Merge "Optimizing the code and uploading ftrace logs to artifacts"
[kvmfornfv.git] / kernel / drivers / block / xen-blkback / common.h
1 /*
2  * This program is free software; you can redistribute it and/or
3  * modify it under the terms of the GNU General Public License version 2
4  * as published by the Free Software Foundation; or, when distributed
5  * separately from the Linux kernel or incorporated into other
6  * software packages, subject to the following license:
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this source file (the "Software"), to deal in the Software without
10  * restriction, including without limitation the rights to use, copy, modify,
11  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
12  * and to permit persons to whom the Software is furnished to do so, subject to
13  * the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
23  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24  * IN THE SOFTWARE.
25  */
26
27 #ifndef __XEN_BLKIF__BACKEND__COMMON_H__
28 #define __XEN_BLKIF__BACKEND__COMMON_H__
29
30 #include <linux/module.h>
31 #include <linux/interrupt.h>
32 #include <linux/slab.h>
33 #include <linux/blkdev.h>
34 #include <linux/vmalloc.h>
35 #include <linux/wait.h>
36 #include <linux/io.h>
37 #include <linux/rbtree.h>
38 #include <asm/setup.h>
39 #include <asm/pgalloc.h>
40 #include <asm/hypervisor.h>
41 #include <xen/grant_table.h>
42 #include <xen/page.h>
43 #include <xen/xenbus.h>
44 #include <xen/interface/io/ring.h>
45 #include <xen/interface/io/blkif.h>
46 #include <xen/interface/io/protocols.h>
47
48 extern unsigned int xen_blkif_max_ring_order;
49 /*
50  * This is the maximum number of segments that would be allowed in indirect
51  * requests. This value will also be passed to the frontend.
52  */
53 #define MAX_INDIRECT_SEGMENTS 256
54
55 /*
56  * Xen use 4K pages. The guest may use different page size (4K or 64K)
57  * Number of Xen pages per segment
58  */
59 #define XEN_PAGES_PER_SEGMENT   (PAGE_SIZE / XEN_PAGE_SIZE)
60
61 #define XEN_PAGES_PER_INDIRECT_FRAME \
62         (XEN_PAGE_SIZE/sizeof(struct blkif_request_segment))
63 #define SEGS_PER_INDIRECT_FRAME \
64         (XEN_PAGES_PER_INDIRECT_FRAME / XEN_PAGES_PER_SEGMENT)
65
66 #define MAX_INDIRECT_PAGES \
67         ((MAX_INDIRECT_SEGMENTS + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
68 #define INDIRECT_PAGES(_segs) DIV_ROUND_UP(_segs, XEN_PAGES_PER_INDIRECT_FRAME)
69
70 /* Not a real protocol.  Used to generate ring structs which contain
71  * the elements common to all protocols only.  This way we get a
72  * compiler-checkable way to use common struct elements, so we can
73  * avoid using switch(protocol) in a number of places.  */
74 struct blkif_common_request {
75         char dummy;
76 };
77 struct blkif_common_response {
78         char dummy;
79 };
80
81 struct blkif_x86_32_request_rw {
82         uint8_t        nr_segments;  /* number of segments                   */
83         blkif_vdev_t   handle;       /* only for read/write requests         */
84         uint64_t       id;           /* private guest value, echoed in resp  */
85         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
86         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
87 } __attribute__((__packed__));
88
89 struct blkif_x86_32_request_discard {
90         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
91         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
92         uint64_t       id;           /* private guest value, echoed in resp  */
93         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
94         uint64_t       nr_sectors;
95 } __attribute__((__packed__));
96
97 struct blkif_x86_32_request_other {
98         uint8_t        _pad1;
99         blkif_vdev_t   _pad2;
100         uint64_t       id;           /* private guest value, echoed in resp  */
101 } __attribute__((__packed__));
102
103 struct blkif_x86_32_request_indirect {
104         uint8_t        indirect_op;
105         uint16_t       nr_segments;
106         uint64_t       id;
107         blkif_sector_t sector_number;
108         blkif_vdev_t   handle;
109         uint16_t       _pad1;
110         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
111         /*
112          * The maximum number of indirect segments (and pages) that will
113          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
114          * is also exported to the guest (via xenstore
115          * feature-max-indirect-segments entry), so the frontend knows how
116          * many indirect segments the backend supports.
