1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2015-2018 Mellanox Technologies. All rights reserved */
3 
4 #include <linux/kernel.h>
5 #include <linux/types.h>
6 #include <linux/dcbnl.h>
7 #include <linux/if_ether.h>
8 #include <linux/list.h>
9 #include <linux/netlink.h>
10 
11 #include "spectrum.h"
12 #include "core.h"
13 #include "port.h"
14 #include "reg.h"
15 
16 struct mlxsw_sp_sb_pr {
17 	enum mlxsw_reg_sbpr_mode mode;
18 	u32 size;
19 	u8 freeze_mode:1,
20 	   freeze_size:1;
21 };
22 
23 struct mlxsw_cp_sb_occ {
24 	u32 cur;
25 	u32 max;
26 };
27 
28 struct mlxsw_sp_sb_cm {
29 	u32 min_buff;
30 	u32 max_buff;
31 	u16 pool_index;
32 	struct mlxsw_cp_sb_occ occ;
33 	u8 freeze_pool:1,
34 	   freeze_thresh:1;
35 };
36 
37 #define MLXSW_SP_SB_INFI -1U
38 #define MLXSW_SP_SB_REST -2U
39 
40 struct mlxsw_sp_sb_pm {
41 	u32 min_buff;
42 	u32 max_buff;
43 	struct mlxsw_cp_sb_occ occ;
44 };
45 
46 struct mlxsw_sp_sb_mm {
47 	u32 min_buff;
48 	u32 max_buff;
49 	u16 pool_index;
50 };
51 
52 struct mlxsw_sp_sb_pool_des {
53 	enum mlxsw_reg_sbxx_dir dir;
54 	u8 pool;
55 };
56 
57 #define MLXSW_SP_SB_POOL_ING		0
58 #define MLXSW_SP_SB_POOL_EGR		4
59 #define MLXSW_SP_SB_POOL_EGR_MC		8
60 #define MLXSW_SP_SB_POOL_ING_CPU	9
61 #define MLXSW_SP_SB_POOL_EGR_CPU	10
62 
63 static const struct mlxsw_sp_sb_pool_des mlxsw_sp1_sb_pool_dess[] = {
64 	{MLXSW_REG_SBXX_DIR_INGRESS, 0},
65 	{MLXSW_REG_SBXX_DIR_INGRESS, 1},
66 	{MLXSW_REG_SBXX_DIR_INGRESS, 2},
67 	{MLXSW_REG_SBXX_DIR_INGRESS, 3},
68 	{MLXSW_REG_SBXX_DIR_EGRESS, 0},
69 	{MLXSW_REG_SBXX_DIR_EGRESS, 1},
70 	{MLXSW_REG_SBXX_DIR_EGRESS, 2},
71 	{MLXSW_REG_SBXX_DIR_EGRESS, 3},
72 	{MLXSW_REG_SBXX_DIR_EGRESS, 15},
73 	{MLXSW_REG_SBXX_DIR_INGRESS, 4},
74 	{MLXSW_REG_SBXX_DIR_EGRESS, 4},
75 };
76 
77 static const struct mlxsw_sp_sb_pool_des mlxsw_sp2_sb_pool_dess[] = {
78 	{MLXSW_REG_SBXX_DIR_INGRESS, 0},
79 	{MLXSW_REG_SBXX_DIR_INGRESS, 1},
80 	{MLXSW_REG_SBXX_DIR_INGRESS, 2},
81 	{MLXSW_REG_SBXX_DIR_INGRESS, 3},
82 	{MLXSW_REG_SBXX_DIR_EGRESS, 0},
83 	{MLXSW_REG_SBXX_DIR_EGRESS, 1},
84 	{MLXSW_REG_SBXX_DIR_EGRESS, 2},
85 	{MLXSW_REG_SBXX_DIR_EGRESS, 3},
86 	{MLXSW_REG_SBXX_DIR_EGRESS, 15},
87 	{MLXSW_REG_SBXX_DIR_INGRESS, 4},
88 	{MLXSW_REG_SBXX_DIR_EGRESS, 4},
89 };
90 
91 #define MLXSW_SP_SB_ING_TC_COUNT 8
92 #define MLXSW_SP_SB_EG_TC_COUNT 16
93 
94 struct mlxsw_sp_sb_port {
95 	struct mlxsw_sp_sb_cm ing_cms[MLXSW_SP_SB_ING_TC_COUNT];
96 	struct mlxsw_sp_sb_cm eg_cms[MLXSW_SP_SB_EG_TC_COUNT];
97 	struct mlxsw_sp_sb_pm *pms;
98 };
99 
100 struct mlxsw_sp_sb {
101 	struct mlxsw_sp_sb_pr *prs;
102 	struct mlxsw_sp_sb_port *ports;
103 	u32 cell_size;
104 	u32 max_headroom_cells;
105 	u64 sb_size;
106 };
107 
108 struct mlxsw_sp_sb_vals {
109 	unsigned int pool_count;
110 	const struct mlxsw_sp_sb_pool_des *pool_dess;
111 	const struct mlxsw_sp_sb_pm *pms;
112 	const struct mlxsw_sp_sb_pm *pms_cpu;
113 	const struct mlxsw_sp_sb_pr *prs;
114 	const struct mlxsw_sp_sb_mm *mms;
115 	const struct mlxsw_sp_sb_cm *cms_ingress;
116 	const struct mlxsw_sp_sb_cm *cms_egress;
117 	const struct mlxsw_sp_sb_cm *cms_cpu;
118 	unsigned int mms_count;
119 	unsigned int cms_ingress_count;
120 	unsigned int cms_egress_count;
121 	unsigned int cms_cpu_count;
122 };
123 
124 struct mlxsw_sp_sb_ops {
125 	u32 (*int_buf_size_get)(int mtu, u32 speed);
126 };
127 
128 u32 mlxsw_sp_cells_bytes(const struct mlxsw_sp *mlxsw_sp, u32 cells)
129 {
130 	return mlxsw_sp->sb->cell_size * cells;
131 }
132 
133 u32 mlxsw_sp_bytes_cells(const struct mlxsw_sp *mlxsw_sp, u32 bytes)
134 {
135 	return DIV_ROUND_UP(bytes, mlxsw_sp->sb->cell_size);
136 }
137 
138 static u32 mlxsw_sp_port_headroom_8x_adjust(const struct mlxsw_sp_port *mlxsw_sp_port,
139 					    u32 size_cells)
140 {
141 	/* Ports with eight lanes use two headroom buffers between which the
142 	 * configured headroom size is split. Therefore, multiply the calculated
143 	 * headroom size by two.
144 	 */
145 	return mlxsw_sp_port->mapping.width == 8 ? 2 * size_cells : size_cells;
146 }
147 
148 static struct mlxsw_sp_sb_pr *mlxsw_sp_sb_pr_get(struct mlxsw_sp *mlxsw_sp,
149 						 u16 pool_index)
150 {
151 	return &mlxsw_sp->sb->prs[pool_index];
152 }
153 
154 static bool mlxsw_sp_sb_cm_exists(u8 pg_buff, enum mlxsw_reg_sbxx_dir dir)
155 {
156 	if (dir == MLXSW_REG_SBXX_DIR_INGRESS)
157 		return pg_buff < MLXSW_SP_SB_ING_TC_COUNT;
158 	else
159 		return pg_buff < MLXSW_SP_SB_EG_TC_COUNT;
160 }
161 
162 static struct mlxsw_sp_sb_cm *mlxsw_sp_sb_cm_get(struct mlxsw_sp *mlxsw_sp,
163 						 u16 local_port, u8 pg_buff,
164 						 enum mlxsw_reg_sbxx_dir dir)
165 {
166 	struct mlxsw_sp_sb_port *sb_port = &mlxsw_sp->sb->ports[local_port];
167 
168 	WARN_ON(!mlxsw_sp_sb_cm_exists(pg_buff, dir));
169 	if (dir == MLXSW_REG_SBXX_DIR_INGRESS)
170 		return &sb_port->ing_cms[pg_buff];
171 	else
172 		return &sb_port->eg_cms[pg_buff];
173 }
174 
175 static struct mlxsw_sp_sb_pm *mlxsw_sp_sb_pm_get(struct mlxsw_sp *mlxsw_sp,
176 						 u16 local_port, u16 pool_index)
177 {
178 	return &mlxsw_sp->sb->ports[local_port].pms[pool_index];
179 }
180 
181 static int mlxsw_sp_sb_pr_write(struct mlxsw_sp *mlxsw_sp, u16 pool_index,
182 				enum mlxsw_reg_sbpr_mode mode,
183 				u32 size, bool infi_size)
184 {
185 	const struct mlxsw_sp_sb_pool_des *des =
186 		&mlxsw_sp->sb_vals->pool_dess[pool_index];
187 	char sbpr_pl[MLXSW_REG_SBPR_LEN];
188 	struct mlxsw_sp_sb_pr *pr;
189 	int err;
190 
191 	mlxsw_reg_sbpr_pack(sbpr_pl, des->pool, des->dir, mode,
192 			    size, infi_size);
193 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sbpr), sbpr_pl);
194 	if (err)
195 		return err;
196 
197 	if (infi_size)
198 		size = mlxsw_sp_bytes_cells(mlxsw_sp, mlxsw_sp->sb->sb_size);
199 	pr = mlxsw_sp_sb_pr_get(mlxsw_sp, pool_index);
200 	pr->mode = mode;
201 	pr->size = size;
202 	return 0;
203 }
204 
205 static int mlxsw_sp_sb_pr_desc_write(struct mlxsw_sp *mlxsw_sp,
206 				     enum mlxsw_reg_sbxx_dir dir,
207 				     enum mlxsw_reg_sbpr_mode mode,
208 				     u32 size, bool infi_size)
209 {
210 	char sbpr_pl[MLXSW_REG_SBPR_LEN];
211 
212 	/* The FW default descriptor buffer configuration uses only pool 14 for
213 	 * descriptors.
