
mr, &mr->out);
if (err) {
+ spin_lock_irq(&ent->lock);
+ ent->pending--;
+ spin_unlock_irq(&ent->lock);
mlx5_ib_warn(dev, "create mkey failed %d\n", err);
kfree(mr);
break;
static int use_umr(int order)
{
- return order <= 17;
+ return order <= MLX5_MAX_UMR_SHIFT;
}
static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr,
{
struct mlx5_ib_dev *dev = to_mdev(pd->device);
struct ib_mr *mr = dev->umrc.mr;
+ struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
sg->addr = dma;
sg->length = ALIGN(sizeof(u64) * n, 64);
wr->num_sge = 0;
wr->opcode = MLX5_IB_WR_UMR;
- wr->wr.fast_reg.page_list_len = n;
- wr->wr.fast_reg.page_shift = page_shift;
- wr->wr.fast_reg.rkey = key;
- wr->wr.fast_reg.iova_start = virt_addr;
- wr->wr.fast_reg.length = len;
- wr->wr.fast_reg.access_flags = access_flags;
- wr->wr.fast_reg.page_list = (struct ib_fast_reg_page_list *)pd;
+
+ umrwr->npages = n;
+ umrwr->page_shift = page_shift;
+ umrwr->mkey = key;
+ umrwr->target.virt_addr = virt_addr;
+ umrwr->length = len;
+ umrwr->access_flags = access_flags;
+ umrwr->pd = pd;
}
static void prep_umr_unreg_wqe(struct mlx5_ib_dev *dev,
struct ib_send_wr *wr, u32 key)
{
- wr->send_flags = MLX5_IB_SEND_UMR_UNREG;
+ struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+
+ wr->send_flags = MLX5_IB_SEND_UMR_UNREG | MLX5_IB_SEND_UMR_FAIL_IF_FREE;
wr->opcode = MLX5_IB_WR_UMR;
- wr->wr.fast_reg.rkey = key;
+ umrwr->mkey = key;
}
void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context)
struct ib_send_wr wr, *bad;
struct mlx5_ib_mr *mr;
struct ib_sge sg;
- int size = sizeof(u64) * npages;
+ int size;
+ __be64 *mr_pas;
+ __be64 *pas;
+ dma_addr_t dma;
int err = 0;
int i;
if (!mr)
return ERR_PTR(-EAGAIN);
- mr->pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL);
- if (!mr->pas) {
+ /* UMR copies MTTs in units of MLX5_UMR_MTT_ALIGNMENT bytes.
+ * To avoid copying garbage after the pas array, we allocate
+ * a little more. */
+ size = ALIGN(sizeof(u64) * npages, MLX5_UMR_MTT_ALIGNMENT);
+ mr_pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL);
+ if (!mr_pas) {
err = -ENOMEM;
goto free_mr;
}
- mlx5_ib_populate_pas(dev, umem, page_shift,
- mr_align(mr->pas, MLX5_UMR_ALIGN), 1);
+ pas = PTR_ALIGN(mr_pas, MLX5_UMR_ALIGN);
+ mlx5_ib_populate_pas(dev, umem, page_shift, pas, MLX5_IB_MTT_PRESENT);
+ /* Clear padding after the actual pages. */
+ memset(pas + npages, 0, size - npages * sizeof(u64));
- mr->dma = dma_map_single(ddev, mr_align(mr->pas, MLX5_UMR_ALIGN), size,
- DMA_TO_DEVICE);
- if (dma_mapping_error(ddev, mr->dma)) {
+ dma = dma_map_single(ddev, pas, size, DMA_TO_DEVICE);
+ if (dma_mapping_error(ddev, dma)) {
err = -ENOMEM;
goto free_pas;
}
memset(&wr, 0, sizeof(wr));
wr.wr_id = (u64)(unsigned long)&umr_context;
- prep_umr_reg_wqe(pd, &wr, &sg, mr->dma, npages, mr->mmr.key, page_shift, virt_addr, len, access_flags);
+ prep_umr_reg_wqe(pd, &wr, &sg, dma, npages, mr->mmr.key, page_shift,
+ virt_addr, len, access_flags);
mlx5_ib_init_umr_context(&umr_context);
down(&umrc->sem);
unmap_dma:
up(&umrc->sem);
- dma_unmap_single(ddev, mr->dma, size, DMA_TO_DEVICE);
+ dma_unmap_single(ddev, dma, size, DMA_TO_DEVICE);
free_pas:
- kfree(mr->pas);
+ kfree(mr_pas);
free_mr:
if (err) {
struct mlx5_ib_mr *mr;
int inlen;
int err;
+ bool pg_cap = !!(dev->mdev->caps.gen.flags &
+ MLX5_DEV_CAP_FLAG_ON_DMND_PG);
mr = kzalloc(sizeof(*mr), GFP_KERNEL);
if (!mr)
err = -ENOMEM;
goto err_1;
}
- mlx5_ib_populate_pas(dev, umem, page_shift, in->pas, 0);
+ mlx5_ib_populate_pas(dev, umem, page_shift, in->pas,
+ pg_cap ? MLX5_IB_MTT_PRESENT : 0);
+ /* The MLX5_MKEY_INBOX_PG_ACCESS bit allows setting the access flags
+ * in the page list submitted with the command. */
+ in->flags = pg_cap ? cpu_to_be32(MLX5_MKEY_INBOX_PG_ACCESS) : 0;
in->seg.flags = convert_access(access_flags) |
MLX5_ACCESS_MODE_MTT;
in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
goto err_free;
}
- in->seg.status = 1 << 6; /* free */
+ in->seg.status = MLX5_MKEY_STATUS_FREE;
in->seg.xlt_oct_size = cpu_to_be32(ndescs);
in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
goto err_free;
}
- in->seg.status = 1 << 6; /* free */
+ in->seg.status = MLX5_MKEY_STATUS_FREE;
in->seg.xlt_oct_size = cpu_to_be32((max_page_list_len + 1) / 2);
in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
in->seg.flags = MLX5_PERM_UMR_EN | MLX5_ACCESS_MODE_MTT;