117          */
118         uint64_t       _pad2;        /* make it 64 byte aligned */
119 } __attribute__((__packed__));
120
121 struct blkif_x86_32_request {
122         uint8_t        operation;    /* BLKIF_OP_???                         */
123         union {
124                 struct blkif_x86_32_request_rw rw;
125                 struct blkif_x86_32_request_discard discard;
126                 struct blkif_x86_32_request_other other;
127                 struct blkif_x86_32_request_indirect indirect;
128         } u;
129 } __attribute__((__packed__));
130
131 /* i386 protocol version */
132 #pragma pack(push, 4)
133 struct blkif_x86_32_response {
134         uint64_t        id;              /* copied from request */
135         uint8_t         operation;       /* copied from request */
136         int16_t         status;          /* BLKIF_RSP_???       */
137 };
138 #pragma pack(pop)
139 /* x86_64 protocol version */
140
141 struct blkif_x86_64_request_rw {
142         uint8_t        nr_segments;  /* number of segments                   */
143         blkif_vdev_t   handle;       /* only for read/write requests         */
144         uint32_t       _pad1;        /* offsetof(blkif_reqest..,u.rw.id)==8  */
145         uint64_t       id;
146         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
147         struct blkif_request_segment seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
148 } __attribute__((__packed__));
149
150 struct blkif_x86_64_request_discard {
151         uint8_t        flag;         /* BLKIF_DISCARD_SECURE or zero         */
152         blkif_vdev_t   _pad1;        /* was "handle" for read/write requests */
153         uint32_t       _pad2;        /* offsetof(blkif_..,u.discard.id)==8   */
154         uint64_t       id;
155         blkif_sector_t sector_number;/* start sector idx on disk (r/w only)  */
156         uint64_t       nr_sectors;
157 } __attribute__((__packed__));
158
159 struct blkif_x86_64_request_other {
160         uint8_t        _pad1;
161         blkif_vdev_t   _pad2;
162         uint32_t       _pad3;        /* offsetof(blkif_..,u.discard.id)==8   */
163         uint64_t       id;           /* private guest value, echoed in resp  */
164 } __attribute__((__packed__));
165
166 struct blkif_x86_64_request_indirect {
167         uint8_t        indirect_op;
168         uint16_t       nr_segments;
169         uint32_t       _pad1;        /* offsetof(blkif_..,u.indirect.id)==8   */
170         uint64_t       id;
171         blkif_sector_t sector_number;
172         blkif_vdev_t   handle;
173         uint16_t       _pad2;
174         grant_ref_t    indirect_grefs[BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST];
175         /*
176          * The maximum number of indirect segments (and pages) that will
177          * be used is determined by MAX_INDIRECT_SEGMENTS, this value
178          * is also exported to the guest (via xenstore
179          * feature-max-indirect-segments entry), so the frontend knows how
180          * many indirect segments the backend supports.
181          */
182         uint32_t       _pad3;        /* make it 64 byte aligned */
183 } __attribute__((__packed__));
184
185 struct blkif_x86_64_request {
186         uint8_t        operation;    /* BLKIF_OP_???                         */
187         union {
188                 struct blkif_x86_64_request_rw rw;
189                 struct blkif_x86_64_request_discard discard;
190                 struct blkif_x86_64_request_other other;
191                 struct blkif_x86_64_request_indirect indirect;
192         } u;
193 } __attribute__((__packed__));
194
195 struct blkif_x86_64_response {
196         uint64_t       __attribute__((__aligned__(8))) id;
197         uint8_t         operation;       /* copied from request */
198         int16_t         status;          /* BLKIF_RSP_???       */
199 };
200
201 DEFINE_RING_TYPES(blkif_common, struct blkif_common_request,
202                   struct blkif_common_response);
203 DEFINE_RING_TYPES(blkif_x86_32, struct blkif_x86_32_request,
204                   struct blkif_x86_32_response);
205 DEFINE_RING_TYPES(blkif_x86_64, struct blkif_x86_64_request,
206                   struct blkif_x86_64_response);
207
208 union blkif_back_rings {
209         struct blkif_back_ring        native;
210         struct blkif_common_back_ring common;
211         struct blkif_x86_32_back_ring x86_32;
212         struct blkif_x86_64_back_ring x86_64;
213 };
214
215 enum blkif_protocol {
216         BLKIF_PROTOCOL_NATIVE = 1,
217         BLKIF_PROTOCOL_X86_32 = 2,
218         BLKIF_PROTOCOL_X86_64 = 3,
219 };
220
221 /*
222  * Default protocol if the frontend doesn't specify one.
