1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
2 /* QLogic qed NIC Driver
3  * Copyright (c) 2015-2017  QLogic Corporation
4  * Copyright (c) 2019-2021 Marvell International Ltd.
5  */
6 
7 #include <linux/types.h>
8 #include <linux/crc8.h>
9 #include <linux/delay.h>
10 #include <linux/kernel.h>
11 #include <linux/slab.h>
12 #include <linux/string.h>
13 #include "qed_hsi.h"
14 #include "qed_hw.h"
15 #include "qed_init_ops.h"
16 #include "qed_iro_hsi.h"
17 #include "qed_reg_addr.h"
18 
19 #define CDU_VALIDATION_DEFAULT_CFG CDU_CONTEXT_VALIDATION_DEFAULT_CFG
20 
21 static u16 con_region_offsets[3][NUM_OF_CONNECTION_TYPES] = {
22 	{400, 336, 352, 368, 304, 384, 416, 352},	/* region 3 offsets */
23 	{528, 496, 416, 512, 448, 512, 544, 480},	/* region 4 offsets */
24 	{608, 544, 496, 576, 576, 592, 624, 560}	/* region 5 offsets */
25 };
26 
27 static u16 task_region_offsets[1][NUM_OF_CONNECTION_TYPES] = {
28 	{240, 240, 112, 0, 0, 0, 0, 96}	/* region 1 offsets */
29 };
30 
31 /* General constants */
32 #define QM_PQ_MEM_4KB(pq_size)	(pq_size ? DIV_ROUND_UP((pq_size + 1) *	\
33 							QM_PQ_ELEMENT_SIZE, \
34 							0x1000) : 0)
35 #define QM_PQ_SIZE_256B(pq_size)	(pq_size ? DIV_ROUND_UP(pq_size, \
36 								0x100) - 1 : 0)
37 #define QM_INVALID_PQ_ID		0xffff
38 
39 /* Max link speed (in Mbps) */
40 #define QM_MAX_LINK_SPEED               100000
41 
42 /* Feature enable */
43 #define QM_BYPASS_EN	1
44 #define QM_BYTE_CRD_EN	1
45 
46 /* Initial VOQ byte credit */
47 #define QM_INITIAL_VOQ_BYTE_CRD         98304
48 /* Other PQ constants */
49 #define QM_OTHER_PQS_PER_PF	4
50 
51 /* VOQ constants */
52 #define MAX_NUM_VOQS	(MAX_NUM_PORTS_K2 * NUM_TCS_4PORT_K2)
53 #define VOQS_BIT_MASK	(BIT(MAX_NUM_VOQS) - 1)
54 
55 /* WFQ constants */
56 
57 /* PF WFQ increment value, 0x9000 = 4*9*1024 */
58 #define QM_PF_WFQ_INC_VAL(weight)       ((weight) * 0x9000)
59 
60 /* PF WFQ Upper bound, in MB, 10 * burst size of 1ms in 50Gbps */
61 #define QM_PF_WFQ_UPPER_BOUND           62500000
62 
63 /* PF WFQ max increment value, 0.7 * upper bound */
64 #define QM_PF_WFQ_MAX_INC_VAL           ((QM_PF_WFQ_UPPER_BOUND * 7) / 10)
65 
66 /* Number of VOQs in E5 PF WFQ credit register (QmWfqCrd) */
67 #define QM_PF_WFQ_CRD_E5_NUM_VOQS       16
68 
69 /* VP WFQ increment value */
70 #define QM_VP_WFQ_INC_VAL(weight)       ((weight) * QM_VP_WFQ_MIN_INC_VAL)
71 
72 /* VP WFQ min increment value */
73 #define QM_VP_WFQ_MIN_INC_VAL           10800
74 
75 /* VP WFQ max increment value, 2^30 */
76 #define QM_VP_WFQ_MAX_INC_VAL           0x40000000
77 
78 /* VP WFQ bypass threshold */
79 #define QM_VP_WFQ_BYPASS_THRESH         (QM_VP_WFQ_MIN_INC_VAL - 100)
80 
81 /* VP RL credit task cost */
82 #define QM_VP_RL_CRD_TASK_COST          9700
83 
84 /* Bit of VOQ in VP WFQ PQ map */
85 #define QM_VP_WFQ_PQ_VOQ_SHIFT          0
86 
87 /* Bit of PF in VP WFQ PQ map */
88 #define QM_VP_WFQ_PQ_PF_SHIFT   5
89 
90 /* RL constants */
91 
92 /* Period in us */
93 #define QM_RL_PERIOD	5
94 
95 /* Period in 25MHz cycles */
96 #define QM_RL_PERIOD_CLK_25M	(25 * QM_RL_PERIOD)
97 
98 /* RL increment value - rate is specified in mbps */
99 #define QM_RL_INC_VAL(rate)                     ({	\
100 						typeof(rate) __rate = (rate); \
101 						max_t(u32,		\
102 						(u32)(((__rate ? __rate : \
103 						100000) *		\
104 						QM_RL_PERIOD *		\
105 						101) / (8 * 100)), 1); })
106 
107 /* PF RL Upper bound is set to 10 * burst size of 1ms in 50Gbps */
108 #define QM_PF_RL_UPPER_BOUND	62500000
109 
110 /* Max PF RL increment value is 0.7 * upper bound */
111 #define QM_PF_RL_MAX_INC_VAL	((QM_PF_RL_UPPER_BOUND * 7) / 10)
112 
113 /* QCN RL Upper bound, speed is in Mpbs */
114 #define QM_GLOBAL_RL_UPPER_BOUND(speed)         ((u32)max_t( \
115 		u32,					    \
116 		(u32)(((speed) *			    \
117 		       QM_RL_PERIOD * 101) / (8 * 100)),    \
118 		QM_VP_RL_CRD_TASK_COST			    \
119 		+ 1000))
120 
121 /* AFullOprtnstcCrdMask constants */
122 #define QM_OPPOR_LINE_VOQ_DEF	1
123 #define QM_OPPOR_FW_STOP_DEF	0
124 #define QM_OPPOR_PQ_EMPTY_DEF	1
125 
126 /* Command Queue constants */
127 
128 /* Pure LB CmdQ lines (+spare) */
129 #define PBF_CMDQ_PURE_LB_LINES	150
130 
131 #define PBF_CMDQ_LINES_RT_OFFSET(ext_voq) \
132 	(PBF_REG_YCMD_QS_NUM_LINES_VOQ0_RT_OFFSET + \
133 	 (ext_voq) * (PBF_REG_YCMD_QS_NUM_LINES_VOQ1_RT_OFFSET - \
134 		PBF_REG_YCMD_QS_NUM_LINES_VOQ0_RT_OFFSET))
135 
136 #define PBF_BTB_GUARANTEED_RT_OFFSET(ext_voq) \
137 	(PBF_REG_BTB_GUARANTEED_VOQ0_RT_OFFSET + \
138 	 (ext_voq) * (PBF_REG_BTB_GUARANTEED_VOQ1_RT_OFFSET - \
139 		PBF_REG_BTB_GUARANTEED_VOQ0_RT_OFFSET))
140 
141 /* Returns the VOQ line credit for the specified number of PBF command lines.
142  * PBF lines are specified in 256b units.
143  */
144 #define QM_VOQ_LINE_CRD(pbf_cmd_lines) \
145 	((((pbf_cmd_lines) - 4) * 2) | QM_LINE_CRD_REG_SIGN_BIT)
146 
147 /* BTB: blocks constants (block size = 256B) */
148 
149 /* 256B blocks in 9700B packet */
150 #define BTB_JUMBO_PKT_BLOCKS	38
151 
152 /* Headroom per-port */
153 #define BTB_HEADROOM_BLOCKS	BTB_JUMBO_PKT_BLOCKS
154 #define BTB_PURE_LB_FACTOR	10
155 
156 /* Factored (hence really 0.7) */
157 #define BTB_PURE_LB_RATIO	7
158 
159 /* QM stop command constants */
160 #define QM_STOP_PQ_MASK_WIDTH		32
161 #define QM_STOP_CMD_ADDR		2
162 #define QM_STOP_CMD_STRUCT_SIZE		2
163 #define QM_STOP_CMD_PAUSE_MASK_OFFSET	0
164 #define QM_STOP_CMD_PAUSE_MASK_SHIFT	0
165 #define QM_STOP_CMD_PAUSE_MASK_MASK	-1
166 #define QM_STOP_CMD_GROUP_ID_OFFSET	1
167 #define QM_STOP_CMD_GROUP_ID_SHIFT	16
168 #define QM_STOP_CMD_GROUP_ID_MASK	15
169 #define QM_STOP_CMD_PQ_TYPE_OFFSET	1
170 #define QM_STOP_CMD_PQ_TYPE_SHIFT	24
171 #define QM_STOP_CMD_PQ_TYPE_MASK	1
172 #define QM_STOP_CMD_MAX_POLL_COUNT	100
173 #define QM_STOP_CMD_POLL_PERIOD_US	500
174 
175 /* QM command macros */
176 #define QM_CMD_STRUCT_SIZE(cmd)	cmd ## _STRUCT_SIZE
177 #define QM_CMD_SET_FIELD(var, cmd, field, value) \
178 	SET_FIELD(var[cmd ## _ ## field ## _OFFSET], \
179 		  cmd ## _ ## field, \
180 		  value)
181 
182 #define QM_INIT_TX_PQ_MAP(p_hwfn, map, pq_id, vp_pq_id, rl_valid,	      \
183 			  rl_id, ext_voq, wrr)				      \
184 	do {								      \
185 		u32 __reg = 0;						      \
186 									      \
187 		BUILD_BUG_ON(sizeof((map).reg) != sizeof(__reg));	      \
188 		memset(&(map), 0, sizeof(map));				      \
189 		SET_FIELD(__reg, QM_RF_PQ_MAP_PQ_VALID, 1);	      \
190 		SET_FIELD(__reg, QM_RF_PQ_MAP_RL_VALID,	      \
191 			  !!(rl_valid));				      \
192 		SET_FIELD(__reg, QM_RF_PQ_MAP_VP_PQ_ID, (vp_pq_id)); \
193 		SET_FIELD(__reg, QM_RF_PQ_MAP_RL_ID, (rl_id));	      \
194 		SET_FIELD(__reg, QM_RF_PQ_MAP_VOQ, (ext_voq));	      \
195 		SET_FIELD(__reg, QM_RF_PQ_MAP_WRR_WEIGHT_GROUP,      \
196 			  (wrr));					      \
197 									      \
198 		STORE_RT_REG((p_hwfn), QM_REG_TXPQMAP_RT_OFFSET + (pq_id),    \
199 			     __reg);					      \
200 		(map).reg = cpu_to_le32(__reg);				      \
201 	} while (0)
202 
203 #define WRITE_PQ_INFO_TO_RAM	1
204 #define PQ_INFO_ELEMENT(vp, pf, tc, port, rl_valid, rl) \
205 	(((vp) << 0) | ((pf) << 12) | ((tc) << 16) | ((port) << 20) | \
206 	((rl_valid ? 1 : 0) << 22) | (((rl) & 255) << 24) | \
207 	(((rl) >> 8) << 9))
208 
209 #define PQ_INFO_RAM_GRC_ADDRESS(pq_id) \
210 	(XSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM + \
211 	XSTORM_PQ_INFO_OFFSET(pq_id))
212 
213 static const char * const s_protocol_types[] = {
214 	"PROTOCOLID_ISCSI", "PROTOCOLID_FCOE", "PROTOCOLID_ROCE",
215 	"PROTOCOLID_CORE", "PROTOCOLID_ETH", "PROTOCOLID_IWARP",
216 	"PROTOCOLID_TOE", "PROTOCOLID_PREROCE", "PROTOCOLID_COMMON",
217 	"PROTOCOLID_TCP", "PROTOCOLID_RDMA", "PROTOCOLID_SCSI",
218 };
219 
220 static const char *s_ramrod_cmd_ids[][28] = {
221 	{
222 	"ISCSI_RAMROD_CMD_ID_UNUSED", "ISCSI_RAMROD_CMD_ID_INIT_FUNC",