214 	 */
215 	mlxsw_reg_sbpr_pack(sbpr_pl, 14, dir, mode, size, infi_size);
216 	mlxsw_reg_sbpr_desc_set(sbpr_pl, true);
217 	return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sbpr), sbpr_pl);
218 }
219 
220 static int mlxsw_sp_sb_cm_write(struct mlxsw_sp *mlxsw_sp, u16 local_port,
221 				u8 pg_buff, u32 min_buff, u32 max_buff,
222 				bool infi_max, u16 pool_index)
223 {
224 	const struct mlxsw_sp_sb_pool_des *des =
225 		&mlxsw_sp->sb_vals->pool_dess[pool_index];
226 	char sbcm_pl[MLXSW_REG_SBCM_LEN];
227 	struct mlxsw_sp_sb_cm *cm;
228 	int err;
229 
230 	mlxsw_reg_sbcm_pack(sbcm_pl, local_port, pg_buff, des->dir,
231 			    min_buff, max_buff, infi_max, des->pool);
232 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sbcm), sbcm_pl);
233 	if (err)
234 		return err;
235 
236 	if (mlxsw_sp_sb_cm_exists(pg_buff, des->dir)) {
237 		if (infi_max)
238 			max_buff = mlxsw_sp_bytes_cells(mlxsw_sp,
239 							mlxsw_sp->sb->sb_size);
240 
241 		cm = mlxsw_sp_sb_cm_get(mlxsw_sp, local_port, pg_buff,
242 					des->dir);
243 		cm->min_buff = min_buff;
244 		cm->max_buff = max_buff;
245 		cm->pool_index = pool_index;
246 	}
247 	return 0;
248 }
249 
250 static int mlxsw_sp_sb_pm_write(struct mlxsw_sp *mlxsw_sp, u16 local_port,
251 				u16 pool_index, u32 min_buff, u32 max_buff)
252 {
253 	const struct mlxsw_sp_sb_pool_des *des =
254 		&mlxsw_sp->sb_vals->pool_dess[pool_index];
255 	char sbpm_pl[MLXSW_REG_SBPM_LEN];
256 	struct mlxsw_sp_sb_pm *pm;
257 	int err;
258 
259 	mlxsw_reg_sbpm_pack(sbpm_pl, local_port, des->pool, des->dir, false,
260 			    min_buff, max_buff);
261 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sbpm), sbpm_pl);
262 	if (err)
263 		return err;
264 
265 	pm = mlxsw_sp_sb_pm_get(mlxsw_sp, local_port, pool_index);
266 	pm->min_buff = min_buff;
267 	pm->max_buff = max_buff;
268 	return 0;
269 }
270 
271 static int mlxsw_sp_sb_pm_occ_clear(struct mlxsw_sp *mlxsw_sp, u16 local_port,
272 				    u16 pool_index, struct list_head *bulk_list)
273 {
274 	const struct mlxsw_sp_sb_pool_des *des =
275 		&mlxsw_sp->sb_vals->pool_dess[pool_index];
276 	char sbpm_pl[MLXSW_REG_SBPM_LEN];
277 
278 	if (local_port == MLXSW_PORT_CPU_PORT &&
279 	    des->dir == MLXSW_REG_SBXX_DIR_INGRESS)
280 		return 0;
281 
282 	mlxsw_reg_sbpm_pack(sbpm_pl, local_port, des->pool, des->dir,
283 			    true, 0, 0);
284 	return mlxsw_reg_trans_query(mlxsw_sp->core, MLXSW_REG(sbpm), sbpm_pl,
285 				     bulk_list, NULL, 0);
286 }
287 
288 static void mlxsw_sp_sb_pm_occ_query_cb(struct mlxsw_core *mlxsw_core,
289 					char *sbpm_pl, size_t sbpm_pl_len,
290 					unsigned long cb_priv)
291 {
292 	struct mlxsw_sp_sb_pm *pm = (struct mlxsw_sp_sb_pm *) cb_priv;
293 
294 	mlxsw_reg_sbpm_unpack(sbpm_pl, &pm->occ.cur, &pm->occ.max);
295 }
296 
297 static int mlxsw_sp_sb_pm_occ_query(struct mlxsw_sp *mlxsw_sp, u16 local_port,
298 				    u16 pool_index, struct list_head *bulk_list)
299 {
300 	const struct mlxsw_sp_sb_pool_des *des =
301 		&mlxsw_sp->sb_vals->pool_dess[pool_index];
302 	char sbpm_pl[MLXSW_REG_SBPM_LEN];
303 	struct mlxsw_sp_sb_pm *pm;
304 
305 	if (local_port == MLXSW_PORT_CPU_PORT &&
306 	    des->dir == MLXSW_REG_SBXX_DIR_INGRESS)
307 		return 0;
308 
309 	pm = mlxsw_sp_sb_pm_get(mlxsw_sp, local_port, pool_index);
310 	mlxsw_reg_sbpm_pack(sbpm_pl, local_port, des->pool, des->dir,
311 			    false, 0, 0);
312 	return mlxsw_reg_trans_query(mlxsw_sp->core, MLXSW_REG(sbpm), sbpm_pl,
313 				     bulk_list,
314 				     mlxsw_sp_sb_pm_occ_query_cb,
315 				     (unsigned long) pm);
316 }
317 
318 void mlxsw_sp_hdroom_prios_reset_buf_idx(struct mlxsw_sp_hdroom *hdroom)
319 {
320 	int prio;
321 
322 	for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; prio++) {
323 		switch (hdroom->mode) {
324 		case MLXSW_SP_HDROOM_MODE_DCB:
325 			hdroom->prios.prio[prio].buf_idx = hdroom->prios.prio[prio].ets_buf_idx;
326 			break;
327 		case MLXSW_SP_HDROOM_MODE_TC:
328 			hdroom->prios.prio[prio].buf_idx = hdroom->prios.prio[prio].set_buf_idx;
329 			break;
330 		}
331 	}
332 }
333 
334 void mlxsw_sp_hdroom_bufs_reset_lossiness(struct mlxsw_sp_hdroom *hdroom)
335 {
336 	int prio;
337 	int i;
338 
339 	for (i = 0; i < DCBX_MAX_BUFFERS; i++)
340 		hdroom->bufs.buf[i].lossy = true;
341 
342 	for (prio = 0; prio < IEEE_8021Q_MAX_PRIORITIES; prio++) {
343 		if (!hdroom->prios.prio[prio].lossy)
344 			hdroom->bufs.buf[hdroom->prios.prio[prio].buf_idx].lossy = false;
345 	}
346 }
347 
348 static u16 mlxsw_sp_hdroom_buf_threshold_get(const struct mlxsw_sp *mlxsw_sp, int mtu)
349 {
350 	return 2 * mlxsw_sp_bytes_cells(mlxsw_sp, mtu);
351 }
352 
353 static void mlxsw_sp_hdroom_buf_pack(char *pbmc_pl, int index, u16 size, u16 thres, bool lossy)
354 {
355 	if (lossy)
356 		mlxsw_reg_pbmc_lossy_buffer_pack(pbmc_pl, index, size);
357 	else
358 		mlxsw_reg_pbmc_lossless_buffer_pack(pbmc_pl, index, size,
359 						    thres);
360 }
361 
362 static u16 mlxsw_sp_hdroom_buf_delay_get(const struct mlxsw_sp *mlxsw_sp,
363 					 const struct mlxsw_sp_hdroom *hdroom)
364 {
365 	u16 delay_cells;
366 
367 	delay_cells = mlxsw_sp_bytes_cells(mlxsw_sp, hdroom->delay_bytes);
368 
369 	/* In the worst case scenario the delay will be made up of packets that
370 	 * are all of size CELL_SIZE + 1, which means each packet will require
371 	 * almost twice its true size when buffered in the switch. We therefore
372 	 * multiply this value by the "cell factor", which is close to 2.
373 	 *
374 	 * Another MTU is added in case the transmitting host already started
375 	 * transmitting a maximum length frame when the PFC packet was received.
376 	 */
377 	return 2 * delay_cells + mlxsw_sp_bytes_cells(mlxsw_sp, hdroom->mtu);
378 }
379 
380 static u32 mlxsw_sp_hdroom_int_buf_size_get(struct mlxsw_sp *mlxsw_sp, int mtu, u32 speed)
381 {
382 	u32 buffsize = mlxsw_sp->sb_ops->int_buf_size_get(mtu, speed);
383 
384 	return mlxsw_sp_bytes_cells(mlxsw_sp, buffsize) + 1;
385 }
386 
387 static bool mlxsw_sp_hdroom_buf_is_used(const struct mlxsw_sp_hdroom *hdroom, int buf)
388 {
389 	int prio;
390 
391 	for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; prio++) {
392 		if (hdroom->prios.prio[prio].buf_idx == buf)
393 			return true;
394 	}
395 	return false;
396 }
397 
398 void mlxsw_sp_hdroom_bufs_reset_sizes(struct mlxsw_sp_port *mlxsw_sp_port,
399 				      struct mlxsw_sp_hdroom *hdroom)
400 {
401 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
402 	u16 reserve_cells;
403 	int i;
404 
405 	/* Internal buffer. */
406 	reserve_cells = mlxsw_sp_hdroom_int_buf_size_get(mlxsw_sp, mlxsw_sp_port->max_mtu,
407 							 mlxsw_sp_port->max_speed);
408 	reserve_cells = mlxsw_sp_port_headroom_8x_adjust(mlxsw_sp_port, reserve_cells);
409 	hdroom->int_buf.reserve_cells = reserve_cells;
410 
411 	if (hdroom->int_buf.enable)
412 		hdroom->int_buf.size_cells = reserve_cells;
413 	else
414 		hdroom->int_buf.size_cells = 0;
415 
416 	/* PG buffers. */
417 	for (i = 0; i < DCBX_MAX_BUFFERS; i++) {
418 		struct mlxsw_sp_hdroom_buf *buf = &hdroom->bufs.buf[i];
419 		u16 thres_cells;
420 		u16 delay_cells;
421 
422 		if (!mlxsw_sp_hdroom_buf_is_used(hdroom, i)) {
423 			thres_cells = 0;
424 			delay_cells = 0;
425 		} else if (buf->lossy) {
426 			thres_cells = mlxsw_sp_hdroom_buf_threshold_get(mlxsw_sp, hdroom->mtu);
427 			delay_cells = 0;
428 		} else {
429 			thres_cells = mlxsw_sp_hdroom_buf_threshold_get(mlxsw_sp, hdroom->mtu);
430 			delay_cells = mlxsw_sp_hdroom_buf_delay_get(mlxsw_sp, hdroom);
431 		}
432 
433 		thres_cells = mlxsw_sp_port_headroom_8x_adjust(mlxsw_sp_port, thres_cells);
434 		delay_cells = mlxsw_sp_port_headroom_8x_adjust(mlxsw_sp_port, delay_cells);
435 
436 		buf->thres_cells = thres_cells;
437 		if (hdroom->mode == MLXSW_SP_HDROOM_MODE_DCB) {
438 			buf->size_cells = thres_cells + delay_cells;
439 		} else {
440 			/* Do not allow going below the minimum size, even if
441 			 * the user requested it.
