2 * Copyright (c) 2018, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 #include "port_buffer.h"
34 int mlx5e_port_query_buffer(struct mlx5e_priv *priv,
35 struct mlx5e_port_buffer *port_buffer)
37 u16 port_buff_cell_sz = priv->dcbx.port_buff_cell_sz;
38 struct mlx5_core_dev *mdev = priv->mdev;
39 int sz = MLX5_ST_SZ_BYTES(pbmc_reg);
46 out = kzalloc(sz, GFP_KERNEL);
50 err = mlx5e_port_query_pbmc(mdev, out);
54 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
55 buffer = MLX5_ADDR_OF(pbmc_reg, out, buffer[i]);
56 port_buffer->buffer[i].lossy =
57 MLX5_GET(bufferx_reg, buffer, lossy);
58 port_buffer->buffer[i].epsb =
59 MLX5_GET(bufferx_reg, buffer, epsb);
60 port_buffer->buffer[i].size =
61 MLX5_GET(bufferx_reg, buffer, size) * port_buff_cell_sz;
62 port_buffer->buffer[i].xon =
63 MLX5_GET(bufferx_reg, buffer, xon_threshold) * port_buff_cell_sz;
64 port_buffer->buffer[i].xoff =
65 MLX5_GET(bufferx_reg, buffer, xoff_threshold) * port_buff_cell_sz;
66 total_used += port_buffer->buffer[i].size;
68 netdev_dbg(priv->netdev, "buffer %d: size=%d, xon=%d, xoff=%d, epsb=%d, lossy=%d\n",
70 port_buffer->buffer[i].size,
71 port_buffer->buffer[i].xon,
72 port_buffer->buffer[i].xoff,
73 port_buffer->buffer[i].epsb,
74 port_buffer->buffer[i].lossy);
77 port_buffer->internal_buffers_size = 0;
78 for (i = MLX5E_MAX_NETWORK_BUFFER; i < MLX5E_TOTAL_BUFFERS; i++) {
79 buffer = MLX5_ADDR_OF(pbmc_reg, out, buffer[i]);
80 port_buffer->internal_buffers_size +=
81 MLX5_GET(bufferx_reg, buffer, size) * port_buff_cell_sz;
84 port_buffer->port_buffer_size =
85 MLX5_GET(pbmc_reg, out, port_buffer_size) * port_buff_cell_sz;
86 port_buffer->headroom_size = total_used;
87 port_buffer->spare_buffer_size = port_buffer->port_buffer_size -
88 port_buffer->internal_buffers_size -
89 port_buffer->headroom_size;
91 netdev_dbg(priv->netdev,
92 "total buffer size=%u, headroom buffer size=%u, internal buffers size=%u, spare buffer size=%u\n",
93 port_buffer->port_buffer_size, port_buffer->headroom_size,
94 port_buffer->internal_buffers_size,
95 port_buffer->spare_buffer_size);
101 struct mlx5e_buffer_pool {
107 static int mlx5e_port_query_pool(struct mlx5_core_dev *mdev,
108 struct mlx5e_buffer_pool *buffer_pool,
109 u32 desc, u8 dir, u8 pool_idx)
111 u32 out[MLX5_ST_SZ_DW(sbpr_reg)] = {};
114 err = mlx5e_port_query_sbpr(mdev, desc, dir, pool_idx, out,
119 buffer_pool->size = MLX5_GET(sbpr_reg, out, size);
120 buffer_pool->infi_size = MLX5_GET(sbpr_reg, out, infi_size);
121 buffer_pool->buff_occupancy = MLX5_GET(sbpr_reg, out, buff_occupancy);
127 MLX5_INGRESS_DIR = 0,
133 MLX5_LOSSLESS_POOL = 1,
136 /* No limit on usage of shared buffer pool (max_buff=0) */
137 #define MLX5_SB_POOL_NO_THRESHOLD 0
138 /* Shared buffer pool usage threshold when calculated
139 * dynamically in alpha units. alpha=13 is equivalent to
140 * HW_alpha of [(1/128) * 2 ^ (alpha-1)] = 32, where HW_alpha
141 * equates to the following portion of the shared buffer pool:
142 * [32 / (1 + n * 32)] While *n* is the number of buffers
143 * that are using the shared buffer pool.
