1 // SPDX-License-Identifier: GPL-2.0-only OR Linux-OpenIB
2 /*
3  * Mellanox BlueField Performance Monitoring Counters driver
4  *
5  * This driver provides a sysfs interface for monitoring
6  * performance statistics in BlueField SoC.
7  *
8  * Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
9  */
10 
11 #include <linux/acpi.h>
12 #include <linux/arm-smccc.h>
13 #include <linux/bitfield.h>
14 #include <linux/errno.h>
15 #include <linux/hwmon.h>
16 #include <linux/platform_device.h>
17 #include <linux/string.h>
18 #include <uapi/linux/psci.h>
19 
20 #define MLXBF_PMC_WRITE_REG_32 0x82000009
21 #define MLXBF_PMC_READ_REG_32 0x8200000A
22 #define MLXBF_PMC_WRITE_REG_64 0x8200000B
23 #define MLXBF_PMC_READ_REG_64 0x8200000C
24 #define MLXBF_PMC_SIP_SVC_UID 0x8200ff01
25 #define MLXBF_PMC_SIP_SVC_VERSION 0x8200ff03
26 #define MLXBF_PMC_SVC_REQ_MAJOR 0
27 #define MLXBF_PMC_SVC_MIN_MINOR 3
28 
29 #define MLXBF_PMC_SMCCC_ACCESS_VIOLATION -4
30 
31 #define MLXBF_PMC_EVENT_SET_BF1 0
32 #define MLXBF_PMC_EVENT_SET_BF2 1
33 #define MLXBF_PMC_EVENT_INFO_LEN 100
34 
35 #define MLXBF_PMC_MAX_BLOCKS 30
36 #define MLXBF_PMC_MAX_ATTRS 30
37 #define MLXBF_PMC_INFO_SZ 4
38 #define MLXBF_PMC_REG_SIZE 8
39 #define MLXBF_PMC_L3C_REG_SIZE 4
40 
41 #define MLXBF_PMC_TYPE_COUNTER 1
42 #define MLXBF_PMC_TYPE_REGISTER 0
43 
44 #define MLXBF_PMC_PERFCTL 0
45 #define MLXBF_PMC_PERFEVT 1
46 #define MLXBF_PMC_PERFACC0 4
47 
48 #define MLXBF_PMC_PERFMON_CONFIG_WR_R_B BIT(0)
49 #define MLXBF_PMC_PERFMON_CONFIG_STROBE BIT(1)
50 #define MLXBF_PMC_PERFMON_CONFIG_ADDR GENMASK_ULL(4, 2)
51 #define MLXBF_PMC_PERFMON_CONFIG_WDATA GENMASK_ULL(60, 5)
52 
53 #define MLXBF_PMC_PERFCTL_FM0 GENMASK_ULL(18, 16)
54 #define MLXBF_PMC_PERFCTL_MS0 GENMASK_ULL(21, 20)
55 #define MLXBF_PMC_PERFCTL_ACCM0 GENMASK_ULL(26, 24)
56 #define MLXBF_PMC_PERFCTL_AD0 BIT(27)
57 #define MLXBF_PMC_PERFCTL_ETRIG0 GENMASK_ULL(29, 28)
58 #define MLXBF_PMC_PERFCTL_EB0 BIT(30)
59 #define MLXBF_PMC_PERFCTL_EN0 BIT(31)
60 
61 #define MLXBF_PMC_PERFEVT_EVTSEL GENMASK_ULL(31, 24)
62 
63 #define MLXBF_PMC_L3C_PERF_CNT_CFG 0x0
64 #define MLXBF_PMC_L3C_PERF_CNT_SEL 0x10
65 #define MLXBF_PMC_L3C_PERF_CNT_SEL_1 0x14
66 #define MLXBF_PMC_L3C_PERF_CNT_LOW 0x40
67 #define MLXBF_PMC_L3C_PERF_CNT_HIGH 0x60
68 
69 #define MLXBF_PMC_L3C_PERF_CNT_CFG_EN BIT(0)
70 #define MLXBF_PMC_L3C_PERF_CNT_CFG_RST BIT(1)
71 #define MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_0 GENMASK(5, 0)
72 #define MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_1 GENMASK(13, 8)
73 #define MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_2 GENMASK(21, 16)
74 #define MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_3 GENMASK(29, 24)
75 
76 #define MLXBF_PMC_L3C_PERF_CNT_SEL_1_CNT_4 GENMASK(5, 0)
77 
78 #define MLXBF_PMC_L3C_PERF_CNT_LOW_VAL GENMASK(31, 0)
79 #define MLXBF_PMC_L3C_PERF_CNT_HIGH_VAL GENMASK(24, 0)
80 
81 /**
82  * struct mlxbf_pmc_attribute - Structure to hold attribute and block info
83  * for each sysfs entry
84  * @dev_attr: Device attribute struct
85  * @index: index to identify counter number within a block
86  * @nr: block number to which the sysfs belongs
87  */
88 struct mlxbf_pmc_attribute {
89 	struct device_attribute dev_attr;
90 	int index;
91 	int nr;
92 };
93 
94 /**
95  * struct mlxbf_pmc_block_info - Structure to hold info for each HW block
96  *
97  * @mmio_base: The VA at which the PMC block is mapped
98  * @blk_size: Size of each mapped region
99  * @counters: Number of counters in the block
100  * @type: Type of counters in the block
101  * @attr_counter: Attributes for "counter" sysfs files
102  * @attr_event: Attributes for "event" sysfs files
103  * @attr_event_list: Attributes for "event_list" sysfs files
104  * @attr_enable: Attributes for "enable" sysfs files
105  * @block_attr: All attributes needed for the block
106  * @block_attr_grp: Attribute group for the block
107  */
108 struct mlxbf_pmc_block_info {
109 	void __iomem *mmio_base;
110 	size_t blk_size;
111 	size_t counters;
112 	int type;
113 	struct mlxbf_pmc_attribute *attr_counter;
114 	struct mlxbf_pmc_attribute *attr_event;
115 	struct mlxbf_pmc_attribute attr_event_list;
116 	struct mlxbf_pmc_attribute attr_enable;
117 	struct attribute *block_attr[MLXBF_PMC_MAX_ATTRS];
118 	struct attribute_group block_attr_grp;
119 };
120 
121 /**
122  * struct mlxbf_pmc_context - Structure to hold PMC context info
123  *
124  * @pdev: The kernel structure representing the device
125  * @total_blocks: Total number of blocks
126  * @tile_count: Number of tiles in the system
127  * @hwmon_dev: Hwmon device for bfperf
128  * @block_name: Block name
129  * @block:  Block info
130  * @groups:  Attribute groups from each block
131  * @svc_sreg_support: Whether SMCs are used to access performance registers
132  * @sreg_tbl_perf: Secure register access table number
133  * @event_set: Event set to use
134  */
135 struct mlxbf_pmc_context {
136 	struct platform_device *pdev;
137 	uint32_t total_blocks;
138 	uint32_t tile_count;
139 	struct device *hwmon_dev;
140 	const char *block_name[MLXBF_PMC_MAX_BLOCKS];
141 	struct mlxbf_pmc_block_info block[MLXBF_PMC_MAX_BLOCKS];
142 	const struct attribute_group *groups[MLXBF_PMC_MAX_BLOCKS];
143 	bool svc_sreg_support;
144 	uint32_t sreg_tbl_perf;
145 	unsigned int event_set;
146 };
147 
148 /**
149  * struct mlxbf_pmc_events - Structure to hold supported events for each block
150  * @evt_num: Event number used to program counters
151  * @evt_name: Name of the event
152  */
153 struct mlxbf_pmc_events {
154 	int evt_num;
155 	char *evt_name;
156 };
157 
158 static const struct mlxbf_pmc_events mlxbf_pmc_pcie_events[] = {
159 	{ 0x0, "IN_P_PKT_CNT" },
160 	{ 0x10, "IN_NP_PKT_CNT" },
161 	{ 0x18, "IN_C_PKT_CNT" },
162 	{ 0x20, "OUT_P_PKT_CNT" },
163 	{ 0x28, "OUT_NP_PKT_CNT" },
164 	{ 0x30, "OUT_C_PKT_CNT" },
165 	{ 0x38, "IN_P_BYTE_CNT" },
166 	{ 0x40, "IN_NP_BYTE_CNT" },
167 	{ 0x48, "IN_C_BYTE_CNT" },
168 	{ 0x50, "OUT_P_BYTE_CNT" },
169 	{ 0x58, "OUT_NP_BYTE_CNT" },