223 	 "ISCSI_RAMROD_CMD_ID_DESTROY_FUNC",
224 	 "ISCSI_RAMROD_CMD_ID_OFFLOAD_CONN",
225 	 "ISCSI_RAMROD_CMD_ID_UPDATE_CONN",
226 	 "ISCSI_RAMROD_CMD_ID_TERMINATION_CONN",
227 	 "ISCSI_RAMROD_CMD_ID_CLEAR_SQ", "ISCSI_RAMROD_CMD_ID_MAC_UPDATE",
228 	 "ISCSI_RAMROD_CMD_ID_CONN_STATS", },
229 	{ "FCOE_RAMROD_CMD_ID_INIT_FUNC", "FCOE_RAMROD_CMD_ID_DESTROY_FUNC",
230 	 "FCOE_RAMROD_CMD_ID_STAT_FUNC",
231 	 "FCOE_RAMROD_CMD_ID_OFFLOAD_CONN",
232 	 "FCOE_RAMROD_CMD_ID_TERMINATE_CONN", },
233 	{ "RDMA_RAMROD_UNUSED", "RDMA_RAMROD_FUNC_INIT",
234 	 "RDMA_RAMROD_FUNC_CLOSE", "RDMA_RAMROD_REGISTER_MR",
235 	 "RDMA_RAMROD_DEREGISTER_MR", "RDMA_RAMROD_CREATE_CQ",
236 	 "RDMA_RAMROD_RESIZE_CQ", "RDMA_RAMROD_DESTROY_CQ",
237 	 "RDMA_RAMROD_CREATE_SRQ", "RDMA_RAMROD_MODIFY_SRQ",
238 	 "RDMA_RAMROD_DESTROY_SRQ", "RDMA_RAMROD_START_NS_TRACKING",
239 	 "RDMA_RAMROD_STOP_NS_TRACKING", "ROCE_RAMROD_CREATE_QP",
240 	 "ROCE_RAMROD_MODIFY_QP", "ROCE_RAMROD_QUERY_QP",
241 	 "ROCE_RAMROD_DESTROY_QP", "ROCE_RAMROD_CREATE_UD_QP",
242 	 "ROCE_RAMROD_DESTROY_UD_QP", "ROCE_RAMROD_FUNC_UPDATE",
243 	 "ROCE_RAMROD_SUSPEND_QP", "ROCE_RAMROD_QUERY_SUSPENDED_QP",
244 	 "ROCE_RAMROD_CREATE_SUSPENDED_QP", "ROCE_RAMROD_RESUME_QP",
245 	 "ROCE_RAMROD_SUSPEND_UD_QP", "ROCE_RAMROD_RESUME_UD_QP",
246 	 "ROCE_RAMROD_CREATE_SUSPENDED_UD_QP", "ROCE_RAMROD_FLUSH_DPT_QP", },
247 	{ "CORE_RAMROD_UNUSED", "CORE_RAMROD_RX_QUEUE_START",
248 	 "CORE_RAMROD_TX_QUEUE_START", "CORE_RAMROD_RX_QUEUE_STOP",
249 	 "CORE_RAMROD_TX_QUEUE_STOP",
250 	 "CORE_RAMROD_RX_QUEUE_FLUSH",
251 	 "CORE_RAMROD_TX_QUEUE_UPDATE", "CORE_RAMROD_QUEUE_STATS_QUERY", },
252 	{ "ETH_RAMROD_UNUSED", "ETH_RAMROD_VPORT_START",
253 	 "ETH_RAMROD_VPORT_UPDATE", "ETH_RAMROD_VPORT_STOP",
254 	 "ETH_RAMROD_RX_QUEUE_START", "ETH_RAMROD_RX_QUEUE_STOP",
255 	 "ETH_RAMROD_TX_QUEUE_START", "ETH_RAMROD_TX_QUEUE_STOP",
256 	 "ETH_RAMROD_FILTERS_UPDATE", "ETH_RAMROD_RX_QUEUE_UPDATE",
257 	 "ETH_RAMROD_RX_CREATE_OPENFLOW_ACTION",
258 	 "ETH_RAMROD_RX_ADD_OPENFLOW_FILTER",
259 	 "ETH_RAMROD_RX_DELETE_OPENFLOW_FILTER",
260 	 "ETH_RAMROD_RX_ADD_UDP_FILTER",
261 	 "ETH_RAMROD_RX_DELETE_UDP_FILTER",
262 	 "ETH_RAMROD_RX_CREATE_GFT_ACTION",
263 	 "ETH_RAMROD_RX_UPDATE_GFT_FILTER", "ETH_RAMROD_TX_QUEUE_UPDATE",
264 	 "ETH_RAMROD_RGFS_FILTER_ADD", "ETH_RAMROD_RGFS_FILTER_DEL",
265 	 "ETH_RAMROD_TGFS_FILTER_ADD", "ETH_RAMROD_TGFS_FILTER_DEL",
266 	 "ETH_RAMROD_GFS_COUNTERS_REPORT_REQUEST", },
267 	{ "RDMA_RAMROD_UNUSED", "RDMA_RAMROD_FUNC_INIT",
268 	 "RDMA_RAMROD_FUNC_CLOSE", "RDMA_RAMROD_REGISTER_MR",
269 	 "RDMA_RAMROD_DEREGISTER_MR", "RDMA_RAMROD_CREATE_CQ",
270 	 "RDMA_RAMROD_RESIZE_CQ", "RDMA_RAMROD_DESTROY_CQ",
271 	 "RDMA_RAMROD_CREATE_SRQ", "RDMA_RAMROD_MODIFY_SRQ",
272 	 "RDMA_RAMROD_DESTROY_SRQ", "RDMA_RAMROD_START_NS_TRACKING",
273 	 "RDMA_RAMROD_STOP_NS_TRACKING",
274 	 "IWARP_RAMROD_CMD_ID_TCP_OFFLOAD",
275 	 "IWARP_RAMROD_CMD_ID_MPA_OFFLOAD",
276 	 "IWARP_RAMROD_CMD_ID_MPA_OFFLOAD_SEND_RTR",
277 	 "IWARP_RAMROD_CMD_ID_CREATE_QP", "IWARP_RAMROD_CMD_ID_QUERY_QP",
278 	 "IWARP_RAMROD_CMD_ID_MODIFY_QP",
279 	 "IWARP_RAMROD_CMD_ID_DESTROY_QP",
280 	 "IWARP_RAMROD_CMD_ID_ABORT_TCP_OFFLOAD", },
281 	{ NULL }, /*TOE*/
282 	{ NULL }, /*PREROCE*/
283 	{ "COMMON_RAMROD_UNUSED", "COMMON_RAMROD_PF_START",
284 	     "COMMON_RAMROD_PF_STOP", "COMMON_RAMROD_VF_START",
285 	     "COMMON_RAMROD_VF_STOP", "COMMON_RAMROD_PF_UPDATE",
286 	     "COMMON_RAMROD_RL_UPDATE", "COMMON_RAMROD_EMPTY", }
287 };
288 
289 /******************** INTERNAL IMPLEMENTATION *********************/
290 
291 /* Returns the external VOQ number */
292 static u8 qed_get_ext_voq(struct qed_hwfn *p_hwfn,
293 			  u8 port_id, u8 tc, u8 max_phys_tcs_per_port)
294 {
295 	if (tc == PURE_LB_TC)
296 		return NUM_OF_PHYS_TCS * MAX_NUM_PORTS_BB + port_id;
297 	else
298 		return port_id * max_phys_tcs_per_port + tc;
299 }
300 
301 /* Prepare PF RL enable/disable runtime init values */
302 static void qed_enable_pf_rl(struct qed_hwfn *p_hwfn, bool pf_rl_en)
303 {
304 	STORE_RT_REG(p_hwfn, QM_REG_RLPFENABLE_RT_OFFSET, pf_rl_en ? 1 : 0);
305 	if (pf_rl_en) {
306 		u8 num_ext_voqs = MAX_NUM_VOQS;
307 		u64 voq_bit_mask = ((u64)1 << num_ext_voqs) - 1;
308 
309 		/* Enable RLs for all VOQs */
310 		STORE_RT_REG(p_hwfn,
311 			     QM_REG_RLPFVOQENABLE_RT_OFFSET,
312 			     (u32)voq_bit_mask);
313 
314 		/* Write RL period */
315 		STORE_RT_REG(p_hwfn,
316 			     QM_REG_RLPFPERIOD_RT_OFFSET, QM_RL_PERIOD_CLK_25M);
317 		STORE_RT_REG(p_hwfn,
318 			     QM_REG_RLPFPERIODTIMER_RT_OFFSET,
319 			     QM_RL_PERIOD_CLK_25M);
320 
321 		/* Set credit threshold for QM bypass flow */
322 		if (QM_BYPASS_EN)
323 			STORE_RT_REG(p_hwfn,
324 				     QM_REG_AFULLQMBYPTHRPFRL_RT_OFFSET,
325 				     QM_PF_RL_UPPER_BOUND);
326 	}
327 }
328 
329 /* Prepare PF WFQ enable/disable runtime init values */
330 static void qed_enable_pf_wfq(struct qed_hwfn *p_hwfn, bool pf_wfq_en)
331 {
332 	STORE_RT_REG(p_hwfn, QM_REG_WFQPFENABLE_RT_OFFSET, pf_wfq_en ? 1 : 0);
333 
334 	/* Set credit threshold for QM bypass flow */
335 	if (pf_wfq_en && QM_BYPASS_EN)
336 		STORE_RT_REG(p_hwfn,
337 			     QM_REG_AFULLQMBYPTHRPFWFQ_RT_OFFSET,
338 			     QM_PF_WFQ_UPPER_BOUND);
339 }
340 
341 /* Prepare global RL enable/disable runtime init values */
342 static void qed_enable_global_rl(struct qed_hwfn *p_hwfn, bool global_rl_en)
343 {
344 	STORE_RT_REG(p_hwfn, QM_REG_RLGLBLENABLE_RT_OFFSET,
345 		     global_rl_en ? 1 : 0);
346 	if (global_rl_en) {
347 		/* Write RL period (use timer 0 only) */
348 		STORE_RT_REG(p_hwfn,
349 			     QM_REG_RLGLBLPERIOD_0_RT_OFFSET,
350 			     QM_RL_PERIOD_CLK_25M);
351 		STORE_RT_REG(p_hwfn,
352 			     QM_REG_RLGLBLPERIODTIMER_0_RT_OFFSET,
353 			     QM_RL_PERIOD_CLK_25M);
354 
355 		/* Set credit threshold for QM bypass flow */
356 		if (QM_BYPASS_EN)
357 			STORE_RT_REG(p_hwfn,
358 				     QM_REG_AFULLQMBYPTHRGLBLRL_RT_OFFSET,
359 				     QM_GLOBAL_RL_UPPER_BOUND(10000) - 1);
360 	}
361 }
362 
363 /* Prepare VPORT WFQ enable/disable runtime init values */
364 static void qed_enable_vport_wfq(struct qed_hwfn *p_hwfn, bool vport_wfq_en)
365 {
366 	STORE_RT_REG(p_hwfn, QM_REG_WFQVPENABLE_RT_OFFSET,
367 		     vport_wfq_en ? 1 : 0);
368 
369 	/* Set credit threshold for QM bypass flow */
370 	if (vport_wfq_en && QM_BYPASS_EN)
371 		STORE_RT_REG(p_hwfn,
372 			     QM_REG_AFULLQMBYPTHRVPWFQ_RT_OFFSET,
373 			     QM_VP_WFQ_BYPASS_THRESH);
374 }
375 
376 /* Prepare runtime init values to allocate PBF command queue lines for
377  * the specified VOQ.
378  */
379 static void qed_cmdq_lines_voq_rt_init(struct qed_hwfn *p_hwfn,
380 				       u8 ext_voq, u16 cmdq_lines)
381 {
382 	u32 qm_line_crd = QM_VOQ_LINE_CRD(cmdq_lines);
383 
384 	OVERWRITE_RT_REG(p_hwfn, PBF_CMDQ_LINES_RT_OFFSET(ext_voq),
385 			 (u32)cmdq_lines);
386 	STORE_RT_REG(p_hwfn, QM_REG_VOQCRDLINE_RT_OFFSET + ext_voq,
387 		     qm_line_crd);
388 	STORE_RT_REG(p_hwfn, QM_REG_VOQINITCRDLINE_RT_OFFSET + ext_voq,
389 		     qm_line_crd);
390 }
391 
392 /* Prepare runtime init values to allocate PBF command queue lines. */
393 static void
394 qed_cmdq_lines_rt_init(struct qed_hwfn *p_hwfn,
395 		       u8 max_ports_per_engine,
396 		       u8 max_phys_tcs_per_port,
397 		       struct init_qm_port_params port_params[MAX_NUM_PORTS])
398 {
399 	u8 tc, ext_voq, port_id, num_tcs_in_port;
400 	u8 num_ext_voqs = MAX_NUM_VOQS;
401 
402 	/* Clear PBF lines of all VOQs */
403 	for (ext_voq = 0; ext_voq < num_ext_voqs; ext_voq++)
404 		STORE_RT_REG(p_hwfn, PBF_CMDQ_LINES_RT_OFFSET(ext_voq), 0);
405 
406 	for (port_id = 0; port_id < max_ports_per_engine; port_id++) {
407 		u16 phys_lines, phys_lines_per_tc;
408 
409 		if (!port_params[port_id].active)
410 			continue;
411 
412 		/* Find number of command queue lines to divide between the
413 		 * active physical TCs.