442 			 */
443 			buf->size_cells = max(buf->set_size_cells, buf->thres_cells);
444 		}
445 	}
446 }
447 
448 #define MLXSW_SP_PB_UNUSED 8
449 
450 static int mlxsw_sp_hdroom_configure_buffers(struct mlxsw_sp_port *mlxsw_sp_port,
451 					     const struct mlxsw_sp_hdroom *hdroom, bool force)
452 {
453 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
454 	char pbmc_pl[MLXSW_REG_PBMC_LEN];
455 	bool dirty;
456 	int err;
457 	int i;
458 
459 	dirty = memcmp(&mlxsw_sp_port->hdroom->bufs, &hdroom->bufs, sizeof(hdroom->bufs));
460 	if (!dirty && !force)
461 		return 0;
462 
463 	mlxsw_reg_pbmc_pack(pbmc_pl, mlxsw_sp_port->local_port, 0xffff, 0xffff / 2);
464 	for (i = 0; i < MLXSW_SP_PB_COUNT; i++) {
465 		const struct mlxsw_sp_hdroom_buf *buf = &hdroom->bufs.buf[i];
466 
467 		if (i == MLXSW_SP_PB_UNUSED)
468 			continue;
469 
470 		mlxsw_sp_hdroom_buf_pack(pbmc_pl, i, buf->size_cells, buf->thres_cells, buf->lossy);
471 	}
472 
473 	mlxsw_reg_pbmc_lossy_buffer_pack(pbmc_pl, MLXSW_REG_PBMC_PORT_SHARED_BUF_IDX, 0);
474 	err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(pbmc), pbmc_pl);
475 	if (err)
476 		return err;
477 
478 	mlxsw_sp_port->hdroom->bufs = hdroom->bufs;
479 	return 0;
480 }
481 
482 static int mlxsw_sp_hdroom_configure_priomap(struct mlxsw_sp_port *mlxsw_sp_port,
483 					     const struct mlxsw_sp_hdroom *hdroom, bool force)
484 {
485 	char pptb_pl[MLXSW_REG_PPTB_LEN];
486 	bool dirty;
487 	int prio;
488 	int err;
489 
490 	dirty = memcmp(&mlxsw_sp_port->hdroom->prios, &hdroom->prios, sizeof(hdroom->prios));
491 	if (!dirty && !force)
492 		return 0;
493 
494 	mlxsw_reg_pptb_pack(pptb_pl, mlxsw_sp_port->local_port);
495 	for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; prio++)
496 		mlxsw_reg_pptb_prio_to_buff_pack(pptb_pl, prio, hdroom->prios.prio[prio].buf_idx);
497 
498 	err = mlxsw_reg_write(mlxsw_sp_port->mlxsw_sp->core, MLXSW_REG(pptb), pptb_pl);
499 	if (err)
500 		return err;
501 
502 	mlxsw_sp_port->hdroom->prios = hdroom->prios;
503 	return 0;
504 }
505 
506 static int mlxsw_sp_hdroom_configure_int_buf(struct mlxsw_sp_port *mlxsw_sp_port,
507 					     const struct mlxsw_sp_hdroom *hdroom, bool force)
508 {
509 	char sbib_pl[MLXSW_REG_SBIB_LEN];
510 	bool dirty;
511 	int err;
512 
513 	dirty = memcmp(&mlxsw_sp_port->hdroom->int_buf, &hdroom->int_buf, sizeof(hdroom->int_buf));
514 	if (!dirty && !force)
515 		return 0;
516 
517 	mlxsw_reg_sbib_pack(sbib_pl, mlxsw_sp_port->local_port, hdroom->int_buf.size_cells);
518 	err = mlxsw_reg_write(mlxsw_sp_port->mlxsw_sp->core, MLXSW_REG(sbib), sbib_pl);
519 	if (err)
520 		return err;
521 
522 	mlxsw_sp_port->hdroom->int_buf = hdroom->int_buf;
523 	return 0;
524 }
525 
526 static bool mlxsw_sp_hdroom_bufs_fit(struct mlxsw_sp *mlxsw_sp,
527 				     const struct mlxsw_sp_hdroom *hdroom)
528 {
529 	u32 taken_headroom_cells = 0;
530 	int i;
531 
532 	for (i = 0; i < MLXSW_SP_PB_COUNT; i++)
533 		taken_headroom_cells += hdroom->bufs.buf[i].size_cells;
534 
535 	taken_headroom_cells += hdroom->int_buf.reserve_cells;
536 	return taken_headroom_cells <= mlxsw_sp->sb->max_headroom_cells;
537 }
538 
539 static int __mlxsw_sp_hdroom_configure(struct mlxsw_sp_port *mlxsw_sp_port,
540 				       const struct mlxsw_sp_hdroom *hdroom, bool force)
541 {
542 	struct mlxsw_sp_hdroom orig_hdroom;
543 	struct mlxsw_sp_hdroom tmp_hdroom;
544 	int err;
545 	int i;
546 
547 	/* Port buffers need to be configured in three steps. First, all buffers
548 	 * with non-zero size are configured. Then, prio-to-buffer map is
549 	 * updated, allowing traffic to flow to the now non-zero buffers.
550 	 * Finally, zero-sized buffers are configured, because now no traffic
551 	 * should be directed to them anymore. This way, in a non-congested
552 	 * system, no packet drops are introduced by the reconfiguration.
553 	 */
554 
555 	orig_hdroom = *mlxsw_sp_port->hdroom;
556 	tmp_hdroom = orig_hdroom;
557 	for (i = 0; i < MLXSW_SP_PB_COUNT; i++) {
558 		if (hdroom->bufs.buf[i].size_cells)
559 			tmp_hdroom.bufs.buf[i] = hdroom->bufs.buf[i];
560 	}
561 
562 	if (!mlxsw_sp_hdroom_bufs_fit(mlxsw_sp_port->mlxsw_sp, &tmp_hdroom) ||
563 	    !mlxsw_sp_hdroom_bufs_fit(mlxsw_sp_port->mlxsw_sp, hdroom))
564 		return -ENOBUFS;
565 
566 	err = mlxsw_sp_hdroom_configure_buffers(mlxsw_sp_port, &tmp_hdroom, force);
567 	if (err)
568 		return err;
569 
570 	err = mlxsw_sp_hdroom_configure_priomap(mlxsw_sp_port, hdroom, force);
571 	if (err)
572 		goto err_configure_priomap;
573 
574 	err = mlxsw_sp_hdroom_configure_buffers(mlxsw_sp_port, hdroom, false);
575 	if (err)
576 		goto err_configure_buffers;
577 
578 	err = mlxsw_sp_hdroom_configure_int_buf(mlxsw_sp_port, hdroom, false);
579 	if (err)
580 		goto err_configure_int_buf;
581 
582 	*mlxsw_sp_port->hdroom = *hdroom;
583 	return 0;
584 
585 err_configure_int_buf:
586 	mlxsw_sp_hdroom_configure_buffers(mlxsw_sp_port, &tmp_hdroom, false);
587 err_configure_buffers:
588 	mlxsw_sp_hdroom_configure_priomap(mlxsw_sp_port, &tmp_hdroom, false);
589 err_configure_priomap:
590 	mlxsw_sp_hdroom_configure_buffers(mlxsw_sp_port, &orig_hdroom, false);
591 	return err;
592 }
593 
594 int mlxsw_sp_hdroom_configure(struct mlxsw_sp_port *mlxsw_sp_port,
595 			      const struct mlxsw_sp_hdroom *hdroom)
596 {
597 	return __mlxsw_sp_hdroom_configure(mlxsw_sp_port, hdroom, false);
598 }
599 
600 static int mlxsw_sp_port_headroom_init(struct mlxsw_sp_port *mlxsw_sp_port)
601 {
602 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
603 	struct mlxsw_sp_hdroom hdroom = {};
604 	u32 size9;
605 	int prio;
606 
607 	hdroom.mtu = mlxsw_sp_port->dev->mtu;
608 	hdroom.mode = MLXSW_SP_HDROOM_MODE_DCB;
609 	for (prio = 0; prio < IEEE_8021QAZ_MAX_TCS; prio++)
610 		hdroom.prios.prio[prio].lossy = true;
611 
612 	mlxsw_sp_hdroom_bufs_reset_lossiness(&hdroom);
613 	mlxsw_sp_hdroom_bufs_reset_sizes(mlxsw_sp_port, &hdroom);
614 
615 	/* Buffer 9 is used for control traffic. */
616 	size9 = mlxsw_sp_port_headroom_8x_adjust(mlxsw_sp_port, mlxsw_sp_port->max_mtu);
617 	hdroom.bufs.buf[9].size_cells = mlxsw_sp_bytes_cells(mlxsw_sp, size9);
618 
619 	return __mlxsw_sp_hdroom_configure(mlxsw_sp_port, &hdroom, true);
620 }
621 
622 static int mlxsw_sp_sb_port_init(struct mlxsw_sp *mlxsw_sp,
623 				 struct mlxsw_sp_sb_port *sb_port)
624 {
625 	struct mlxsw_sp_sb_pm *pms;
626 
627 	pms = kcalloc(mlxsw_sp->sb_vals->pool_count, sizeof(*pms),
628 		      GFP_KERNEL);
629 	if (!pms)
630 		return -ENOMEM;
631 	sb_port->pms = pms;
632 	return 0;
633 }
634 
635 static void mlxsw_sp_sb_port_fini(struct mlxsw_sp_sb_port *sb_port)
636 {
637 	kfree(sb_port->pms);
638 }
639 
640 static int mlxsw_sp_sb_ports_init(struct mlxsw_sp *mlxsw_sp)
641 {
642 	unsigned int max_ports = mlxsw_core_max_ports(mlxsw_sp->core);
643 	struct mlxsw_sp_sb_pr *prs;
644 	int i;
645 	int err;
646 
647 	mlxsw_sp->sb->ports = kcalloc(max_ports,
648 				      sizeof(struct mlxsw_sp_sb_port),
649 				      GFP_KERNEL);
650 	if (!mlxsw_sp->sb->ports)
651 		return -ENOMEM;
652 
653 	prs = kcalloc(mlxsw_sp->sb_vals->pool_count, sizeof(*prs),
654 		      GFP_KERNEL);
655 	if (!prs) {
656 		err = -ENOMEM;
657 		goto err_alloc_prs;
658 	}
659 	mlxsw_sp->sb->prs = prs;
660 
661 	for (i = 0; i < max_ports; i++) {
662 		err = mlxsw_sp_sb_port_init(mlxsw_sp, &mlxsw_sp->sb->ports[i]);
663 		if (err)
664 			goto err_sb_port_init;
665 	}
666 
667 	return 0;
668 
669 err_sb_port_init:
670 	for (i--; i >= 0; i--)
671 		mlxsw_sp_sb_port_fini(&mlxsw_sp->sb->ports[i]);
672 	kfree(mlxsw_sp->sb->prs);
673 err_alloc_prs:
674 	kfree(mlxsw_sp->sb->ports);
675 	return err;
676 }
677 
678 static void mlxsw_sp_sb_ports_fini(struct mlxsw_sp *mlxsw_sp)
679 {
680 	int max_ports = mlxsw_core_max_ports(mlxsw_sp->core);
681 	int i;
682 
683 	for (i = max_ports - 1; i >= 0; i--)
684 		mlxsw_sp_sb_port_fini(&mlxsw_sp->sb->ports[i]);
685 	kfree(mlxsw_sp->sb->prs);
686 	kfree(mlxsw_sp->sb->ports);
687 }
688 
689 #define MLXSW_SP_SB_PR(_mode, _size)	\
690 	{				\
691 		.mode = _mode,		\
692 		.size = _size,		\
693 	}
694 
695 #define MLXSW_SP_SB_PR_EXT(_mode, _size, _freeze_mode, _freeze_size)	\
696 	{								\
697 		.mode = _mode,						\
698 		.size = _size,						\
699 		.freeze_mode = _freeze_mode,				\
700 		.freeze_size = _freeze_size,				\
701 	}
702 
703 #define MLXSW_SP1_SB_PR_CPU_SIZE	(256 * 1000)
704 
705 /* Order according to mlxsw_sp1_sb_pool_dess */
706 static const struct mlxsw_sp_sb_pr mlxsw_sp1_sb_prs[] = {
707 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, MLXSW_SP_SB_REST),
708 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
709 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
710 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
711 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC, MLXSW_SP_SB_REST,
712 			   true, false),
713 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
714 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
715 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, 0),
716 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_STATIC, MLXSW_SP_SB_INFI,
717 			   true, true),
718 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC,
719 			   MLXSW_SP1_SB_PR_CPU_SIZE, true, false),
720 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC,
721 			   MLXSW_SP1_SB_PR_CPU_SIZE, true, false),
722 };
723 
724 #define MLXSW_SP2_SB_PR_CPU_SIZE	(256 * 1000)
725 
726 /* Order according to mlxsw_sp2_sb_pool_dess */
727 static const struct mlxsw_sp_sb_pr mlxsw_sp2_sb_prs[] = {
728 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_DYNAMIC, MLXSW_SP_SB_REST),
729 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
730 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
731 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
732 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC, MLXSW_SP_SB_REST,
733 			   true, false),
734 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
735 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
736 	MLXSW_SP_SB_PR(MLXSW_REG_SBPR_MODE_STATIC, 0),
737 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_STATIC, MLXSW_SP_SB_INFI,
738 			   true, true),
739 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC,
740 			   MLXSW_SP2_SB_PR_CPU_SIZE, true, false),
741 	MLXSW_SP_SB_PR_EXT(MLXSW_REG_SBPR_MODE_DYNAMIC,
742 			   MLXSW_SP2_SB_PR_CPU_SIZE, true, false),
743 };
744 
745 static int mlxsw_sp_sb_prs_init(struct mlxsw_sp *mlxsw_sp,
746 				const struct mlxsw_sp_sb_pr *prs,
747 				const struct mlxsw_sp_sb_pool_des *pool_dess,
748 				size_t prs_len)
749 {
750 	/* Round down, unlike mlxsw_sp_bytes_cells(). */
751 	u32 sb_cells = div_u64(mlxsw_sp->sb->sb_size, mlxsw_sp->sb->cell_size);
752 	u32 rest_cells[2] = {sb_cells, sb_cells};
753 	int i;
754 	int err;
755 
756 	/* Calculate how much space to give to the "REST" pools in either
757 	 * direction.
