linux/drivers/infiniband/hw/mlx5/srq.c
Jason Gunthorpe 8a7904a672 RDMA/mlx5: Lower setting the umem's PAS for SRQ
Some of the SRQ types are created using a WQ, and the WQ requires a
different parameter set to mlx5_umem_find_best_quantized_pgoff() as it has
a 5 bit page_offset.

Add the umem to the mlx5_srq_attr and defer computing the PAS data until
the code has figured out what kind of mailbox to use. Compute the PAS
directly from the umem for each of the four unique mailbox types.

This also avoids allocating memory to store the user PAS, instead it is
written directly to the mailbox as in most other cases.

Fixes: 01949d0109 ("net/mlx5_core: Enable XRCs and SRQs when using ISSI > 0")
Link: https://lore.kernel.org/r/20201115114311.136250-8-leon@kernel.org
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
2020-11-16 16:53:30 -04:00

471 lines
11 KiB
C

// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
/*
* Copyright (c) 2013-2018, Mellanox Technologies inc. All rights reserved.
*/
#include <linux/module.h>
#include <linux/mlx5/qp.h>
#include <linux/slab.h>
#include <rdma/ib_umem.h>
#include <rdma/ib_user_verbs.h>
#include "mlx5_ib.h"
#include "srq.h"
static void *get_wqe(struct mlx5_ib_srq *srq, int n)
{
return mlx5_frag_buf_get_wqe(&srq->fbc, n);
}
static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
{
struct ib_event event;
struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
if (ibsrq->event_handler) {
event.device = ibsrq->device;
event.element.srq = ibsrq;
switch (type) {
case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
event.event = IB_EVENT_SRQ_LIMIT_REACHED;
break;
case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
event.event = IB_EVENT_SRQ_ERR;
break;
default:
pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
type, srq->srqn);
return;
}
ibsrq->event_handler(&event, ibsrq->srq_context);
}
}
static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq,
struct mlx5_srq_attr *in,
struct ib_udata *udata, int buf_size)
{
struct mlx5_ib_dev *dev = to_mdev(pd->device);
struct mlx5_ib_create_srq ucmd = {};
struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
udata, struct mlx5_ib_ucontext, ibucontext);
size_t ucmdlen;
int err;
u32 uidx = MLX5_IB_DEFAULT_UIDX;
ucmdlen = min(udata->inlen, sizeof(ucmd));
if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
mlx5_ib_dbg(dev, "failed copy udata\n");
return -EFAULT;
}
if (ucmd.reserved0 || ucmd.reserved1)
return -EINVAL;
if (udata->inlen > sizeof(ucmd) &&
!ib_is_udata_cleared(udata, sizeof(ucmd),
udata->inlen - sizeof(ucmd)))
return -EINVAL;
if (in->type != IB_SRQT_BASIC) {
err = get_srq_user_index(ucontext, &ucmd, udata->inlen, &uidx);
if (err)
return err;
}
srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
srq->umem = ib_umem_get(pd->device, ucmd.buf_addr, buf_size, 0);
if (IS_ERR(srq->umem)) {
mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
err = PTR_ERR(srq->umem);
return err;
}
in->umem = srq->umem;
err = mlx5_ib_db_map_user(ucontext, udata, ucmd.db_addr, &srq->db);
if (err) {
mlx5_ib_dbg(dev, "map doorbell failed\n");
goto err_umem;
}
in->uid = (in->type != IB_SRQT_XRC) ? to_mpd(pd)->uid : 0;
if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
in->type != IB_SRQT_BASIC)
in->user_index = uidx;
return 0;
err_umem:
ib_umem_release(srq->umem);
return err;
}
static int create_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq,
struct mlx5_srq_attr *in, int buf_size)
{
int err;
int i;
struct mlx5_wqe_srq_next_seg *next;
err = mlx5_db_alloc(dev->mdev, &srq->db);
if (err) {
mlx5_ib_warn(dev, "alloc dbell rec failed\n");
return err;
}
if (mlx5_frag_buf_alloc_node(dev->mdev, buf_size, &srq->buf,
dev->mdev->priv.numa_node)) {
mlx5_ib_dbg(dev, "buf alloc failed\n");
err = -ENOMEM;
goto err_db;
}