223  */
224 #ifdef CONFIG_X86
225 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_X86_32
226 #else
227 #  define BLKIF_PROTOCOL_DEFAULT BLKIF_PROTOCOL_NATIVE
228 #endif
229
230 struct xen_vbd {
231         /* What the domain refers to this vbd as. */
232         blkif_vdev_t            handle;
233         /* Non-zero -> read-only */
234         unsigned char           readonly;
235         /* VDISK_xxx */
236         unsigned char           type;
237         /* phys device that this vbd maps to. */
238         u32                     pdevice;
239         struct block_device     *bdev;
240         /* Cached size parameter. */
241         sector_t                size;
242         unsigned int            flush_support:1;
243         unsigned int            discard_secure:1;
244         unsigned int            feature_gnt_persistent:1;
245         unsigned int            overflow_max_grants:1;
246 };
247
248 struct backend_info;
249
250 /* Number of available flags */
251 #define PERSISTENT_GNT_FLAGS_SIZE       2
252 /* This persistent grant is currently in use */
253 #define PERSISTENT_GNT_ACTIVE           0
254 /*
255  * This persistent grant has been used, this flag is set when we remove the
256  * PERSISTENT_GNT_ACTIVE, to know that this grant has been used recently.
257  */
258 #define PERSISTENT_GNT_WAS_ACTIVE       1
259
260 /* Number of requests that we can fit in a ring */
261 #define XEN_BLKIF_REQS_PER_PAGE         32
262
263 struct persistent_gnt {
264         struct page *page;
265         grant_ref_t gnt;
266         grant_handle_t handle;
267         DECLARE_BITMAP(flags, PERSISTENT_GNT_FLAGS_SIZE);
268         struct rb_node node;
269         struct list_head remove_node;
270 };
271
272 struct xen_blkif {
273         /* Unique identifier for this interface. */
274         domid_t                 domid;
275         unsigned int            handle;
276         /* Physical parameters of the comms window. */
277         unsigned int            irq;
278         /* Comms information. */
279         enum blkif_protocol     blk_protocol;
280         union blkif_back_rings  blk_rings;
281         void                    *blk_ring;
282         /* The VBD attached to this interface. */
283         struct xen_vbd          vbd;
284         /* Back pointer to the backend_info. */
285         struct backend_info     *be;
286         /* Private fields. */
287         spinlock_t              blk_ring_lock;
288         atomic_t                refcnt;
289
290         wait_queue_head_t       wq;
291         /* for barrier (drain) requests */
292         struct completion       drain_complete;
293         atomic_t                drain;
294         atomic_t                inflight;
295         /* One thread per one blkif. */
296         struct task_struct      *xenblkd;
297         unsigned int            waiting_reqs;
298
299         /* tree to store persistent grants */
300         struct rb_root          persistent_gnts;
301         unsigned int            persistent_gnt_c;
302         atomic_t                persistent_gnt_in_use;
303         unsigned long           next_lru;
304
305         /* used by the kworker that offload work from the persistent purge */
306         struct list_head        persistent_purge_list;
307         struct work_struct      persistent_purge_work;
308
309         /* buffer of free pages to map grant refs */
310         spinlock_t              free_pages_lock;
311         int                     free_pages_num;
312         struct list_head        free_pages;
313
314         /* List of all 'pending_req' available */
315         struct list_head        pending_free;
316         /* And its spinlock. */
317         spinlock_t              pending_free_lock;
318         wait_queue_head_t       pending_free_wq;
319
320         /* statistics */
321         unsigned long           st_print;
322         unsigned long long                      st_rd_req;
323         unsigned long long                      st_wr_req;
324         unsigned long long                      st_oo_req;