414 		 */
415 		phys_lines = port_params[port_id].num_pbf_cmd_lines;
416 		phys_lines -= PBF_CMDQ_PURE_LB_LINES;
417 
418 		/* Find #lines per active physical TC */
419 		num_tcs_in_port = 0;
420 		for (tc = 0; tc < max_phys_tcs_per_port; tc++)
421 			if (((port_params[port_id].active_phys_tcs >>
422 			      tc) & 0x1) == 1)
423 				num_tcs_in_port++;
424 		phys_lines_per_tc = phys_lines / num_tcs_in_port;
425 
426 		/* Init registers per active TC */
427 		for (tc = 0; tc < max_phys_tcs_per_port; tc++) {
428 			ext_voq = qed_get_ext_voq(p_hwfn,
429 						  port_id,
430 						  tc, max_phys_tcs_per_port);
431 			if (((port_params[port_id].active_phys_tcs >>
432 			      tc) & 0x1) == 1)
433 				qed_cmdq_lines_voq_rt_init(p_hwfn,
434 							   ext_voq,
435 							   phys_lines_per_tc);
436 		}
437 
438 		/* Init registers for pure LB TC */
439 		ext_voq = qed_get_ext_voq(p_hwfn,
440 					  port_id,
441 					  PURE_LB_TC, max_phys_tcs_per_port);
442 		qed_cmdq_lines_voq_rt_init(p_hwfn, ext_voq,
443 					   PBF_CMDQ_PURE_LB_LINES);
444 	}
445 }
446 
447 /* Prepare runtime init values to allocate guaranteed BTB blocks for the
448  * specified port. The guaranteed BTB space is divided between the TCs as
449  * follows (shared space Is currently not used):
450  * 1. Parameters:
451  *    B - BTB blocks for this port
452  *    C - Number of physical TCs for this port
453  * 2. Calculation:
454  *    a. 38 blocks (9700B jumbo frame) are allocated for global per port
455  *	 headroom.
456  *    b. B = B - 38 (remainder after global headroom allocation).
457  *    c. MAX(38,B/(C+0.7)) blocks are allocated for the pure LB VOQ.
458  *    d. B = B - MAX(38, B/(C+0.7)) (remainder after pure LB allocation).
459  *    e. B/C blocks are allocated for each physical TC.
460  * Assumptions:
461  * - MTU is up to 9700 bytes (38 blocks)
462  * - All TCs are considered symmetrical (same rate and packet size)
463  * - No optimization for lossy TC (all are considered lossless). Shared space
464  *   is not enabled and allocated for each TC.
465  */
466 static void
467 qed_btb_blocks_rt_init(struct qed_hwfn *p_hwfn,
468 		       u8 max_ports_per_engine,
469 		       u8 max_phys_tcs_per_port,
470 		       struct init_qm_port_params port_params[MAX_NUM_PORTS])
471 {
472 	u32 usable_blocks, pure_lb_blocks, phys_blocks;
473 	u8 tc, ext_voq, port_id, num_tcs_in_port;
474 
475 	for (port_id = 0; port_id < max_ports_per_engine; port_id++) {
476 		if (!port_params[port_id].active)
477 			continue;
478 
479 		/* Subtract headroom blocks */
480 		usable_blocks = port_params[port_id].num_btb_blocks -
481 				BTB_HEADROOM_BLOCKS;
482 
483 		/* Find blocks per physical TC. Use factor to avoid floating
484 		 * arithmethic.
485 		 */
486 		num_tcs_in_port = 0;
487 		for (tc = 0; tc < NUM_OF_PHYS_TCS; tc++)
488 			if (((port_params[port_id].active_phys_tcs >>
489 			      tc) & 0x1) == 1)
490 				num_tcs_in_port++;
491 
492 		pure_lb_blocks = (usable_blocks * BTB_PURE_LB_FACTOR) /
493 				 (num_tcs_in_port * BTB_PURE_LB_FACTOR +
494 				  BTB_PURE_LB_RATIO);
495 		pure_lb_blocks = max_t(u32, BTB_JUMBO_PKT_BLOCKS,
496 				       pure_lb_blocks / BTB_PURE_LB_FACTOR);
497 		phys_blocks = (usable_blocks - pure_lb_blocks) /
498 			      num_tcs_in_port;
499 
500 		/* Init physical TCs */
501 		for (tc = 0; tc < NUM_OF_PHYS_TCS; tc++) {
502 			if (((port_params[port_id].active_phys_tcs >>
503 			      tc) & 0x1) == 1) {
504 				ext_voq =
505 					qed_get_ext_voq(p_hwfn,
506 							port_id,
507 							tc,
508 							max_phys_tcs_per_port);
509 				STORE_RT_REG(p_hwfn,
510 					     PBF_BTB_GUARANTEED_RT_OFFSET
511 					     (ext_voq), phys_blocks);
512 			}
513 		}
514 
515 		/* Init pure LB TC */
516 		ext_voq = qed_get_ext_voq(p_hwfn,
517 					  port_id,
518 					  PURE_LB_TC, max_phys_tcs_per_port);
519 		STORE_RT_REG(p_hwfn, PBF_BTB_GUARANTEED_RT_OFFSET(ext_voq),
520 			     pure_lb_blocks);
521 	}
522 }
523 
524 /* Prepare runtime init values for the specified RL.
525  * Set max link speed (100Gbps) per rate limiter.
526  * Return -1 on error.
527  */
528 static int qed_global_rl_rt_init(struct qed_hwfn *p_hwfn)
529 {
530 	u32 upper_bound = QM_GLOBAL_RL_UPPER_BOUND(QM_MAX_LINK_SPEED) |
531 			  (u32)QM_RL_CRD_REG_SIGN_BIT;
532 	u32 inc_val;
533 	u16 rl_id;
534 
535 	/* Go over all global RLs */
536 	for (rl_id = 0; rl_id < MAX_QM_GLOBAL_RLS; rl_id++) {
537 		inc_val = QM_RL_INC_VAL(QM_MAX_LINK_SPEED);
538 
539 		STORE_RT_REG(p_hwfn,
540 			     QM_REG_RLGLBLCRD_RT_OFFSET + rl_id,
541 			     (u32)QM_RL_CRD_REG_SIGN_BIT);
542 		STORE_RT_REG(p_hwfn,
543 			     QM_REG_RLGLBLUPPERBOUND_RT_OFFSET + rl_id,
544 			     upper_bound);
545 		STORE_RT_REG(p_hwfn,
546 			     QM_REG_RLGLBLINCVAL_RT_OFFSET + rl_id, inc_val);
547 	}
548 
549 	return 0;
550 }
551 
552 /* Returns the upper bound for the specified Vport RL parameters.
553  * link_speed is in Mbps.
554  * Returns 0 in case of error.
555  */
556 static u32 qed_get_vport_rl_upper_bound(enum init_qm_rl_type vport_rl_type,
557 					u32 link_speed)
558 {
559 	switch (vport_rl_type) {
560 	case QM_RL_TYPE_NORMAL:
561 		return QM_INITIAL_VOQ_BYTE_CRD;
562 	case QM_RL_TYPE_QCN:
563 		return QM_GLOBAL_RL_UPPER_BOUND(link_speed);
564 	default:
565 		return 0;
566 	}
567 }
568 
569 /* Prepare VPORT RL runtime init values.
570  * Return -1 on error.
571  */
572 static int qed_vport_rl_rt_init(struct qed_hwfn *p_hwfn,
573 				u16 start_rl,
574 				u16 num_rls,
575 				u32 link_speed,
576 				struct init_qm_rl_params *rl_params)
577 {
578 	u16 i, rl_id;
579 
580 	if (num_rls && start_rl + num_rls >= MAX_QM_GLOBAL_RLS) {
581 		DP_NOTICE(p_hwfn, "Invalid rate limiter configuration\n");
582 		return -1;
583 	}
584 
585 	/* Go over all PF VPORTs */
586 	for (i = 0, rl_id = start_rl; i < num_rls; i++, rl_id++) {
587 		u32 upper_bound, inc_val;
588 
589 		upper_bound =
590 		    qed_get_vport_rl_upper_bound((enum init_qm_rl_type)
591 						 rl_params[i].vport_rl_type,
592 						 link_speed);
593 
594 		inc_val =
595 		    QM_RL_INC_VAL(rl_params[i].vport_rl ?
596 				  rl_params[i].vport_rl : link_speed);
597 		if (inc_val > upper_bound) {
598 			DP_NOTICE(p_hwfn,
599 				  "Invalid RL rate - limit configuration\n");
600 			return -1;
601 		}
602 
603 		STORE_RT_REG(p_hwfn, QM_REG_RLGLBLCRD_RT_OFFSET + rl_id,
604 			     (u32)QM_RL_CRD_REG_SIGN_BIT);
605 		STORE_RT_REG(p_hwfn, QM_REG_RLGLBLUPPERBOUND_RT_OFFSET + rl_id,
606 			     upper_bound | (u32)QM_RL_CRD_REG_SIGN_BIT);
607 		STORE_RT_REG(p_hwfn, QM_REG_RLGLBLINCVAL_RT_OFFSET + rl_id,
608 			     inc_val);
609 	}
610 
611 	return 0;
612 }
613 
614 /* Prepare Tx PQ mapping runtime init values for the specified PF */
615 static int qed_tx_pq_map_rt_init(struct qed_hwfn *p_hwfn,
616 				 struct qed_ptt *p_ptt,
617 				 struct qed_qm_pf_rt_init_params *p_params,
618 				 u32 base_mem_addr_4kb)
619 {
620 	u32 tx_pq_vf_mask[MAX_QM_TX_QUEUES / QM_PF_QUEUE_GROUP_SIZE] = { 0 };
621 	struct init_qm_vport_params *vport_params = p_params->vport_params;
622 	u32 num_tx_pq_vf_masks = MAX_QM_TX_QUEUES / QM_PF_QUEUE_GROUP_SIZE;
623 	u16 num_pqs, first_pq_group, last_pq_group, i, j, pq_id, pq_group;
624 	struct init_qm_pq_params *pq_params = p_params->pq_params;
625 	u32 pq_mem_4kb, vport_pq_mem_4kb, mem_addr_4kb;
626 
627 	num_pqs = p_params->num_pf_pqs + p_params->num_vf_pqs;
628 
629 	first_pq_group = p_params->start_pq / QM_PF_QUEUE_GROUP_SIZE;
630 	last_pq_group = (p_params->start_pq + num_pqs - 1) /
631 			QM_PF_QUEUE_GROUP_SIZE;
632 
633 	pq_mem_4kb = QM_PQ_MEM_4KB(p_params->num_pf_cids);
634 	vport_pq_mem_4kb = QM_PQ_MEM_4KB(p_params->num_vf_cids);
635 	mem_addr_4kb = base_mem_addr_4kb;
636 
637 	/* Set mapping from PQ group to PF */
638 	for (pq_group = first_pq_group; pq_group <= last_pq_group; pq_group++)
639 		STORE_RT_REG(p_hwfn, QM_REG_PQTX2PF_0_RT_OFFSET + pq_group,
640 			     (u32)(p_params->pf_id));
641 
642 	/* Set PQ sizes */
643 	STORE_RT_REG(p_hwfn, QM_REG_MAXPQSIZE_0_RT_OFFSET,
644 		     QM_PQ_SIZE_256B(p_params->num_pf_cids));
645 	STORE_RT_REG(p_hwfn, QM_REG_MAXPQSIZE_1_RT_OFFSET,
646 		     QM_PQ_SIZE_256B(p_params->num_vf_cids));
647 
648 	/* Go over all Tx PQs */
649 	for (i = 0, pq_id = p_params->start_pq; i < num_pqs; i++, pq_id++) {
650 		u16 *p_first_tx_pq_id, vport_id_in_pf;
651 		struct qm_rf_pq_map tx_pq_map;
652 		u8 tc_id = pq_params[i].tc_id;
653 		bool is_vf_pq;
654 		u8 ext_voq;
655 
656 		ext_voq = qed_get_ext_voq(p_hwfn,
657 					  pq_params[i].port_id,
658 					  tc_id,
659 					  p_params->max_phys_tcs_per_port);
660 		is_vf_pq = (i >= p_params->num_pf_pqs);
661 
662 		/* Update first Tx PQ of VPORT/TC */
663 		vport_id_in_pf = pq_params[i].vport_id - p_params->start_vport;
664 		p_first_tx_pq_id =
665 		    &vport_params[vport_id_in_pf].first_tx_pq_id[tc_id];
666 		if (*p_first_tx_pq_id == QM_INVALID_PQ_ID) {
667 			u32 map_val =
668 				(ext_voq << QM_VP_WFQ_PQ_VOQ_SHIFT) |
669 				(p_params->pf_id << QM_VP_WFQ_PQ_PF_SHIFT);
670 
671 			/* Create new VP PQ */
672 			*p_first_tx_pq_id = pq_id;
673 
674 			/* Map VP PQ to VOQ and PF */
675 			STORE_RT_REG(p_hwfn,
676 				     QM_REG_WFQVPMAP_RT_OFFSET +
677 				     *p_first_tx_pq_id,
678 				     map_val);
679 		}
680 
681 		/* Prepare PQ map entry */
682 		QM_INIT_TX_PQ_MAP(p_hwfn,
683 				  tx_pq_map,
684 				  pq_id,
685 				  *p_first_tx_pq_id,
686 				  pq_params[i].rl_valid,
687 				  pq_params[i].rl_id,
688 				  ext_voq, pq_params[i].wrr_group);
689 
690 		/* Set PQ base address */
691 		STORE_RT_REG(p_hwfn,
692 			     QM_REG_BASEADDRTXPQ_RT_OFFSET + pq_id,
693 			     mem_addr_4kb);
694 
695 		/* Clear PQ pointer table entry (64 bit) */
696 		if (p_params->is_pf_loading)
697 			for (j = 0; j < 2; j++)
698 				STORE_RT_REG(p_hwfn,
699 					     QM_REG_PTRTBLTX_RT_OFFSET +
700 					     (pq_id * 2) + j, 0);
701 
702 		/* Write PQ info to RAM */
703 		if (WRITE_PQ_INFO_TO_RAM != 0) {
704 			u32 pq_info = 0;
705 
706 			pq_info = PQ_INFO_ELEMENT(*p_first_tx_pq_id,
707 						  p_params->pf_id,
708 						  tc_id,
709 						  pq_params[i].port_id,
710 						  pq_params[i].rl_valid,
711 						  pq_params[i].rl_id);
712 			qed_wr(p_hwfn, p_ptt, PQ_INFO_RAM_GRC_ADDRESS(pq_id),
713 			       pq_info);
714 		}
715 
716 		/* If VF PQ, add indication to PQ VF mask */
717 		if (is_vf_pq) {
718 			tx_pq_vf_mask[pq_id /
719 				      QM_PF_QUEUE_GROUP_SIZE] |=
720 			    BIT((pq_id % QM_PF_QUEUE_GROUP_SIZE));
721 			mem_addr_4kb += vport_pq_mem_4kb;
722 		} else {
723 			mem_addr_4kb += pq_mem_4kb;
724 		}
725 	}
726 
727 	/* Store Tx PQ VF mask to size select register */
728 	for (i = 0; i < num_tx_pq_vf_masks; i++)
729 		if (tx_pq_vf_mask[i])
730 			STORE_RT_REG(p_hwfn,
731 				     QM_REG_MAXPQSIZETXSEL_0_RT_OFFSET + i,
732 				     tx_pq_vf_mask[i]);
733 
734 	return 0;
735 }
736 
737 /* Prepare Other PQ mapping runtime init values for the specified PF */
738 static void qed_other_pq_map_rt_init(struct qed_hwfn *p_hwfn,
739 				     u8 pf_id,
740 				     bool is_pf_loading,
741 				     u32 num_pf_cids,
742 				     u32 num_tids, u32 base_mem_addr_4kb)
743 {
744 	u32 pq_size, pq_mem_4kb, mem_addr_4kb;
745 	u16 i, j, pq_id, pq_group;
746 
747 	/* A single other PQ group is used in each PF, where PQ group i is used
748 	 * in PF i.
