Subject: Re: [Patch v7 12/12] RDMA/mana_ib: Add a driver for Microsoft
Azure Network Adapter

On Mon, 2022-10-17 at 12:20 -0700, longli@linuxonhyperv.com wrote:

quoted

diff --git a/drivers/infiniband/hw/mana/main.c

b/drivers/infiniband/hw/mana/main.c
new file mode 100644
index 000000000000..57e5f9dca454

--- /dev/null
+++ b/drivers/infiniband/hw/mana/main.c

[...]

quoted

+static int mana_gd_destroy_doorbell_page(struct gdma_context *gc,
+					 int doorbell_page)
+{
+	struct gdma_destroy_resource_range_req req = {};
+	struct gdma_resp_hdr resp = {};
+	int err;
+
+	mana_gd_init_req_hdr(&req.hdr,

GDMA_DESTROY_RESOURCE_RANGE,

quoted

+			     sizeof(req), sizeof(resp));
+
+	req.resource_type = GDMA_RESOURCE_DOORBELL_PAGE;
+	req.num_resources = 1;
+	req.allocated_resources = doorbell_page;
+
+	err = mana_gd_send_request(gc, sizeof(req), &req, sizeof(resp),

&resp);

quoted

+	if (err || resp.status) {
+		dev_err(gc->dev,
+			"Failed to destroy doorbell page: ret %d, 0x%x\n",
+			err, resp.status);
+		return err ? err : -EPROTO;

Minor nit: the preferred style is:
		return err ?: -EPROTO;

a few other occurences below.

Will change this.

quoted

+	}
+
+	return 0;
+}

[...]

quoted

+int mana_ib_gd_create_dma_region(struct mana_ib_dev *dev, struct

ib_umem *umem,

quoted

+				 mana_handle_t *gdma_region)
+{
+	struct gdma_dma_region_add_pages_req *add_req = NULL;
+	struct gdma_create_dma_region_resp create_resp = {};
+	struct gdma_create_dma_region_req *create_req;
+	size_t num_pages_cur, num_pages_to_handle;
+	unsigned int create_req_msg_size;
+	struct hw_channel_context *hwc;
+	struct ib_block_iter biter;
+	size_t max_pgs_create_cmd;
+	struct gdma_context *gc;
+	size_t num_pages_total;
+	struct gdma_dev *mdev;
+	unsigned long page_sz;
+	void *request_buf;
+	unsigned int i;
+	int err;
+
+	mdev = dev->gdma_dev;
+	gc = mdev->gdma_context;
+	hwc = gc->hwc.driver_data;
+
+	/* Hardware requires dma region to align to chosen page size */
+	page_sz = ib_umem_find_best_pgsz(umem, PAGE_SZ_BM, 0);
+	if (!page_sz) {
+		ibdev_dbg(&dev->ib_dev, "failed to find page size.\n");
+		return -ENOMEM;
+	}
+	num_pages_total = ib_umem_num_dma_blocks(umem, page_sz);
+
+	max_pgs_create_cmd =
+		(hwc->max_req_msg_size - sizeof(*create_req)) /

sizeof(u64);

quoted

+	num_pages_to_handle =
+		min_t(size_t, num_pages_total, max_pgs_create_cmd);
+	create_req_msg_size =
+		struct_size(create_req, page_addr_list,

num_pages_to_handle);

quoted

+
+	request_buf = kzalloc(hwc->max_req_msg_size, GFP_KERNEL);
+	if (!request_buf)
+		return -ENOMEM;
+
+	create_req = request_buf;
+	mana_gd_init_req_hdr(&create_req->hdr,

GDMA_CREATE_DMA_REGION,

quoted

+			     create_req_msg_size, sizeof(create_resp));
+
+	create_req->length = umem->length;
+	create_req->offset_in_page = umem->address & (page_sz - 1);
+	create_req->gdma_page_type = order_base_2(page_sz) -