170 	{ 0x60, "OUT_C_BYTE_CNT" },
171 };
172 
173 static const struct mlxbf_pmc_events mlxbf_pmc_smgen_events[] = {
174 	{ 0x0, "AW_REQ" },
175 	{ 0x1, "AW_BEATS" },
176 	{ 0x2, "AW_TRANS" },
177 	{ 0x3, "AW_RESP" },
178 	{ 0x4, "AW_STL" },
179 	{ 0x5, "AW_LAT" },
180 	{ 0x6, "AW_REQ_TBU" },
181 	{ 0x8, "AR_REQ" },
182 	{ 0x9, "AR_BEATS" },
183 	{ 0xa, "AR_TRANS" },
184 	{ 0xb, "AR_STL" },
185 	{ 0xc, "AR_LAT" },
186 	{ 0xd, "AR_REQ_TBU" },
187 	{ 0xe, "TBU_MISS" },
188 	{ 0xf, "TX_DAT_AF" },
189 	{ 0x10, "RX_DAT_AF" },
190 	{ 0x11, "RETRYQ_CRED" },
191 };
192 
193 static const struct mlxbf_pmc_events mlxbf_pmc_trio_events_1[] = {
194 	{ 0x0, "DISABLE" },
195 	{ 0xa0, "TPIO_DATA_BEAT" },
196 	{ 0xa1, "TDMA_DATA_BEAT" },
197 	{ 0xa2, "MAP_DATA_BEAT" },
198 	{ 0xa3, "TXMSG_DATA_BEAT" },
199 	{ 0xa4, "TPIO_DATA_PACKET" },
200 	{ 0xa5, "TDMA_DATA_PACKET" },
201 	{ 0xa6, "MAP_DATA_PACKET" },
202 	{ 0xa7, "TXMSG_DATA_PACKET" },
203 	{ 0xa8, "TDMA_RT_AF" },
204 	{ 0xa9, "TDMA_PBUF_MAC_AF" },
205 	{ 0xaa, "TRIO_MAP_WRQ_BUF_EMPTY" },
206 	{ 0xab, "TRIO_MAP_CPL_BUF_EMPTY" },
207 	{ 0xac, "TRIO_MAP_RDQ0_BUF_EMPTY" },
208 	{ 0xad, "TRIO_MAP_RDQ1_BUF_EMPTY" },
209 	{ 0xae, "TRIO_MAP_RDQ2_BUF_EMPTY" },
210 	{ 0xaf, "TRIO_MAP_RDQ3_BUF_EMPTY" },
211 	{ 0xb0, "TRIO_MAP_RDQ4_BUF_EMPTY" },
212 	{ 0xb1, "TRIO_MAP_RDQ5_BUF_EMPTY" },
213 	{ 0xb2, "TRIO_MAP_RDQ6_BUF_EMPTY" },
214 	{ 0xb3, "TRIO_MAP_RDQ7_BUF_EMPTY" },
215 };
216 
217 static const struct mlxbf_pmc_events mlxbf_pmc_trio_events_2[] = {
218 	{ 0x0, "DISABLE" },
219 	{ 0xa0, "TPIO_DATA_BEAT" },
220 	{ 0xa1, "TDMA_DATA_BEAT" },
221 	{ 0xa2, "MAP_DATA_BEAT" },
222 	{ 0xa3, "TXMSG_DATA_BEAT" },
223 	{ 0xa4, "TPIO_DATA_PACKET" },
224 	{ 0xa5, "TDMA_DATA_PACKET" },
225 	{ 0xa6, "MAP_DATA_PACKET" },
226 	{ 0xa7, "TXMSG_DATA_PACKET" },
227 	{ 0xa8, "TDMA_RT_AF" },
228 	{ 0xa9, "TDMA_PBUF_MAC_AF" },
229 	{ 0xaa, "TRIO_MAP_WRQ_BUF_EMPTY" },
230 	{ 0xab, "TRIO_MAP_CPL_BUF_EMPTY" },
231 	{ 0xac, "TRIO_MAP_RDQ0_BUF_EMPTY" },
232 	{ 0xad, "TRIO_MAP_RDQ1_BUF_EMPTY" },
233 	{ 0xae, "TRIO_MAP_RDQ2_BUF_EMPTY" },
234 	{ 0xaf, "TRIO_MAP_RDQ3_BUF_EMPTY" },
235 	{ 0xb0, "TRIO_MAP_RDQ4_BUF_EMPTY" },
236 	{ 0xb1, "TRIO_MAP_RDQ5_BUF_EMPTY" },
237 	{ 0xb2, "TRIO_MAP_RDQ6_BUF_EMPTY" },
238 	{ 0xb3, "TRIO_MAP_RDQ7_BUF_EMPTY" },
239 	{ 0xb4, "TRIO_RING_TX_FLIT_CH0" },
240 	{ 0xb5, "TRIO_RING_TX_FLIT_CH1" },
241 	{ 0xb6, "TRIO_RING_TX_FLIT_CH2" },
242 	{ 0xb7, "TRIO_RING_TX_FLIT_CH3" },
243 	{ 0xb8, "TRIO_RING_TX_FLIT_CH4" },
244 	{ 0xb9, "TRIO_RING_RX_FLIT_CH0" },
245 	{ 0xba, "TRIO_RING_RX_FLIT_CH1" },
246 	{ 0xbb, "TRIO_RING_RX_FLIT_CH2" },
247 	{ 0xbc, "TRIO_RING_RX_FLIT_CH3" },
248 };
249 
250 static const struct mlxbf_pmc_events mlxbf_pmc_ecc_events[] = {
251 	{ 0x0, "DISABLE" },
252 	{ 0x100, "ECC_SINGLE_ERROR_CNT" },
253 	{ 0x104, "ECC_DOUBLE_ERROR_CNT" },
254 	{ 0x114, "SERR_INJ" },
255 	{ 0x118, "DERR_INJ" },
256 	{ 0x124, "ECC_SINGLE_ERROR_0" },
257 	{ 0x164, "ECC_DOUBLE_ERROR_0" },
258 	{ 0x340, "DRAM_ECC_COUNT" },
259 	{ 0x344, "DRAM_ECC_INJECT" },
260 	{ 0x348, "DRAM_ECC_ERROR" },
261 };
262 
263 static const struct mlxbf_pmc_events mlxbf_pmc_mss_events[] = {
264 	{ 0x0, "DISABLE" },
265 	{ 0xc0, "RXREQ_MSS" },
266 	{ 0xc1, "RXDAT_MSS" },
267 	{ 0xc2, "TXRSP_MSS" },
268 	{ 0xc3, "TXDAT_MSS" },
269 };
270 
271 static const struct mlxbf_pmc_events mlxbf_pmc_hnf_events[] = {
272 	{ 0x0, "DISABLE" },
273 	{ 0x45, "HNF_REQUESTS" },
274 	{ 0x46, "HNF_REJECTS" },
275 	{ 0x47, "ALL_BUSY" },
276 	{ 0x48, "MAF_BUSY" },
277 	{ 0x49, "MAF_REQUESTS" },
278 	{ 0x4a, "RNF_REQUESTS" },
279 	{ 0x4b, "REQUEST_TYPE" },
280 	{ 0x4c, "MEMORY_READS" },
281 	{ 0x4d, "MEMORY_WRITES" },
282 	{ 0x4e, "VICTIM_WRITE" },
283 	{ 0x4f, "POC_FULL" },
284 	{ 0x50, "POC_FAIL" },
285 	{ 0x51, "POC_SUCCESS" },
286 	{ 0x52, "POC_WRITES" },
287 	{ 0x53, "POC_READS" },
288 	{ 0x54, "FORWARD" },
289 	{ 0x55, "RXREQ_HNF" },
290 	{ 0x56, "RXRSP_HNF" },
291 	{ 0x57, "RXDAT_HNF" },
292 	{ 0x58, "TXREQ_HNF" },
293 	{ 0x59, "TXRSP_HNF" },
294 	{ 0x5a, "TXDAT_HNF" },
295 	{ 0x5b, "TXSNP_HNF" },
296 	{ 0x5c, "INDEX_MATCH" },
297 	{ 0x5d, "A72_ACCESS" },
298 	{ 0x5e, "IO_ACCESS" },
299 	{ 0x5f, "TSO_WRITE" },
300 	{ 0x60, "TSO_CONFLICT" },
301 	{ 0x61, "DIR_HIT" },
302 	{ 0x62, "HNF_ACCEPTS" },
303 	{ 0x63, "REQ_BUF_EMPTY" },
304 	{ 0x64, "REQ_BUF_IDLE_MAF" },
305 	{ 0x65, "TSO_NOARB" },
306 	{ 0x66, "TSO_NOARB_CYCLES" },
307 	{ 0x67, "MSS_NO_CREDIT" },
308 	{ 0x68, "TXDAT_NO_LCRD" },
309 	{ 0x69, "TXSNP_NO_LCRD" },
310 	{ 0x6a, "TXRSP_NO_LCRD" },
311 	{ 0x6b, "TXREQ_NO_LCRD" },
312 	{ 0x6c, "TSO_CL_MATCH" },
313 	{ 0x6d, "MEMORY_READS_BYPASS" },
314 	{ 0x6e, "TSO_NOARB_TIMEOUT" },
315 	{ 0x6f, "ALLOCATE" },
316 	{ 0x70, "VICTIM" },
317 	{ 0x71, "A72_WRITE" },
318 	{ 0x72, "A72_READ" },
319 	{ 0x73, "IO_WRITE" },
320 	{ 0x74, "IO_READ" },
321 	{ 0x75, "TSO_REJECT" },
322 	{ 0x80, "TXREQ_RN" },
323 	{ 0x81, "TXRSP_RN" },
324 	{ 0x82, "TXDAT_RN" },
325 	{ 0x83, "RXSNP_RN" },
326 	{ 0x84, "RXRSP_RN" },
327 	{ 0x85, "RXDAT_RN" },
328 };
329 
330 static const struct mlxbf_pmc_events mlxbf_pmc_hnfnet_events[] = {
331 	{ 0x0, "DISABLE" },
332 	{ 0x12, "CDN_REQ" },
333 	{ 0x13, "DDN_REQ" },
334 	{ 0x14, "NDN_REQ" },
335 	{ 0x15, "CDN_DIAG_N_OUT_OF_CRED" },
336 	{ 0x16, "CDN_DIAG_S_OUT_OF_CRED" },
337 	{ 0x17, "CDN_DIAG_E_OUT_OF_CRED" },
338 	{ 0x18, "CDN_DIAG_W_OUT_OF_CRED" },
339 	{ 0x19, "CDN_DIAG_C_OUT_OF_CRED" },
340 	{ 0x1a, "CDN_DIAG_N_EGRESS" },
341 	{ 0x1b, "CDN_DIAG_S_EGRESS" },
342 	{ 0x1c, "CDN_DIAG_E_EGRESS" },
343 	{ 0x1d, "CDN_DIAG_W_EGRESS" },
344 	{ 0x1e, "CDN_DIAG_C_EGRESS" },
345 	{ 0x1f, "CDN_DIAG_N_INGRESS" },
346 	{ 0x20, "CDN_DIAG_S_INGRESS" },
347 	{ 0x21, "CDN_DIAG_E_INGRESS" },
348 	{ 0x22, "CDN_DIAG_W_INGRESS" },
349 	{ 0x23, "CDN_DIAG_C_INGRESS" },
350 	{ 0x24, "CDN_DIAG_CORE_SENT" },
351 	{ 0x25, "DDN_DIAG_N_OUT_OF_CRED" },
352 	{ 0x26, "DDN_DIAG_S_OUT_OF_CRED" },
353 	{ 0x27, "DDN_DIAG_E_OUT_OF_CRED" },
354 	{ 0x28, "DDN_DIAG_W_OUT_OF_CRED" },
355 	{ 0x29, "DDN_DIAG_C_OUT_OF_CRED" },