758 	 */
759 	for (i = 0; i < prs_len; i++) {
760 		enum mlxsw_reg_sbxx_dir dir = pool_dess[i].dir;
761 		u32 size = prs[i].size;
762 		u32 size_cells;
763 
764 		if (size == MLXSW_SP_SB_INFI || size == MLXSW_SP_SB_REST)
765 			continue;
766 
767 		size_cells = mlxsw_sp_bytes_cells(mlxsw_sp, size);
768 		if (WARN_ON_ONCE(size_cells > rest_cells[dir]))
769 			continue;
770 
771 		rest_cells[dir] -= size_cells;
772 	}
773 
774 	for (i = 0; i < prs_len; i++) {
775 		u32 size = prs[i].size;
776 		u32 size_cells;
777 
778 		if (size == MLXSW_SP_SB_INFI) {
779 			err = mlxsw_sp_sb_pr_write(mlxsw_sp, i, prs[i].mode,
780 						   0, true);
781 		} else if (size == MLXSW_SP_SB_REST) {
782 			size_cells = rest_cells[pool_dess[i].dir];
783 			err = mlxsw_sp_sb_pr_write(mlxsw_sp, i, prs[i].mode,
784 						   size_cells, false);
785 		} else {
786 			size_cells = mlxsw_sp_bytes_cells(mlxsw_sp, size);
787 			err = mlxsw_sp_sb_pr_write(mlxsw_sp, i, prs[i].mode,
788 						   size_cells, false);
789 		}
790 		if (err)
791 			return err;
792 	}
793 
794 	err = mlxsw_sp_sb_pr_desc_write(mlxsw_sp, MLXSW_REG_SBXX_DIR_INGRESS,
795 					MLXSW_REG_SBPR_MODE_DYNAMIC, 0, true);
796 	if (err)
797 		return err;
798 
799 	err = mlxsw_sp_sb_pr_desc_write(mlxsw_sp, MLXSW_REG_SBXX_DIR_EGRESS,
800 					MLXSW_REG_SBPR_MODE_DYNAMIC, 0, true);
801 	if (err)
802 		return err;
803 
804 	return 0;
805 }
806 
807 #define MLXSW_SP_SB_CM(_min_buff, _max_buff, _pool)	\
808 	{						\
809 		.min_buff = _min_buff,			\
810 		.max_buff = _max_buff,			\
811 		.pool_index = _pool,			\
812 	}
813 
814 #define MLXSW_SP_SB_CM_ING(_min_buff, _max_buff)	\
815 	{						\
816 		.min_buff = _min_buff,			\
817 		.max_buff = _max_buff,			\
818 		.pool_index = MLXSW_SP_SB_POOL_ING,	\
819 	}
820 
821 #define MLXSW_SP_SB_CM_EGR(_min_buff, _max_buff)	\
822 	{						\
823 		.min_buff = _min_buff,			\
824 		.max_buff = _max_buff,			\
825 		.pool_index = MLXSW_SP_SB_POOL_EGR,	\
826 	}
827 
828 #define MLXSW_SP_SB_CM_EGR_MC(_min_buff, _max_buff)	\
829 	{						\
830 		.min_buff = _min_buff,			\
831 		.max_buff = _max_buff,			\
832 		.pool_index = MLXSW_SP_SB_POOL_EGR_MC,	\
833 		.freeze_pool = true,			\
834 		.freeze_thresh = true,			\
835 	}
836 
837 static const struct mlxsw_sp_sb_cm mlxsw_sp1_sb_cms_ingress[] = {
838 	MLXSW_SP_SB_CM_ING(10000, 8),
839 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
840 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
841 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
842 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
843 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
844 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
845 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
846 	MLXSW_SP_SB_CM_ING(0, 0), /* dummy, this PG does not exist */
847 	MLXSW_SP_SB_CM(10000, 8, MLXSW_SP_SB_POOL_ING_CPU),
848 };
849 
850 static const struct mlxsw_sp_sb_cm mlxsw_sp2_sb_cms_ingress[] = {
851 	MLXSW_SP_SB_CM_ING(0, 7),
852 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
853 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
854 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
855 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
856 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
857 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
858 	MLXSW_SP_SB_CM_ING(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
859 	MLXSW_SP_SB_CM_ING(0, 0), /* dummy, this PG does not exist */
860 	MLXSW_SP_SB_CM(10000, 8, MLXSW_SP_SB_POOL_ING_CPU),
861 };
862 
863 static const struct mlxsw_sp_sb_cm mlxsw_sp1_sb_cms_egress[] = {
864 	MLXSW_SP_SB_CM_EGR(1500, 9),
865 	MLXSW_SP_SB_CM_EGR(1500, 9),
866 	MLXSW_SP_SB_CM_EGR(1500, 9),
867 	MLXSW_SP_SB_CM_EGR(1500, 9),
868 	MLXSW_SP_SB_CM_EGR(1500, 9),
869 	MLXSW_SP_SB_CM_EGR(1500, 9),
870 	MLXSW_SP_SB_CM_EGR(1500, 9),
871 	MLXSW_SP_SB_CM_EGR(1500, 9),
872 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
873 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
874 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
875 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
876 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
877 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
878 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
879 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
880 	MLXSW_SP_SB_CM_EGR(1, 0xff),
881 };
882 
883 static const struct mlxsw_sp_sb_cm mlxsw_sp2_sb_cms_egress[] = {
884 	MLXSW_SP_SB_CM_EGR(0, 7),
885 	MLXSW_SP_SB_CM_EGR(0, 7),
886 	MLXSW_SP_SB_CM_EGR(0, 7),
887 	MLXSW_SP_SB_CM_EGR(0, 7),
888 	MLXSW_SP_SB_CM_EGR(0, 7),
889 	MLXSW_SP_SB_CM_EGR(0, 7),
890 	MLXSW_SP_SB_CM_EGR(0, 7),
891 	MLXSW_SP_SB_CM_EGR(0, 7),
892 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
893 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
894 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
895 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
896 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
897 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
898 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
899 	MLXSW_SP_SB_CM_EGR_MC(0, MLXSW_SP_SB_INFI),
900 	MLXSW_SP_SB_CM_EGR(1, 0xff),
901 };
902 
903 #define MLXSW_SP_CPU_PORT_SB_CM MLXSW_SP_SB_CM(0, 0, MLXSW_SP_SB_POOL_EGR_CPU)
904 
905 static const struct mlxsw_sp_sb_cm mlxsw_sp_cpu_port_sb_cms[] = {
906 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
907 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
908 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
909 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
910 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
911 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
912 	MLXSW_SP_CPU_PORT_SB_CM,
913 	MLXSW_SP_SB_CM(1000, 8, MLXSW_SP_SB_POOL_EGR_CPU),
914 	MLXSW_SP_CPU_PORT_SB_CM,
915 	MLXSW_SP_CPU_PORT_SB_CM,
916 	MLXSW_SP_CPU_PORT_SB_CM,
917 	MLXSW_SP_CPU_PORT_SB_CM,
918 	MLXSW_SP_CPU_PORT_SB_CM,
919 	MLXSW_SP_CPU_PORT_SB_CM,
920 	MLXSW_SP_CPU_PORT_SB_CM,
921 	MLXSW_SP_CPU_PORT_SB_CM,
922 	MLXSW_SP_CPU_PORT_SB_CM,
923 	MLXSW_SP_CPU_PORT_SB_CM,
924 	MLXSW_SP_CPU_PORT_SB_CM,
925 	MLXSW_SP_CPU_PORT_SB_CM,
926 	MLXSW_SP_CPU_PORT_SB_CM,
927 	MLXSW_SP_CPU_PORT_SB_CM,
928 	MLXSW_SP_CPU_PORT_SB_CM,
929 	MLXSW_SP_CPU_PORT_SB_CM,
930 	MLXSW_SP_CPU_PORT_SB_CM,
931 	MLXSW_SP_CPU_PORT_SB_CM,
932 	MLXSW_SP_CPU_PORT_SB_CM,
933 	MLXSW_SP_CPU_PORT_SB_CM,
934 	MLXSW_SP_CPU_PORT_SB_CM,
935 	MLXSW_SP_CPU_PORT_SB_CM,
936 	MLXSW_SP_CPU_PORT_SB_CM,
937 	MLXSW_SP_CPU_PORT_SB_CM,
938 };
939 
940 static bool
941 mlxsw_sp_sb_pool_is_static(struct mlxsw_sp *mlxsw_sp, u16 pool_index)
942 {
943 	struct mlxsw_sp_sb_pr *pr = mlxsw_sp_sb_pr_get(mlxsw_sp, pool_index);
944 
