4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
28 #include <sys/types.h>
29 #include <sys/debug.h>
30 #include <sys/kstat.h>
41 #define PARITY_PQ (2U)
42 #define PARITY_PQR (3U)
49 * Parity generation methods indexes
51 enum raidz_math_gen_op {
58 * Data reconstruction methods indexes
71 extern const char *raidz_gen_name[RAIDZ_GEN_NUM];
72 extern const char *raidz_rec_name[RAIDZ_REC_NUM];
75 * Methods used to define raidz implementation
77 * @raidz_gen_f Parity generation function
78 * @par1 pointer to raidz_map
79 * @raidz_rec_f Data reconstruction function
80 * @par1 pointer to raidz_map
81 * @par2 array of reconstruction targets
82 * @will_work_f Function returns TRUE if impl. is supported on the system
83 * @init_impl_f Function is called once on init
84 * @fini_impl_f Function is called once on fini
86 typedef void (*raidz_gen_f)(void *);
87 typedef int (*raidz_rec_f)(void *, const int *);
88 typedef boolean_t (*will_work_f)(void);
89 typedef void (*init_impl_f)(void);
90 typedef void (*fini_impl_f)(void);
92 #define RAIDZ_IMPL_NAME_MAX (16)
94 typedef struct raidz_impl_ops {
97 raidz_gen_f gen[RAIDZ_GEN_NUM]; /* Parity generate functions */
98 raidz_rec_f rec[RAIDZ_REC_NUM]; /* Data reconstruction functions */
99 will_work_f is_supported; /* Support check function */
100 char name[RAIDZ_IMPL_NAME_MAX]; /* Name of the implementation */
103 typedef struct raidz_col {
104 size_t rc_devidx; /* child device index for I/O */
105 size_t rc_offset; /* device offset */
106 size_t rc_size; /* I/O size */
107 void *rc_data; /* I/O data */
108 void *rc_gdata; /* used to store the "good" version */
109 int rc_error; /* I/O error for this device */
110 unsigned int rc_tried; /* Did we attempt this I/O column? */
111 unsigned int rc_skipped; /* Did we skip this I/O column? */
114 typedef struct raidz_map {
115 size_t rm_cols; /* Regular column count */
116 size_t rm_scols; /* Count including skipped columns */
117 size_t rm_bigcols; /* Number of oversized columns */
118 size_t rm_asize; /* Actual total I/O size */
119 size_t rm_missingdata; /* Count of missing data devices */
120 size_t rm_missingparity; /* Count of missing parity devices */
121 size_t rm_firstdatacol; /* First data column/parity count */
122 size_t rm_nskip; /* Skipped sectors for padding */
123 size_t rm_skipstart; /* Column index of padding start */
124 void *rm_datacopy; /* rm_asize-buffer of copied data */
125 size_t rm_reports; /* # of referencing checksum reports */
126 unsigned int rm_freed; /* map no longer has referencing ZIO */
127 unsigned int rm_ecksuminjected; /* checksum error was injected */
128 raidz_impl_ops_t *rm_ops; /* RAIDZ math operations */
129 raidz_col_t rm_col[1]; /* Flexible array of I/O columns */
132 #define RAIDZ_ORIGINAL_IMPL (INT_MAX)
134 extern const raidz_impl_ops_t vdev_raidz_scalar_impl;
135 #if defined(__x86_64) && defined(HAVE_SSE2) /* only x86_64 for now */
136 extern const raidz_impl_ops_t vdev_raidz_sse2_impl;
138 #if defined(__x86_64) && defined(HAVE_SSSE3) /* only x86_64 for now */
139 extern const raidz_impl_ops_t vdev_raidz_ssse3_impl;
141 #if defined(__x86_64) && defined(HAVE_AVX2) /* only x86_64 for now */
142 extern const raidz_impl_ops_t vdev_raidz_avx2_impl;
146 * Commonly used raidz_map helpers
148 * raidz_parity Returns parity of the RAIDZ block
149 * raidz_ncols Returns number of columns the block spans
150 * raidz_nbigcols Returns number of big columns columns
151 * raidz_col_p Returns pointer to a column
152 * raidz_col_size Returns size of a column
153 * raidz_big_size Returns size of big columns
154 * raidz_short_size Returns size of short columns
156 #define raidz_parity(rm) ((rm)->rm_firstdatacol)
157 #define raidz_ncols(rm) ((rm)->rm_cols)