356 	{ 0x2a, "DDN_DIAG_N_EGRESS" },
357 	{ 0x2b, "DDN_DIAG_S_EGRESS" },
358 	{ 0x2c, "DDN_DIAG_E_EGRESS" },
359 	{ 0x2d, "DDN_DIAG_W_EGRESS" },
360 	{ 0x2e, "DDN_DIAG_C_EGRESS" },
361 	{ 0x2f, "DDN_DIAG_N_INGRESS" },
362 	{ 0x30, "DDN_DIAG_S_INGRESS" },
363 	{ 0x31, "DDN_DIAG_E_INGRESS" },
364 	{ 0x32, "DDN_DIAG_W_INGRESS" },
365 	{ 0x33, "DDN_DIAG_C_INGRESS" },
366 	{ 0x34, "DDN_DIAG_CORE_SENT" },
367 	{ 0x35, "NDN_DIAG_N_OUT_OF_CRED" },
368 	{ 0x36, "NDN_DIAG_S_OUT_OF_CRED" },
369 	{ 0x37, "NDN_DIAG_E_OUT_OF_CRED" },
370 	{ 0x38, "NDN_DIAG_W_OUT_OF_CRED" },
371 	{ 0x39, "NDN_DIAG_C_OUT_OF_CRED" },
372 	{ 0x3a, "NDN_DIAG_N_EGRESS" },
373 	{ 0x3b, "NDN_DIAG_S_EGRESS" },
374 	{ 0x3c, "NDN_DIAG_E_EGRESS" },
375 	{ 0x3d, "NDN_DIAG_W_EGRESS" },
376 	{ 0x3e, "NDN_DIAG_C_EGRESS" },
377 	{ 0x3f, "NDN_DIAG_N_INGRESS" },
378 	{ 0x40, "NDN_DIAG_S_INGRESS" },
379 	{ 0x41, "NDN_DIAG_E_INGRESS" },
380 	{ 0x42, "NDN_DIAG_W_INGRESS" },
381 	{ 0x43, "NDN_DIAG_C_INGRESS" },
382 	{ 0x44, "NDN_DIAG_CORE_SENT" },
383 };
384 
385 static const struct mlxbf_pmc_events mlxbf_pmc_l3c_events[] = {
386 	{ 0x00, "DISABLE" },
387 	{ 0x01, "CYCLES" },
388 	{ 0x02, "TOTAL_RD_REQ_IN" },
389 	{ 0x03, "TOTAL_WR_REQ_IN" },
390 	{ 0x04, "TOTAL_WR_DBID_ACK" },
391 	{ 0x05, "TOTAL_WR_DATA_IN" },
392 	{ 0x06, "TOTAL_WR_COMP" },
393 	{ 0x07, "TOTAL_RD_DATA_OUT" },
394 	{ 0x08, "TOTAL_CDN_REQ_IN_BANK0" },
395 	{ 0x09, "TOTAL_CDN_REQ_IN_BANK1" },
396 	{ 0x0a, "TOTAL_DDN_REQ_IN_BANK0" },
397 	{ 0x0b, "TOTAL_DDN_REQ_IN_BANK1" },
398 	{ 0x0c, "TOTAL_EMEM_RD_RES_IN_BANK0" },
399 	{ 0x0d, "TOTAL_EMEM_RD_RES_IN_BANK1" },
400 	{ 0x0e, "TOTAL_CACHE_RD_RES_IN_BANK0" },
401 	{ 0x0f, "TOTAL_CACHE_RD_RES_IN_BANK1" },
402 	{ 0x10, "TOTAL_EMEM_RD_REQ_BANK0" },
403 	{ 0x11, "TOTAL_EMEM_RD_REQ_BANK1" },
404 	{ 0x12, "TOTAL_EMEM_WR_REQ_BANK0" },
405 	{ 0x13, "TOTAL_EMEM_WR_REQ_BANK1" },
406 	{ 0x14, "TOTAL_RD_REQ_OUT" },
407 	{ 0x15, "TOTAL_WR_REQ_OUT" },
408 	{ 0x16, "TOTAL_RD_RES_IN" },
409 	{ 0x17, "HITS_BANK0" },
410 	{ 0x18, "HITS_BANK1" },
411 	{ 0x19, "MISSES_BANK0" },
412 	{ 0x1a, "MISSES_BANK1" },
413 	{ 0x1b, "ALLOCATIONS_BANK0" },
414 	{ 0x1c, "ALLOCATIONS_BANK1" },
415 	{ 0x1d, "EVICTIONS_BANK0" },
416 	{ 0x1e, "EVICTIONS_BANK1" },
417 	{ 0x1f, "DBID_REJECT" },
418 	{ 0x20, "WRDB_REJECT_BANK0" },
419 	{ 0x21, "WRDB_REJECT_BANK1" },
420 	{ 0x22, "CMDQ_REJECT_BANK0" },
421 	{ 0x23, "CMDQ_REJECT_BANK1" },
422 	{ 0x24, "COB_REJECT_BANK0" },
423 	{ 0x25, "COB_REJECT_BANK1" },
424 	{ 0x26, "TRB_REJECT_BANK0" },
425 	{ 0x27, "TRB_REJECT_BANK1" },
426 	{ 0x28, "TAG_REJECT_BANK0" },
427 	{ 0x29, "TAG_REJECT_BANK1" },
428 	{ 0x2a, "ANY_REJECT_BANK0" },
429 	{ 0x2b, "ANY_REJECT_BANK1" },
430 };
431 
432 static struct mlxbf_pmc_context *pmc;
433 
434 /* UUID used to probe ATF service. */
435 static const char *mlxbf_pmc_svc_uuid_str = "89c036b4-e7d7-11e6-8797-001aca00bfc4";
436 
437 /* Calls an SMC to access a performance register */
438 static int mlxbf_pmc_secure_read(void __iomem *addr, uint32_t command,
439 				 uint64_t *result)
440 {
441 	struct arm_smccc_res res;
442 	int status, err = 0;
443 
444 	arm_smccc_smc(command, pmc->sreg_tbl_perf, (uintptr_t)addr, 0, 0, 0, 0,
445 		      0, &res);
446 
447 	status = res.a0;
448 
449 	switch (status) {
450 	case PSCI_RET_NOT_SUPPORTED:
451 		err = -EINVAL;
452 		break;
453 	case MLXBF_PMC_SMCCC_ACCESS_VIOLATION:
454 		err = -EACCES;
455 		break;
456 	default:
457 		*result = res.a1;
458 		break;
459 	}
460 
461 	return err;
462 }
463 
464 /* Read from a performance counter */
465 static int mlxbf_pmc_read(void __iomem *addr, uint32_t command,
466 			  uint64_t *result)
467 {
468 	if (pmc->svc_sreg_support)
469 		return mlxbf_pmc_secure_read(addr, command, result);
470 
471 	if (command == MLXBF_PMC_READ_REG_32)
472 		*result = readl(addr);
473 	else
474 		*result = readq(addr);
475 
476 	return 0;
477 }
478 
479 /* Convenience function for 32-bit reads */
480 static int mlxbf_pmc_readl(void __iomem *addr, uint32_t *result)
481 {
482 	uint64_t read_out;
483 	int status;
484 
485 	status = mlxbf_pmc_read(addr, MLXBF_PMC_READ_REG_32, &read_out);
486 	if (status)
487 		return status;
488 	*result = (uint32_t)read_out;
489 
490 	return 0;
491 }
492 
493 /* Calls an SMC to access a performance register */
494 static int mlxbf_pmc_secure_write(void __iomem *addr, uint32_t command,
495 				  uint64_t value)
496 {
497 	struct arm_smccc_res res;
498 	int status, err = 0;
499 
500 	arm_smccc_smc(command, pmc->sreg_tbl_perf, value, (uintptr_t)addr, 0, 0,
501 		      0, 0, &res);
502 
503 	status = res.a0;
504 
505 	switch (status) {
506 	case PSCI_RET_NOT_SUPPORTED:
507 		err = -EINVAL;
508 		break;
509 	case MLXBF_PMC_SMCCC_ACCESS_VIOLATION:
510 		err = -EACCES;
511 		break;
512 	}
513 
514 	return err;
515 }
516 
517 /* Write to a performance counter */
518 static int mlxbf_pmc_write(void __iomem *addr, int command, uint64_t value)
519 {
520 	if (pmc->svc_sreg_support)
521 		return mlxbf_pmc_secure_write(addr, command, value);
522 
523 	if (command == MLXBF_PMC_WRITE_REG_32)
524 		writel(value, addr);
525 	else
526 		writeq(value, addr);
527 
528 	return 0;
529 }
530 
531 /* Check if the register offset is within the mapped region for the block */
532 static bool mlxbf_pmc_valid_range(int blk_num, uint32_t offset)
533 {
534 	if ((offset >= 0) && !(offset % MLXBF_PMC_REG_SIZE) &&
535 	    (offset + MLXBF_PMC_REG_SIZE <= pmc->block[blk_num].blk_size))
536 		return true; /* inside the mapped PMC space */
537 
538 	return false;
539 }
540 
541 /* Get the event list corresponding to a certain block */
542 static const struct mlxbf_pmc_events *mlxbf_pmc_event_list(const char *blk,
543 							   int *size)
544 {
545 	const struct mlxbf_pmc_events *events;
546 
547 	if (strstr(blk, "tilenet")) {
548 		events = mlxbf_pmc_hnfnet_events;
549 		*size = ARRAY_SIZE(mlxbf_pmc_hnfnet_events);
550 	} else if (strstr(blk, "tile")) {
551 		events = mlxbf_pmc_hnf_events;
552 		*size = ARRAY_SIZE(mlxbf_pmc_hnf_events);
553 	} else if (strstr(blk, "triogen")) {
