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>
32 #include <sys/vdev_impl.h>
43 #define PARITY_PQ (2U)
44 #define PARITY_PQR (3U)
51 * Parity generation methods indexes
53 enum raidz_math_gen_op {
60 * Data reconstruction methods indexes
73 extern const char *raidz_gen_name[RAIDZ_GEN_NUM];
74 extern const char *raidz_rec_name[RAIDZ_REC_NUM];
77 * Methods used to define raidz implementation
79 * @raidz_gen_f Parity generation function
80 * @par1 pointer to raidz_map
81 * @raidz_rec_f Data reconstruction function
82 * @par1 pointer to raidz_map
83 * @par2 array of reconstruction targets
84 * @will_work_f Function returns TRUE if impl. is supported on the system
85 * @init_impl_f Function is called once on init
86 * @fini_impl_f Function is called once on fini
88 typedef void (*raidz_gen_f)(void *);
89 typedef int (*raidz_rec_f)(void *, const int *);
90 typedef boolean_t (*will_work_f)(void);
91 typedef void (*init_impl_f)(void);
92 typedef void (*fini_impl_f)(void);
94 #define RAIDZ_IMPL_NAME_MAX (20)
96 typedef struct raidz_impl_ops {
99 raidz_gen_f gen[RAIDZ_GEN_NUM]; /* Parity generate functions */
100 raidz_rec_f rec[RAIDZ_REC_NUM]; /* Data reconstruction functions */
101 will_work_f is_supported; /* Support check function */
102 char name[RAIDZ_IMPL_NAME_MAX]; /* Name of the implementation */
105 typedef struct raidz_col {
106 uint64_t rc_devidx; /* child device index for I/O */
107 uint64_t rc_offset; /* device offset */
108 uint64_t rc_size; /* I/O size */
109 abd_t rc_abdstruct; /* rc_abd probably points here */
110 abd_t *rc_abd; /* I/O data */
111 void *rc_orig_data; /* pre-reconstruction */
112 abd_t *rc_gdata; /* used to store the "good" version */
113 int rc_error; /* I/O error for this device */
114 uint8_t rc_tried; /* Did we attempt this I/O column? */
115 uint8_t rc_skipped; /* Did we skip this I/O column? */
116 uint8_t rc_need_orig_restore; /* need to restore from orig_data? */
117 uint8_t rc_repair; /* Write good data to this column */
120 typedef struct raidz_row {
121 uint64_t rr_cols; /* Regular column count */
122 uint64_t rr_scols; /* Count including skipped columns */
123 uint64_t rr_bigcols; /* Remainder data column count */
124 uint64_t rr_missingdata; /* Count of missing data devices */
125 uint64_t rr_missingparity; /* Count of missing parity devices */
126 uint64_t rr_firstdatacol; /* First data column/parity count */
127 abd_t *rr_abd_copy; /* rm_asize-buffer of copied data */
128 abd_t *rr_abd_empty; /* dRAID empty sector buffer */
129 int rr_nempty; /* empty sectors included in parity */
130 int rr_code; /* reconstruction code (unused) */
132 uint64_t rr_offset; /* Logical offset for *_io_verify() */
133 uint64_t rr_size; /* Physical size for *_io_verify() */
135 raidz_col_t rr_col[0]; /* Flexible array of I/O columns */
138 typedef struct raidz_map {
139 uintptr_t rm_reports; /* # of referencing checksum reports */
140 boolean_t rm_freed; /* map no longer has referencing ZIO */
141 boolean_t rm_ecksuminjected; /* checksum error was injected */
142 int rm_nrows; /* Regular row count */
143 int rm_nskip; /* RAIDZ sectors skipped for padding */
144 int rm_skipstart; /* Column index of padding start */
145 const raidz_impl_ops_t *rm_ops; /* RAIDZ math operations */
146 raidz_row_t *rm_row[0]; /* flexible array of rows */
150 #define RAIDZ_ORIGINAL_IMPL (INT_MAX)
152 extern const raidz_impl_ops_t vdev_raidz_scalar_impl;
153 extern boolean_t raidz_will_scalar_work(void);
155 #if defined(__x86_64) && defined(HAVE_SSE2) /* only x86_64 for now */
156 extern const raidz_impl_ops_t vdev_raidz_sse2_impl;
158 #if defined(__x86_64) && defined(HAVE_SSSE3) /* only x86_64 for now */
159 extern const raidz_impl_ops_t vdev_raidz_ssse3_impl;
161 #if defined(__x86_64) && defined(HAVE_AVX2) /* only x86_64 for now */
162 extern const raidz_impl_ops_t vdev_raidz_avx2_impl;
164 #if defined(__x86_64) && defined(HAVE_AVX512F) /* only x86_64 for now */
165 extern const raidz_impl_ops_t vdev_raidz_avx512f_impl;
167 #if defined(__x86_64) && defined(HAVE_AVX512BW) /* only x86_64 for now */
168 extern const raidz_impl_ops_t vdev_raidz_avx512bw_impl;
170 #if defined(__aarch64__)
171 extern const raidz_impl_ops_t vdev_raidz_aarch64_neon_impl;
172 extern const raidz_impl_ops_t vdev_raidz_aarch64_neonx2_impl;
174 #if defined(__powerpc__)
175 extern const raidz_impl_ops_t vdev_raidz_powerpc_altivec_impl;
179 * Commonly used raidz_map helpers
181 * raidz_parity Returns parity of the RAIDZ block
182 * raidz_ncols Returns number of columns the block spans
183 * Note, all rows have the same number of columns.