145 #define MLX5_SB_POOL_THRESHOLD 13
147 /* Shared buffer class management parameters */
148 struct mlx5_sbcm_params {
154 static const struct mlx5_sbcm_params sbcm_default = {
155 .pool_idx = MLX5_LOSSY_POOL,
156 .max_buff = MLX5_SB_POOL_NO_THRESHOLD,
160 static const struct mlx5_sbcm_params sbcm_lossy = {
161 .pool_idx = MLX5_LOSSY_POOL,
162 .max_buff = MLX5_SB_POOL_NO_THRESHOLD,
166 static const struct mlx5_sbcm_params sbcm_lossless = {
167 .pool_idx = MLX5_LOSSLESS_POOL,
168 .max_buff = MLX5_SB_POOL_THRESHOLD,
172 static const struct mlx5_sbcm_params sbcm_lossless_no_threshold = {
173 .pool_idx = MLX5_LOSSLESS_POOL,
174 .max_buff = MLX5_SB_POOL_NO_THRESHOLD,
179 * select_sbcm_params() - selects the shared buffer pool configuration
181 * @buffer: <input> port buffer to retrieve params of
182 * @lossless_buff_count: <input> number of lossless buffers in total
184 * The selection is based on the following rules:
185 * 1. If buffer size is 0, no shared buffer pool is used.
186 * 2. If buffer is lossy, use lossy shared buffer pool.
187 * 3. If there are more than 1 lossless buffers, use lossless shared buffer pool
189 * 4. If there is only 1 lossless buffer, use lossless shared buffer pool
192 * @return const struct mlx5_sbcm_params* selected values
194 static const struct mlx5_sbcm_params *
195 select_sbcm_params(struct mlx5e_bufferx_reg *buffer, u8 lossless_buff_count)
197 if (buffer->size == 0)
198 return &sbcm_default;
203 if (lossless_buff_count > 1)
204 return &sbcm_lossless;
206 return &sbcm_lossless_no_threshold;
209 static int port_update_pool_cfg(struct mlx5_core_dev *mdev,
210 struct mlx5e_port_buffer *port_buffer)
212 const struct mlx5_sbcm_params *p;
213 u8 lossless_buff_count = 0;
217 if (!MLX5_CAP_GEN(mdev, sbcam_reg))
220 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++)
221 lossless_buff_count += ((port_buffer->buffer[i].size) &&
222 (!(port_buffer->buffer[i].lossy)));
224 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
225 p = select_sbcm_params(&port_buffer->buffer[i], lossless_buff_count);
226 err = mlx5e_port_set_sbcm(mdev, 0, i,
238 static int port_update_shared_buffer(struct mlx5_core_dev *mdev,
239 u32 current_headroom_size,
240 u32 new_headroom_size)
242 struct mlx5e_buffer_pool lossless_ipool;
243 struct mlx5e_buffer_pool lossy_epool;
244 u32 lossless_ipool_size;
245 u32 shared_buffer_size;
246 u32 total_buffer_size;
247 u32 lossy_epool_size;
250 if (!MLX5_CAP_GEN(mdev, sbcam_reg))
253 err = mlx5e_port_query_pool(mdev, &lossy_epool, 0, MLX5_EGRESS_DIR,
258 err = mlx5e_port_query_pool(mdev, &lossless_ipool, 0, MLX5_INGRESS_DIR,
263 total_buffer_size = current_headroom_size + lossy_epool.size +
265 shared_buffer_size = total_buffer_size - new_headroom_size;
267 if (shared_buffer_size < 4) {
268 pr_err("Requested port buffer is too large, not enough space left for shared buffer\n");
272 /* Total shared buffer size is split in a ratio of 3:1 between
273 * lossy and lossless pools respectively.