554 		events = mlxbf_pmc_smgen_events;
555 		*size = ARRAY_SIZE(mlxbf_pmc_smgen_events);
556 	} else if (strstr(blk, "trio")) {
557 		switch (pmc->event_set) {
558 		case MLXBF_PMC_EVENT_SET_BF1:
559 			events = mlxbf_pmc_trio_events_1;
560 			*size = ARRAY_SIZE(mlxbf_pmc_trio_events_1);
561 			break;
562 		case MLXBF_PMC_EVENT_SET_BF2:
563 			events = mlxbf_pmc_trio_events_2;
564 			*size = ARRAY_SIZE(mlxbf_pmc_trio_events_2);
565 			break;
566 		default:
567 			events = NULL;
568 			*size = 0;
569 			break;
570 		}
571 	} else if (strstr(blk, "mss")) {
572 		events = mlxbf_pmc_mss_events;
573 		*size = ARRAY_SIZE(mlxbf_pmc_mss_events);
574 	} else if (strstr(blk, "ecc")) {
575 		events = mlxbf_pmc_ecc_events;
576 		*size = ARRAY_SIZE(mlxbf_pmc_ecc_events);
577 	} else if (strstr(blk, "pcie")) {
578 		events = mlxbf_pmc_pcie_events;
579 		*size = ARRAY_SIZE(mlxbf_pmc_pcie_events);
580 	} else if (strstr(blk, "l3cache")) {
581 		events = mlxbf_pmc_l3c_events;
582 		*size = ARRAY_SIZE(mlxbf_pmc_l3c_events);
583 	} else if (strstr(blk, "gic")) {
584 		events = mlxbf_pmc_smgen_events;
585 		*size = ARRAY_SIZE(mlxbf_pmc_smgen_events);
586 	} else if (strstr(blk, "smmu")) {
587 		events = mlxbf_pmc_smgen_events;
588 		*size = ARRAY_SIZE(mlxbf_pmc_smgen_events);
589 	} else {
590 		events = NULL;
591 		*size = 0;
592 	}
593 
594 	return events;
595 }
596 
597 /* Get the event number given the name */
598 static int mlxbf_pmc_get_event_num(const char *blk, const char *evt)
599 {
600 	const struct mlxbf_pmc_events *events;
601 	int i, size;
602 
603 	events = mlxbf_pmc_event_list(blk, &size);
604 	if (!events)
605 		return -EINVAL;
606 
607 	for (i = 0; i < size; ++i) {
608 		if (!strcmp(evt, events[i].evt_name))
609 			return events[i].evt_num;
610 	}
611 
612 	return -ENODEV;
613 }
614 
615 /* Get the event number given the name */
616 static char *mlxbf_pmc_get_event_name(const char *blk, int evt)
617 {
618 	const struct mlxbf_pmc_events *events;
619 	int i, size;
620 
621 	events = mlxbf_pmc_event_list(blk, &size);
622 	if (!events)
623 		return NULL;
624 
625 	for (i = 0; i < size; ++i) {
626 		if (evt == events[i].evt_num)
627 			return events[i].evt_name;
628 	}
629 
630 	return NULL;
631 }
632 
633 /* Method to enable/disable/reset l3cache counters */
634 static int mlxbf_pmc_config_l3_counters(int blk_num, bool enable, bool reset)
635 {
636 	uint32_t perfcnt_cfg = 0;
637 
638 	if (enable)
639 		perfcnt_cfg |= MLXBF_PMC_L3C_PERF_CNT_CFG_EN;
640 	if (reset)
641 		perfcnt_cfg |= MLXBF_PMC_L3C_PERF_CNT_CFG_RST;
642 
643 	return mlxbf_pmc_write(pmc->block[blk_num].mmio_base +
644 				       MLXBF_PMC_L3C_PERF_CNT_CFG,
645 			       MLXBF_PMC_WRITE_REG_32, perfcnt_cfg);
646 }
647 
648 /* Method to handle l3cache counter programming */
649 static int mlxbf_pmc_program_l3_counter(int blk_num, uint32_t cnt_num,
650 					uint32_t evt)
651 {
652 	uint32_t perfcnt_sel_1 = 0;
653 	uint32_t perfcnt_sel = 0;
654 	uint32_t *wordaddr;
655 	void __iomem *pmcaddr;
656 	int ret;
657 
658 	/* Disable all counters before programming them */
659 	if (mlxbf_pmc_config_l3_counters(blk_num, false, false))
660 		return -EINVAL;
661 
662 	/* Select appropriate register information */
663 	switch (cnt_num) {
664 	case 0 ... 3:
665 		pmcaddr = pmc->block[blk_num].mmio_base +
666 			  MLXBF_PMC_L3C_PERF_CNT_SEL;
667 		wordaddr = &perfcnt_sel;
668 		break;
669 	case 4:
670 		pmcaddr = pmc->block[blk_num].mmio_base +
671 			  MLXBF_PMC_L3C_PERF_CNT_SEL_1;
672 		wordaddr = &perfcnt_sel_1;
673 		break;
674 	default:
675 		return -EINVAL;
676 	}
677 
678 	ret = mlxbf_pmc_readl(pmcaddr, wordaddr);
679 	if (ret)
680 		return ret;
681 
682 	switch (cnt_num) {
683 	case 0:
684 		perfcnt_sel &= ~MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_0;
685 		perfcnt_sel |= FIELD_PREP(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_0,
686 					  evt);
687 		break;
688 	case 1:
689 		perfcnt_sel &= ~MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_1;
690 		perfcnt_sel |= FIELD_PREP(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_1,
691 					  evt);
692 		break;
693 	case 2:
694 		perfcnt_sel &= ~MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_2;
695 		perfcnt_sel |= FIELD_PREP(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_2,
696 					  evt);
697 		break;
698 	case 3:
699 		perfcnt_sel &= ~MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_3;
700 		perfcnt_sel |= FIELD_PREP(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_3,
701 					  evt);
702 		break;
703 	case 4:
704 		perfcnt_sel_1 &= ~MLXBF_PMC_L3C_PERF_CNT_SEL_1_CNT_4;
705 		perfcnt_sel_1 |= FIELD_PREP(MLXBF_PMC_L3C_PERF_CNT_SEL_1_CNT_4,
706 					    evt);
707 		break;
708 	default:
709 		return -EINVAL;
710 	}
711 
712 	return mlxbf_pmc_write(pmcaddr, MLXBF_PMC_WRITE_REG_32, *wordaddr);
713 }
714 
715 /* Method to program a counter to monitor an event */
716 static int mlxbf_pmc_program_counter(int blk_num, uint32_t cnt_num,
717 				     uint32_t evt, bool is_l3)
718 {
719 	uint64_t perfctl, perfevt, perfmon_cfg;
720 
721 	if (cnt_num >= pmc->block[blk_num].counters)
722 		return -ENODEV;
723 
724 	if (is_l3)
725 		return mlxbf_pmc_program_l3_counter(blk_num, cnt_num, evt);
726 
727 	/* Configure the counter */
728 	perfctl = FIELD_PREP(MLXBF_PMC_PERFCTL_EN0, 1);
729 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_EB0, 0);
730 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_ETRIG0, 1);
731 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_AD0, 0);
732 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_ACCM0, 0);
733 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_MS0, 0);
734 	perfctl |= FIELD_PREP(MLXBF_PMC_PERFCTL_FM0, 0);
735 
736 	perfmon_cfg = FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WDATA, perfctl);
737 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_ADDR,
738 				  MLXBF_PMC_PERFCTL);
739 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_STROBE, 1);