PAGE_SHIFT;

quoted

+	create_req->page_count = num_pages_total;
+	create_req->page_addr_list_len = num_pages_to_handle;
+
+	ibdev_dbg(&dev->ib_dev, "size_dma_region %lu

num_pages_total %lu\n",

quoted

+		  umem->length, num_pages_total);
+
+	ibdev_dbg(&dev->ib_dev, "page_sz %lu offset_in_page %u\n",
+		  page_sz, create_req->offset_in_page);
+
+	ibdev_dbg(&dev->ib_dev, "num_pages_to_handle %lu,

gdma_page_type %u",

quoted

+		  num_pages_to_handle, create_req->gdma_page_type);
+
+	__rdma_umem_block_iter_start(&biter, umem, page_sz);
+
+	for (i = 0; i < num_pages_to_handle; ++i) {
+		dma_addr_t cur_addr;
+
+		__rdma_block_iter_next(&biter);
+		cur_addr = rdma_block_iter_dma_address(&biter);
+
+		create_req->page_addr_list[i] = cur_addr;
+	}
+
+	err = mana_gd_send_request(gc, create_req_msg_size, create_req,
+				   sizeof(create_resp), &create_resp);
+	if (err || create_resp.hdr.status) {
+		ibdev_dbg(&dev->ib_dev,
+			  "Failed to create DMA region: %d, 0x%x\n", err,
+			  create_resp.hdr.status);
+		if (!err)
+			err = -EPROTO;
+
+		kfree(request_buf);
+		return err;

Minor nit: you can avoid a little code doplication replacing the above
2 lines with:
		goto out;

Good suggestion, will make the change.

and ...

quoted

+	}
+
+	*gdma_region = create_resp.dma_region_handle;
+	ibdev_dbg(&dev->ib_dev, "Created DMA region with handle

0x%llx\n",

quoted

+		  *gdma_region);
+
+	num_pages_cur = num_pages_to_handle;
+
+	if (num_pages_cur < num_pages_total) {
+		unsigned int add_req_msg_size;
+		size_t max_pgs_add_cmd =
+			(hwc->max_req_msg_size - sizeof(*add_req)) /
+			sizeof(u64);
+
+		num_pages_to_handle =
+			min_t(size_t, num_pages_total - num_pages_cur,
+			      max_pgs_add_cmd);
+
+		/* Calculate the max num of pages that will be handled */
+		add_req_msg_size = struct_size(add_req, page_addr_list,
+					       num_pages_to_handle);
+		add_req = request_buf;
+
+		while (num_pages_cur < num_pages_total) {
+			struct gdma_general_resp add_resp = {};
+			u32 expected_status = 0;
+
+			if (num_pages_cur + num_pages_to_handle <
+			    num_pages_total) {
+				/* Status indicating more pages are needed

*/

quoted

+				expected_status =

GDMA_STATUS_MORE_ENTRIES;

quoted

+			}
+
+			memset(add_req, 0, add_req_msg_size);
+
+			mana_gd_init_req_hdr(&add_req->hdr,
+

GDMA_DMA_REGION_ADD_PAGES,

quoted

+					     add_req_msg_size,
+					     sizeof(add_resp));
+			add_req->dma_region_handle = *gdma_region;
+			add_req->page_addr_list_len =

num_pages_to_handle;

quoted

+
+			for (i = 0; i < num_pages_to_handle; ++i) {
+				dma_addr_t cur_addr =
+

	rdma_block_iter_dma_address(&biter);

quoted

+				add_req->page_addr_list[i] = cur_addr;
+				__rdma_block_iter_next(&biter);
+
+				ibdev_dbg(&dev->ib_dev,
+					  "page_addr_list %lu addr 0x%llx\n",
+					  num_pages_cur + i, cur_addr);
+			}
+
+			err = mana_gd_send_request(gc, add_req_msg_size,
+						   add_req, sizeof(add_resp),
+						   &add_resp);
+			if (err || add_resp.hdr.status != expected_status) {
+				ibdev_dbg(&dev->ib_dev,
+					  "Failed put DMA

pages %u: %d,0x%x\n",

quoted

+					  i, err, add_resp.hdr.status);
+				err = -EPROTO;
+				break;
+			}
+
+			num_pages_cur += num_pages_to_handle;
+			num_pages_to_handle =
+				min_t(size_t, num_pages_total -

num_pages_cur,

quoted

+				      max_pgs_add_cmd);
+			add_req_msg_size = sizeof(*add_req) +
+					   num_pages_to_handle *

sizeof(u64);

quoted

+		}
+	}
+
+	kfree(request_buf);
+
+	if (err)
+		mana_ib_gd_destroy_dma_region(dev,

create_resp.dma_region_handle);