749 	 */
750 	pq_group = pf_id;
751 	pq_size = num_pf_cids + num_tids;
752 	pq_mem_4kb = QM_PQ_MEM_4KB(pq_size);
753 	mem_addr_4kb = base_mem_addr_4kb;
754 
755 	/* Map PQ group to PF */
756 	STORE_RT_REG(p_hwfn, QM_REG_PQOTHER2PF_0_RT_OFFSET + pq_group,
757 		     (u32)(pf_id));
758 
759 	/* Set PQ sizes */
760 	STORE_RT_REG(p_hwfn, QM_REG_MAXPQSIZE_2_RT_OFFSET,
761 		     QM_PQ_SIZE_256B(pq_size));
762 
763 	for (i = 0, pq_id = pf_id * QM_PF_QUEUE_GROUP_SIZE;
764 	     i < QM_OTHER_PQS_PER_PF; i++, pq_id++) {
765 		/* Set PQ base address */
766 		STORE_RT_REG(p_hwfn,
767 			     QM_REG_BASEADDROTHERPQ_RT_OFFSET + pq_id,
768 			     mem_addr_4kb);
769 
770 		/* Clear PQ pointer table entry */
771 		if (is_pf_loading)
772 			for (j = 0; j < 2; j++)
773 				STORE_RT_REG(p_hwfn,
774 					     QM_REG_PTRTBLOTHER_RT_OFFSET +
775 					     (pq_id * 2) + j, 0);
776 
777 		mem_addr_4kb += pq_mem_4kb;
778 	}
779 }
780 
781 /* Prepare PF WFQ runtime init values for the specified PF.
782  * Return -1 on error.
783  */
784 static int qed_pf_wfq_rt_init(struct qed_hwfn *p_hwfn,
785 			      struct qed_qm_pf_rt_init_params *p_params)
786 {
787 	u16 num_tx_pqs = p_params->num_pf_pqs + p_params->num_vf_pqs;
788 	struct init_qm_pq_params *pq_params = p_params->pq_params;
789 	u32 inc_val, crd_reg_offset;
790 	u8 ext_voq;
791 	u16 i;
792 
793 	inc_val = QM_PF_WFQ_INC_VAL(p_params->pf_wfq);
794 	if (!inc_val || inc_val > QM_PF_WFQ_MAX_INC_VAL) {
795 		DP_NOTICE(p_hwfn, "Invalid PF WFQ weight configuration\n");
796 		return -1;
797 	}
798 
799 	for (i = 0; i < num_tx_pqs; i++) {
800 		ext_voq = qed_get_ext_voq(p_hwfn,
801 					  pq_params[i].port_id,
802 					  pq_params[i].tc_id,
803 					  p_params->max_phys_tcs_per_port);
804 		crd_reg_offset =
805 			(p_params->pf_id < MAX_NUM_PFS_BB ?
806 			 QM_REG_WFQPFCRD_RT_OFFSET :
807 			 QM_REG_WFQPFCRD_MSB_RT_OFFSET) +
808 			ext_voq * MAX_NUM_PFS_BB +
809 			(p_params->pf_id % MAX_NUM_PFS_BB);
810 		OVERWRITE_RT_REG(p_hwfn,
811 				 crd_reg_offset, (u32)QM_WFQ_CRD_REG_SIGN_BIT);
812 	}
813 
814 	STORE_RT_REG(p_hwfn,
815 		     QM_REG_WFQPFUPPERBOUND_RT_OFFSET + p_params->pf_id,
816 		     QM_PF_WFQ_UPPER_BOUND | (u32)QM_WFQ_CRD_REG_SIGN_BIT);
817 	STORE_RT_REG(p_hwfn, QM_REG_WFQPFWEIGHT_RT_OFFSET + p_params->pf_id,
818 		     inc_val);
819 
820 	return 0;
821 }
822 
823 /* Prepare PF RL runtime init values for the specified PF.
824  * Return -1 on error.
825  */
826 static int qed_pf_rl_rt_init(struct qed_hwfn *p_hwfn, u8 pf_id, u32 pf_rl)
827 {
828 	u32 inc_val = QM_RL_INC_VAL(pf_rl);
829 
830 	if (inc_val > QM_PF_RL_MAX_INC_VAL) {
831 		DP_NOTICE(p_hwfn, "Invalid PF rate limit configuration\n");
832 		return -1;
833 	}
834 
835 	STORE_RT_REG(p_hwfn,
836 		     QM_REG_RLPFCRD_RT_OFFSET + pf_id,
837 		     (u32)QM_RL_CRD_REG_SIGN_BIT);
838 	STORE_RT_REG(p_hwfn,
839 		     QM_REG_RLPFUPPERBOUND_RT_OFFSET + pf_id,
840 		     QM_PF_RL_UPPER_BOUND | (u32)QM_RL_CRD_REG_SIGN_BIT);
841 	STORE_RT_REG(p_hwfn, QM_REG_RLPFINCVAL_RT_OFFSET + pf_id, inc_val);
842 
843 	return 0;
844 }
845 
846 /* Prepare VPORT WFQ runtime init values for the specified VPORTs.
847  * Return -1 on error.
848  */
849 static int qed_vp_wfq_rt_init(struct qed_hwfn *p_hwfn,
850 			      u16 num_vports,
851 			      struct init_qm_vport_params *vport_params)
852 {
853 	u16 vport_pq_id, wfq, i;
854 	u32 inc_val;
855 	u8 tc;
856 
857 	/* Go over all PF VPORTs */
858 	for (i = 0; i < num_vports; i++) {
859 		/* Each VPORT can have several VPORT PQ IDs for various TCs */
860 		for (tc = 0; tc < NUM_OF_TCS; tc++) {
861 			/* Check if VPORT/TC is valid */
862 			vport_pq_id = vport_params[i].first_tx_pq_id[tc];
863 			if (vport_pq_id == QM_INVALID_PQ_ID)
864 				continue;
865 
866 			/* Find WFQ weight (per VPORT or per VPORT+TC) */
867 			wfq = vport_params[i].wfq;
868 			wfq = wfq ? wfq : vport_params[i].tc_wfq[tc];
869 			inc_val = QM_VP_WFQ_INC_VAL(wfq);
870 			if (inc_val > QM_VP_WFQ_MAX_INC_VAL) {
871 				DP_NOTICE(p_hwfn,
872 					  "Invalid VPORT WFQ weight configuration\n");
873 				return -1;
874 			}
875 
876 			/* Config registers */
877 			STORE_RT_REG(p_hwfn, QM_REG_WFQVPCRD_RT_OFFSET +
878 				     vport_pq_id,
879 				     (u32)QM_WFQ_CRD_REG_SIGN_BIT);
880 			STORE_RT_REG(p_hwfn, QM_REG_WFQVPUPPERBOUND_RT_OFFSET +
881 				     vport_pq_id,
882 				     inc_val | QM_WFQ_CRD_REG_SIGN_BIT);
883 			STORE_RT_REG(p_hwfn, QM_REG_WFQVPWEIGHT_RT_OFFSET +
884 				     vport_pq_id, inc_val);
885 		}
886 	}
887 
888 	return 0;
889 }
890 
891 static bool qed_poll_on_qm_cmd_ready(struct qed_hwfn *p_hwfn,
892 				     struct qed_ptt *p_ptt)
893 {
894 	u32 reg_val, i;
895 
896 	for (i = 0, reg_val = 0; i < QM_STOP_CMD_MAX_POLL_COUNT && !reg_val;
897 	     i++) {
898 		udelay(QM_STOP_CMD_POLL_PERIOD_US);
899 		reg_val = qed_rd(p_hwfn, p_ptt, QM_REG_SDMCMDREADY);
900 	}
901 
902 	/* Check if timeout while waiting for SDM command ready */
903 	if (i == QM_STOP_CMD_MAX_POLL_COUNT) {
904 		DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
905 			   "Timeout when waiting for QM SDM command ready signal\n");
906 		return false;
907 	}
908 
909 	return true;
910 }
911 
912 static bool qed_send_qm_cmd(struct qed_hwfn *p_hwfn,
913 			    struct qed_ptt *p_ptt,
914 			    u32 cmd_addr, u32 cmd_data_lsb, u32 cmd_data_msb)
915 {
916 	if (!qed_poll_on_qm_cmd_ready(p_hwfn, p_ptt))
917 		return false;
918 
919 	qed_wr(p_hwfn, p_ptt, QM_REG_SDMCMDADDR, cmd_addr);
920 	qed_wr(p_hwfn, p_ptt, QM_REG_SDMCMDDATALSB, cmd_data_lsb);
921 	qed_wr(p_hwfn, p_ptt, QM_REG_SDMCMDDATAMSB, cmd_data_msb);
922 	qed_wr(p_hwfn, p_ptt, QM_REG_SDMCMDGO, 1);
923 	qed_wr(p_hwfn, p_ptt, QM_REG_SDMCMDGO, 0);
924 
925 	return qed_poll_on_qm_cmd_ready(p_hwfn, p_ptt);
926 }
927 
928 /******************** INTERFACE IMPLEMENTATION *********************/
929 
930 u32 qed_qm_pf_mem_size(u32 num_pf_cids,
931 		       u32 num_vf_cids,
932 		       u32 num_tids, u16 num_pf_pqs, u16 num_vf_pqs)
933 {
934 	return QM_PQ_MEM_4KB(num_pf_cids) * num_pf_pqs +
935 	       QM_PQ_MEM_4KB(num_vf_cids) * num_vf_pqs +
936 	       QM_PQ_MEM_4KB(num_pf_cids + num_tids) * QM_OTHER_PQS_PER_PF;
937 }
938 
939 int qed_qm_common_rt_init(struct qed_hwfn *p_hwfn,
940 			  struct qed_qm_common_rt_init_params *p_params)
941 {
942 	u32 mask = 0;
943 
944 	/* Init AFullOprtnstcCrdMask */
945 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_LINEVOQ,
946 		  QM_OPPOR_LINE_VOQ_DEF);
947 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_BYTEVOQ, QM_BYTE_CRD_EN);
948 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_PFWFQ,
949 		  p_params->pf_wfq_en ? 1 : 0);
950 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_VPWFQ,
951 		  p_params->vport_wfq_en ? 1 : 0);
952 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_PFRL,
953 		  p_params->pf_rl_en ? 1 : 0);
954 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_VPQCNRL,
955 		  p_params->global_rl_en ? 1 : 0);
956 	SET_FIELD(mask, QM_RF_OPPORTUNISTIC_MASK_FWPAUSE, QM_OPPOR_FW_STOP_DEF);
957 	SET_FIELD(mask,
958 		  QM_RF_OPPORTUNISTIC_MASK_QUEUEEMPTY, QM_OPPOR_PQ_EMPTY_DEF);
959 	STORE_RT_REG(p_hwfn, QM_REG_AFULLOPRTNSTCCRDMASK_RT_OFFSET, mask);
960 
961 	/* Enable/disable PF RL */
962 	qed_enable_pf_rl(p_hwfn, p_params->pf_rl_en);
963 
964 	/* Enable/disable PF WFQ */
965 	qed_enable_pf_wfq(p_hwfn, p_params->pf_wfq_en);
966 