740 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WR_R_B, 1);
741 
742 	if (mlxbf_pmc_write(pmc->block[blk_num].mmio_base +
743 				    cnt_num * MLXBF_PMC_REG_SIZE,
744 			    MLXBF_PMC_WRITE_REG_64, perfmon_cfg))
745 		return -EFAULT;
746 
747 	/* Select the event */
748 	perfevt = FIELD_PREP(MLXBF_PMC_PERFEVT_EVTSEL, evt);
749 
750 	perfmon_cfg = FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WDATA, perfevt);
751 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_ADDR,
752 				  MLXBF_PMC_PERFEVT);
753 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_STROBE, 1);
754 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WR_R_B, 1);
755 
756 	if (mlxbf_pmc_write(pmc->block[blk_num].mmio_base +
757 				    cnt_num * MLXBF_PMC_REG_SIZE,
758 			    MLXBF_PMC_WRITE_REG_64, perfmon_cfg))
759 		return -EFAULT;
760 
761 	/* Clear the accumulator */
762 	perfmon_cfg = FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_ADDR,
763 				 MLXBF_PMC_PERFACC0);
764 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_STROBE, 1);
765 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WR_R_B, 1);
766 
767 	if (mlxbf_pmc_write(pmc->block[blk_num].mmio_base +
768 				    cnt_num * MLXBF_PMC_REG_SIZE,
769 			    MLXBF_PMC_WRITE_REG_64, perfmon_cfg))
770 		return -EFAULT;
771 
772 	return 0;
773 }
774 
775 /* Method to handle l3 counter reads */
776 static int mlxbf_pmc_read_l3_counter(int blk_num, uint32_t cnt_num,
777 				     uint64_t *result)
778 {
779 	uint32_t perfcnt_low = 0, perfcnt_high = 0;
780 	uint64_t value;
781 	int status = 0;
782 
783 	status = mlxbf_pmc_readl(pmc->block[blk_num].mmio_base +
784 					 MLXBF_PMC_L3C_PERF_CNT_LOW +
785 					 cnt_num * MLXBF_PMC_L3C_REG_SIZE,
786 				 &perfcnt_low);
787 
788 	if (status)
789 		return status;
790 
791 	status = mlxbf_pmc_readl(pmc->block[blk_num].mmio_base +
792 					 MLXBF_PMC_L3C_PERF_CNT_HIGH +
793 					 cnt_num * MLXBF_PMC_L3C_REG_SIZE,
794 				 &perfcnt_high);
795 
796 	if (status)
797 		return status;
798 
799 	value = perfcnt_high;
800 	value = value << 32;
801 	value |= perfcnt_low;
802 	*result = value;
803 
804 	return 0;
805 }
806 
807 /* Method to read the counter value */
808 static int mlxbf_pmc_read_counter(int blk_num, uint32_t cnt_num, bool is_l3,
809 				  uint64_t *result)
810 {
811 	uint32_t perfcfg_offset, perfval_offset;
812 	uint64_t perfmon_cfg;
813 	int status;
814 
815 	if (cnt_num >= pmc->block[blk_num].counters)
816 		return -EINVAL;
817 
818 	if (is_l3)
819 		return mlxbf_pmc_read_l3_counter(blk_num, cnt_num, result);
820 
821 	perfcfg_offset = cnt_num * MLXBF_PMC_REG_SIZE;
822 	perfval_offset = perfcfg_offset +
823 			 pmc->block[blk_num].counters * MLXBF_PMC_REG_SIZE;
824 
825 	/* Set counter in "read" mode */
826 	perfmon_cfg = FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_ADDR,
827 				 MLXBF_PMC_PERFACC0);
828 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_STROBE, 1);
829 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WR_R_B, 0);
830 
831 	status = mlxbf_pmc_write(pmc->block[blk_num].mmio_base + perfcfg_offset,
832 				 MLXBF_PMC_WRITE_REG_64, perfmon_cfg);
833 
834 	if (status)
835 		return status;
836 
837 	/* Get the counter value */
838 	return mlxbf_pmc_read(pmc->block[blk_num].mmio_base + perfval_offset,
839 			      MLXBF_PMC_READ_REG_64, result);
840 }
841 
842 /* Method to read L3 block event */
843 static int mlxbf_pmc_read_l3_event(int blk_num, uint32_t cnt_num,
844 				   uint64_t *result)
845 {
846 	uint32_t perfcnt_sel = 0, perfcnt_sel_1 = 0;
847 	uint32_t *wordaddr;
848 	void __iomem *pmcaddr;
849 	uint64_t evt;
850 
851 	/* Select appropriate register information */
852 	switch (cnt_num) {
853 	case 0 ... 3:
854 		pmcaddr = pmc->block[blk_num].mmio_base +
855 			  MLXBF_PMC_L3C_PERF_CNT_SEL;
856 		wordaddr = &perfcnt_sel;
857 		break;
858 	case 4:
859 		pmcaddr = pmc->block[blk_num].mmio_base +
860 			  MLXBF_PMC_L3C_PERF_CNT_SEL_1;
861 		wordaddr = &perfcnt_sel_1;
862 		break;
863 	default:
864 		return -EINVAL;
865 	}
866 
867 	if (mlxbf_pmc_readl(pmcaddr, wordaddr))
868 		return -EINVAL;
869 
870 	/* Read from appropriate register field for the counter */
871 	switch (cnt_num) {
872 	case 0:
873 		evt = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_0, perfcnt_sel);
874 		break;
875 	case 1:
876 		evt = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_1, perfcnt_sel);
877 		break;
878 	case 2:
879 		evt = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_2, perfcnt_sel);
880 		break;
881 	case 3:
882 		evt = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_SEL_CNT_3, perfcnt_sel);
883 		break;
884 	case 4:
885 		evt = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_SEL_1_CNT_4,
886 				perfcnt_sel_1);
887 		break;
888 	default:
889 		return -EINVAL;
890 	}
891 	*result = evt;
892 
893 	return 0;
894 }
895 
896 /* Method to find the event currently being monitored by a counter */
897 static int mlxbf_pmc_read_event(int blk_num, uint32_t cnt_num, bool is_l3,
898 				uint64_t *result)
899 {
900 	uint32_t perfcfg_offset, perfval_offset;
901 	uint64_t perfmon_cfg, perfevt;
902 
903 	if (cnt_num >= pmc->block[blk_num].counters)
904 		return -EINVAL;
905 
906 	if (is_l3)
907 		return mlxbf_pmc_read_l3_event(blk_num, cnt_num, result);
908 
909 	perfcfg_offset = cnt_num * MLXBF_PMC_REG_SIZE;
910 	perfval_offset = perfcfg_offset +
911 			 pmc->block[blk_num].counters * MLXBF_PMC_REG_SIZE;
912 
913 	/* Set counter in "read" mode */
914 	perfmon_cfg = FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_ADDR,
915 				 MLXBF_PMC_PERFEVT);
916 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_STROBE, 1);
917 	perfmon_cfg |= FIELD_PREP(MLXBF_PMC_PERFMON_CONFIG_WR_R_B, 0);
918 
919 	if (mlxbf_pmc_write(pmc->block[blk_num].mmio_base + perfcfg_offset,
920 			    MLXBF_PMC_WRITE_REG_64, perfmon_cfg))
921 		return -EFAULT;
922 
923 	/* Get the event number */
924 	if (mlxbf_pmc_read(pmc->block[blk_num].mmio_base + perfval_offset,
925 			   MLXBF_PMC_READ_REG_64, &perfevt))
