]> git.proxmox.com Git - mirror_zfs.git/blame - include/sys/vdev_raidz_impl.h
RAIDZ parity kstat rework
[mirror_zfs.git] / include / sys / vdev_raidz_impl.h
CommitLineData
ab9f4b0b
GN
1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (C) 2016 Gvozden Nešković. All rights reserved.
23 */
24
25#ifndef _VDEV_RAIDZ_H
26#define _VDEV_RAIDZ_H
27
28#include <sys/types.h>
29#include <sys/debug.h>
30#include <sys/kstat.h>
31
32#ifdef __cplusplus
33extern "C" {
34#endif
35
36#define CODE_P (0U)
37#define CODE_Q (1U)
38#define CODE_R (2U)
39
40#define PARITY_P (1U)
41#define PARITY_PQ (2U)
42#define PARITY_PQR (3U)
43
44#define TARGET_X (0U)
45#define TARGET_Y (1U)
46#define TARGET_Z (2U)
47
48/*
49 * Parity generation methods indexes
50 */
51enum raidz_math_gen_op {
52 RAIDZ_GEN_P = 0,
53 RAIDZ_GEN_PQ,
54 RAIDZ_GEN_PQR,
55 RAIDZ_GEN_NUM = 3
56};
57/*
58 * Data reconstruction methods indexes
59 */
60enum raidz_rec_op {
61 RAIDZ_REC_P = 0,
62 RAIDZ_REC_Q,
63 RAIDZ_REC_R,
64 RAIDZ_REC_PQ,
65 RAIDZ_REC_PR,
66 RAIDZ_REC_QR,
67 RAIDZ_REC_PQR,
68 RAIDZ_REC_NUM = 7
69};
70
71extern const char *raidz_gen_name[RAIDZ_GEN_NUM];
72extern const char *raidz_rec_name[RAIDZ_REC_NUM];
73
74/*
75 * Methods used to define raidz implementation
76 *
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
85 */
86typedef void (*raidz_gen_f)(void *);
87typedef int (*raidz_rec_f)(void *, const int *);
88typedef boolean_t (*will_work_f)(void);
89typedef void (*init_impl_f)(void);
90typedef void (*fini_impl_f)(void);
91
c9187d86
GN
92#define RAIDZ_IMPL_NAME_MAX (16)
93
ab9f4b0b
GN
94typedef struct raidz_impl_ops {
95 init_impl_f init;
96 fini_impl_f fini;
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 */
c9187d86 100 char name[RAIDZ_IMPL_NAME_MAX]; /* Name of the implementation */
ab9f4b0b
GN
101} raidz_impl_ops_t;
102
103typedef 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? */
112} raidz_col_t;
113
114typedef 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 */
130} raidz_map_t;
131
c9187d86
GN
132#define RAIDZ_ORIGINAL_IMPL (INT_MAX)
133
ae25d222
GN
134extern const raidz_impl_ops_t vdev_raidz_scalar_impl;
135#if defined(__x86_64) && defined(HAVE_SSE2) /* only x86_64 for now */
136extern const raidz_impl_ops_t vdev_raidz_sse2_impl;
137#endif
138#if defined(__x86_64) && defined(HAVE_SSSE3) /* only x86_64 for now */
139extern const raidz_impl_ops_t vdev_raidz_ssse3_impl;
140#endif
141#if defined(__x86_64) && defined(HAVE_AVX2) /* only x86_64 for now */
142extern const raidz_impl_ops_t vdev_raidz_avx2_impl;
143#endif
144
ab9f4b0b
GN
145/*
146 * Commonly used raidz_map helpers
147 *
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
155 */
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))
163
164/*
165 * Macro defines an RAIDZ parity generation method
166 *
167 * @code parity the function produce
168 * @impl name of the implementation
169 */
170#define _RAIDZ_GEN_WRAP(code, impl) \
171static void \
172impl ## _gen_ ## code(void *rmp) \
173{ \
174 raidz_map_t *rm = (raidz_map_t *) rmp; \
175 raidz_generate_## code ## _impl(rm); \
176}
177
178/*
179 * Macro defines an RAIDZ data reconstruction method
180 *
181 * @code parity the function produce
182 * @impl name of the implementation
183 */
184#define _RAIDZ_REC_WRAP(code, impl) \
185static int \
186impl ## _rec_ ## code(void *rmp, const int *tgtidx) \
187{ \
188 raidz_map_t *rm = (raidz_map_t *) rmp; \
189 return (raidz_reconstruct_## code ## _impl(rm, tgtidx)); \
190}
191
192/*