945 	return pr->mode == MLXSW_REG_SBPR_MODE_STATIC;
946 }
947 
948 static int __mlxsw_sp_sb_cms_init(struct mlxsw_sp *mlxsw_sp, u16 local_port,
949 				  enum mlxsw_reg_sbxx_dir dir,
950 				  const struct mlxsw_sp_sb_cm *cms,
951 				  size_t cms_len)
952 {
953 	const struct mlxsw_sp_sb_vals *sb_vals = mlxsw_sp->sb_vals;
954 	int i;
955 	int err;
956 
957 	for (i = 0; i < cms_len; i++) {
958 		const struct mlxsw_sp_sb_cm *cm;
959 		u32 min_buff;
960 		u32 max_buff;
961 
962 		if (i == 8 && dir == MLXSW_REG_SBXX_DIR_INGRESS)
963 			continue; /* PG number 8 does not exist, skip it */
964 		cm = &cms[i];
965 		if (WARN_ON(sb_vals->pool_dess[cm->pool_index].dir != dir))
966 			continue;
967 
968 		min_buff = mlxsw_sp_bytes_cells(mlxsw_sp, cm->min_buff);
969 		max_buff = cm->max_buff;
970 		if (max_buff == MLXSW_SP_SB_INFI) {
971 			err = mlxsw_sp_sb_cm_write(mlxsw_sp, local_port, i,
972 						   min_buff, 0,
973 						   true, cm->pool_index);
974 		} else {
975 			if (mlxsw_sp_sb_pool_is_static(mlxsw_sp,
976 						       cm->pool_index))
977 				max_buff = mlxsw_sp_bytes_cells(mlxsw_sp,
978 								max_buff);
979 			err = mlxsw_sp_sb_cm_write(mlxsw_sp, local_port, i,
980 						   min_buff, max_buff,
981 						   false, cm->pool_index);
982 		}
983 		if (err)
984 			return err;
985 	}
986 	return 0;
987 }
988 
989 static int mlxsw_sp_port_sb_cms_init(struct mlxsw_sp_port *mlxsw_sp_port)
990 {
991 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
992 	int err;
993 
994 	err = __mlxsw_sp_sb_cms_init(mlxsw_sp,
995 				     mlxsw_sp_port->local_port,
996 				     MLXSW_REG_SBXX_DIR_INGRESS,
997 				     mlxsw_sp->sb_vals->cms_ingress,
998 				     mlxsw_sp->sb_vals->cms_ingress_count);
999 	if (err)
1000 		return err;
1001 	return __mlxsw_sp_sb_cms_init(mlxsw_sp_port->mlxsw_sp,
1002 				      mlxsw_sp_port->local_port,
1003 				      MLXSW_REG_SBXX_DIR_EGRESS,
1004 				      mlxsw_sp->sb_vals->cms_egress,
1005 				      mlxsw_sp->sb_vals->cms_egress_count);
1006 }
1007 
1008 static int mlxsw_sp_cpu_port_sb_cms_init(struct mlxsw_sp *mlxsw_sp)
1009 {
1010 	return __mlxsw_sp_sb_cms_init(mlxsw_sp, 0, MLXSW_REG_SBXX_DIR_EGRESS,
1011 				      mlxsw_sp->sb_vals->cms_cpu,
1012 				      mlxsw_sp->sb_vals->cms_cpu_count);
1013 }
1014 
1015 #define MLXSW_SP_SB_PM(_min_buff, _max_buff)	\
1016 	{					\
1017 		.min_buff = _min_buff,		\
1018 		.max_buff = _max_buff,		\
1019 	}
1020 
1021 /* Order according to mlxsw_sp1_sb_pool_dess */
1022 static const struct mlxsw_sp_sb_pm mlxsw_sp1_sb_pms[] = {
1023 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MAX),
1024 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1025 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1026 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1027 	MLXSW_SP_SB_PM(0, 7),
1028 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1029 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1030 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1031 	MLXSW_SP_SB_PM(10000, 90000),
1032 	MLXSW_SP_SB_PM(0, 8),	/* 50% occupancy */
1033 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1034 };
1035 
1036 /* Order according to mlxsw_sp2_sb_pool_dess */
1037 static const struct mlxsw_sp_sb_pm mlxsw_sp2_sb_pms[] = {
1038 	MLXSW_SP_SB_PM(0, 7),
1039 	MLXSW_SP_SB_PM(0, 0),
1040 	MLXSW_SP_SB_PM(0, 0),
1041 	MLXSW_SP_SB_PM(0, 0),
1042 	MLXSW_SP_SB_PM(0, 7),
1043 	MLXSW_SP_SB_PM(0, 0),
1044 	MLXSW_SP_SB_PM(0, 0),
1045 	MLXSW_SP_SB_PM(0, 0),
1046 	MLXSW_SP_SB_PM(10000, 90000),
1047 	MLXSW_SP_SB_PM(0, 8),	/* 50% occupancy */
1048 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN),
1049 };
1050 
1051 /* Order according to mlxsw_sp*_sb_pool_dess */
1052 static const struct mlxsw_sp_sb_pm mlxsw_sp_cpu_port_sb_pms[] = {
1053 	MLXSW_SP_SB_PM(0, 0),
1054 	MLXSW_SP_SB_PM(0, 0),
1055 	MLXSW_SP_SB_PM(0, 0),
1056 	MLXSW_SP_SB_PM(0, 0),
1057 	MLXSW_SP_SB_PM(0, 0),
1058 	MLXSW_SP_SB_PM(0, 0),
1059 	MLXSW_SP_SB_PM(0, 0),
1060 	MLXSW_SP_SB_PM(0, 0),
1061 	MLXSW_SP_SB_PM(0, 90000),
1062 	MLXSW_SP_SB_PM(0, 0),
1063 	MLXSW_SP_SB_PM(0, MLXSW_REG_SBXX_DYN_MAX_BUFF_MAX),
1064 };
1065 
1066 static int mlxsw_sp_sb_pms_init(struct mlxsw_sp *mlxsw_sp, u16 local_port,
1067 				const struct mlxsw_sp_sb_pm *pms,
1068 				bool skip_ingress)
1069 {
1070 	int i, err;
1071 
1072 	for (i = 0; i < mlxsw_sp->sb_vals->pool_count; i++) {
1073 		const struct mlxsw_sp_sb_pm *pm = &pms[i];
1074 		const struct mlxsw_sp_sb_pool_des *des;
1075 		u32 max_buff;
1076 		u32 min_buff;
1077 
1078 		des = &mlxsw_sp->sb_vals->pool_dess[i];
1079 		if (skip_ingress && des->dir == MLXSW_REG_SBXX_DIR_INGRESS)
1080 			continue;
1081 
1082 		min_buff = mlxsw_sp_bytes_cells(mlxsw_sp, pm->min_buff);
1083 		max_buff = pm->max_buff;
1084 		if (mlxsw_sp_sb_pool_is_static(mlxsw_sp, i))
1085 			max_buff = mlxsw_sp_bytes_cells(mlxsw_sp, max_buff);
1086 		err = mlxsw_sp_sb_pm_write(mlxsw_sp, local_port, i, min_buff,
1087 					   max_buff);
1088 		if (err)
1089 			return err;
1090 	}
1091 	return 0;
1092 }
1093 
1094 static int mlxsw_sp_port_sb_pms_init(struct mlxsw_sp_port *mlxsw_sp_port)
1095 {
1096 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1097 
1098 	return mlxsw_sp_sb_pms_init(mlxsw_sp, mlxsw_sp_port->local_port,
1099 				    mlxsw_sp->sb_vals->pms, false);
1100 }
1101 
1102 static int mlxsw_sp_cpu_port_sb_pms_init(struct mlxsw_sp *mlxsw_sp)
1103 {
1104 	return mlxsw_sp_sb_pms_init(mlxsw_sp, 0, mlxsw_sp->sb_vals->pms_cpu,
1105 				    true);
1106 }
1107 
1108 #define MLXSW_SP_SB_MM(_min_buff, _max_buff)		\
1109 	{						\
1110 		.min_buff = _min_buff,			\
1111 		.max_buff = _max_buff,			\
1112 		.pool_index = MLXSW_SP_SB_POOL_EGR,	\
1113 	}
1114 
1115 static const struct mlxsw_sp_sb_mm mlxsw_sp_sb_mms[] = {
1116 	MLXSW_SP_SB_MM(0, 6),
1117 	MLXSW_SP_SB_MM(0, 6),
1118 	MLXSW_SP_SB_MM(0, 6),
1119 	MLXSW_SP_SB_MM(0, 6),
1120 	MLXSW_SP_SB_MM(0, 6),
1121 	MLXSW_SP_SB_MM(0, 6),
1122 	MLXSW_SP_SB_MM(0, 6),
1123 	MLXSW_SP_SB_MM(0, 6),
1124 	MLXSW_SP_SB_MM(0, 6),
1125 	MLXSW_SP_SB_MM(0, 6),
1126 	MLXSW_SP_SB_MM(0, 6),
1127 	MLXSW_SP_SB_MM(0, 6),
1128 	MLXSW_SP_SB_MM(0, 6),
1129 	MLXSW_SP_SB_MM(0, 6),
1130 	MLXSW_SP_SB_MM(0, 6),
1131 };
1132 
1133 static int mlxsw_sp_sb_mms_init(struct mlxsw_sp *mlxsw_sp)
1134 {
1135 	char sbmm_pl[MLXSW_REG_SBMM_LEN];
1136 	int i;
1137 	int err;
1138 
1139 	for (i = 0; i < mlxsw_sp->sb_vals->mms_count; i++) {
1140 		const struct mlxsw_sp_sb_pool_des *des;
1141 		const struct mlxsw_sp_sb_mm *mc;
1142 		u32 min_buff;
1143 
1144 		mc = &mlxsw_sp->sb_vals->mms[i];
1145 		des = &mlxsw_sp->sb_vals->pool_dess[mc->pool_index];
1146 		/* All pools used by sb_mm's are initialized using dynamic
1147 		 * thresholds, therefore 'max_buff' isn't specified in cells.