926 		return -EFAULT;
927 
928 	*result = FIELD_GET(MLXBF_PMC_PERFEVT_EVTSEL, perfevt);
929 
930 	return 0;
931 }
932 
933 /* Method to read a register */
934 static int mlxbf_pmc_read_reg(int blk_num, uint32_t offset, uint64_t *result)
935 {
936 	uint32_t ecc_out;
937 
938 	if (strstr(pmc->block_name[blk_num], "ecc")) {
939 		if (mlxbf_pmc_readl(pmc->block[blk_num].mmio_base + offset,
940 				    &ecc_out))
941 			return -EFAULT;
942 
943 		*result = ecc_out;
944 		return 0;
945 	}
946 
947 	if (mlxbf_pmc_valid_range(blk_num, offset))
948 		return mlxbf_pmc_read(pmc->block[blk_num].mmio_base + offset,
949 				      MLXBF_PMC_READ_REG_64, result);
950 
951 	return -EINVAL;
952 }
953 
954 /* Method to write to a register */
955 static int mlxbf_pmc_write_reg(int blk_num, uint32_t offset, uint64_t data)
956 {
957 	if (strstr(pmc->block_name[blk_num], "ecc")) {
958 		return mlxbf_pmc_write(pmc->block[blk_num].mmio_base + offset,
959 				       MLXBF_PMC_WRITE_REG_32, data);
960 	}
961 
962 	if (mlxbf_pmc_valid_range(blk_num, offset))
963 		return mlxbf_pmc_write(pmc->block[blk_num].mmio_base + offset,
964 				       MLXBF_PMC_WRITE_REG_64, data);
965 
966 	return -EINVAL;
967 }
968 
969 /* Show function for "counter" sysfs files */
970 static ssize_t mlxbf_pmc_counter_show(struct device *dev,
971 				      struct device_attribute *attr, char *buf)
972 {
973 	struct mlxbf_pmc_attribute *attr_counter = container_of(
974 		attr, struct mlxbf_pmc_attribute, dev_attr);
975 	int blk_num, cnt_num, offset;
976 	bool is_l3 = false;
977 	uint64_t value;
978 
979 	blk_num = attr_counter->nr;
980 	cnt_num = attr_counter->index;
981 
982 	if (strstr(pmc->block_name[blk_num], "l3cache"))
983 		is_l3 = true;
984 
985 	if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_COUNTER) {
986 		if (mlxbf_pmc_read_counter(blk_num, cnt_num, is_l3, &value))
987 			return -EINVAL;
988 	} else if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_REGISTER) {
989 		offset = mlxbf_pmc_get_event_num(pmc->block_name[blk_num],
990 						 attr->attr.name);
991 		if (offset < 0)
992 			return -EINVAL;
993 		if (mlxbf_pmc_read_reg(blk_num, offset, &value))
994 			return -EINVAL;
995 	} else
996 		return -EINVAL;
997 
998 	return sysfs_emit(buf, "0x%llx\n", value);
999 }
1000 
1001 /* Store function for "counter" sysfs files */
1002 static ssize_t mlxbf_pmc_counter_store(struct device *dev,
1003 				       struct device_attribute *attr,
1004 				       const char *buf, size_t count)
1005 {
1006 	struct mlxbf_pmc_attribute *attr_counter = container_of(
1007 		attr, struct mlxbf_pmc_attribute, dev_attr);
1008 	int blk_num, cnt_num, offset, err, data;
1009 	bool is_l3 = false;
1010 	uint64_t evt_num;
1011 
1012 	blk_num = attr_counter->nr;
1013 	cnt_num = attr_counter->index;
1014 
1015 	err = kstrtoint(buf, 0, &data);
1016 	if (err < 0)
1017 		return err;
1018 
1019 	/* Allow non-zero writes only to the ecc regs */
1020 	if (!(strstr(pmc->block_name[blk_num], "ecc")) && data)
1021 		return -EINVAL;
1022 
1023 	/* Do not allow writes to the L3C regs */
1024 	if (strstr(pmc->block_name[blk_num], "l3cache"))
1025 		return -EINVAL;
1026 
1027 	if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_COUNTER) {
1028 		err = mlxbf_pmc_read_event(blk_num, cnt_num, is_l3, &evt_num);
1029 		if (err)
1030 			return err;
1031 		err = mlxbf_pmc_program_counter(blk_num, cnt_num, evt_num,
1032 						is_l3);
1033 		if (err)
1034 			return err;
1035 	} else if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_REGISTER) {
1036 		offset = mlxbf_pmc_get_event_num(pmc->block_name[blk_num],
1037 						 attr->attr.name);
1038 		if (offset < 0)
1039 			return -EINVAL;
1040 		err = mlxbf_pmc_write_reg(blk_num, offset, data);
1041 		if (err)
1042 			return err;
1043 	} else
1044 		return -EINVAL;
1045 
1046 	return count;
1047 }
1048 
1049 /* Show function for "event" sysfs files */
1050 static ssize_t mlxbf_pmc_event_show(struct device *dev,
1051 				    struct device_attribute *attr, char *buf)
1052 {
1053 	struct mlxbf_pmc_attribute *attr_event = container_of(
1054 		attr, struct mlxbf_pmc_attribute, dev_attr);
1055 	int blk_num, cnt_num, err;
1056 	bool is_l3 = false;
1057 	uint64_t evt_num;
1058 	char *evt_name;
1059 
1060 	blk_num = attr_event->nr;
1061 	cnt_num = attr_event->index;
1062 
1063 	if (strstr(pmc->block_name[blk_num], "l3cache"))
1064 		is_l3 = true;
1065 
1066 	err = mlxbf_pmc_read_event(blk_num, cnt_num, is_l3, &evt_num);
1067 	if (err)
1068 		return sysfs_emit(buf, "No event being monitored\n");
1069 
1070 	evt_name = mlxbf_pmc_get_event_name(pmc->block_name[blk_num], evt_num);
1071 	if (!evt_name)
1072 		return -EINVAL;
1073 
1074 	return sysfs_emit(buf, "0x%llx: %s\n", evt_num, evt_name);
1075 }
1076 
1077 /* Store function for "event" sysfs files */
1078 static ssize_t mlxbf_pmc_event_store(struct device *dev,
1079 				     struct device_attribute *attr,
1080 				     const char *buf, size_t count)
1081 {
1082 	struct mlxbf_pmc_attribute *attr_event = container_of(
1083 		attr, struct mlxbf_pmc_attribute, dev_attr);
1084 	int blk_num, cnt_num, evt_num, err;
1085 	bool is_l3 = false;
1086 
1087 	blk_num = attr_event->nr;
1088 	cnt_num = attr_event->index;
1089 
1090 	if (isalpha(buf[0])) {
1091 		evt_num = mlxbf_pmc_get_event_num(pmc->block_name[blk_num],
1092 						  buf);
1093 		if (evt_num < 0)
1094 			return -EINVAL;
1095 	} else {
1096 		err = kstrtoint(buf, 0, &evt_num);
1097 		if (err < 0)
1098 			return err;
1099 	}
1100 
1101 	if (strstr(pmc->block_name[blk_num], "l3cache"))
1102 		is_l3 = true;
1103 
1104 	err = mlxbf_pmc_program_counter(blk_num, cnt_num, evt_num, is_l3);
1105 	if (err)
1106 		return err;
1107 
1108 	return count;
1109 }
1110 
1111 /* Show function for "event_list" sysfs files */
1112 static ssize_t mlxbf_pmc_event_list_show(struct device *dev,
1113 					 struct device_attribute *attr,
1114 					 char *buf)
1115 {
1116 	struct mlxbf_pmc_attribute *attr_event_list = container_of(
1117 		attr, struct mlxbf_pmc_attribute, dev_attr);