mlx5_init_fbc(srq->buf.frags, srq->msrq.wqe_shift, ilog2(srq->msrq.max),
&srq->fbc);
srq->head = 0;
srq->tail = srq->msrq.max - 1;
srq->wqe_ctr = 0;
for (i = 0; i < srq->msrq.max; i++) {
next = get_wqe(srq, i);
next->next_wqe_index =
cpu_to_be16((i + 1) & (srq->msrq.max - 1));
}
mlx5_ib_dbg(dev, "srq->buf.page_shift = %d\n", srq->buf.page_shift);
in->pas = kvcalloc(srq->buf.npages, sizeof(*in->pas), GFP_KERNEL);
if (!in->pas) {
err = -ENOMEM;
goto err_buf;
}
mlx5_fill_page_frag_array(&srq->buf, in->pas);
srq->wrid = kvmalloc_array(srq->msrq.max, sizeof(u64), GFP_KERNEL);
if (!srq->wrid) {
err = -ENOMEM;
goto err_in;
}
srq->wq_sig = 0;
in->log_page_size = srq->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
if (MLX5_CAP_GEN(dev->mdev, cqe_version) == MLX5_CQE_VERSION_V1 &&
in->type != IB_SRQT_BASIC)
in->user_index = MLX5_IB_DEFAULT_UIDX;
return 0;
err_in:
kvfree(in->pas);
err_buf:
mlx5_frag_buf_free(dev->mdev, &srq->buf);
err_db:
mlx5_db_free(dev->mdev, &srq->db);
return err;
}
static void destroy_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq,
struct ib_udata *udata)
{
mlx5_ib_db_unmap_user(
rdma_udata_to_drv_context(
udata,
struct mlx5_ib_ucontext,
ibucontext),
&srq->db);
ib_umem_release(srq->umem);
}
static void destroy_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq)
{
kvfree(srq->wrid);
mlx5_frag_buf_free(dev->mdev, &srq->buf);
mlx5_db_free(dev->mdev, &srq->db);
}
int mlx5_ib_create_srq(struct ib_srq *ib_srq,
struct ib_srq_init_attr *init_attr,
struct ib_udata *udata)
{
struct mlx5_ib_dev *dev = to_mdev(ib_srq->device);
struct mlx5_ib_srq *srq = to_msrq(ib_srq);
size_t desc_size;
size_t buf_size;
int err;
struct mlx5_srq_attr in = {};
__u32 max_srq_wqes = 1 << MLX5_CAP_GEN(dev->mdev, log_max_srq_sz);
if (init_attr->srq_type != IB_SRQT_BASIC &&
init_attr->srq_type != IB_SRQT_XRC &&
init_attr->srq_type != IB_SRQT_TM)
return -EOPNOTSUPP;
/* Sanity check SRQ size before proceeding */
if (init_attr->attr.max_wr >= max_srq_wqes) {
mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
init_attr->attr.max_wr,
max_srq_wqes);
return -EINVAL;
}
mutex_init(&srq->mutex);
spin_lock_init(&srq->lock);
srq->msrq.max = roundup_pow_of_two(init_attr->attr.max_wr + 1);
srq->msrq.max_gs = init_attr->attr.max_sge;
desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
if (desc_size == 0 || srq->msrq.max_gs > desc_size)
return -EINVAL;
desc_size = roundup_pow_of_two(desc_size);
desc_size = max_t(size_t, 32, desc_size);
if (desc_size < sizeof(struct mlx5_wqe_srq_next_seg))
return -EINVAL;
srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
sizeof(struct mlx5_wqe_data_seg);
srq->msrq.wqe_shift = ilog2(desc_size);
buf_size = srq->msrq.max * desc_size;
if (buf_size < desc_size)
return -EINVAL;
in.type = init_attr->srq_type;
if (udata)
err = create_srq_user(ib_srq->pd, srq, &in, udata, buf_size);
else
err = create_srq_kernel(dev, srq, &in, buf_size);
if (err) {
mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
udata ? "user" : "kernel", err);
return err;
}
in.log_size = ilog2(srq->msrq.max);
in.wqe_shift = srq->msrq.wqe_shift - 4;
if (srq->wq_sig)
in.flags |= MLX5_SRQ_FLAG_WQ_SIG;
if (init_attr->srq_type == IB_SRQT_XRC && init_attr->ext.xrc.xrcd)
in.xrcd = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
else
in.xrcd = dev->devr.xrcdn0;
if (init_attr->srq_type == IB_SRQT_TM) {
in.tm_log_list_size =
ilog2(init_attr->ext.tag_matching.max_num_tags) + 1;
if (in.tm_log_list_size >
MLX5_CAP_GEN(dev->mdev, log_tag_matching_list_sz)) {
mlx5_ib_dbg(dev, "TM SRQ max_num_tags exceeding limit\n");
err = -EINVAL;
goto err_usr_kern_srq;
}
in.flags |= MLX5_SRQ_FLAG_RNDV;
}
if (ib_srq_has_cq(init_attr->srq_type))
in.cqn = to_mcq(init_attr->ext.cq)->mcq.cqn;
else
in.cqn = to_mcq(dev->devr.c0)->mcq.cqn;
in.pd = to_mpd(ib_srq->pd)->pdn;
in.db_record = srq->db.dma;