1148 		 */
1149 		min_buff = mlxsw_sp_bytes_cells(mlxsw_sp, mc->min_buff);
1150 		mlxsw_reg_sbmm_pack(sbmm_pl, i, min_buff, mc->max_buff,
1151 				    des->pool);
1152 		err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(sbmm), sbmm_pl);
1153 		if (err)
1154 			return err;
1155 	}
1156 	return 0;
1157 }
1158 
1159 static void mlxsw_sp_pool_count(struct mlxsw_sp *mlxsw_sp,
1160 				u16 *p_ingress_len, u16 *p_egress_len)
1161 {
1162 	int i;
1163 
1164 	for (i = 0; i < mlxsw_sp->sb_vals->pool_count; ++i) {
1165 		if (mlxsw_sp->sb_vals->pool_dess[i].dir ==
1166 		    MLXSW_REG_SBXX_DIR_INGRESS)
1167 			(*p_ingress_len)++;
1168 		else
1169 			(*p_egress_len)++;
1170 	}
1171 
1172 	WARN(*p_egress_len == 0, "No egress pools\n");
1173 }
1174 
1175 const struct mlxsw_sp_sb_vals mlxsw_sp1_sb_vals = {
1176 	.pool_count = ARRAY_SIZE(mlxsw_sp1_sb_pool_dess),
1177 	.pool_dess = mlxsw_sp1_sb_pool_dess,
1178 	.pms = mlxsw_sp1_sb_pms,
1179 	.pms_cpu = mlxsw_sp_cpu_port_sb_pms,
1180 	.prs = mlxsw_sp1_sb_prs,
1181 	.mms = mlxsw_sp_sb_mms,
1182 	.cms_ingress = mlxsw_sp1_sb_cms_ingress,
1183 	.cms_egress = mlxsw_sp1_sb_cms_egress,
1184 	.cms_cpu = mlxsw_sp_cpu_port_sb_cms,
1185 	.mms_count = ARRAY_SIZE(mlxsw_sp_sb_mms),
1186 	.cms_ingress_count = ARRAY_SIZE(mlxsw_sp1_sb_cms_ingress),
1187 	.cms_egress_count = ARRAY_SIZE(mlxsw_sp1_sb_cms_egress),
1188 	.cms_cpu_count = ARRAY_SIZE(mlxsw_sp_cpu_port_sb_cms),
1189 };
1190 
1191 const struct mlxsw_sp_sb_vals mlxsw_sp2_sb_vals = {
1192 	.pool_count = ARRAY_SIZE(mlxsw_sp2_sb_pool_dess),
1193 	.pool_dess = mlxsw_sp2_sb_pool_dess,
1194 	.pms = mlxsw_sp2_sb_pms,
1195 	.pms_cpu = mlxsw_sp_cpu_port_sb_pms,
1196 	.prs = mlxsw_sp2_sb_prs,
1197 	.mms = mlxsw_sp_sb_mms,
1198 	.cms_ingress = mlxsw_sp2_sb_cms_ingress,
1199 	.cms_egress = mlxsw_sp2_sb_cms_egress,
1200 	.cms_cpu = mlxsw_sp_cpu_port_sb_cms,
1201 	.mms_count = ARRAY_SIZE(mlxsw_sp_sb_mms),
1202 	.cms_ingress_count = ARRAY_SIZE(mlxsw_sp2_sb_cms_ingress),
1203 	.cms_egress_count = ARRAY_SIZE(mlxsw_sp2_sb_cms_egress),
1204 	.cms_cpu_count = ARRAY_SIZE(mlxsw_sp_cpu_port_sb_cms),
1205 };
1206 
1207 static u32 mlxsw_sp1_pb_int_buf_size_get(int mtu, u32 speed)
1208 {
1209 	return mtu * 5 / 2;
1210 }
1211 
1212 static u32 __mlxsw_sp_pb_int_buf_size_get(int mtu, u32 speed, u32 buffer_factor)
1213 {
1214 	return 3 * mtu + buffer_factor * speed / 1000;
1215 }
1216 
1217 #define MLXSW_SP2_SPAN_EG_MIRROR_BUFFER_FACTOR 38
1218 
1219 static u32 mlxsw_sp2_pb_int_buf_size_get(int mtu, u32 speed)
1220 {
1221 	int factor = MLXSW_SP2_SPAN_EG_MIRROR_BUFFER_FACTOR;
1222 
1223 	return __mlxsw_sp_pb_int_buf_size_get(mtu, speed, factor);
1224 }
1225 
1226 #define MLXSW_SP3_SPAN_EG_MIRROR_BUFFER_FACTOR 50
1227 
1228 static u32 mlxsw_sp3_pb_int_buf_size_get(int mtu, u32 speed)
1229 {
1230 	int factor = MLXSW_SP3_SPAN_EG_MIRROR_BUFFER_FACTOR;
1231 
1232 	return __mlxsw_sp_pb_int_buf_size_get(mtu, speed, factor);
1233 }
1234 
1235 const struct mlxsw_sp_sb_ops mlxsw_sp1_sb_ops = {
1236 	.int_buf_size_get = mlxsw_sp1_pb_int_buf_size_get,
1237 };
1238 
1239 const struct mlxsw_sp_sb_ops mlxsw_sp2_sb_ops = {
1240 	.int_buf_size_get = mlxsw_sp2_pb_int_buf_size_get,
1241 };
1242 
1243 const struct mlxsw_sp_sb_ops mlxsw_sp3_sb_ops = {
1244 	.int_buf_size_get = mlxsw_sp3_pb_int_buf_size_get,
1245 };
1246 
1247 int mlxsw_sp_buffers_init(struct mlxsw_sp *mlxsw_sp)
1248 {
1249 	u32 max_headroom_size;
1250 	u16 ing_pool_count = 0;
1251 	u16 eg_pool_count = 0;
1252 	int err;
1253 
1254 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, CELL_SIZE))
1255 		return -EIO;
1256 
1257 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, GUARANTEED_SHARED_BUFFER))
1258 		return -EIO;
1259 
1260 	if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, MAX_HEADROOM_SIZE))
1261 		return -EIO;
1262 
1263 	mlxsw_sp->sb = kzalloc(sizeof(*mlxsw_sp->sb), GFP_KERNEL);
1264 	if (!mlxsw_sp->sb)
1265 		return -ENOMEM;
1266 	mlxsw_sp->sb->cell_size = MLXSW_CORE_RES_GET(mlxsw_sp->core, CELL_SIZE);
1267 	mlxsw_sp->sb->sb_size = MLXSW_CORE_RES_GET(mlxsw_sp->core,
1268 						   GUARANTEED_SHARED_BUFFER);
1269 	max_headroom_size = MLXSW_CORE_RES_GET(mlxsw_sp->core,
1270 					       MAX_HEADROOM_SIZE);
1271 	/* Round down, because this limit must not be overstepped. */
1272 	mlxsw_sp->sb->max_headroom_cells = max_headroom_size /
1273 						mlxsw_sp->sb->cell_size;
1274 
1275 	err = mlxsw_sp_sb_ports_init(mlxsw_sp);
1276 	if (err)
1277 		goto err_sb_ports_init;
1278 	err = mlxsw_sp_sb_prs_init(mlxsw_sp, mlxsw_sp->sb_vals->prs,
1279 				   mlxsw_sp->sb_vals->pool_dess,
1280 				   mlxsw_sp->sb_vals->pool_count);
1281 	if (err)
1282 		goto err_sb_prs_init;
1283 	err = mlxsw_sp_cpu_port_sb_cms_init(mlxsw_sp);
1284 	if (err)
1285 		goto err_sb_cpu_port_sb_cms_init;
1286 	err = mlxsw_sp_cpu_port_sb_pms_init(mlxsw_sp);
1287 	if (err)
1288 		goto err_sb_cpu_port_pms_init;
1289 	err = mlxsw_sp_sb_mms_init(mlxsw_sp);
1290 	if (err)
1291 		goto err_sb_mms_init;
1292 	mlxsw_sp_pool_count(mlxsw_sp, &ing_pool_count, &eg_pool_count);
1293 	err = devl_sb_register(priv_to_devlink(mlxsw_sp->core), 0,
1294 			       mlxsw_sp->sb->sb_size,
1295 			       ing_pool_count,
1296 			       eg_pool_count,
1297 			       MLXSW_SP_SB_ING_TC_COUNT,
1298 			       MLXSW_SP_SB_EG_TC_COUNT);
1299 	if (err)
1300 		goto err_devlink_sb_register;
1301 
1302 	return 0;
1303 
1304 err_devlink_sb_register:
1305 err_sb_mms_init:
1306 err_sb_cpu_port_pms_init:
1307 err_sb_cpu_port_sb_cms_init:
1308 err_sb_prs_init:
1309 	mlxsw_sp_sb_ports_fini(mlxsw_sp);
1310 err_sb_ports_init:
1311 	kfree(mlxsw_sp->sb);
1312 	return err;
1313 }
1314 
1315 void mlxsw_sp_buffers_fini(struct mlxsw_sp *mlxsw_sp)
1316 {
1317 	devl_sb_unregister(priv_to_devlink(mlxsw_sp->core), 0);
1318 	mlxsw_sp_sb_ports_fini(mlxsw_sp);
1319 	kfree(mlxsw_sp->sb);
1320 }
1321 
1322 int mlxsw_sp_port_buffers_init(struct mlxsw_sp_port *mlxsw_sp_port)
1323 {
1324 	int err;
1325 
1326 	mlxsw_sp_port->hdroom = kzalloc(sizeof(*mlxsw_sp_port->hdroom), GFP_KERNEL);
1327 	if (!mlxsw_sp_port->hdroom)
1328 		return -ENOMEM;
1329 	mlxsw_sp_port->hdroom->mtu = mlxsw_sp_port->dev->mtu;
1330 
1331 	err = mlxsw_sp_port_headroom_init(mlxsw_sp_port);
1332 	if (err)
1333 		goto err_headroom_init;
1334 	err = mlxsw_sp_port_sb_cms_init(mlxsw_sp_port);
1335 	if (err)
1336 		goto err_port_sb_cms_init;
1337 	err = mlxsw_sp_port_sb_pms_init(mlxsw_sp_port);
1338 	if (err)
1339 		goto err_port_sb_pms_init;
1340 	return 0;
1341 
1342 err_port_sb_pms_init:
1343 err_port_sb_cms_init:
1344 err_headroom_init:
1345 	kfree(mlxsw_sp_port->hdroom);
1346 	return err;
1347 }
1348 
1349 void mlxsw_sp_port_buffers_fini(struct mlxsw_sp_port *mlxsw_sp_port)
1350 {
1351 	kfree(mlxsw_sp_port->hdroom);
1352 }
1353 
1354 int mlxsw_sp_sb_pool_get(struct mlxsw_core *mlxsw_core,
1355 			 unsigned int sb_index, u16 pool_index,
1356 			 struct devlink_sb_pool_info *pool_info)
1357 {
1358 	struct mlxsw_sp *mlxsw_sp = mlxsw_core_driver_priv(mlxsw_core);
1359 	enum mlxsw_reg_sbxx_dir dir;
1360 	struct mlxsw_sp_sb_pr *pr;
1361 
1362 	dir = mlxsw_sp->sb_vals->pool_dess[pool_index].dir;
1363 	pr = mlxsw_sp_sb_pr_get(mlxsw_sp, pool_index);
1364 	pool_info->pool_type = (enum devlink_sb_pool_type) dir;
1365 	pool_info->size = mlxsw_sp_cells_bytes(mlxsw_sp, pr->size);
1366 	pool_info->threshold_type = (enum devlink_sb_threshold_type) pr->mode;
1367 	pool_info->cell_size = mlxsw_sp->sb->cell_size;
1368 	return 0;
1369 }
1370 
1371 int mlxsw_sp_sb_pool_set(struct mlxsw_core *mlxsw_core,
