1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2018 Mellanox Technologies. All rights reserved */
3
4 #include <linux/errno.h>
5 #include <linux/gfp.h>
6 #include <linux/kernel.h>
7 #include <linux/refcount.h>
8 #include <linux/mutex.h>
9
10 #include "spectrum.h"
11 #include "spectrum_acl_tcam.h"
12
13 struct mlxsw_sp_acl_bf {
14 struct mutex lock; /* Protects Bloom Filter updates. */
15 unsigned int bank_size;
16 refcount_t refcnt[];
17 };
18
19 /* Bloom filter uses a crc-16 hash over chunks of data which contain 4 key
20 * blocks, eRP ID and region ID. In Spectrum-2, region key is combined of up to
21 * 12 key blocks, so there can be up to 3 chunks in the Bloom filter key,
22 * depending on the actual number of key blocks used in the region.
23 * The layout of the Bloom filter key is as follows:
24 *
25 * +-------------------------+------------------------+------------------------+
26 * | Chunk 2 Key blocks 11-8 | Chunk 1 Key blocks 7-4 | Chunk 0 Key blocks 3-0 |
27 * +-------------------------+------------------------+------------------------+
28 */
29 #define MLXSW_BLOOM_KEY_CHUNKS 3
30 #define MLXSW_BLOOM_KEY_LEN 69
31
32 /* Each chunk size is 23 bytes. 18 bytes of it contain 4 key blocks, each is
33 * 36 bits, 2 bytes which hold eRP ID and region ID, and 3 bytes of zero
34 * padding.
35 * The layout of each chunk is as follows:
36 *
37 * +---------+----------------------+-----------------------------------+
38 * | 3 bytes | 2 bytes | 18 bytes |
39 * +---------+-----------+----------+-----------------------------------+
40 * | 183:158 | 157:148 | 147:144 | 143:0 |
41 * +---------+-----------+----------+-----------------------------------+
42 * | 0 | region ID | eRP ID | 4 Key blocks (18 Bytes) |
43 * +---------+-----------+----------+-----------------------------------+
44 */
45 #define MLXSW_BLOOM_CHUNK_PAD_BYTES 3
46 #define MLXSW_BLOOM_CHUNK_KEY_BYTES 18
47 #define MLXSW_BLOOM_KEY_CHUNK_BYTES 23
48
49 /* The offset of the key block within a chunk is 5 bytes as it comes after
50 * 3 bytes of zero padding and 16 bits of region ID and eRP ID.
51 */
52 #define MLXSW_BLOOM_CHUNK_KEY_OFFSET 5
53
54 /* Each chunk contains 4 key blocks. Chunk 2 uses key blocks 11-8,
55 * and we need to populate it with 4 key blocks copied from the entry encoded
56 * key. Since the encoded key contains a padding, key block 11 starts at offset
57 * 2. block 7 that is used in chunk 1 starts at offset 20 as 4 key blocks take
58 * 18 bytes.
59 * This array defines key offsets for easy access when copying key blocks from
60 * entry key to Bloom filter chunk.
61 */
62 static const u8 chunk_key_offsets[MLXSW_BLOOM_KEY_CHUNKS] = {2, 20, 38};
63
64 /* This table is just the CRC of each possible byte. It is
65 * computed, Msbit first, for the Bloom filter polynomial
66 * which is 0x8529 (1 + x^3 + x^5 + x^8 + x^10 + x^15 and
67 * the implicit x^16).