err = mlx5_cmd_create_srq(dev, &srq->msrq, &in);
kvfree(in.pas);
if (err) {
mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
goto err_usr_kern_srq;
}
mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
srq->msrq.event = mlx5_ib_srq_event;
srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
if (udata) {
struct mlx5_ib_create_srq_resp resp = {
.srqn = srq->msrq.srqn,
};
if (ib_copy_to_udata(udata, &resp, min(udata->outlen,
sizeof(resp)))) {
mlx5_ib_dbg(dev, "copy to user failed\n");
err = -EFAULT;
goto err_core;
}
}
init_attr->attr.max_wr = srq->msrq.max - 1;
return 0;
err_core:
mlx5_cmd_destroy_srq(dev, &srq->msrq);
err_usr_kern_srq:
if (udata)
destroy_srq_user(ib_srq->pd, srq, udata);
else
destroy_srq_kernel(dev, srq);
return err;
}
int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
{
struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
struct mlx5_ib_srq *srq = to_msrq(ibsrq);
int ret;
/* We don't support resizing SRQs yet */
if (attr_mask & IB_SRQ_MAX_WR)
return -EINVAL;
if (attr_mask & IB_SRQ_LIMIT) {
if (attr->srq_limit >= srq->msrq.max)
return -EINVAL;
mutex_lock(&srq->mutex);
ret = mlx5_cmd_arm_srq(dev, &srq->msrq, attr->srq_limit, 1);
mutex_unlock(&srq->mutex);
if (ret)
return ret;
}
return 0;
}
int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
{
struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
struct mlx5_ib_srq *srq = to_msrq(ibsrq);
int ret;
struct mlx5_srq_attr *out;
out = kzalloc(sizeof(*out), GFP_KERNEL);
if (!out)
return -ENOMEM;
ret = mlx5_cmd_query_srq(dev, &srq->msrq, out);
if (ret)
goto out_box;
srq_attr->srq_limit = out->lwm;
srq_attr->max_wr = srq->msrq.max - 1;
srq_attr->max_sge = srq->msrq.max_gs;
out_box:
kfree(out);
return ret;
}
int mlx5_ib_destroy_srq(struct ib_srq *srq, struct ib_udata *udata)
{
struct mlx5_ib_dev *dev = to_mdev(srq->device);
struct mlx5_ib_srq *msrq = to_msrq(srq);
int ret;
ret = mlx5_cmd_destroy_srq(dev, &msrq->msrq);
if (ret)
return ret;
if (udata)
destroy_srq_user(srq->pd, msrq, udata);
else
destroy_srq_kernel(dev, msrq);
return 0;
}
void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
{
struct mlx5_wqe_srq_next_seg *next;
/* always called with interrupts disabled. */
spin_lock(&srq->lock);
next = get_wqe(srq, srq->tail);
next->next_wqe_index = cpu_to_be16(wqe_index);
srq->tail = wqe_index;
spin_unlock(&srq->lock);
}
int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, const struct ib_recv_wr *wr,
const struct ib_recv_wr **bad_wr)
{
struct mlx5_ib_srq *srq = to_msrq(ibsrq);
struct mlx5_wqe_srq_next_seg *next;
struct mlx5_wqe_data_seg *scat;
struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
struct mlx5_core_dev *mdev = dev->mdev;
unsigned long flags;
int err = 0;
int nreq;
int i;
spin_lock_irqsave(&srq->lock, flags);
if (mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) {
err = -EIO;
*bad_wr = wr;
goto out;
}
for (nreq = 0; wr; nreq++, wr = wr->next) {
if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
err = -EINVAL;
*bad_wr = wr;
break;
}
if (unlikely(srq->head == srq->tail)) {
err = -ENOMEM;
*bad_wr = wr;
break;
}
srq->wrid[srq->head] = wr->wr_id;
next = get_wqe(srq, srq->head);
srq->head = be16_to_cpu(next->next_wqe_index);
scat = (struct mlx5_wqe_data_seg *)(next + 1);
for (i = 0; i < wr->num_sge; i++) {
scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
scat[i].lkey = cpu_to_be32(wr->sg_list[i].lkey);
scat[i].addr = cpu_to_be64(wr->sg_list[i].addr);
}
if (i < srq->msrq.max_avail_gather) {
scat[i].byte_count = 0;
scat[i].lkey = cpu_to_be32(MLX5_INVALID_LKEY);
scat[i].addr = 0;
}
}
if (likely(nreq)) {
srq->wqe_ctr += nreq;
/* Make sure that descriptors are written before
* doorbell record.
*/
wmb();
*srq->db.db = cpu_to_be32(srq->wqe_ctr);
}
out:
spin_unlock_irqrestore(&srq->lock, flags);
return err;
}