193 * Define all gen methods for an implementation
194 *
195 * @impl name of the implementation
196 */
197#define DEFINE_GEN_METHODS(impl) \
198 _RAIDZ_GEN_WRAP(p, impl); \
199 _RAIDZ_GEN_WRAP(pq, impl); \
200 _RAIDZ_GEN_WRAP(pqr, impl)
201
202/*
203 * Define all rec functions for an implementation
204 *
205 * @impl name of the implementation
206 */
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)
215
216#define RAIDZ_GEN_METHODS(impl) \
217{ \
218 [RAIDZ_GEN_P] = & impl ## _gen_p, \
219 [RAIDZ_GEN_PQ] = & impl ## _gen_pq, \
220 [RAIDZ_GEN_PQR] = & impl ## _gen_pqr \
221}
222
223#define RAIDZ_REC_METHODS(impl) \
224{ \
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 \
232}
233
234
235typedef struct raidz_impl_kstat {
26a08b5c
GN
236 uint64_t gen[RAIDZ_GEN_NUM]; /* gen method speed B/s */
237 uint64_t rec[RAIDZ_REC_NUM]; /* rec method speed B/s */
ab9f4b0b
GN
238} raidz_impl_kstat_t;
239
240/*
241 * Enumerate various multiplication constants
242 * used in reconstruction methods
243 */
244typedef enum raidz_mul_info {
245 /* Reconstruct Q */
246 MUL_Q_X = 0,
247 /* Reconstruct R */
248 MUL_R_X = 0,
249 /* Reconstruct PQ */
250 MUL_PQ_X = 0,
251 MUL_PQ_Y = 1,
252 /* Reconstruct PR */
253 MUL_PR_X = 0,
254 MUL_PR_Y = 1,
255 /* Reconstruct QR */
256 MUL_QR_XQ = 0,
257 MUL_QR_X = 1,
258 MUL_QR_YQ = 2,
259 MUL_QR_Y = 3,
260 /* Reconstruct PQR */
261 MUL_PQR_XP = 0,
262 MUL_PQR_XQ = 1,
263 MUL_PQR_XR = 2,
264 MUL_PQR_YU = 3,
265 MUL_PQR_YP = 4,
266 MUL_PQR_YQ = 5,
267
268 MUL_CNT = 6
269} raidz_mul_info_t;
270
271/*
272 * Powers of 2 in the Galois field.
273 */
274extern const uint8_t vdev_raidz_pow2[256] __attribute__((aligned(256)));
275/* Logs of 2 in the Galois field defined above. */
276extern const uint8_t vdev_raidz_log2[256] __attribute__((aligned(256)));
277
278/*
279 * Multiply a given number by 2 raised to the given power.
280 */
281static inline uint8_t
282vdev_raidz_exp2(const uint8_t a, const unsigned exp)
283{
284 if (a == 0)
285 return (0);
286
287 return (vdev_raidz_pow2[(exp + (unsigned) vdev_raidz_log2[a]) % 255]);
288}
289
290/*
291 * Galois Field operations.
292 *
293 * gf_exp2 - computes 2 raised to the given power
294 * gf_exp2 - computes 4 raised to the given power
295 * gf_mul - multiplication
296 * gf_div - division
297 * gf_inv - multiplicative inverse
298 */
299typedef unsigned gf_t;
300typedef unsigned gf_log_t;
301
302static inline gf_t
303gf_mul(const gf_t a, const gf_t b)
304{
305 gf_log_t logsum;
306
307 if (a == 0 || b == 0)
308 return (0);
309
310 logsum = (gf_log_t) vdev_raidz_log2[a] + (gf_log_t) vdev_raidz_log2[b];
311
312 return ((gf_t) vdev_raidz_pow2[logsum % 255]);
313}
314
315static inline gf_t
316gf_div(const gf_t a, const gf_t b)
317{
318 gf_log_t logsum;
319
320 ASSERT3U(b, >, 0);
321 if (a == 0)
322 return (0);
323
324 logsum = (gf_log_t) 255 + (gf_log_t) vdev_raidz_log2[a] -
325 (gf_log_t) vdev_raidz_log2[b];
326
327 return ((gf_t) vdev_raidz_pow2[logsum % 255]);
328}
329
330static inline gf_t
331gf_inv(const gf_t a)
332{
333 gf_log_t logsum;
334
335 ASSERT3U(a, >, 0);
336
337 logsum = (gf_log_t) 255 - (gf_log_t) vdev_raidz_log2[a];
338
339 return ((gf_t) vdev_raidz_pow2[logsum]);
340}
341
342static inline gf_t
343gf_exp2(gf_log_t exp)
344{
345 return (vdev_raidz_pow2[exp % 255]);
346}
347
348static inline gf_t
349gf_exp4(gf_log_t exp)
350{
351 ASSERT3U(exp, <=, 255);
352 return ((gf_t) vdev_raidz_pow2[(2 * exp) % 255]);
353}
354
355#ifdef __cplusplus
356}
357#endif
358
359#endif /* _VDEV_RAIDZ_H */