967 	/* Enable/disable global RL */
968 	qed_enable_global_rl(p_hwfn, p_params->global_rl_en);
969 
970 	/* Enable/disable VPORT WFQ */
971 	qed_enable_vport_wfq(p_hwfn, p_params->vport_wfq_en);
972 
973 	/* Init PBF CMDQ line credit */
974 	qed_cmdq_lines_rt_init(p_hwfn,
975 			       p_params->max_ports_per_engine,
976 			       p_params->max_phys_tcs_per_port,
977 			       p_params->port_params);
978 
979 	/* Init BTB blocks in PBF */
980 	qed_btb_blocks_rt_init(p_hwfn,
981 			       p_params->max_ports_per_engine,
982 			       p_params->max_phys_tcs_per_port,
983 			       p_params->port_params);
984 
985 	qed_global_rl_rt_init(p_hwfn);
986 
987 	return 0;
988 }
989 
990 int qed_qm_pf_rt_init(struct qed_hwfn *p_hwfn,
991 		      struct qed_ptt *p_ptt,
992 		      struct qed_qm_pf_rt_init_params *p_params)
993 {
994 	struct init_qm_vport_params *vport_params = p_params->vport_params;
995 	u32 other_mem_size_4kb = QM_PQ_MEM_4KB(p_params->num_pf_cids +
996 					       p_params->num_tids) *
997 				 QM_OTHER_PQS_PER_PF;
998 	u16 i;
999 	u8 tc;
1000 
1001 	/* Clear first Tx PQ ID array for each VPORT */
1002 	for (i = 0; i < p_params->num_vports; i++)
1003 		for (tc = 0; tc < NUM_OF_TCS; tc++)
1004 			vport_params[i].first_tx_pq_id[tc] = QM_INVALID_PQ_ID;
1005 
1006 	/* Map Other PQs (if any) */
1007 	qed_other_pq_map_rt_init(p_hwfn,
1008 				 p_params->pf_id,
1009 				 p_params->is_pf_loading, p_params->num_pf_cids,
1010 				 p_params->num_tids, 0);
1011 
1012 	/* Map Tx PQs */
1013 	if (qed_tx_pq_map_rt_init(p_hwfn, p_ptt, p_params, other_mem_size_4kb))
1014 		return -1;
1015 
1016 	/* Init PF WFQ */
1017 	if (p_params->pf_wfq)
1018 		if (qed_pf_wfq_rt_init(p_hwfn, p_params))
1019 			return -1;
1020 
1021 	/* Init PF RL */
1022 	if (qed_pf_rl_rt_init(p_hwfn, p_params->pf_id, p_params->pf_rl))
1023 		return -1;
1024 
1025 	/* Init VPORT WFQ */
1026 	if (qed_vp_wfq_rt_init(p_hwfn, p_params->num_vports, vport_params))
1027 		return -1;
1028 
1029 	/* Set VPORT RL */
1030 	if (qed_vport_rl_rt_init(p_hwfn, p_params->start_rl,
1031 				 p_params->num_rls, p_params->link_speed,
1032 				 p_params->rl_params))
1033 		return -1;
1034 
1035 	return 0;
1036 }
1037 
1038 int qed_init_pf_wfq(struct qed_hwfn *p_hwfn,
1039 		    struct qed_ptt *p_ptt, u8 pf_id, u16 pf_wfq)
1040 {
1041 	u32 inc_val = QM_PF_WFQ_INC_VAL(pf_wfq);
1042 
1043 	if (!inc_val || inc_val > QM_PF_WFQ_MAX_INC_VAL) {
1044 		DP_NOTICE(p_hwfn, "Invalid PF WFQ weight configuration\n");
1045 		return -1;
1046 	}
1047 
1048 	qed_wr(p_hwfn, p_ptt, QM_REG_WFQPFWEIGHT + pf_id * 4, inc_val);
1049 
1050 	return 0;
1051 }
1052 
1053 int qed_init_pf_rl(struct qed_hwfn *p_hwfn,
1054 		   struct qed_ptt *p_ptt, u8 pf_id, u32 pf_rl)
1055 {
1056 	u32 inc_val = QM_RL_INC_VAL(pf_rl);
1057 
1058 	if (inc_val > QM_PF_RL_MAX_INC_VAL) {
1059 		DP_NOTICE(p_hwfn, "Invalid PF rate limit configuration\n");
1060 		return -1;
1061 	}
1062 
1063 	qed_wr(p_hwfn,
1064 	       p_ptt, QM_REG_RLPFCRD + pf_id * 4, (u32)QM_RL_CRD_REG_SIGN_BIT);
1065 	qed_wr(p_hwfn, p_ptt, QM_REG_RLPFINCVAL + pf_id * 4, inc_val);
1066 
1067 	return 0;
1068 }
1069 
1070 int qed_init_vport_wfq(struct qed_hwfn *p_hwfn,
1071 		       struct qed_ptt *p_ptt,
1072 		       u16 first_tx_pq_id[NUM_OF_TCS], u16 wfq)
1073 {
1074 	int result = 0;
1075 	u16 vport_pq_id;
1076 	u8 tc;
1077 
1078 	for (tc = 0; tc < NUM_OF_TCS && !result; tc++) {
1079 		vport_pq_id = first_tx_pq_id[tc];
1080 		if (vport_pq_id != QM_INVALID_PQ_ID)
1081 			result = qed_init_vport_tc_wfq(p_hwfn, p_ptt,
1082 						       vport_pq_id, wfq);
1083 	}
1084 
1085 	return result;
1086 }
1087 
1088 int qed_init_vport_tc_wfq(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
1089 			  u16 first_tx_pq_id, u16 wfq)
1090 {
1091 	u32 inc_val;
1092 
1093 	if (first_tx_pq_id == QM_INVALID_PQ_ID)
1094 		return -1;
1095 
1096 	inc_val = QM_VP_WFQ_INC_VAL(wfq);
1097 	if (!inc_val || inc_val > QM_VP_WFQ_MAX_INC_VAL) {
1098 		DP_NOTICE(p_hwfn, "Invalid VPORT WFQ configuration.\n");
1099 		return -1;
1100 	}
1101 
1102 	qed_wr(p_hwfn, p_ptt, QM_REG_WFQVPCRD + first_tx_pq_id * 4,
1103 	       (u32)QM_WFQ_CRD_REG_SIGN_BIT);
1104 	qed_wr(p_hwfn, p_ptt, QM_REG_WFQVPUPPERBOUND + first_tx_pq_id * 4,
1105 	       inc_val | QM_WFQ_CRD_REG_SIGN_BIT);
1106 	qed_wr(p_hwfn, p_ptt, QM_REG_WFQVPWEIGHT + first_tx_pq_id * 4,
1107 	       inc_val);
1108 
1109 	return 0;
1110 }
1111 
1112 int qed_init_global_rl(struct qed_hwfn *p_hwfn,
1113 		       struct qed_ptt *p_ptt, u16 rl_id, u32 rate_limit,
1114 		       enum init_qm_rl_type vport_rl_type)
1115 {
1116 	u32 inc_val, upper_bound;
1117 
1118 	upper_bound =
1119 	    (vport_rl_type ==
1120 	     QM_RL_TYPE_QCN) ? QM_GLOBAL_RL_UPPER_BOUND(QM_MAX_LINK_SPEED) :
1121 	    QM_INITIAL_VOQ_BYTE_CRD;
1122 	inc_val = QM_RL_INC_VAL(rate_limit);
1123 	if (inc_val > upper_bound) {
1124 		DP_NOTICE(p_hwfn, "Invalid VPORT rate limit configuration.\n");
1125 		return -1;
1126 	}
1127 
1128 	qed_wr(p_hwfn, p_ptt,
1129 	       QM_REG_RLGLBLCRD + rl_id * 4, (u32)QM_RL_CRD_REG_SIGN_BIT);
1130 	qed_wr(p_hwfn,
1131 	       p_ptt,
1132 	       QM_REG_RLGLBLUPPERBOUND + rl_id * 4,
1133 	       upper_bound | (u32)QM_RL_CRD_REG_SIGN_BIT);
1134 	qed_wr(p_hwfn, p_ptt, QM_REG_RLGLBLINCVAL + rl_id * 4, inc_val);
1135 
1136 	return 0;
1137 }
1138 
1139 bool qed_send_qm_stop_cmd(struct qed_hwfn *p_hwfn,
1140 			  struct qed_ptt *p_ptt,
1141 			  bool is_release_cmd,
1142 			  bool is_tx_pq, u16 start_pq, u16 num_pqs)
1143 {
1144 	u32 cmd_arr[QM_CMD_STRUCT_SIZE(QM_STOP_CMD)] = { 0 };
1145 	u32 pq_mask = 0, last_pq, pq_id;
1146 
1147 	last_pq = start_pq + num_pqs - 1;
1148 
1149 	/* Set command's PQ type */
1150 	QM_CMD_SET_FIELD(cmd_arr, QM_STOP_CMD, PQ_TYPE, is_tx_pq ? 0 : 1);
1151 
1152 	/* Go over requested PQs */
1153 	for (pq_id = start_pq; pq_id <= last_pq; pq_id++) {
1154 		/* Set PQ bit in mask (stop command only) */
1155 		if (!is_release_cmd)
1156 			pq_mask |= BIT((pq_id % QM_STOP_PQ_MASK_WIDTH));
1157 
1158 		/* If last PQ or end of PQ mask, write command */
1159 		if ((pq_id == last_pq) ||
1160 		    (pq_id % QM_STOP_PQ_MASK_WIDTH ==
1161 		     (QM_STOP_PQ_MASK_WIDTH - 1))) {
1162 			QM_CMD_SET_FIELD(cmd_arr,
1163 					 QM_STOP_CMD, PAUSE_MASK, pq_mask);
1164 			QM_CMD_SET_FIELD(cmd_arr,
1165 					 QM_STOP_CMD,
1166 					 GROUP_ID,
1167 					 pq_id / QM_STOP_PQ_MASK_WIDTH);
1168 			if (!qed_send_qm_cmd(p_hwfn, p_ptt, QM_STOP_CMD_ADDR,
1169 					     cmd_arr[0], cmd_arr[1]))
1170 				return false;
1171 			pq_mask = 0;
1172 		}
1173 	}
1174 
1175 	return true;
1176 }
1177 
1178 #define SET_TUNNEL_TYPE_ENABLE_BIT(var, offset, enable) \
1179 	do { \
1180 		typeof(var) *__p_var = &(var); \
1181 		typeof(offset) __offset = offset; \
1182 		*__p_var = (*__p_var & ~BIT(__offset)) | \
1183 			   ((enable) ? BIT(__offset) : 0); \
1184 	} while (0)
1185 
1186 #define PRS_ETH_TUNN_OUTPUT_FORMAT     0xF4DAB910
1187 #define PRS_ETH_OUTPUT_FORMAT          0xFFFF4910
1188 
1189 #define ARR_REG_WR(dev, ptt, addr, arr,	arr_size) \
1190 	do { \
1191 		u32 i; \
1192 		\
1193 		for (i = 0; i < (arr_size); i++) \
1194 			qed_wr(dev, ptt, \
1195 			       ((addr) + (4 * i)), \
1196 			       ((u32 *)&(arr))[i]); \
1197 	} while (0)
1198 
1199 /**
1200  * qed_dmae_to_grc() - Internal function for writing from host to
1201  * wide-bus registers (split registers are not supported yet).