1118 	int blk_num, i, size, len = 0, ret = 0;
1119 	const struct mlxbf_pmc_events *events;
1120 	char e_info[MLXBF_PMC_EVENT_INFO_LEN];
1121 
1122 	blk_num = attr_event_list->nr;
1123 
1124 	events = mlxbf_pmc_event_list(pmc->block_name[blk_num], &size);
1125 	if (!events)
1126 		return -EINVAL;
1127 
1128 	for (i = 0, buf[0] = '\0'; i < size; ++i) {
1129 		len += snprintf(e_info, sizeof(e_info), "0x%x: %s\n",
1130 				events[i].evt_num, events[i].evt_name);
1131 		if (len >= PAGE_SIZE)
1132 			break;
1133 		strcat(buf, e_info);
1134 		ret = len;
1135 	}
1136 
1137 	return ret;
1138 }
1139 
1140 /* Show function for "enable" sysfs files - only for l3cache */
1141 static ssize_t mlxbf_pmc_enable_show(struct device *dev,
1142 				     struct device_attribute *attr, char *buf)
1143 {
1144 	struct mlxbf_pmc_attribute *attr_enable = container_of(
1145 		attr, struct mlxbf_pmc_attribute, dev_attr);
1146 	uint32_t perfcnt_cfg;
1147 	int blk_num, value;
1148 
1149 	blk_num = attr_enable->nr;
1150 
1151 	if (mlxbf_pmc_readl(pmc->block[blk_num].mmio_base +
1152 				    MLXBF_PMC_L3C_PERF_CNT_CFG,
1153 			    &perfcnt_cfg))
1154 		return -EINVAL;
1155 
1156 	value = FIELD_GET(MLXBF_PMC_L3C_PERF_CNT_CFG_EN, perfcnt_cfg);
1157 
1158 	return sysfs_emit(buf, "%d\n", value);
1159 }
1160 
1161 /* Store function for "enable" sysfs files - only for l3cache */
1162 static ssize_t mlxbf_pmc_enable_store(struct device *dev,
1163 				      struct device_attribute *attr,
1164 				      const char *buf, size_t count)
1165 {
1166 	struct mlxbf_pmc_attribute *attr_enable = container_of(
1167 		attr, struct mlxbf_pmc_attribute, dev_attr);
1168 	int err, en, blk_num;
1169 
1170 	blk_num = attr_enable->nr;
1171 
1172 	err = kstrtoint(buf, 0, &en);
1173 	if (err < 0)
1174 		return err;
1175 
1176 	if (!en) {
1177 		err = mlxbf_pmc_config_l3_counters(blk_num, false, false);
1178 		if (err)
1179 			return err;
1180 	} else if (en == 1) {
1181 		err = mlxbf_pmc_config_l3_counters(blk_num, false, true);
1182 		if (err)
1183 			return err;
1184 		err = mlxbf_pmc_config_l3_counters(blk_num, true, false);
1185 		if (err)
1186 			return err;
1187 	} else
1188 		return -EINVAL;
1189 
1190 	return count;
1191 }
1192 
1193 /* Populate attributes for blocks with counters to monitor performance */
1194 static int mlxbf_pmc_init_perftype_counter(struct device *dev, int blk_num)
1195 {
1196 	struct mlxbf_pmc_attribute *attr;
1197 	int i = 0, j = 0;
1198 
1199 	/* "event_list" sysfs to list events supported by the block */
1200 	attr = &pmc->block[blk_num].attr_event_list;
1201 	attr->dev_attr.attr.mode = 0444;
1202 	attr->dev_attr.show = mlxbf_pmc_event_list_show;
1203 	attr->nr = blk_num;
1204 	attr->dev_attr.attr.name = devm_kasprintf(dev, GFP_KERNEL, "event_list");
1205 	if (!attr->dev_attr.attr.name)
1206 		return -ENOMEM;
1207 	pmc->block[blk_num].block_attr[i] = &attr->dev_attr.attr;
1208 	attr = NULL;
1209 
1210 	/* "enable" sysfs to start/stop the counters. Only in L3C blocks */
1211 	if (strstr(pmc->block_name[blk_num], "l3cache")) {
1212 		attr = &pmc->block[blk_num].attr_enable;
1213 		attr->dev_attr.attr.mode = 0644;
1214 		attr->dev_attr.show = mlxbf_pmc_enable_show;
1215 		attr->dev_attr.store = mlxbf_pmc_enable_store;
1216 		attr->nr = blk_num;
1217 		attr->dev_attr.attr.name = devm_kasprintf(dev, GFP_KERNEL,
1218 							  "enable");
1219 		if (!attr->dev_attr.attr.name)
1220 			return -ENOMEM;
1221 		pmc->block[blk_num].block_attr[++i] = &attr->dev_attr.attr;
1222 		attr = NULL;
1223 	}
1224 
1225 	pmc->block[blk_num].attr_counter = devm_kcalloc(
1226 		dev, pmc->block[blk_num].counters,
1227 		sizeof(struct mlxbf_pmc_attribute), GFP_KERNEL);
1228 	if (!pmc->block[blk_num].attr_counter)
1229 		return -ENOMEM;
1230 
1231 	pmc->block[blk_num].attr_event = devm_kcalloc(
1232 		dev, pmc->block[blk_num].counters,
1233 		sizeof(struct mlxbf_pmc_attribute), GFP_KERNEL);
1234 	if (!pmc->block[blk_num].attr_event)
1235 		return -ENOMEM;
1236 
1237 	/* "eventX" and "counterX" sysfs to program and read counter values */
1238 	for (j = 0; j < pmc->block[blk_num].counters; ++j) {
1239 		attr = &pmc->block[blk_num].attr_counter[j];
1240 		attr->dev_attr.attr.mode = 0644;
1241 		attr->dev_attr.show = mlxbf_pmc_counter_show;
1242 		attr->dev_attr.store = mlxbf_pmc_counter_store;
1243 		attr->index = j;
1244 		attr->nr = blk_num;
1245 		attr->dev_attr.attr.name = devm_kasprintf(dev, GFP_KERNEL,
1246 							  "counter%d", j);
1247 		if (!attr->dev_attr.attr.name)
1248 			return -ENOMEM;
1249 		pmc->block[blk_num].block_attr[++i] = &attr->dev_attr.attr;
1250 		attr = NULL;
1251 
1252 		attr = &pmc->block[blk_num].attr_event[j];
1253 		attr->dev_attr.attr.mode = 0644;
1254 		attr->dev_attr.show = mlxbf_pmc_event_show;
1255 		attr->dev_attr.store = mlxbf_pmc_event_store;
1256 		attr->index = j;
1257 		attr->nr = blk_num;
1258 		attr->dev_attr.attr.name = devm_kasprintf(dev, GFP_KERNEL,
1259 							  "event%d", j);
1260 		if (!attr->dev_attr.attr.name)
1261 			return -ENOMEM;
1262 		pmc->block[blk_num].block_attr[++i] = &attr->dev_attr.attr;
1263 		attr = NULL;
1264 	}
1265 
1266 	return 0;
1267 }
1268 
1269 /* Populate attributes for blocks with registers to monitor performance */
1270 static int mlxbf_pmc_init_perftype_reg(struct device *dev, int blk_num)
1271 {
1272 	struct mlxbf_pmc_attribute *attr;
1273 	const struct mlxbf_pmc_events *events;
1274 	int i = 0, j = 0;
1275 
1276 	events = mlxbf_pmc_event_list(pmc->block_name[blk_num], &j);
1277 	if (!events)
1278 		return -EINVAL;
1279 
1280 	pmc->block[blk_num].attr_event = devm_kcalloc(
1281 		dev, j, sizeof(struct mlxbf_pmc_attribute), GFP_KERNEL);
1282 	if (!pmc->block[blk_num].attr_event)
1283 		return -ENOMEM;
1284 
1285 	while (j > 0) {
1286 		--j;
1287 		attr = &pmc->block[blk_num].attr_event[j];
1288 		attr->dev_attr.attr.mode = 0644;
1289 		attr->dev_attr.show = mlxbf_pmc_counter_show;
1290 		attr->dev_attr.store = mlxbf_pmc_counter_store;
1291 		attr->nr = blk_num;