... here:

	if (err)
		mana_ib_gd_destroy_dma_region(dev,
create_resp.dma_region_handle);

out:
	kfree(request_buf);

quoted

+
+	return err;
+}

[...]

quoted

diff --git a/drivers/infiniband/hw/mana/qp.c

b/drivers/infiniband/hw/mana/qp.c new file mode 100644 index
000000000000..fec7d4a06ace

--- /dev/null
+++ b/drivers/infiniband/hw/mana/qp.c

@@ -0,0 +1,505 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (c) 2022, Microsoft Corporation. All rights reserved.
+ */
+
+#include "mana_ib.h"
+
+static int mana_ib_cfg_vport_steering(struct mana_ib_dev *dev,
+				      struct net_device *ndev,
+				      mana_handle_t default_rxobj,
+				      mana_handle_t ind_table[],
+				      u32 log_ind_tbl_size, u32

rx_hash_key_len,

quoted

+				      u8 *rx_hash_key)
+{
+	struct mana_port_context *mpc = netdev_priv(ndev);
+	struct mana_cfg_rx_steer_req *req = NULL;
+	struct mana_cfg_rx_steer_resp resp = {};
+	mana_handle_t *req_indir_tab;
+	struct gdma_context *gc;
+	struct gdma_dev *mdev;
+	u32 req_buf_size;
+	int i, err;
+
+	mdev = dev->gdma_dev;
+	gc = mdev->gdma_context;
+
+	req_buf_size =
+		sizeof(*req) + sizeof(mana_handle_t) *

MANA_INDIRECT_TABLE_SIZE;

quoted

+	req = kzalloc(req_buf_size, GFP_KERNEL);
+	if (!req)
+		return -ENOMEM;
+
+	mana_gd_init_req_hdr(&req->hdr, MANA_CONFIG_VPORT_RX,

req_buf_size,

quoted

+			     sizeof(resp));
+
+	req->vport = mpc->port_handle;
+	req->rx_enable = 1;
+	req->update_default_rxobj = 1;
+	req->default_rxobj = default_rxobj;
+	req->hdr.dev_id = mdev->dev_id;
+
+	/* If there are more than 1 entries in indirection table, enable RSS */
+	if (log_ind_tbl_size)
+		req->rss_enable = true;
+
+	req->num_indir_entries = MANA_INDIRECT_TABLE_SIZE;
+	req->indir_tab_offset = sizeof(*req);
+	req->update_indir_tab = true;
+
+	req_indir_tab = (mana_handle_t *)(req + 1);
+	/* The ind table passed to the hardware must have
+	 * MANA_INDIRECT_TABLE_SIZE entries. Adjust the verb
+	 * ind_table to MANA_INDIRECT_TABLE_SIZE if required
+	 */
+	ibdev_dbg(&dev->ib_dev, "ind table size %u\n", 1 <<

log_ind_tbl_size);

quoted

+	for (i = 0; i < MANA_INDIRECT_TABLE_SIZE; i++) {
+		req_indir_tab[i] = ind_table[i % (1 << log_ind_tbl_size)];
+		ibdev_dbg(&dev->ib_dev, "index %u handle 0x%llx\n", i,
+			  req_indir_tab[i]);
+	}
+
+	req->update_hashkey = true;
+	if (rx_hash_key_len)
+		memcpy(req->hashkey, rx_hash_key, rx_hash_key_len);
+	else
+		netdev_rss_key_fill(req->hashkey, MANA_HASH_KEY_SIZE);
+
+	ibdev_dbg(&dev->ib_dev, "vport handle %llu default_rxobj