68 */
69 static const u16 mlxsw_sp_acl_bf_crc_tab[256] = {
70 0x0000, 0x8529, 0x8f7b, 0x0a52, 0x9bdf, 0x1ef6, 0x14a4, 0x918d,
71 0xb297, 0x37be, 0x3dec, 0xb8c5, 0x2948, 0xac61, 0xa633, 0x231a,
72 0xe007, 0x652e, 0x6f7c, 0xea55, 0x7bd8, 0xfef1, 0xf4a3, 0x718a,
73 0x5290, 0xd7b9, 0xddeb, 0x58c2, 0xc94f, 0x4c66, 0x4634, 0xc31d,
74 0x4527, 0xc00e, 0xca5c, 0x4f75, 0xdef8, 0x5bd1, 0x5183, 0xd4aa,
75 0xf7b0, 0x7299, 0x78cb, 0xfde2, 0x6c6f, 0xe946, 0xe314, 0x663d,
76 0xa520, 0x2009, 0x2a5b, 0xaf72, 0x3eff, 0xbbd6, 0xb184, 0x34ad,
77 0x17b7, 0x929e, 0x98cc, 0x1de5, 0x8c68, 0x0941, 0x0313, 0x863a,
78 0x8a4e, 0x0f67, 0x0535, 0x801c, 0x1191, 0x94b8, 0x9eea, 0x1bc3,
79 0x38d9, 0xbdf0, 0xb7a2, 0x328b, 0xa306, 0x262f, 0x2c7d, 0xa954,
80 0x6a49, 0xef60, 0xe532, 0x601b, 0xf196, 0x74bf, 0x7eed, 0xfbc4,
81 0xd8de, 0x5df7, 0x57a5, 0xd28c, 0x4301, 0xc628, 0xcc7a, 0x4953,
82 0xcf69, 0x4a40, 0x4012, 0xc53b, 0x54b6, 0xd19f, 0xdbcd, 0x5ee4,
83 0x7dfe, 0xf8d7, 0xf285, 0x77ac, 0xe621, 0x6308, 0x695a, 0xec73,
84 0x2f6e, 0xaa47, 0xa015, 0x253c, 0xb4b1, 0x3198, 0x3bca, 0xbee3,
85 0x9df9, 0x18d0, 0x1282, 0x97ab, 0x0626, 0x830f, 0x895d, 0x0c74,
86 0x91b5, 0x149c, 0x1ece, 0x9be7, 0x0a6a, 0x8f43, 0x8511, 0x0038,
87 0x2322, 0xa60b, 0xac59, 0x2970, 0xb8fd, 0x3dd4, 0x3786, 0xb2af,
88 0x71b2, 0xf49b, 0xfec9, 0x7be0, 0xea6d, 0x6f44, 0x6516, 0xe03f,
89 0xc325, 0x460c, 0x4c5e, 0xc977, 0x58fa, 0xddd3, 0xd781, 0x52a8,
90 0xd492, 0x51bb, 0x5be9, 0xdec0, 0x4f4d, 0xca64, 0xc036, 0x451f,
91 0x6605, 0xe32c, 0xe97e, 0x6c57, 0xfdda, 0x78f3, 0x72a1, 0xf788,
92 0x3495, 0xb1bc, 0xbbee, 0x3ec7, 0xaf4a, 0x2a63, 0x2031, 0xa518,
93 0x8602, 0x032b, 0x0979, 0x8c50, 0x1ddd, 0x98f4, 0x92a6, 0x178f,
94 0x1bfb, 0x9ed2, 0x9480, 0x11a9, 0x8024, 0x050d, 0x0f5f, 0x8a76,
95 0xa96c, 0x2c45, 0x2617, 0xa33e, 0x32b3, 0xb79a, 0xbdc8, 0x38e1,
96 0xfbfc, 0x7ed5, 0x7487, 0xf1ae, 0x6023, 0xe50a, 0xef58, 0x6a71,
97 0x496b, 0xcc42, 0xc610, 0x4339, 0xd2b4, 0x579d, 0x5dcf, 0xd8e6,
98 0x5edc, 0xdbf5, 0xd1a7, 0x548e, 0xc503, 0x402a, 0x4a78, 0xcf51,
99 0xec4b, 0x6962, 0x6330, 0xe619, 0x7794, 0xf2bd, 0xf8ef, 0x7dc6,
100 0xbedb, 0x3bf2, 0x31a0, 0xb489, 0x2504, 0xa02d, 0xaa7f, 0x2f56,
101 0x0c4c, 0x8965, 0x8337, 0x061e, 0x9793, 0x12ba, 0x18e8, 0x9dc1,
102 };
103
mlxsw_sp_acl_bf_crc_byte(u16 crc,u8 c)104 static u16 mlxsw_sp_acl_bf_crc_byte(u16 crc, u8 c)
105 {
106 return (crc << 8) ^ mlxsw_sp_acl_bf_crc_tab[(crc >> 8) ^ c];
107 }
108
mlxsw_sp_acl_bf_crc(const u8 * buffer,size_t len)109 static u16 mlxsw_sp_acl_bf_crc(const u8 *buffer, size_t len)
110 {
111 u16 crc = 0;
112
113 while (len--)
114 crc = mlxsw_sp_acl_bf_crc_byte(crc, *buffer++);
115 return crc;
116 }
117
118 static void