1372 			 unsigned int sb_index, u16 pool_index, u32 size,
1373 			 enum devlink_sb_threshold_type threshold_type,
1374 			 struct netlink_ext_ack *extack)
1375 {
1376 	struct mlxsw_sp *mlxsw_sp = mlxsw_core_driver_priv(mlxsw_core);
1377 	u32 pool_size = mlxsw_sp_bytes_cells(mlxsw_sp, size);
1378 	const struct mlxsw_sp_sb_pr *pr;
1379 	enum mlxsw_reg_sbpr_mode mode;
1380 
1381 	mode = (enum mlxsw_reg_sbpr_mode) threshold_type;
1382 	pr = &mlxsw_sp->sb_vals->prs[pool_index];
1383 
1384 	if (size > MLXSW_CORE_RES_GET(mlxsw_sp->core,
1385 				      GUARANTEED_SHARED_BUFFER)) {
1386 		NL_SET_ERR_MSG_MOD(extack, "Exceeded shared buffer size");
1387 		return -EINVAL;
1388 	}
1389 
1390 	if (pr->freeze_mode && pr->mode != mode) {
1391 		NL_SET_ERR_MSG_MOD(extack, "Changing this pool's threshold type is forbidden");
1392 		return -EINVAL;
1393 	}
1394 
1395 	if (pr->freeze_size && pr->size != size) {
1396 		NL_SET_ERR_MSG_MOD(extack, "Changing this pool's size is forbidden");
1397 		return -EINVAL;
1398 	}
1399 
1400 	return mlxsw_sp_sb_pr_write(mlxsw_sp, pool_index, mode,
1401 				    pool_size, false);
1402 }
1403 
1404 #define MLXSW_SP_SB_THRESHOLD_TO_ALPHA_OFFSET (-2) /* 3->1, 16->14 */
1405 
1406 static u32 mlxsw_sp_sb_threshold_out(struct mlxsw_sp *mlxsw_sp, u16 pool_index,
1407 				     u32 max_buff)
1408 {
1409 	struct mlxsw_sp_sb_pr *pr = mlxsw_sp_sb_pr_get(mlxsw_sp, pool_index);
1410 
1411 	if (pr->mode == MLXSW_REG_SBPR_MODE_DYNAMIC)
1412 		return max_buff - MLXSW_SP_SB_THRESHOLD_TO_ALPHA_OFFSET;
1413 	return mlxsw_sp_cells_bytes(mlxsw_sp, max_buff);
1414 }
1415 
1416 static int mlxsw_sp_sb_threshold_in(struct mlxsw_sp *mlxsw_sp, u16 pool_index,
1417 				    u32 threshold, u32 *p_max_buff,
1418 				    struct netlink_ext_ack *extack)
1419 {
1420 	struct mlxsw_sp_sb_pr *pr = mlxsw_sp_sb_pr_get(mlxsw_sp, pool_index);
1421 
1422 	if (pr->mode == MLXSW_REG_SBPR_MODE_DYNAMIC) {
1423 		int val;
1424 
1425 		val = threshold + MLXSW_SP_SB_THRESHOLD_TO_ALPHA_OFFSET;
1426 		if (val < MLXSW_REG_SBXX_DYN_MAX_BUFF_MIN ||
1427 		    val > MLXSW_REG_SBXX_DYN_MAX_BUFF_MAX) {
1428 			NL_SET_ERR_MSG_MOD(extack, "Invalid dynamic threshold value");
1429 			return -EINVAL;
1430 		}
1431 		*p_max_buff = val;
1432 	} else {
1433 		*p_max_buff = mlxsw_sp_bytes_cells(mlxsw_sp, threshold);
1434 	}
1435 	return 0;
1436 }
1437 
1438 int mlxsw_sp_sb_port_pool_get(struct mlxsw_core_port *mlxsw_core_port,
1439 			      unsigned int sb_index, u16 pool_index,
1440 			      u32 *p_threshold)
1441 {
1442 	struct mlxsw_sp_port *mlxsw_sp_port =
1443 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1444 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1445 	u16 local_port = mlxsw_sp_port->local_port;
1446 	struct mlxsw_sp_sb_pm *pm = mlxsw_sp_sb_pm_get(mlxsw_sp, local_port,
1447 						       pool_index);
1448 
1449 	*p_threshold = mlxsw_sp_sb_threshold_out(mlxsw_sp, pool_index,
1450 						 pm->max_buff);
1451 	return 0;
1452 }
1453 
1454 int mlxsw_sp_sb_port_pool_set(struct mlxsw_core_port *mlxsw_core_port,
1455 			      unsigned int sb_index, u16 pool_index,
1456 			      u32 threshold, struct netlink_ext_ack *extack)
1457 {
1458 	struct mlxsw_sp_port *mlxsw_sp_port =
1459 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1460 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1461 	u16 local_port = mlxsw_sp_port->local_port;
1462 	u32 max_buff;
1463 	int err;
1464 
1465 	if (local_port == MLXSW_PORT_CPU_PORT) {
1466 		NL_SET_ERR_MSG_MOD(extack, "Changing CPU port's threshold is forbidden");
1467 		return -EINVAL;
1468 	}
1469 
1470 	err = mlxsw_sp_sb_threshold_in(mlxsw_sp, pool_index,
1471 				       threshold, &max_buff, extack);
1472 	if (err)
1473 		return err;
1474 
1475 	return mlxsw_sp_sb_pm_write(mlxsw_sp, local_port, pool_index,
1476 				    0, max_buff);
1477 }
1478 
1479 int mlxsw_sp_sb_tc_pool_bind_get(struct mlxsw_core_port *mlxsw_core_port,
1480 				 unsigned int sb_index, u16 tc_index,
1481 				 enum devlink_sb_pool_type pool_type,
1482 				 u16 *p_pool_index, u32 *p_threshold)
1483 {
1484 	struct mlxsw_sp_port *mlxsw_sp_port =
1485 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1486 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1487 	u16 local_port = mlxsw_sp_port->local_port;
1488 	u8 pg_buff = tc_index;
1489 	enum mlxsw_reg_sbxx_dir dir = (enum mlxsw_reg_sbxx_dir) pool_type;
1490 	struct mlxsw_sp_sb_cm *cm = mlxsw_sp_sb_cm_get(mlxsw_sp, local_port,
1491 						       pg_buff, dir);
1492 
1493 	*p_threshold = mlxsw_sp_sb_threshold_out(mlxsw_sp, cm->pool_index,
1494 						 cm->max_buff);
1495 	*p_pool_index = cm->pool_index;
1496 	return 0;
1497 }
1498 
1499 int mlxsw_sp_sb_tc_pool_bind_set(struct mlxsw_core_port *mlxsw_core_port,
1500 				 unsigned int sb_index, u16 tc_index,
1501 				 enum devlink_sb_pool_type pool_type,
1502 				 u16 pool_index, u32 threshold,
1503 				 struct netlink_ext_ack *extack)
1504 {
1505 	struct mlxsw_sp_port *mlxsw_sp_port =
1506 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1507 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1508 	u16 local_port = mlxsw_sp_port->local_port;
1509 	const struct mlxsw_sp_sb_cm *cm;
1510 	u8 pg_buff = tc_index;
1511 	enum mlxsw_reg_sbxx_dir dir = (enum mlxsw_reg_sbxx_dir) pool_type;
1512 	u32 max_buff;
1513 	int err;
1514 
1515 	if (local_port == MLXSW_PORT_CPU_PORT) {
1516 		NL_SET_ERR_MSG_MOD(extack, "Changing CPU port's binding is forbidden");
1517 		return -EINVAL;
1518 	}
1519 
1520 	if (dir != mlxsw_sp->sb_vals->pool_dess[pool_index].dir) {
1521 		NL_SET_ERR_MSG_MOD(extack, "Binding egress TC to ingress pool and vice versa is forbidden");
1522 		return -EINVAL;
1523 	}
1524 
1525 	if (dir == MLXSW_REG_SBXX_DIR_INGRESS)
1526 		cm = &mlxsw_sp->sb_vals->cms_ingress[tc_index];
1527 	else
1528 		cm = &mlxsw_sp->sb_vals->cms_egress[tc_index];
1529 
1530 	if (cm->freeze_pool && cm->pool_index != pool_index) {
1531 		NL_SET_ERR_MSG_MOD(extack, "Binding this TC to a different pool is forbidden");
1532 		return -EINVAL;
1533 	}
1534 
1535 	if (cm->freeze_thresh && cm->max_buff != threshold) {
1536 		NL_SET_ERR_MSG_MOD(extack, "Changing this TC's threshold is forbidden");
1537 		return -EINVAL;
1538 	}
1539 
1540 	err = mlxsw_sp_sb_threshold_in(mlxsw_sp, pool_index,
1541 				       threshold, &max_buff, extack);
1542 	if (err)
1543 		return err;
1544 
1545 	return mlxsw_sp_sb_cm_write(mlxsw_sp, local_port, pg_buff,
1546 				    0, max_buff, false, pool_index);
1547 }
1548 
1549 #define MASKED_COUNT_MAX \
1550 	(MLXSW_REG_SBSR_REC_MAX_COUNT / \
1551 	 (MLXSW_SP_SB_ING_TC_COUNT + MLXSW_SP_SB_EG_TC_COUNT))
1552 
1553 struct mlxsw_sp_sb_sr_occ_query_cb_ctx {
1554 	u8 masked_count;
1555 	u16 local_port_1;
1556 };
1557 
1558 static void mlxsw_sp_sb_sr_occ_query_cb(struct mlxsw_core *mlxsw_core,
1559 					char *sbsr_pl, size_t sbsr_pl_len,
1560 					unsigned long cb_priv)
1561 {
1562 	struct mlxsw_sp *mlxsw_sp = mlxsw_core_driver_priv(mlxsw_core);
1563 	struct mlxsw_sp_sb_sr_occ_query_cb_ctx cb_ctx;
1564 	u8 masked_count;
1565 	u16 local_port;
1566 	int rec_index = 0;
1567 	struct mlxsw_sp_sb_cm *cm;
1568 	int i;
1569 
1570 	memcpy(&cb_ctx, &cb_priv, sizeof(cb_ctx));
1571 
1572 	masked_count = 0;
1573 	for (local_port = cb_ctx.local_port_1;
1574 	     local_port < mlxsw_core_max_ports(mlxsw_core); local_port++) {
1575 		if (!mlxsw_sp->ports[local_port])
1576 			continue;
1577 		if (local_port == MLXSW_PORT_CPU_PORT) {
1578 			/* Ingress quotas are not supported for the CPU port */
1579 			masked_count++;
1580 			continue;
1581 		}