0x%llx\n",

quoted

+		  req->vport, default_rxobj);
+
+	err = mana_gd_send_request(gc, req_buf_size, req, sizeof(resp),

&resp);

quoted

+	if (err) {
+		netdev_err(ndev, "Failed to configure vPort RX: %d\n", err);
+		goto out;
+	}
+
+	if (resp.hdr.status) {
+		netdev_err(ndev, "vPort RX configuration failed: 0x%x\n",
+			   resp.hdr.status);
+		err = -EPROTO;

This is confusing: if this error condition is reached, both error and succesful
configuration will be logged. I guess an additional:

		goto out;

is needed.

Yes, it's confusing. Will change this.

quoted

+	}
+
+	netdev_info(ndev, "Configured steering vPort %llu

log_entries %u\n",

quoted

+		    mpc->port_handle, log_ind_tbl_size);
+
+out:
+	kfree(req);
+	return err;
+}
+
+static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd,
+				 struct ib_qp_init_attr *attr,
+				 struct ib_udata *udata)
+{
+	struct mana_ib_qp *qp = container_of(ibqp, struct mana_ib_qp,

ibqp);

quoted

+	struct mana_ib_dev *mdev =
+		container_of(pd->device, struct mana_ib_dev, ib_dev);
+	struct ib_rwq_ind_table *ind_tbl = attr->rwq_ind_tbl;
+	struct mana_ib_create_qp_rss_resp resp = {};
+	struct mana_ib_create_qp_rss ucmd = {};
+	struct gdma_dev *gd = mdev->gdma_dev;
+	mana_handle_t *mana_ind_table;
+	struct mana_port_context *mpc;
+	struct mana_context *mc;
+	struct net_device *ndev;
+	struct mana_ib_cq *cq;
+	struct mana_ib_wq *wq;
+	unsigned int ind_tbl_size;
+	struct ib_cq *ibcq;
+	struct ib_wq *ibwq;
+	u32 port;
+	int ret;
+	int i;

This causes a build warning with clang:

../drivers/infiniband/hw/mana/qp.c:172:6: warning: variable 'i' is used
uninitialized whenever 'if' condition is true [-Wsometimes-uninitialized]
        if (!mana_ind_table) {
            ^~~~~~~~~~~~~~~
../drivers/infiniband/hw/mana/qp.c:241:9: note: uninitialized use occurs
here
        while (i-- > 0) {
               ^
../drivers/infiniband/hw/mana/qp.c:172:2: note: remove the 'if' if its
condition is always false
        if (!mana_ind_table) {
        ^~~~~~~~~~~~~~~~~~~~~~
../drivers/infiniband/hw/mana/qp.c:113:7: note: initialize the variable 'i' to
silence this warning
        int i;

Thank you. Will fix it.

quoted

+
+	mc = gd->driver_data;
+
+	if (!udata || udata->inlen < sizeof(ucmd))
+		return -EINVAL;
+
+	ret = ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata-
inlen));
+	if (ret) {
+		ibdev_dbg(&mdev->ib_dev,
+			  "Failed copy from udata for create rss-qp, err %d\n",
+			  ret);
+		return -EFAULT;
+	}
+
+	if (attr->cap.max_recv_wr > MAX_SEND_BUFFERS_PER_QUEUE) {
+		ibdev_dbg(&mdev->ib_dev,
+			  "Requested max_recv_wr %d exceeding limit\n",
+			  attr->cap.max_recv_wr);
+		return -EINVAL;
+	}
+
+	if (attr->cap.max_recv_sge > MAX_RX_WQE_SGL_ENTRIES) {
+		ibdev_dbg(&mdev->ib_dev,
+			  "Requested max_recv_sge %d exceeding limit\n",
+			  attr->cap.max_recv_sge);
+		return -EINVAL;
+	}
+
+	ind_tbl_size = 1 << ind_tbl->log_ind_tbl_size;
+	if (ind_tbl_size > MANA_INDIRECT_TABLE_SIZE) {
+		ibdev_dbg(&mdev->ib_dev,
+			  "Indirect table size %d exceeding limit\n",
+			  ind_tbl_size);
+		return -EINVAL;
+	}
+
+	if (ucmd.rx_hash_function != MANA_IB_RX_HASH_FUNC_TOEPLITZ)