mlxsw_sp_acl_bf_key_encode(struct mlxsw_sp_acl_atcam_region * aregion,struct mlxsw_sp_acl_atcam_entry * aentry,char * output,u8 * len)119 mlxsw_sp_acl_bf_key_encode(struct mlxsw_sp_acl_atcam_region *aregion,
120 struct mlxsw_sp_acl_atcam_entry *aentry,
121 char *output, u8 *len)
122 {
123 struct mlxsw_afk_key_info *key_info = aregion->region->key_info;
124 u8 chunk_index, chunk_count, block_count;
125 char *chunk = output;
126 __be16 erp_region_id;
127
128 block_count = mlxsw_afk_key_info_blocks_count_get(key_info);
129 chunk_count = 1 + ((block_count - 1) >> 2);
130 erp_region_id = cpu_to_be16(aentry->ht_key.erp_id |
131 (aregion->region->id << 4));
132 for (chunk_index = MLXSW_BLOOM_KEY_CHUNKS - chunk_count;
133 chunk_index < MLXSW_BLOOM_KEY_CHUNKS; chunk_index++) {
134 memset(chunk, 0, MLXSW_BLOOM_CHUNK_PAD_BYTES);
135 memcpy(chunk + MLXSW_BLOOM_CHUNK_PAD_BYTES, &erp_region_id,
136 sizeof(erp_region_id));
137 memcpy(chunk + MLXSW_BLOOM_CHUNK_KEY_OFFSET,
138 &aentry->enc_key[chunk_key_offsets[chunk_index]],
139 MLXSW_BLOOM_CHUNK_KEY_BYTES);
140 chunk += MLXSW_BLOOM_KEY_CHUNK_BYTES;
141 }
142 *len = chunk_count * MLXSW_BLOOM_KEY_CHUNK_BYTES;
143 }
144
145 static unsigned int
mlxsw_sp_acl_bf_rule_count_index_get(struct mlxsw_sp_acl_bf * bf,unsigned int erp_bank,unsigned int bf_index)146 mlxsw_sp_acl_bf_rule_count_index_get(struct mlxsw_sp_acl_bf *bf,
147 unsigned int erp_bank,
148 unsigned int bf_index)
149 {
150 return erp_bank * bf->bank_size + bf_index;
151 }
152
153 static unsigned int
mlxsw_sp_acl_bf_index_get(struct mlxsw_sp_acl_bf * bf,struct mlxsw_sp_acl_atcam_region * aregion,struct mlxsw_sp_acl_atcam_entry * aentry)154 mlxsw_sp_acl_bf_index_get(struct mlxsw_sp_acl_bf *bf,
155 struct mlxsw_sp_acl_atcam_region *aregion,
156 struct mlxsw_sp_acl_atcam_entry *aentry)
157 {
158 char bf_key[MLXSW_BLOOM_KEY_LEN];
159 u8 bf_size;
160
161 mlxsw_sp_acl_bf_key_encode(aregion, aentry, bf_key, &bf_size);
162 return mlxsw_sp_acl_bf_crc(bf_key, bf_size);
163 }
164
165 int
mlxsw_sp_acl_bf_entry_add(struct mlxsw_sp * mlxsw_sp,struct mlxsw_sp_acl_bf * bf,struct mlxsw_sp_acl_atcam_region * aregion,unsigned int erp_bank,struct mlxsw_sp_acl_atcam_entry * aentry)166 mlxsw_sp_acl_bf_entry_add(struct mlxsw_sp *mlxsw_sp,
167 struct mlxsw_sp_acl_bf *bf,
168 struct mlxsw_sp_acl_atcam_region *aregion,
169 unsigned int erp_bank,
170 struct mlxsw_sp_acl_atcam_entry *aentry)
171 {
172 unsigned int rule_index;
173 char *peabfe_pl;
174 u16 bf_index;
175 int err;
176
177 mutex_lock(&bf->lock);
178
179 bf_index = mlxsw_sp_acl_bf_index_get(bf, aregion, aentry);
180 rule_index = mlxsw_sp_acl_bf_rule_count_index_get(bf, erp_bank,