1582 		for (i = 0; i < MLXSW_SP_SB_ING_TC_COUNT; i++) {
1583 			cm = mlxsw_sp_sb_cm_get(mlxsw_sp, local_port, i,
1584 						MLXSW_REG_SBXX_DIR_INGRESS);
1585 			mlxsw_reg_sbsr_rec_unpack(sbsr_pl, rec_index++,
1586 						  &cm->occ.cur, &cm->occ.max);
1587 		}
1588 		if (++masked_count == cb_ctx.masked_count)
1589 			break;
1590 	}
1591 	masked_count = 0;
1592 	for (local_port = cb_ctx.local_port_1;
1593 	     local_port < mlxsw_core_max_ports(mlxsw_core); local_port++) {
1594 		if (!mlxsw_sp->ports[local_port])
1595 			continue;
1596 		for (i = 0; i < MLXSW_SP_SB_EG_TC_COUNT; i++) {
1597 			cm = mlxsw_sp_sb_cm_get(mlxsw_sp, local_port, i,
1598 						MLXSW_REG_SBXX_DIR_EGRESS);
1599 			mlxsw_reg_sbsr_rec_unpack(sbsr_pl, rec_index++,
1600 						  &cm->occ.cur, &cm->occ.max);
1601 		}
1602 		if (++masked_count == cb_ctx.masked_count)
1603 			break;
1604 	}
1605 }
1606 
1607 int mlxsw_sp_sb_occ_snapshot(struct mlxsw_core *mlxsw_core,
1608 			     unsigned int sb_index)
1609 {
1610 	u16 local_port, local_port_1, first_local_port, last_local_port;
1611 	struct mlxsw_sp *mlxsw_sp = mlxsw_core_driver_priv(mlxsw_core);
1612 	struct mlxsw_sp_sb_sr_occ_query_cb_ctx cb_ctx;
1613 	u8 masked_count, current_page = 0;
1614 	unsigned long cb_priv = 0;
1615 	LIST_HEAD(bulk_list);
1616 	char *sbsr_pl;
1617 	int i;
1618 	int err;
1619 	int err2;
1620 
1621 	sbsr_pl = kmalloc(MLXSW_REG_SBSR_LEN, GFP_KERNEL);
1622 	if (!sbsr_pl)
1623 		return -ENOMEM;
1624 
1625 	local_port = MLXSW_PORT_CPU_PORT;
1626 next_batch:
1627 	local_port_1 = local_port;
1628 	masked_count = 0;
1629 	mlxsw_reg_sbsr_pack(sbsr_pl, false);
1630 	mlxsw_reg_sbsr_port_page_set(sbsr_pl, current_page);
1631 	first_local_port = current_page * MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE;
1632 	last_local_port = current_page * MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE +
1633 			  MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE - 1;
1634 
1635 	for (i = 0; i < MLXSW_SP_SB_ING_TC_COUNT; i++)
1636 		mlxsw_reg_sbsr_pg_buff_mask_set(sbsr_pl, i, 1);
1637 	for (i = 0; i < MLXSW_SP_SB_EG_TC_COUNT; i++)
1638 		mlxsw_reg_sbsr_tclass_mask_set(sbsr_pl, i, 1);
1639 	for (; local_port < mlxsw_core_max_ports(mlxsw_core); local_port++) {
1640 		if (!mlxsw_sp->ports[local_port])
1641 			continue;
1642 		if (local_port > last_local_port) {
1643 			current_page++;
1644 			goto do_query;
1645 		}
1646 		if (local_port != MLXSW_PORT_CPU_PORT) {
1647 			/* Ingress quotas are not supported for the CPU port */
1648 			mlxsw_reg_sbsr_ingress_port_mask_set(sbsr_pl,
1649 							     local_port - first_local_port,
1650 							     1);
1651 		}
1652 		mlxsw_reg_sbsr_egress_port_mask_set(sbsr_pl,
1653 						    local_port - first_local_port,
1654 						    1);
1655 		for (i = 0; i < mlxsw_sp->sb_vals->pool_count; i++) {
1656 			err = mlxsw_sp_sb_pm_occ_query(mlxsw_sp, local_port, i,
1657 						       &bulk_list);
1658 			if (err)
1659 				goto out;
1660 		}
1661 		if (++masked_count == MASKED_COUNT_MAX)
1662 			goto do_query;
1663 	}
1664 
1665 do_query:
1666 	cb_ctx.masked_count = masked_count;
1667 	cb_ctx.local_port_1 = local_port_1;
1668 	memcpy(&cb_priv, &cb_ctx, sizeof(cb_ctx));
1669 	err = mlxsw_reg_trans_query(mlxsw_core, MLXSW_REG(sbsr), sbsr_pl,
1670 				    &bulk_list, mlxsw_sp_sb_sr_occ_query_cb,
1671 				    cb_priv);
1672 	if (err)
1673 		goto out;
1674 	if (local_port < mlxsw_core_max_ports(mlxsw_core)) {
1675 		local_port++;
1676 		goto next_batch;
1677 	}
1678 
1679 out:
1680 	err2 = mlxsw_reg_trans_bulk_wait(&bulk_list);
1681 	if (!err)
1682 		err = err2;
1683 	kfree(sbsr_pl);
1684 	return err;
1685 }
1686 
1687 int mlxsw_sp_sb_occ_max_clear(struct mlxsw_core *mlxsw_core,
1688 			      unsigned int sb_index)
1689 {
1690 	struct mlxsw_sp *mlxsw_sp = mlxsw_core_driver_priv(mlxsw_core);
1691 	u16 local_port, first_local_port, last_local_port;
1692 	LIST_HEAD(bulk_list);
1693 	unsigned int masked_count;
1694 	u8 current_page = 0;
1695 	char *sbsr_pl;
1696 	int i;
1697 	int err;
1698 	int err2;
1699 
1700 	sbsr_pl = kmalloc(MLXSW_REG_SBSR_LEN, GFP_KERNEL);
1701 	if (!sbsr_pl)
1702 		return -ENOMEM;
1703 
1704 	local_port = MLXSW_PORT_CPU_PORT;
1705 next_batch:
1706 	masked_count = 0;
1707 	mlxsw_reg_sbsr_pack(sbsr_pl, true);
1708 	mlxsw_reg_sbsr_port_page_set(sbsr_pl, current_page);
1709 	first_local_port = current_page * MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE;
1710 	last_local_port = current_page * MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE +
1711 			  MLXSW_REG_SBSR_NUM_PORTS_IN_PAGE - 1;
1712 
1713 	for (i = 0; i < MLXSW_SP_SB_ING_TC_COUNT; i++)
1714 		mlxsw_reg_sbsr_pg_buff_mask_set(sbsr_pl, i, 1);
1715 	for (i = 0; i < MLXSW_SP_SB_EG_TC_COUNT; i++)
1716 		mlxsw_reg_sbsr_tclass_mask_set(sbsr_pl, i, 1);
1717 	for (; local_port < mlxsw_core_max_ports(mlxsw_core); local_port++) {
1718 		if (!mlxsw_sp->ports[local_port])
1719 			continue;
1720 		if (local_port > last_local_port) {
1721 			current_page++;
1722 			goto do_query;
1723 		}
1724 		if (local_port != MLXSW_PORT_CPU_PORT) {
1725 			/* Ingress quotas are not supported for the CPU port */
1726 			mlxsw_reg_sbsr_ingress_port_mask_set(sbsr_pl,
1727 							     local_port - first_local_port,
1728 							     1);
1729 		}
1730 		mlxsw_reg_sbsr_egress_port_mask_set(sbsr_pl,
1731 						    local_port - first_local_port,
1732 						    1);
1733 		for (i = 0; i < mlxsw_sp->sb_vals->pool_count; i++) {
1734 			err = mlxsw_sp_sb_pm_occ_clear(mlxsw_sp, local_port, i,
1735 						       &bulk_list);
1736 			if (err)
1737 				goto out;
1738 		}
1739 		if (++masked_count == MASKED_COUNT_MAX)
1740 			goto do_query;
1741 	}
1742 
1743 do_query:
1744 	err = mlxsw_reg_trans_query(mlxsw_core, MLXSW_REG(sbsr), sbsr_pl,
1745 				    &bulk_list, NULL, 0);
1746 	if (err)
1747 		goto out;
1748 	if (local_port < mlxsw_core_max_ports(mlxsw_core)) {
1749 		local_port++;
1750 		goto next_batch;
1751 	}
1752 
1753 out:
1754 	err2 = mlxsw_reg_trans_bulk_wait(&bulk_list);
1755 	if (!err)
1756 		err = err2;
1757 	kfree(sbsr_pl);
1758 	return err;
1759 }
1760 
1761 int mlxsw_sp_sb_occ_port_pool_get(struct mlxsw_core_port *mlxsw_core_port,
1762 				  unsigned int sb_index, u16 pool_index,
1763 				  u32 *p_cur, u32 *p_max)
1764 {
1765 	struct mlxsw_sp_port *mlxsw_sp_port =
1766 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1767 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1768 	u16 local_port = mlxsw_sp_port->local_port;
1769 	struct mlxsw_sp_sb_pm *pm = mlxsw_sp_sb_pm_get(mlxsw_sp, local_port,
1770 						       pool_index);
1771 
1772 	*p_cur = mlxsw_sp_cells_bytes(mlxsw_sp, pm->occ.cur);
1773 	*p_max = mlxsw_sp_cells_bytes(mlxsw_sp, pm->occ.max);
1774 	return 0;
1775 }
1776 
1777 int mlxsw_sp_sb_occ_tc_port_bind_get(struct mlxsw_core_port *mlxsw_core_port,
1778 				     unsigned int sb_index, u16 tc_index,
1779 				     enum devlink_sb_pool_type pool_type,
1780 				     u32 *p_cur, u32 *p_max)
1781 {
1782 	struct mlxsw_sp_port *mlxsw_sp_port =
1783 			mlxsw_core_port_driver_priv(mlxsw_core_port);
1784 	struct mlxsw_sp *mlxsw_sp = mlxsw_sp_port->mlxsw_sp;
1785 	u16 local_port = mlxsw_sp_port->local_port;
1786 	u8 pg_buff = tc_index;
1787 	enum mlxsw_reg_sbxx_dir dir = (enum mlxsw_reg_sbxx_dir) pool_type;
1788 	struct mlxsw_sp_sb_cm *cm = mlxsw_sp_sb_cm_get(mlxsw_sp, local_port,
1789 						       pg_buff, dir);
1790 
1791 	*p_cur = mlxsw_sp_cells_bytes(mlxsw_sp, cm->occ.cur);
1792 	*p_max = mlxsw_sp_cells_bytes(mlxsw_sp, cm->occ.max);
1793 	return 0;
1794 }
1795