1202  *
1203  * @p_hwfn: HW device data.
1204  * @p_ptt: PTT window used for writing the registers.
1205  * @p_data: Pointer to source data.
1206  * @addr: Destination register address.
1207  * @len_in_dwords: Data length in dwords (u32).
1208  *
1209  * Return: Length of the written data in dwords (u32) or -1 on invalid
1210  *         input.
1211  */
1212 static int qed_dmae_to_grc(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
1213 			   __le32 *p_data, u32 addr, u32 len_in_dwords)
1214 {
1215 	struct qed_dmae_params params = { 0 };
1216 	u32 *data_cpu;
1217 	int rc;
1218 
1219 	if (!p_data)
1220 		return -1;
1221 
1222 	/* Set DMAE params */
1223 	SET_FIELD(params.flags, QED_DMAE_PARAMS_COMPLETION_DST, 1);
1224 
1225 	/* Execute DMAE command */
1226 	rc = qed_dmae_host2grc(p_hwfn, p_ptt,
1227 			       (u64)(uintptr_t)(p_data),
1228 			       addr, len_in_dwords, &params);
1229 
1230 	/* If not read using DMAE, read using GRC */
1231 	if (rc) {
1232 		DP_VERBOSE(p_hwfn,
1233 			   QED_MSG_DEBUG,
1234 			   "Failed writing to chip using DMAE, using GRC instead\n");
1235 
1236 		/* Swap to CPU byteorder and write to registers using GRC */
1237 		data_cpu = (__force u32 *)p_data;
1238 		le32_to_cpu_array(data_cpu, len_in_dwords);
1239 
1240 		ARR_REG_WR(p_hwfn, p_ptt, addr, data_cpu, len_in_dwords);
1241 		cpu_to_le32_array(data_cpu, len_in_dwords);
1242 	}
1243 
1244 	return len_in_dwords;
1245 }
1246 
1247 void qed_set_vxlan_dest_port(struct qed_hwfn *p_hwfn,
1248 			     struct qed_ptt *p_ptt, u16 dest_port)
1249 {
1250 	/* Update PRS register */
1251 	qed_wr(p_hwfn, p_ptt, PRS_REG_VXLAN_PORT, dest_port);
1252 
1253 	/* Update NIG register */
1254 	qed_wr(p_hwfn, p_ptt, NIG_REG_VXLAN_CTRL, dest_port);
1255 
1256 	/* Update PBF register */
1257 	qed_wr(p_hwfn, p_ptt, PBF_REG_VXLAN_PORT, dest_port);
1258 }
1259 
1260 void qed_set_vxlan_enable(struct qed_hwfn *p_hwfn,
1261 			  struct qed_ptt *p_ptt, bool vxlan_enable)
1262 {
1263 	u32 reg_val;
1264 	u8 shift;
1265 
1266 	/* Update PRS register */
1267 	reg_val = qed_rd(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN);
1268 	SET_FIELD(reg_val,
1269 		  PRS_REG_ENCAPSULATION_TYPE_EN_VXLAN_ENABLE, vxlan_enable);
1270 	qed_wr(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN, reg_val);
1271 	if (reg_val) {
1272 		reg_val =
1273 		    qed_rd(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0);
1274 
1275 		/* Update output  only if tunnel blocks not included. */
1276 		if (reg_val == (u32)PRS_ETH_OUTPUT_FORMAT)
1277 			qed_wr(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0,
1278 			       (u32)PRS_ETH_TUNN_OUTPUT_FORMAT);
1279 	}
1280 
1281 	/* Update NIG register */
1282 	reg_val = qed_rd(p_hwfn, p_ptt, NIG_REG_ENC_TYPE_ENABLE);
1283 	shift = NIG_REG_ENC_TYPE_ENABLE_VXLAN_ENABLE_SHIFT;
1284 	SET_TUNNEL_TYPE_ENABLE_BIT(reg_val, shift, vxlan_enable);
1285 	qed_wr(p_hwfn, p_ptt, NIG_REG_ENC_TYPE_ENABLE, reg_val);
1286 
1287 	/* Update DORQ register */
1288 	qed_wr(p_hwfn,
1289 	       p_ptt, DORQ_REG_L2_EDPM_TUNNEL_VXLAN_EN, vxlan_enable ? 1 : 0);
1290 }
1291 
1292 void qed_set_gre_enable(struct qed_hwfn *p_hwfn,
1293 			struct qed_ptt *p_ptt,
1294 			bool eth_gre_enable, bool ip_gre_enable)
1295 {
1296 	u32 reg_val;
1297 	u8 shift;
1298 
1299 	/* Update PRS register */
1300 	reg_val = qed_rd(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN);
1301 	SET_FIELD(reg_val,
1302 		  PRS_REG_ENCAPSULATION_TYPE_EN_ETH_OVER_GRE_ENABLE,
1303 		  eth_gre_enable);
1304 	SET_FIELD(reg_val,
1305 		  PRS_REG_ENCAPSULATION_TYPE_EN_IP_OVER_GRE_ENABLE,
1306 		  ip_gre_enable);
1307 	qed_wr(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN, reg_val);
1308 	if (reg_val) {
1309 		reg_val =
1310 		    qed_rd(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0);
1311 
1312 		/* Update output  only if tunnel blocks not included. */
1313 		if (reg_val == (u32)PRS_ETH_OUTPUT_FORMAT)
1314 			qed_wr(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0,
1315 			       (u32)PRS_ETH_TUNN_OUTPUT_FORMAT);
1316 	}
1317 
1318 	/* Update NIG register */
1319 	reg_val = qed_rd(p_hwfn, p_ptt, NIG_REG_ENC_TYPE_ENABLE);
1320 	shift = NIG_REG_ENC_TYPE_ENABLE_ETH_OVER_GRE_ENABLE_SHIFT;
1321 	SET_TUNNEL_TYPE_ENABLE_BIT(reg_val, shift, eth_gre_enable);
1322 	shift = NIG_REG_ENC_TYPE_ENABLE_IP_OVER_GRE_ENABLE_SHIFT;
1323 	SET_TUNNEL_TYPE_ENABLE_BIT(reg_val, shift, ip_gre_enable);
1324 	qed_wr(p_hwfn, p_ptt, NIG_REG_ENC_TYPE_ENABLE, reg_val);
1325 
1326 	/* Update DORQ registers */
1327 	qed_wr(p_hwfn,
1328 	       p_ptt,
1329 	       DORQ_REG_L2_EDPM_TUNNEL_GRE_ETH_EN, eth_gre_enable ? 1 : 0);
1330 	qed_wr(p_hwfn,
1331 	       p_ptt, DORQ_REG_L2_EDPM_TUNNEL_GRE_IP_EN, ip_gre_enable ? 1 : 0);
1332 }
1333 
1334 void qed_set_geneve_dest_port(struct qed_hwfn *p_hwfn,
1335 			      struct qed_ptt *p_ptt, u16 dest_port)
1336 {
1337 	/* Update PRS register */
1338 	qed_wr(p_hwfn, p_ptt, PRS_REG_NGE_PORT, dest_port);
1339 
1340 	/* Update NIG register */
1341 	qed_wr(p_hwfn, p_ptt, NIG_REG_NGE_PORT, dest_port);
1342 
1343 	/* Update PBF register */
1344 	qed_wr(p_hwfn, p_ptt, PBF_REG_NGE_PORT, dest_port);
1345 }
1346 
1347 void qed_set_geneve_enable(struct qed_hwfn *p_hwfn,
1348 			   struct qed_ptt *p_ptt,
1349 			   bool eth_geneve_enable, bool ip_geneve_enable)
1350 {
1351 	u32 reg_val;
1352 
1353 	/* Update PRS register */
1354 	reg_val = qed_rd(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN);
1355 	SET_FIELD(reg_val,
1356 		  PRS_REG_ENCAPSULATION_TYPE_EN_ETH_OVER_GENEVE_ENABLE,
1357 		  eth_geneve_enable);
1358 	SET_FIELD(reg_val,
1359 		  PRS_REG_ENCAPSULATION_TYPE_EN_IP_OVER_GENEVE_ENABLE,
1360 		  ip_geneve_enable);
1361 	qed_wr(p_hwfn, p_ptt, PRS_REG_ENCAPSULATION_TYPE_EN, reg_val);
1362 	if (reg_val) {
1363 		reg_val =
1364 		    qed_rd(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0);
1365 
1366 		/* Update output  only if tunnel blocks not included. */
1367 		if (reg_val == (u32)PRS_ETH_OUTPUT_FORMAT)
1368 			qed_wr(p_hwfn, p_ptt, PRS_REG_OUTPUT_FORMAT_4_0,
1369 			       (u32)PRS_ETH_TUNN_OUTPUT_FORMAT);
1370 	}
1371 
1372 	/* Update NIG register */
1373 	qed_wr(p_hwfn, p_ptt, NIG_REG_NGE_ETH_ENABLE,
1374 	       eth_geneve_enable ? 1 : 0);
1375 	qed_wr(p_hwfn, p_ptt, NIG_REG_NGE_IP_ENABLE, ip_geneve_enable ? 1 : 0);
1376 
1377 	/* EDPM with geneve tunnel not supported in BB */
1378 	if (QED_IS_BB_B0(p_hwfn->cdev))
1379 		return;
1380 
1381 	/* Update DORQ registers */
1382 	qed_wr(p_hwfn,
1383 	       p_ptt,
1384 	       DORQ_REG_L2_EDPM_TUNNEL_NGE_ETH_EN_K2,
1385 	       eth_geneve_enable ? 1 : 0);
1386 	qed_wr(p_hwfn,
1387 	       p_ptt,
1388 	       DORQ_REG_L2_EDPM_TUNNEL_NGE_IP_EN_K2,
1389 	       ip_geneve_enable ? 1 : 0);
1390 }
1391 
1392 #define PRS_ETH_VXLAN_NO_L2_ENABLE_OFFSET      3
1393 #define PRS_ETH_VXLAN_NO_L2_OUTPUT_FORMAT   0xC8DAB910
1394 
1395 void qed_set_vxlan_no_l2_enable(struct qed_hwfn *p_hwfn,
1396 				struct qed_ptt *p_ptt, bool enable)
1397 {
1398 	u32 reg_val, cfg_mask;
1399 
1400 	/* read PRS config register */
1401 	reg_val = qed_rd(p_hwfn, p_ptt, PRS_REG_MSG_INFO);
1402 
1403 	/* set VXLAN_NO_L2_ENABLE mask */
1404 	cfg_mask = BIT(PRS_ETH_VXLAN_NO_L2_ENABLE_OFFSET);
1405 
1406 	if (enable) {
1407 		/* set VXLAN_NO_L2_ENABLE flag */
1408 		reg_val |= cfg_mask;
1409 
1410 		/* update PRS FIC  register */
1411 		qed_wr(p_hwfn,
1412 		       p_ptt,
1413 		       PRS_REG_OUTPUT_FORMAT_4_0,
1414 		       (u32)PRS_ETH_VXLAN_NO_L2_OUTPUT_FORMAT);
1415 	} else {
1416 		/* clear VXLAN_NO_L2_ENABLE flag */
1417 		reg_val &= ~cfg_mask;
1418 	}
1419 
1420 	/* write PRS config register */
1421 	qed_wr(p_hwfn, p_ptt, PRS_REG_MSG_INFO, reg_val);
1422 }
1423 
1424 #define T_ETH_PACKET_ACTION_GFT_EVENTID  23
1425 #define PARSER_ETH_CONN_GFT_ACTION_CM_HDR  272
1426 #define T_ETH_PACKET_MATCH_RFS_EVENTID 25
1427 #define PARSER_ETH_CONN_CM_HDR 0
1428 #define CAM_LINE_SIZE sizeof(u32)
1429 #define RAM_LINE_SIZE sizeof(u64)
1430 #define REG_SIZE sizeof(u32)
1431 
1432 void qed_gft_disable(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt, u16 pf_id)
1433 {
1434 	struct regpair ram_line = { 0 };
1435 
1436 	/* Disable gft search for PF */
1437 	qed_wr(p_hwfn, p_ptt, PRS_REG_SEARCH_GFT, 0);
1438 
1439 	/* Clean ram & cam for next gft session */
1440 
1441 	/* Zero camline */
1442 	qed_wr(p_hwfn, p_ptt, PRS_REG_GFT_CAM + CAM_LINE_SIZE * pf_id, 0);
1443 
1444 	/* Zero ramline */
1445 	qed_dmae_to_grc(p_hwfn, p_ptt, &ram_line.lo,
1446 			PRS_REG_GFT_PROFILE_MASK_RAM + RAM_LINE_SIZE * pf_id,
1447 			sizeof(ram_line) / REG_SIZE);