275 lossy_epool_size = (shared_buffer_size / 4) * 3;
276 lossless_ipool_size = shared_buffer_size / 4;
278 mlx5e_port_set_sbpr(mdev, 0, MLX5_EGRESS_DIR, MLX5_LOSSY_POOL, 0,
280 mlx5e_port_set_sbpr(mdev, 0, MLX5_INGRESS_DIR, MLX5_LOSSLESS_POOL, 0,
281 lossless_ipool_size);
285 static int port_set_buffer(struct mlx5e_priv *priv,
286 struct mlx5e_port_buffer *port_buffer)
288 u16 port_buff_cell_sz = priv->dcbx.port_buff_cell_sz;
289 struct mlx5_core_dev *mdev = priv->mdev;
290 int sz = MLX5_ST_SZ_BYTES(pbmc_reg);
291 u32 new_headroom_size = 0;
292 u32 current_headroom_size;
297 current_headroom_size = port_buffer->headroom_size;
299 in = kzalloc(sz, GFP_KERNEL);
303 err = mlx5e_port_query_pbmc(mdev, in);
307 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
308 void *buffer = MLX5_ADDR_OF(pbmc_reg, in, buffer[i]);
309 u64 size = port_buffer->buffer[i].size;
310 u64 xoff = port_buffer->buffer[i].xoff;
311 u64 xon = port_buffer->buffer[i].xon;
313 new_headroom_size += size;
314 do_div(size, port_buff_cell_sz);
315 do_div(xoff, port_buff_cell_sz);
316 do_div(xon, port_buff_cell_sz);
317 MLX5_SET(bufferx_reg, buffer, size, size);
318 MLX5_SET(bufferx_reg, buffer, lossy, port_buffer->buffer[i].lossy);
319 MLX5_SET(bufferx_reg, buffer, xoff_threshold, xoff);
320 MLX5_SET(bufferx_reg, buffer, xon_threshold, xon);
323 new_headroom_size /= port_buff_cell_sz;
324 current_headroom_size /= port_buff_cell_sz;
325 err = port_update_shared_buffer(priv->mdev, current_headroom_size,
330 err = port_update_pool_cfg(priv->mdev, port_buffer);
334 err = mlx5e_port_set_pbmc(mdev, in);
340 /* xoff = ((301+2.16 * len [m]) * speed [Gbps] + 2.72 MTU [B])
341 * minimum speed value is 40Gbps
343 static u32 calculate_xoff(struct mlx5e_priv *priv, unsigned int mtu)
349 err = mlx5e_port_linkspeed(priv->mdev, &speed);
352 speed = max_t(u32, speed, SPEED_40000);
354 xoff = (301 + 216 * priv->dcbx.cable_len / 100) * speed / 1000 + 272 * mtu / 100;
356 netdev_dbg(priv->netdev, "%s: xoff=%d\n", __func__, xoff);
360 static int update_xoff_threshold(struct mlx5e_port_buffer *port_buffer,
361 u32 xoff, unsigned int max_mtu, u16 port_buff_cell_sz)
365 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
366 if (port_buffer->buffer[i].lossy) {
367 port_buffer->buffer[i].xoff = 0;
368 port_buffer->buffer[i].xon = 0;
372 if (port_buffer->buffer[i].size <
373 (xoff + max_mtu + port_buff_cell_sz)) {
374 pr_err("buffer_size[%d]=%d is not enough for lossless buffer\n",
375 i, port_buffer->buffer[i].size);
379 port_buffer->buffer[i].xoff = port_buffer->buffer[i].size - xoff;
380 port_buffer->buffer[i].xon =
381 port_buffer->buffer[i].xoff - max_mtu;
388 * update_buffer_lossy - Update buffer configuration based on pfc
389 * @mdev: port function core device
390 * @max_mtu: netdev's max_mtu
391 * @pfc_en: <input> current pfc configuration
392 * @buffer: <input> current prio to buffer mapping
393 * @xoff: <input> xoff value
394 * @port_buff_cell_sz: <input> port buffer cell_size
395 * @port_buffer: <output> port receive buffer configuration
398 * Update buffer configuration based on pfc configuration and
399 * priority to buffer mapping.
400 * Buffer's lossy bit is changed to:
401 * lossless if there is at least one PFC enabled priority
402 * mapped to this buffer lossy if all priorities mapped to
403 * this buffer are PFC disabled
405 * @return: 0 if no error,
406 * sets change to true if buffer configuration was modified.
408 static int update_buffer_lossy(struct mlx5_core_dev *mdev,
409 unsigned int max_mtu,
410 u8 pfc_en, u8 *buffer, u32 xoff, u16 port_buff_cell_sz,
411 struct mlx5e_port_buffer *port_buffer,
414 bool changed = false;
422 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
426 for (prio = 0; prio < MLX5E_MAX_PRIORITY; prio++) {
427 if (buffer[prio] != i)
431 lossy_count += !(pfc_en & (1 << prio));
434 if (lossy_count == prio_count)
436 else /* lossy_count < prio_count */
439 if (lossy != port_buffer->buffer[i].lossy) {
440 port_buffer->buffer[i].lossy = lossy;
446 err = update_xoff_threshold(port_buffer, xoff, max_mtu, port_buff_cell_sz);
450 err = port_update_pool_cfg(mdev, port_buffer);
460 static int fill_pfc_en(struct mlx5_core_dev *mdev, u8 *pfc_en)
462 u32 g_rx_pause, g_tx_pause;
465 err = mlx5_query_port_pause(mdev, &g_rx_pause, &g_tx_pause);
469 /* If global pause enabled, set all active buffers to lossless.