{

quoted

+		ibdev_dbg(&mdev->ib_dev,
+			  "RX Hash function is not supported, %d\n",
+			  ucmd.rx_hash_function);
+		return -EINVAL;
+	}
+
+	/* IB ports start with 1, MANA start with 0 */
+	port = ucmd.port;
+	if (port < 1 || port > mc->num_ports) {
+		ibdev_dbg(&mdev->ib_dev, "Invalid port %u in creating

qp\n",

quoted

+			  port);
+		return -EINVAL;
+	}
+	ndev = mc->ports[port - 1];
+	mpc = netdev_priv(ndev);
+
+	ibdev_dbg(&mdev->ib_dev, "rx_hash_function %d port %d\n",
+		  ucmd.rx_hash_function, port);
+
+	mana_ind_table = kcalloc(ind_tbl_size, sizeof(mana_handle_t),
+				 GFP_KERNEL);
+	if (!mana_ind_table) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+
+	qp->port = port;
+
+	for (i = 0; i < ind_tbl_size; i++) {
+		struct mana_obj_spec wq_spec = {};
+		struct mana_obj_spec cq_spec = {};
+
+		ibwq = ind_tbl->ind_tbl[i];
+		wq = container_of(ibwq, struct mana_ib_wq, ibwq);
+
+		ibcq = ibwq->cq;
+		cq = container_of(ibcq, struct mana_ib_cq, ibcq);
+
+		wq_spec.gdma_region = wq->gdma_region;
+		wq_spec.queue_size = wq->wq_buf_size;
+
+		cq_spec.gdma_region = cq->gdma_region;
+		cq_spec.queue_size = cq->cqe * COMP_ENTRY_SIZE;
+		cq_spec.modr_ctx_id = 0;
+		cq_spec.attached_eq = GDMA_CQ_NO_EQ;
+
+		ret = mana_create_wq_obj(mpc, mpc->port_handle,

GDMA_RQ,

quoted

+					 &wq_spec, &cq_spec, &wq-
rx_object);
+		if (ret)
+			goto fail;
+
+		/* The GDMA regions are now owned by the WQ object */
+		wq->gdma_region = GDMA_INVALID_DMA_REGION;
+		cq->gdma_region = GDMA_INVALID_DMA_REGION;
+
+		wq->id = wq_spec.queue_index;
+		cq->id = cq_spec.queue_index;
+
+		ibdev_dbg(&mdev->ib_dev,
+			  "ret %d rx_object 0x%llx wq id %llu cq id %llu\n",
+			  ret, wq->rx_object, wq->id, cq->id);
+
+		resp.entries[i].cqid = cq->id;
+		resp.entries[i].wqid = wq->id;
+
+		mana_ind_table[i] = wq->rx_object;
+	}
+	resp.num_entries = i;
+
+	ret = mana_ib_cfg_vport_steering(mdev, ndev, wq->rx_object,
+					 mana_ind_table,
+					 ind_tbl->log_ind_tbl_size,
+					 ucmd.rx_hash_key_len,
+					 ucmd.rx_hash_key);
+	if (ret)
+		goto fail;
+
+	ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
+	if (ret) {
+		ibdev_dbg(&mdev->ib_dev,
+			  "Failed to copy to udata create rss-qp, %d\n",
+			  ret);
+		goto fail;
+	}
+
+	kfree(mana_ind_table);
+
+	return 0;
+
+fail:
+	while (i-- > 0) {
+		ibwq = ind_tbl->ind_tbl[i];
+		wq = container_of(ibwq, struct mana_ib_wq, ibwq);
+		mana_destroy_wq_obj(mpc, GDMA_RQ, wq->rx_object);
+	}
+
+	kfree(mana_ind_table);
+
+	return ret;
+}

Cheers,

Paolo