158 #define raidz_nbigcols(rm) ((rm)->rm_bigcols)
159 #define raidz_col_p(rm, c) ((rm)->rm_col + (c))
160 #define raidz_col_size(rm, c) ((rm)->rm_col[c].rc_size)
161 #define raidz_big_size(rm) (raidz_col_size(rm, CODE_P))
162 #define raidz_short_size(rm) (raidz_col_size(rm, raidz_ncols(rm)-1))
165 * Macro defines an RAIDZ parity generation method
167 * @code parity the function produce
168 * @impl name of the implementation
170 #define _RAIDZ_GEN_WRAP(code, impl) \
172 impl ## _gen_ ## code(void *rmp) \
174 raidz_map_t *rm = (raidz_map_t *) rmp; \
175 raidz_generate_## code ## _impl(rm); \
179 * Macro defines an RAIDZ data reconstruction method
181 * @code parity the function produce
182 * @impl name of the implementation
184 #define _RAIDZ_REC_WRAP(code, impl) \
186 impl ## _rec_ ## code(void *rmp, const int *tgtidx) \
188 raidz_map_t *rm = (raidz_map_t *) rmp; \
189 return (raidz_reconstruct_## code ## _impl(rm, tgtidx)); \
193 * Define all gen methods for an implementation
195 * @impl name of the implementation
197 #define DEFINE_GEN_METHODS(impl) \
198 _RAIDZ_GEN_WRAP(p, impl); \
199 _RAIDZ_GEN_WRAP(pq, impl); \
200 _RAIDZ_GEN_WRAP(pqr, impl)
203 * Define all rec functions for an implementation
205 * @impl name of the implementation
207 #define DEFINE_REC_METHODS(impl) \
208 _RAIDZ_REC_WRAP(p, impl); \
209 _RAIDZ_REC_WRAP(q, impl); \
210 _RAIDZ_REC_WRAP(r, impl); \
211 _RAIDZ_REC_WRAP(pq, impl); \
212 _RAIDZ_REC_WRAP(pr, impl); \
213 _RAIDZ_REC_WRAP(qr, impl); \
214 _RAIDZ_REC_WRAP(pqr, impl)
216 #define RAIDZ_GEN_METHODS(impl) \
218 [RAIDZ_GEN_P] = & impl ## _gen_p, \
219 [RAIDZ_GEN_PQ] = & impl ## _gen_pq, \
220 [RAIDZ_GEN_PQR] = & impl ## _gen_pqr \
223 #define RAIDZ_REC_METHODS(impl) \
225 [RAIDZ_REC_P] = & impl ## _rec_p, \
226 [RAIDZ_REC_Q] = & impl ## _rec_q, \
227 [RAIDZ_REC_R] = & impl ## _rec_r, \
228 [RAIDZ_REC_PQ] = & impl ## _rec_pq, \
229 [RAIDZ_REC_PR] = & impl ## _rec_pr, \
230 [RAIDZ_REC_QR] = & impl ## _rec_qr, \
231 [RAIDZ_REC_PQR] = & impl ## _rec_pqr \
235 typedef struct raidz_impl_kstat {
236 kstat_named_t gen[RAIDZ_GEN_NUM]; /* gen method speed kiB/s */
237 kstat_named_t rec[RAIDZ_REC_NUM]; /* rec method speed kiB/s */
238 } raidz_impl_kstat_t;
241 * Enumerate various multiplication constants
242 * used in reconstruction methods
244 typedef enum raidz_mul_info {
260 /* Reconstruct PQR */
272 * Powers of 2 in the Galois field.
274 extern const uint8_t vdev_raidz_pow2[256] __attribute__((aligned(256)));
275 /* Logs of 2 in the Galois field defined above. */
276 extern const uint8_t vdev_raidz_log2[256] __attribute__((aligned(256)));
279 * Multiply a given number by 2 raised to the given power.
281 static inline uint8_t
282 vdev_raidz_exp2(const uint8_t a, const unsigned exp)
287 return (vdev_raidz_pow2[(exp + (unsigned) vdev_raidz_log2[a]) % 255]);
291 * Galois Field operations.
293 * gf_exp2 - computes 2 raised to the given power
294 * gf_exp2 - computes 4 raised to the given power
295 * gf_mul - multiplication
297 * gf_inv - multiplicative inverse
299 typedef unsigned gf_t;
300 typedef unsigned gf_log_t;
303 gf_mul(const gf_t a, const gf_t b)
307 if (a == 0 || b == 0)
310 logsum = (gf_log_t) vdev_raidz_log2[a] + (gf_log_t) vdev_raidz_log2[b];
312 return ((gf_t) vdev_raidz_pow2[logsum % 255]);
316 gf_div(const gf_t a, const gf_t b)
324 logsum = (gf_log_t) 255 + (gf_log_t) vdev_raidz_log2[a] -
325 (gf_log_t) vdev_raidz_log2[b];
327 return ((gf_t) vdev_raidz_pow2[logsum % 255]);
337 logsum = (gf_log_t) 255 - (gf_log_t) vdev_raidz_log2[a];
339 return ((gf_t) vdev_raidz_pow2[logsum]);
343 gf_exp2(gf_log_t exp)
345 return (vdev_raidz_pow2[exp % 255]);
349 gf_exp4(gf_log_t exp)
351 ASSERT3U(exp, <=, 255);
352 return ((gf_t) vdev_raidz_pow2[(2 * exp) % 255]);
359 #endif /* _VDEV_RAIDZ_H */