470 * Otherwise, check PFC setting.
472 if (g_rx_pause || g_tx_pause)
475 err = mlx5_query_port_pfc(mdev, pfc_en, NULL);
480 #define MINIMUM_MAX_MTU 9216
481 int mlx5e_port_manual_buffer_config(struct mlx5e_priv *priv,
482 u32 change, unsigned int mtu,
483 struct ieee_pfc *pfc,
487 u16 port_buff_cell_sz = priv->dcbx.port_buff_cell_sz;
488 struct net_device *netdev = priv->netdev;
489 struct mlx5e_port_buffer port_buffer;
490 u32 xoff = calculate_xoff(priv, mtu);
491 bool update_prio2buffer = false;
492 u8 buffer[MLX5E_MAX_PRIORITY];
493 bool update_buffer = false;
494 unsigned int max_mtu;
500 netdev_dbg(netdev, "%s: change=%x\n", __func__, change);
501 max_mtu = max_t(unsigned int, priv->netdev->max_mtu, MINIMUM_MAX_MTU);
503 err = mlx5e_port_query_buffer(priv, &port_buffer);
507 if (change & MLX5E_PORT_BUFFER_CABLE_LEN) {
508 update_buffer = true;
509 err = update_xoff_threshold(&port_buffer, xoff, max_mtu, port_buff_cell_sz);
514 if (change & MLX5E_PORT_BUFFER_PFC) {
515 netdev_dbg(netdev, "%s: requested PFC per priority bitmask: 0x%x\n",
516 __func__, pfc->pfc_en);
517 err = mlx5e_port_query_priority2buffer(priv->mdev, buffer);
521 err = update_buffer_lossy(priv->mdev, max_mtu, pfc->pfc_en, buffer, xoff,
522 port_buff_cell_sz, &port_buffer,
528 if (change & MLX5E_PORT_BUFFER_PRIO2BUFFER) {
529 update_prio2buffer = true;
530 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++)
531 netdev_dbg(priv->netdev, "%s: requested to map prio[%d] to buffer %d\n",
532 __func__, i, prio2buffer[i]);
534 err = fill_pfc_en(priv->mdev, &curr_pfc_en);
538 err = update_buffer_lossy(priv->mdev, max_mtu, curr_pfc_en, prio2buffer, xoff,
539 port_buff_cell_sz, &port_buffer, &update_buffer);
544 if (change & MLX5E_PORT_BUFFER_SIZE) {
545 for (i = 0; i < MLX5E_MAX_NETWORK_BUFFER; i++) {
546 netdev_dbg(priv->netdev, "%s: buffer[%d]=%d\n", __func__, i, buffer_size[i]);
547 if (!port_buffer.buffer[i].lossy && !buffer_size[i]) {
548 netdev_dbg(priv->netdev, "%s: lossless buffer[%d] size cannot be zero\n",
553 port_buffer.buffer[i].size = buffer_size[i];
554 total_used += buffer_size[i];
557 netdev_dbg(priv->netdev, "%s: total buffer requested=%d\n", __func__, total_used);
559 if (total_used > port_buffer.headroom_size &&
560 (total_used - port_buffer.headroom_size) >
561 port_buffer.spare_buffer_size)
564 update_buffer = true;
565 err = update_xoff_threshold(&port_buffer, xoff, max_mtu, port_buff_cell_sz);
570 /* Need to update buffer configuration if xoff value is changed */
571 if (!update_buffer && xoff != priv->dcbx.xoff) {
572 update_buffer = true;
573 err = update_xoff_threshold(&port_buffer, xoff, max_mtu, port_buff_cell_sz);
577 priv->dcbx.xoff = xoff;
579 /* Apply the settings */
581 err = port_set_buffer(priv, &port_buffer);
586 if (update_prio2buffer)
587 err = mlx5e_port_set_priority2buffer(priv->mdev, prio2buffer);