1448 }
1449 
1450 void qed_gft_config(struct qed_hwfn *p_hwfn,
1451 		    struct qed_ptt *p_ptt,
1452 		    u16 pf_id,
1453 		    bool tcp,
1454 		    bool udp,
1455 		    bool ipv4, bool ipv6, enum gft_profile_type profile_type)
1456 {
1457 	struct regpair ram_line;
1458 	u32 search_non_ip_as_gft;
1459 	u32 reg_val, cam_line;
1460 	u32 lo = 0, hi = 0;
1461 
1462 	if (!ipv6 && !ipv4)
1463 		DP_NOTICE(p_hwfn,
1464 			  "gft_config: must accept at least on of - ipv4 or ipv6'\n");
1465 	if (!tcp && !udp)
1466 		DP_NOTICE(p_hwfn,
1467 			  "gft_config: must accept at least on of - udp or tcp\n");
1468 	if (profile_type >= MAX_GFT_PROFILE_TYPE)
1469 		DP_NOTICE(p_hwfn, "gft_config: unsupported gft_profile_type\n");
1470 
1471 	/* Set RFS event ID to be awakened i Tstorm By Prs */
1472 	reg_val = T_ETH_PACKET_MATCH_RFS_EVENTID <<
1473 		  PRS_REG_CM_HDR_GFT_EVENT_ID_SHIFT;
1474 	reg_val |= PARSER_ETH_CONN_CM_HDR << PRS_REG_CM_HDR_GFT_CM_HDR_SHIFT;
1475 	qed_wr(p_hwfn, p_ptt, PRS_REG_CM_HDR_GFT, reg_val);
1476 
1477 	/* Do not load context only cid in PRS on match. */
1478 	qed_wr(p_hwfn, p_ptt, PRS_REG_LOAD_L2_FILTER, 0);
1479 
1480 	/* Do not use tenant ID exist bit for gft search */
1481 	qed_wr(p_hwfn, p_ptt, PRS_REG_SEARCH_TENANT_ID, 0);
1482 
1483 	/* Set Cam */
1484 	cam_line = 0;
1485 	SET_FIELD(cam_line, GFT_CAM_LINE_MAPPED_VALID, 1);
1486 
1487 	/* Filters are per PF!! */
1488 	SET_FIELD(cam_line,
1489 		  GFT_CAM_LINE_MAPPED_PF_ID_MASK,
1490 		  GFT_CAM_LINE_MAPPED_PF_ID_MASK_MASK);
1491 	SET_FIELD(cam_line, GFT_CAM_LINE_MAPPED_PF_ID, pf_id);
1492 
1493 	if (!(tcp && udp)) {
1494 		SET_FIELD(cam_line,
1495 			  GFT_CAM_LINE_MAPPED_UPPER_PROTOCOL_TYPE_MASK,
1496 			  GFT_CAM_LINE_MAPPED_UPPER_PROTOCOL_TYPE_MASK_MASK);
1497 		if (tcp)
1498 			SET_FIELD(cam_line,
1499 				  GFT_CAM_LINE_MAPPED_UPPER_PROTOCOL_TYPE,
1500 				  GFT_PROFILE_TCP_PROTOCOL);
1501 		else
1502 			SET_FIELD(cam_line,
1503 				  GFT_CAM_LINE_MAPPED_UPPER_PROTOCOL_TYPE,
1504 				  GFT_PROFILE_UDP_PROTOCOL);
1505 	}
1506 
1507 	if (!(ipv4 && ipv6)) {
1508 		SET_FIELD(cam_line, GFT_CAM_LINE_MAPPED_IP_VERSION_MASK, 1);
1509 		if (ipv4)
1510 			SET_FIELD(cam_line,
1511 				  GFT_CAM_LINE_MAPPED_IP_VERSION,
1512 				  GFT_PROFILE_IPV4);
1513 		else
1514 			SET_FIELD(cam_line,
1515 				  GFT_CAM_LINE_MAPPED_IP_VERSION,
1516 				  GFT_PROFILE_IPV6);
1517 	}
1518 
1519 	/* Write characteristics to cam */
1520 	qed_wr(p_hwfn, p_ptt, PRS_REG_GFT_CAM + CAM_LINE_SIZE * pf_id,
1521 	       cam_line);
1522 	cam_line =
1523 	    qed_rd(p_hwfn, p_ptt, PRS_REG_GFT_CAM + CAM_LINE_SIZE * pf_id);
1524 
1525 	/* Write line to RAM - compare to filter 4 tuple */
1526 
1527 	/* Search no IP as GFT */
1528 	search_non_ip_as_gft = 0;
1529 
1530 	/* Tunnel type */
1531 	SET_FIELD(lo, GFT_RAM_LINE_TUNNEL_DST_PORT, 1);
1532 	SET_FIELD(lo, GFT_RAM_LINE_TUNNEL_OVER_IP_PROTOCOL, 1);
1533 
1534 	if (profile_type == GFT_PROFILE_TYPE_4_TUPLE) {
1535 		SET_FIELD(hi, GFT_RAM_LINE_DST_IP, 1);
1536 		SET_FIELD(hi, GFT_RAM_LINE_SRC_IP, 1);
1537 		SET_FIELD(hi, GFT_RAM_LINE_OVER_IP_PROTOCOL, 1);
1538 		SET_FIELD(lo, GFT_RAM_LINE_ETHERTYPE, 1);
1539 		SET_FIELD(lo, GFT_RAM_LINE_SRC_PORT, 1);
1540 		SET_FIELD(lo, GFT_RAM_LINE_DST_PORT, 1);
1541 	} else if (profile_type == GFT_PROFILE_TYPE_L4_DST_PORT) {
1542 		SET_FIELD(hi, GFT_RAM_LINE_OVER_IP_PROTOCOL, 1);
1543 		SET_FIELD(lo, GFT_RAM_LINE_ETHERTYPE, 1);
1544 		SET_FIELD(lo, GFT_RAM_LINE_DST_PORT, 1);
1545 	} else if (profile_type == GFT_PROFILE_TYPE_IP_DST_ADDR) {
1546 		SET_FIELD(hi, GFT_RAM_LINE_DST_IP, 1);
1547 		SET_FIELD(lo, GFT_RAM_LINE_ETHERTYPE, 1);
1548 	} else if (profile_type == GFT_PROFILE_TYPE_IP_SRC_ADDR) {
1549 		SET_FIELD(hi, GFT_RAM_LINE_SRC_IP, 1);
1550 		SET_FIELD(lo, GFT_RAM_LINE_ETHERTYPE, 1);
1551 	} else if (profile_type == GFT_PROFILE_TYPE_TUNNEL_TYPE) {
1552 		SET_FIELD(lo, GFT_RAM_LINE_TUNNEL_ETHERTYPE, 1);
1553 
1554 		/* Allow tunneled traffic without inner IP */
1555 		search_non_ip_as_gft = 1;
1556 	}
1557 
1558 	ram_line.lo = cpu_to_le32(lo);
1559 	ram_line.hi = cpu_to_le32(hi);
1560 
1561 	qed_wr(p_hwfn,
1562 	       p_ptt, PRS_REG_SEARCH_NON_IP_AS_GFT, search_non_ip_as_gft);
1563 	qed_dmae_to_grc(p_hwfn, p_ptt, &ram_line.lo,
1564 			PRS_REG_GFT_PROFILE_MASK_RAM + RAM_LINE_SIZE * pf_id,
1565 			sizeof(ram_line) / REG_SIZE);
1566 
1567 	/* Set default profile so that no filter match will happen */
1568 	ram_line.lo = cpu_to_le32(0xffffffff);
1569 	ram_line.hi = cpu_to_le32(0x3ff);
1570 	qed_dmae_to_grc(p_hwfn, p_ptt, &ram_line.lo,
1571 			PRS_REG_GFT_PROFILE_MASK_RAM + RAM_LINE_SIZE *
1572 			PRS_GFT_CAM_LINES_NO_MATCH,
1573 			sizeof(ram_line) / REG_SIZE);
1574 
1575 	/* Enable gft search */
1576 	qed_wr(p_hwfn, p_ptt, PRS_REG_SEARCH_GFT, 1);
1577 }
1578 
1579 DECLARE_CRC8_TABLE(cdu_crc8_table);
1580 
1581 /* Calculate and return CDU validation byte per connection type/region/cid */
1582 static u8 qed_calc_cdu_validation_byte(u8 conn_type, u8 region, u32 cid)
1583 {
1584 	const u8 validation_cfg = CDU_VALIDATION_DEFAULT_CFG;
1585 	u8 crc, validation_byte = 0;
1586 	static u8 crc8_table_valid; /* automatically initialized to 0 */
1587 	u32 validation_string = 0;
1588 	__be32 data_to_crc;
1589 
1590 	if (!crc8_table_valid) {
1591 		crc8_populate_msb(cdu_crc8_table, 0x07);
1592 		crc8_table_valid = 1;
1593 	}
1594 
1595 	/* The CRC is calculated on the String-to-compress:
1596 	 * [31:8]  = {CID[31:20],CID[11:0]}
1597 	 * [7:4]   = Region
1598 	 * [3:0]   = Type
1599 	 */
1600 	if ((validation_cfg >> CDU_CONTEXT_VALIDATION_CFG_USE_CID) & 1)
1601 		validation_string |= (cid & 0xFFF00000) | ((cid & 0xFFF) << 8);
1602 
1603 	if ((validation_cfg >> CDU_CONTEXT_VALIDATION_CFG_USE_REGION) & 1)
1604 		validation_string |= ((region & 0xF) << 4);
1605 
1606 	if ((validation_cfg >> CDU_CONTEXT_VALIDATION_CFG_USE_TYPE) & 1)
1607 		validation_string |= (conn_type & 0xF);
1608 
1609 	/* Convert to big-endian and calculate CRC8 */
1610 	data_to_crc = cpu_to_be32(validation_string);
1611 	crc = crc8(cdu_crc8_table, (u8 *)&data_to_crc, sizeof(data_to_crc),
1612 		   CRC8_INIT_VALUE);
1613 
1614 	/* The validation byte [7:0] is composed:
1615 	 * for type A validation
1616 	 * [7]          = active configuration bit
1617 	 * [6:0]        = crc[6:0]
1618 	 *
1619 	 * for type B validation
1620 	 * [7]          = active configuration bit
1621 	 * [6:3]        = connection_type[3:0]
1622 	 * [2:0]        = crc[2:0]
1623 	 */
1624 	validation_byte |=
1625 	    ((validation_cfg >>
1626 	      CDU_CONTEXT_VALIDATION_CFG_USE_ACTIVE) & 1) << 7;
1627 
1628 	if ((validation_cfg >>
1629 	     CDU_CONTEXT_VALIDATION_CFG_VALIDATION_TYPE_SHIFT) & 1)
1630 		validation_byte |= ((conn_type & 0xF) << 3) | (crc & 0x7);
1631 	else
1632 		validation_byte |= crc & 0x7F;
1633 
1634 	return validation_byte;
1635 }
1636 
1637 /* Calcualte and set validation bytes for session context */
1638 void qed_calc_session_ctx_validation(void *p_ctx_mem,
1639 				     u16 ctx_size, u8 ctx_type, u32 cid)
1640 {
1641 	u8 *x_val_ptr, *t_val_ptr, *u_val_ptr, *p_ctx;
1642 
1643 	p_ctx = (u8 * const)p_ctx_mem;
1644 	x_val_ptr = &p_ctx[con_region_offsets[0][ctx_type]];
1645 	t_val_ptr = &p_ctx[con_region_offsets[1][ctx_type]];
1646 	u_val_ptr = &p_ctx[con_region_offsets[2][ctx_type]];
1647 
1648 	memset(p_ctx, 0, ctx_size);
1649 
1650 	*x_val_ptr = qed_calc_cdu_validation_byte(ctx_type, 3, cid);
1651 	*t_val_ptr = qed_calc_cdu_validation_byte(ctx_type, 4, cid);
1652 	*u_val_ptr = qed_calc_cdu_validation_byte(ctx_type, 5, cid);
1653 }
1654 
1655 /* Calcualte and set validation bytes for task context */
1656 void qed_calc_task_ctx_validation(void *p_ctx_mem,
1657 				  u16 ctx_size, u8 ctx_type, u32 tid)
1658 {
1659 	u8 *p_ctx, *region1_val_ptr;
1660 
1661 	p_ctx = (u8 * const)p_ctx_mem;
1662 	region1_val_ptr = &p_ctx[task_region_offsets[0][ctx_type]];
1663 
1664 	memset(p_ctx, 0, ctx_size);
1665 
1666 	*region1_val_ptr = qed_calc_cdu_validation_byte(ctx_type, 1, tid);
1667 }
1668 
1669 /* Memset session context to 0 while preserving validation bytes */
1670 void qed_memset_session_ctx(void *p_ctx_mem, u32 ctx_size, u8 ctx_type)
1671 {
1672 	u8 *x_val_ptr, *t_val_ptr, *u_val_ptr, *p_ctx;
1673 	u8 x_val, t_val, u_val;
1674 
1675 	p_ctx = (u8 * const)p_ctx_mem;
1676 	x_val_ptr = &p_ctx[con_region_offsets[0][ctx_type]];
1677 	t_val_ptr = &p_ctx[con_region_offsets[1][ctx_type]];
1678 	u_val_ptr = &p_ctx[con_region_offsets[2][ctx_type]];
1679 
1680 	x_val = *x_val_ptr;
1681 	t_val = *t_val_ptr;
1682 	u_val = *u_val_ptr;
1683 