181 bf_index);
182
183 if (refcount_inc_not_zero(&bf->refcnt[rule_index])) {
184 err = 0;
185 goto unlock;
186 }
187
188 peabfe_pl = kmalloc(MLXSW_REG_PEABFE_LEN, GFP_KERNEL);
189 if (!peabfe_pl) {
190 err = -ENOMEM;
191 goto unlock;
192 }
193
194 mlxsw_reg_peabfe_pack(peabfe_pl);
195 mlxsw_reg_peabfe_rec_pack(peabfe_pl, 0, 1, erp_bank, bf_index);
196 err = mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(peabfe), peabfe_pl);
197 kfree(peabfe_pl);
198 if (err)
199 goto unlock;
200
201 refcount_set(&bf->refcnt[rule_index], 1);
202 err = 0;
203
204 unlock:
205 mutex_unlock(&bf->lock);
206 return err;
207 }
208
209 void
mlxsw_sp_acl_bf_entry_del(struct mlxsw_sp * mlxsw_sp,struct mlxsw_sp_acl_bf * bf,struct mlxsw_sp_acl_atcam_region * aregion,unsigned int erp_bank,struct mlxsw_sp_acl_atcam_entry * aentry)210 mlxsw_sp_acl_bf_entry_del(struct mlxsw_sp *mlxsw_sp,
211 struct mlxsw_sp_acl_bf *bf,
212 struct mlxsw_sp_acl_atcam_region *aregion,
213 unsigned int erp_bank,
214 struct mlxsw_sp_acl_atcam_entry *aentry)
215 {
216 unsigned int rule_index;
217 char *peabfe_pl;
218 u16 bf_index;
219
220 mutex_lock(&bf->lock);
221
222 bf_index = mlxsw_sp_acl_bf_index_get(bf, aregion, aentry);
223 rule_index = mlxsw_sp_acl_bf_rule_count_index_get(bf, erp_bank,
224 bf_index);
225
226 if (refcount_dec_and_test(&bf->refcnt[rule_index])) {
227 peabfe_pl = kmalloc(MLXSW_REG_PEABFE_LEN, GFP_KERNEL);
228 if (!peabfe_pl)
229 goto unlock;
230
231 mlxsw_reg_peabfe_pack(peabfe_pl);
232 mlxsw_reg_peabfe_rec_pack(peabfe_pl, 0, 0, erp_bank, bf_index);
233 mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(peabfe), peabfe_pl);
234 kfree(peabfe_pl);
235 }
236
237 unlock:
238 mutex_unlock(&bf->lock);
239 }
240
241 struct mlxsw_sp_acl_bf *
mlxsw_sp_acl_bf_init(struct mlxsw_sp * mlxsw_sp,unsigned int num_erp_banks)242 mlxsw_sp_acl_bf_init(struct mlxsw_sp *mlxsw_sp, unsigned int num_erp_banks)
243 {
244 struct mlxsw_sp_acl_bf *bf;
245 unsigned int bf_bank_size;
246
247 if (!MLXSW_CORE_RES_VALID(mlxsw_sp->core, ACL_MAX_BF_LOG))
248 return ERR_PTR(-EIO);
249
250 /* Bloom filter size per erp_table_bank
251 * is 2^ACL_MAX_BF_LOG
252 */
253 bf_bank_size = 1 << MLXSW_CORE_RES_GET(mlxsw_sp->core, ACL_MAX_BF_LOG);
254 bf = kzalloc(struct_size(bf, refcnt, size_mul(bf_bank_size, num_erp_banks)),
255 GFP_KERNEL);
256 if (!bf)
257 return ERR_PTR(-ENOMEM);
258
259 bf->bank_size = bf_bank_size;
260 mutex_init(&bf->lock);
261
262 return bf;
263 }
264
mlxsw_sp_acl_bf_fini(struct mlxsw_sp_acl_bf * bf)265 void mlxsw_sp_acl_bf_fini(struct mlxsw_sp_acl_bf *bf)
266 {
267 mutex_destroy(&bf->lock);
268 kfree(bf);
269 }
270