1292 		attr->dev_attr.attr.name = devm_kasprintf(dev, GFP_KERNEL,
1293 							  events[j].evt_name);
1294 		if (!attr->dev_attr.attr.name)
1295 			return -ENOMEM;
1296 		pmc->block[blk_num].block_attr[i] = &attr->dev_attr.attr;
1297 		attr = NULL;
1298 		i++;
1299 	}
1300 
1301 	return 0;
1302 }
1303 
1304 /* Helper to create the bfperf sysfs sub-directories and files */
1305 static int mlxbf_pmc_create_groups(struct device *dev, int blk_num)
1306 {
1307 	int err;
1308 
1309 	/* Populate attributes based on counter type */
1310 	if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_COUNTER)
1311 		err = mlxbf_pmc_init_perftype_counter(dev, blk_num);
1312 	else if (pmc->block[blk_num].type == MLXBF_PMC_TYPE_REGISTER)
1313 		err = mlxbf_pmc_init_perftype_reg(dev, blk_num);
1314 	else
1315 		err = -EINVAL;
1316 
1317 	if (err)
1318 		return err;
1319 
1320 	/* Add a new attribute_group for the block */
1321 	pmc->block[blk_num].block_attr_grp.attrs = pmc->block[blk_num].block_attr;
1322 	pmc->block[blk_num].block_attr_grp.name = devm_kasprintf(
1323 		dev, GFP_KERNEL, pmc->block_name[blk_num]);
1324 	if (!pmc->block[blk_num].block_attr_grp.name)
1325 		return -ENOMEM;
1326 	pmc->groups[blk_num] = &pmc->block[blk_num].block_attr_grp;
1327 
1328 	return 0;
1329 }
1330 
1331 static bool mlxbf_pmc_guid_match(const guid_t *guid,
1332 				 const struct arm_smccc_res *res)
1333 {
1334 	guid_t id = GUID_INIT(res->a0, res->a1, res->a1 >> 16, res->a2,
1335 			      res->a2 >> 8, res->a2 >> 16, res->a2 >> 24,
1336 			      res->a3, res->a3 >> 8, res->a3 >> 16,
1337 			      res->a3 >> 24);
1338 
1339 	return guid_equal(guid, &id);
1340 }
1341 
1342 /* Helper to map the Performance Counters from the varios blocks */
1343 static int mlxbf_pmc_map_counters(struct device *dev)
1344 {
1345 	uint64_t info[MLXBF_PMC_INFO_SZ];
1346 	int i, tile_num, ret;
1347 
1348 	for (i = 0; i < pmc->total_blocks; ++i) {
1349 		if (strstr(pmc->block_name[i], "tile")) {
1350 			if (sscanf(pmc->block_name[i], "tile%d", &tile_num) != 1)
1351 				return -EINVAL;
1352 
1353 			if (tile_num >= pmc->tile_count)
1354 				continue;
1355 		}
1356 		ret = device_property_read_u64_array(dev, pmc->block_name[i],
1357 						     info, MLXBF_PMC_INFO_SZ);
1358 		if (ret)
1359 			return ret;
1360 
1361 		/*
1362 		 * Do not remap if the proper SMC calls are supported,
1363 		 * since the SMC calls expect physical addresses.
1364 		 */
1365 		if (pmc->svc_sreg_support)
1366 			pmc->block[i].mmio_base = (void __iomem *)info[0];
1367 		else
1368 			pmc->block[i].mmio_base =
1369 				devm_ioremap(dev, info[0], info[1]);
1370 
1371 		pmc->block[i].blk_size = info[1];
1372 		pmc->block[i].counters = info[2];
1373 		pmc->block[i].type = info[3];
1374 
1375 		if (!pmc->block[i].mmio_base)
1376 			return -ENOMEM;
1377 
1378 		ret = mlxbf_pmc_create_groups(dev, i);
1379 		if (ret)
1380 			return ret;
1381 	}
1382 
1383 	return 0;
1384 }
1385 
1386 static int mlxbf_pmc_probe(struct platform_device *pdev)
1387 {
1388 	struct acpi_device *acpi_dev = ACPI_COMPANION(&pdev->dev);
1389 	const char *hid = acpi_device_hid(acpi_dev);
1390 	struct device *dev = &pdev->dev;
1391 	struct arm_smccc_res res;
1392 	guid_t guid;
1393 	int ret;
1394 
1395 	/* Ensure we have the UUID we expect for this service. */
1396 	arm_smccc_smc(MLXBF_PMC_SIP_SVC_UID, 0, 0, 0, 0, 0, 0, 0, &res);
1397 	guid_parse(mlxbf_pmc_svc_uuid_str, &guid);
1398 	if (!mlxbf_pmc_guid_match(&guid, &res))
1399 		return -ENODEV;
1400 
1401 	pmc = devm_kzalloc(dev, sizeof(struct mlxbf_pmc_context), GFP_KERNEL);
1402 	if (!pmc)
1403 		return -ENOMEM;
1404 
1405 	/*
1406 	 * ACPI indicates whether we use SMCs to access registers or not.
1407 	 * If sreg_tbl_perf is not present, just assume we're not using SMCs.
1408 	 */
1409 	ret = device_property_read_u32(dev, "sec_reg_block",
1410 				       &pmc->sreg_tbl_perf);
1411 	if (ret) {
1412 		pmc->svc_sreg_support = false;
1413 	} else {
1414 		/*
1415 		 * Check service version to see if we actually do support the
1416 		 * needed SMCs. If we have the calls we need, mark support for
1417 		 * them in the pmc struct.
1418 		 */
1419 		arm_smccc_smc(MLXBF_PMC_SIP_SVC_VERSION, 0, 0, 0, 0, 0, 0, 0,
1420 			      &res);
1421 		if (res.a0 == MLXBF_PMC_SVC_REQ_MAJOR &&
1422 		    res.a1 >= MLXBF_PMC_SVC_MIN_MINOR)
1423 			pmc->svc_sreg_support = true;
1424 		else
1425 			return -EINVAL;
1426 	}
1427 
1428 	if (!strcmp(hid, "MLNXBFD0"))
1429 		pmc->event_set = MLXBF_PMC_EVENT_SET_BF1;
1430 	else if (!strcmp(hid, "MLNXBFD1"))
1431 		pmc->event_set = MLXBF_PMC_EVENT_SET_BF2;
1432 	else
1433 		return -ENODEV;
1434 
1435 	ret = device_property_read_u32(dev, "block_num", &pmc->total_blocks);
1436 	if (ret)
1437 		return ret;
1438 
1439 	ret = device_property_read_string_array(dev, "block_name",
1440 						pmc->block_name,
1441 						pmc->total_blocks);
1442 	if (ret != pmc->total_blocks)
1443 		return -EFAULT;
1444 
1445 	ret = device_property_read_u32(dev, "tile_num", &pmc->tile_count);
1446 	if (ret)
1447 		return ret;
1448 
1449 	pmc->pdev = pdev;
1450 
1451 	ret = mlxbf_pmc_map_counters(dev);
1452 	if (ret)
1453 		return ret;
1454 
1455 	pmc->hwmon_dev = devm_hwmon_device_register_with_groups(
1456 		dev, "bfperf", pmc, pmc->groups);
1457 	if (IS_ERR(pmc->hwmon_dev))
1458 		return PTR_ERR(pmc->hwmon_dev);
1459 	platform_set_drvdata(pdev, pmc);
1460 
1461 	return 0;
1462 }
1463 
1464 static const struct acpi_device_id mlxbf_pmc_acpi_ids[] = { { "MLNXBFD0", 0 },
1465 							    { "MLNXBFD1", 0 },
1466 							    {}, };
1467 
1468 MODULE_DEVICE_TABLE(acpi, mlxbf_pmc_acpi_ids);
1469 static struct platform_driver pmc_driver = {
1470 	.driver = { .name = "mlxbf-pmc",
1471 		    .acpi_match_table = ACPI_PTR(mlxbf_pmc_acpi_ids), },
1472 	.probe = mlxbf_pmc_probe,
1473 };
1474 
1475 module_platform_driver(pmc_driver);
1476 
1477 MODULE_AUTHOR("Shravan Kumar Ramani <sramani@mellanox.com>");
1478 MODULE_DESCRIPTION("Mellanox PMC driver");
1479 MODULE_LICENSE("Dual BSD/GPL");
1480