1684 	memset(p_ctx, 0, ctx_size);
1685 
1686 	*x_val_ptr = x_val;
1687 	*t_val_ptr = t_val;
1688 	*u_val_ptr = u_val;
1689 }
1690 
1691 /* Memset task context to 0 while preserving validation bytes */
1692 void qed_memset_task_ctx(void *p_ctx_mem, u32 ctx_size, u8 ctx_type)
1693 {
1694 	u8 *p_ctx, *region1_val_ptr;
1695 	u8 region1_val;
1696 
1697 	p_ctx = (u8 * const)p_ctx_mem;
1698 	region1_val_ptr = &p_ctx[task_region_offsets[0][ctx_type]];
1699 
1700 	region1_val = *region1_val_ptr;
1701 
1702 	memset(p_ctx, 0, ctx_size);
1703 
1704 	*region1_val_ptr = region1_val;
1705 }
1706 
1707 /* Enable and configure context validation */
1708 void qed_enable_context_validation(struct qed_hwfn *p_hwfn,
1709 				   struct qed_ptt *p_ptt)
1710 {
1711 	u32 ctx_validation;
1712 
1713 	/* Enable validation for connection region 3: CCFC_CTX_VALID0[31:24] */
1714 	ctx_validation = CDU_VALIDATION_DEFAULT_CFG << 24;
1715 	qed_wr(p_hwfn, p_ptt, CDU_REG_CCFC_CTX_VALID0, ctx_validation);
1716 
1717 	/* Enable validation for connection region 5: CCFC_CTX_VALID1[15:8] */
1718 	ctx_validation = CDU_VALIDATION_DEFAULT_CFG << 8;
1719 	qed_wr(p_hwfn, p_ptt, CDU_REG_CCFC_CTX_VALID1, ctx_validation);
1720 
1721 	/* Enable validation for connection region 1: TCFC_CTX_VALID0[15:8] */
1722 	ctx_validation = CDU_VALIDATION_DEFAULT_CFG << 8;
1723 	qed_wr(p_hwfn, p_ptt, CDU_REG_TCFC_CTX_VALID0, ctx_validation);
1724 }
1725 
1726 const char *qed_get_protocol_type_str(u32 protocol_type)
1727 {
1728 	if (protocol_type >= ARRAY_SIZE(s_protocol_types))
1729 		return "Invalid protocol type";
1730 
1731 	return s_protocol_types[protocol_type];
1732 }
1733 
1734 const char *qed_get_ramrod_cmd_id_str(u32 protocol_type, u32 ramrod_cmd_id)
1735 {
1736 	const char *ramrod_cmd_id_str;
1737 
1738 	if (protocol_type >= ARRAY_SIZE(s_ramrod_cmd_ids))
1739 		return "Invalid protocol type";
1740 
1741 	if (ramrod_cmd_id >= ARRAY_SIZE(s_ramrod_cmd_ids[0]))
1742 		return "Invalid Ramrod command ID";
1743 
1744 	ramrod_cmd_id_str = s_ramrod_cmd_ids[protocol_type][ramrod_cmd_id];
1745 
1746 	if (!ramrod_cmd_id_str)
1747 		return "Invalid Ramrod command ID";
1748 
1749 	return ramrod_cmd_id_str;
1750 }
1751 
1752 static u32 qed_get_rdma_assert_ram_addr(struct qed_hwfn *p_hwfn, u8 storm_id)
1753 {
1754 	switch (storm_id) {
1755 	case 0:
1756 		return TSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1757 		    TSTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1758 	case 1:
1759 		return MSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1760 		    MSTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1761 	case 2:
1762 		return USEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1763 		    USTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1764 	case 3:
1765 		return XSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1766 		    XSTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1767 	case 4:
1768 		return YSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1769 		    YSTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1770 	case 5:
1771 		return PSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1772 		    PSTORM_RDMA_ASSERT_LEVEL_OFFSET(p_hwfn->rel_pf_id);
1773 
1774 	default:
1775 		return 0;
1776 	}
1777 }
1778 
1779 void qed_set_rdma_error_level(struct qed_hwfn *p_hwfn,
1780 			      struct qed_ptt *p_ptt,
1781 			      u8 assert_level[NUM_STORMS])
1782 {
1783 	u8 storm_id;
1784 
1785 	for (storm_id = 0; storm_id < NUM_STORMS; storm_id++) {
1786 		u32 ram_addr = qed_get_rdma_assert_ram_addr(p_hwfn, storm_id);
1787 
1788 		qed_wr(p_hwfn, p_ptt, ram_addr, assert_level[storm_id]);
1789 	}
1790 }
1791 
1792 #define PHYS_ADDR_DWORDS        DIV_ROUND_UP(sizeof(dma_addr_t), 4)
1793 #define OVERLAY_HDR_SIZE_DWORDS (sizeof(struct fw_overlay_buf_hdr) / 4)
1794 
1795 static u32 qed_get_overlay_addr_ram_addr(struct qed_hwfn *p_hwfn, u8 storm_id)
1796 {
1797 	switch (storm_id) {
1798 	case 0:
1799 		return TSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1800 		    TSTORM_OVERLAY_BUF_ADDR_OFFSET;
1801 	case 1:
1802 		return MSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1803 		    MSTORM_OVERLAY_BUF_ADDR_OFFSET;
1804 	case 2:
1805 		return USEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1806 		    USTORM_OVERLAY_BUF_ADDR_OFFSET;
1807 	case 3:
1808 		return XSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1809 		    XSTORM_OVERLAY_BUF_ADDR_OFFSET;
1810 	case 4:
1811 		return YSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1812 		    YSTORM_OVERLAY_BUF_ADDR_OFFSET;
1813 	case 5:
1814 		return PSEM_REG_FAST_MEMORY + SEM_FAST_REG_INT_RAM +
1815 		    PSTORM_OVERLAY_BUF_ADDR_OFFSET;
1816 
1817 	default:
1818 		return 0;
1819 	}
1820 }
1821 
1822 struct phys_mem_desc *qed_fw_overlay_mem_alloc(struct qed_hwfn *p_hwfn,
1823 					       const u32 * const
1824 					       fw_overlay_in_buf,
1825 					       u32 buf_size_in_bytes)
1826 {
1827 	u32 buf_size = buf_size_in_bytes / sizeof(u32), buf_offset = 0;
1828 	struct phys_mem_desc *allocated_mem;
1829 
1830 	if (!buf_size)
1831 		return NULL;
1832 
1833 	allocated_mem = kcalloc(NUM_STORMS, sizeof(struct phys_mem_desc),
1834 				GFP_KERNEL);
1835 	if (!allocated_mem)
1836 		return NULL;
1837 
1838 	memset(allocated_mem, 0, NUM_STORMS * sizeof(struct phys_mem_desc));
1839 
1840 	/* For each Storm, set physical address in RAM */
1841 	while (buf_offset < buf_size) {
1842 		struct phys_mem_desc *storm_mem_desc;
1843 		struct fw_overlay_buf_hdr *hdr;
1844 		u32 storm_buf_size;
1845 		u8 storm_id;
1846 
1847 		hdr =
1848 		    (struct fw_overlay_buf_hdr *)&fw_overlay_in_buf[buf_offset];
1849 		storm_buf_size = GET_FIELD(hdr->data,
1850 					   FW_OVERLAY_BUF_HDR_BUF_SIZE);
1851 		storm_id = GET_FIELD(hdr->data, FW_OVERLAY_BUF_HDR_STORM_ID);
1852 		if (storm_id >= NUM_STORMS)
1853 			break;
1854 		storm_mem_desc = allocated_mem + storm_id;
1855 		storm_mem_desc->size = storm_buf_size * sizeof(u32);
1856 
1857 		/* Allocate physical memory for Storm's overlays buffer */
1858 		storm_mem_desc->virt_addr =
1859 		    dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1860 				       storm_mem_desc->size,
1861 				       &storm_mem_desc->phys_addr, GFP_KERNEL);
1862 		if (!storm_mem_desc->virt_addr)
1863 			break;
1864 
1865 		/* Skip overlays buffer header */
1866 		buf_offset += OVERLAY_HDR_SIZE_DWORDS;
1867 
1868 		/* Copy Storm's overlays buffer to allocated memory */
1869 		memcpy(storm_mem_desc->virt_addr,
1870 		       &fw_overlay_in_buf[buf_offset], storm_mem_desc->size);
1871 
1872 		/* Advance to next Storm */
1873 		buf_offset += storm_buf_size;
1874 	}
1875 
1876 	/* If memory allocation has failed, free all allocated memory */
1877 	if (buf_offset < buf_size) {
1878 		qed_fw_overlay_mem_free(p_hwfn, &allocated_mem);
1879 		return NULL;
1880 	}
1881 
1882 	return allocated_mem;
1883 }
1884 
1885 void qed_fw_overlay_init_ram(struct qed_hwfn *p_hwfn,
1886 			     struct qed_ptt *p_ptt,
1887 			     struct phys_mem_desc *fw_overlay_mem)
1888 {
1889 	u8 storm_id;
1890 
1891 	for (storm_id = 0; storm_id < NUM_STORMS; storm_id++) {
1892 		struct phys_mem_desc *storm_mem_desc =
1893 		    (struct phys_mem_desc *)fw_overlay_mem + storm_id;
1894 		u32 ram_addr, i;
1895 
1896 		/* Skip Storms with no FW overlays */
1897 		if (!storm_mem_desc->virt_addr)
1898 			continue;
1899 
1900 		/* Calculate overlay RAM GRC address of current PF */
1901 		ram_addr = qed_get_overlay_addr_ram_addr(p_hwfn, storm_id) +
1902 			   sizeof(dma_addr_t) * p_hwfn->rel_pf_id;
1903 
1904 		/* Write Storm's overlay physical address to RAM */
1905 		for (i = 0; i < PHYS_ADDR_DWORDS; i++, ram_addr += sizeof(u32))
1906 			qed_wr(p_hwfn, p_ptt, ram_addr,
1907 			       ((u32 *)&storm_mem_desc->phys_addr)[i]);
1908 	}
1909 }
1910 
1911 void qed_fw_overlay_mem_free(struct qed_hwfn *p_hwfn,
1912 			     struct phys_mem_desc **fw_overlay_mem)
1913 {
1914 	u8 storm_id;
1915 
1916 	if (!fw_overlay_mem || !(*fw_overlay_mem))
1917 		return;
1918 
1919 	for (storm_id = 0; storm_id < NUM_STORMS; storm_id++) {
1920 		struct phys_mem_desc *storm_mem_desc =
1921 		    (struct phys_mem_desc *)*fw_overlay_mem + storm_id;
1922 
1923 		/* Free Storm's physical memory */
1924 		if (storm_mem_desc->virt_addr)
1925 			dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1926 					  storm_mem_desc->size,
1927 					  storm_mem_desc->virt_addr,
1928 					  storm_mem_desc->phys_addr);
1929 	}
1930 
1931 	/* Free allocated virtual memory */
1932 	kfree(*fw_overlay_mem);
1933 	*fw_overlay_mem = NULL;
1934 }
1935