184 * raidz_nbigcols Returns number of big columns
185 * raidz_col_p Returns pointer to a column
186 * raidz_col_size Returns size of a column
187 * raidz_big_size Returns size of big columns
188 * raidz_short_size Returns size of short columns
190 #define raidz_parity(rm) ((rm)->rm_row[0]->rr_firstdatacol)
191 #define raidz_ncols(rm) ((rm)->rm_row[0]->rr_cols)
192 #define raidz_nbigcols(rm) ((rm)->rm_bigcols)
193 #define raidz_col_p(rm, c) ((rm)->rm_col + (c))
194 #define raidz_col_size(rm, c) ((rm)->rm_col[c].rc_size)
195 #define raidz_big_size(rm) (raidz_col_size(rm, CODE_P))
196 #define raidz_short_size(rm) (raidz_col_size(rm, raidz_ncols(rm)-1))
199 * Macro defines an RAIDZ parity generation method
201 * @code parity the function produce
202 * @impl name of the implementation
204 #define _RAIDZ_GEN_WRAP(code, impl) \
206 impl ## _gen_ ## code(void *rrp) \
208 raidz_row_t *rr = (raidz_row_t *)rrp; \
209 raidz_generate_## code ## _impl(rr); \
213 * Macro defines an RAIDZ data reconstruction method
215 * @code parity the function produce
216 * @impl name of the implementation
218 #define _RAIDZ_REC_WRAP(code, impl) \
220 impl ## _rec_ ## code(void *rrp, const int *tgtidx) \
222 raidz_row_t *rr = (raidz_row_t *)rrp; \
223 return (raidz_reconstruct_## code ## _impl(rr, tgtidx)); \
227 * Define all gen methods for an implementation
229 * @impl name of the implementation
231 #define DEFINE_GEN_METHODS(impl) \
232 _RAIDZ_GEN_WRAP(p, impl); \
233 _RAIDZ_GEN_WRAP(pq, impl); \
234 _RAIDZ_GEN_WRAP(pqr, impl)
237 * Define all rec functions for an implementation
239 * @impl name of the implementation
241 #define DEFINE_REC_METHODS(impl) \
242 _RAIDZ_REC_WRAP(p, impl); \
243 _RAIDZ_REC_WRAP(q, impl); \
244 _RAIDZ_REC_WRAP(r, impl); \
245 _RAIDZ_REC_WRAP(pq, impl); \
246 _RAIDZ_REC_WRAP(pr, impl); \
247 _RAIDZ_REC_WRAP(qr, impl); \
248 _RAIDZ_REC_WRAP(pqr, impl)
250 #define RAIDZ_GEN_METHODS(impl) \
252 [RAIDZ_GEN_P] = & impl ## _gen_p, \
253 [RAIDZ_GEN_PQ] = & impl ## _gen_pq, \
254 [RAIDZ_GEN_PQR] = & impl ## _gen_pqr \
257 #define RAIDZ_REC_METHODS(impl) \
259 [RAIDZ_REC_P] = & impl ## _rec_p, \
260 [RAIDZ_REC_Q] = & impl ## _rec_q, \
261 [RAIDZ_REC_R] = & impl ## _rec_r, \
262 [RAIDZ_REC_PQ] = & impl ## _rec_pq, \
263 [RAIDZ_REC_PR] = & impl ## _rec_pr, \
264 [RAIDZ_REC_QR] = & impl ## _rec_qr, \
265 [RAIDZ_REC_PQR] = & impl ## _rec_pqr \
269 typedef struct raidz_impl_kstat {
270 uint64_t gen[RAIDZ_GEN_NUM]; /* gen method speed B/s */
271 uint64_t rec[RAIDZ_REC_NUM]; /* rec method speed B/s */
272 } raidz_impl_kstat_t;
275 * Enumerate various multiplication constants
276 * used in reconstruction methods
278 typedef enum raidz_mul_info {
294 /* Reconstruct PQR */
306 * Powers of 2 in the Galois field.
308 extern const uint8_t vdev_raidz_pow2[256] __attribute__((aligned(256)));
309 /* Logs of 2 in the Galois field defined above. */
310 extern const uint8_t vdev_raidz_log2[256] __attribute__((aligned(256)));
313 * Multiply a given number by 2 raised to the given power.
315 static inline uint8_t
316 vdev_raidz_exp2(const uint8_t a, const unsigned exp)
321 return (vdev_raidz_pow2[(exp + (unsigned)vdev_raidz_log2[a]) % 255]);
325 * Galois Field operations.
327 * gf_exp2 - computes 2 raised to the given power
328 * gf_exp2 - computes 4 raised to the given power
329 * gf_mul - multiplication
331 * gf_inv - multiplicative inverse
333 typedef unsigned gf_t;
334 typedef unsigned gf_log_t;
337 gf_mul(const gf_t a, const gf_t b)
341 if (a == 0 || b == 0)
344 logsum = (gf_log_t)vdev_raidz_log2[a] + (gf_log_t)vdev_raidz_log2[b];
346 return ((gf_t)vdev_raidz_pow2[logsum % 255]);
350 gf_div(const gf_t a, const gf_t b)
358 logsum = (gf_log_t)255 + (gf_log_t)vdev_raidz_log2[a] -
359 (gf_log_t)vdev_raidz_log2[b];
361 return ((gf_t)vdev_raidz_pow2[logsum % 255]);
371 logsum = (gf_log_t)255 - (gf_log_t)vdev_raidz_log2[a];
373 return ((gf_t)vdev_raidz_pow2[logsum]);
377 gf_exp2(gf_log_t exp)
379 return (vdev_raidz_pow2[exp % 255]);
383 gf_exp4(gf_log_t exp)
385 ASSERT3U(exp, <=, 255);
386 return ((gf_t)vdev_raidz_pow2[(2 * exp) % 255]);
393 #endif /* _VDEV_RAIDZ_H */