325         unsigned long long                      st_f_req;
326         unsigned long long                      st_ds_req;
327         unsigned long long                      st_rd_sect;
328         unsigned long long                      st_wr_sect;
329
330         struct work_struct      free_work;
331         /* Thread shutdown wait queue. */
332         wait_queue_head_t       shutdown_wq;
333         unsigned int nr_ring_pages;
334 };
335
336 struct seg_buf {
337         unsigned long offset;
338         unsigned int nsec;
339 };
340
341 struct grant_page {
342         struct page             *page;
343         struct persistent_gnt   *persistent_gnt;
344         grant_handle_t          handle;
345         grant_ref_t             gref;
346 };
347
348 /*
349  * Each outstanding request that we've passed to the lower device layers has a
350  * 'pending_req' allocated to it. Each buffer_head that completes decrements
351  * the pendcnt towards zero. When it hits zero, the specified domain has a
352  * response queued for it, with the saved 'id' passed back.
353  */
354 struct pending_req {
355         struct xen_blkif        *blkif;
356         u64                     id;
357         int                     nr_segs;
358         atomic_t                pendcnt;
359         unsigned short          operation;
360         int                     status;
361         struct list_head        free_list;
362         struct grant_page       *segments[MAX_INDIRECT_SEGMENTS];
363         /* Indirect descriptors */
364         struct grant_page       *indirect_pages[MAX_INDIRECT_PAGES];
365         struct seg_buf          seg[MAX_INDIRECT_SEGMENTS];
366         struct bio              *biolist[MAX_INDIRECT_SEGMENTS];
367         struct gnttab_unmap_grant_ref unmap[MAX_INDIRECT_SEGMENTS];
368         struct page                   *unmap_pages[MAX_INDIRECT_SEGMENTS];
369         struct gntab_unmap_queue_data gnttab_unmap_data;
370 };
371
372
373 #define vbd_sz(_v)      ((_v)->bdev->bd_part ? \
374                          (_v)->bdev->bd_part->nr_sects : \
375                           get_capacity((_v)->bdev->bd_disk))
376
377 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
378 #define xen_blkif_put(_b)                               \
379         do {                                            \
380                 if (atomic_dec_and_test(&(_b)->refcnt)) \
381                         schedule_work(&(_b)->free_work);\
382         } while (0)
383
384 struct phys_req {
385         unsigned short          dev;
386         blkif_sector_t          nr_sects;
387         struct block_device     *bdev;
388         blkif_sector_t          sector_number;
389 };
390 int xen_blkif_interface_init(void);
391
392 int xen_blkif_xenbus_init(void);
393
394 irqreturn_t xen_blkif_be_int(int irq, void *dev_id);
395 int xen_blkif_schedule(void *arg);
396 int xen_blkif_purge_persistent(void *arg);
397 void xen_blkbk_free_caches(struct xen_blkif *blkif);
398
399 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
400                               struct backend_info *be, int state);
401
402 int xen_blkbk_barrier(struct xenbus_transaction xbt,
403                       struct backend_info *be, int state);
404 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be);
405 void xen_blkbk_unmap_purged_grants(struct work_struct *work);
406
407 static inline void blkif_get_x86_32_req(struct blkif_request *dst,
408                                         struct blkif_x86_32_request *src)
409 {
410         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
411         dst->operation = READ_ONCE(src->operation);
412         switch (dst->operation) {
413         case BLKIF_OP_READ:
414         case BLKIF_OP_WRITE:
415         case BLKIF_OP_WRITE_BARRIER:
416         case BLKIF_OP_FLUSH_DISKCACHE:
417                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
418                 dst->u.rw.handle = src->u.rw.handle;
419                 dst->u.rw.id = src->u.rw.id;
420                 dst->u.rw.sector_number = src->u.rw.sector_number;
421                 barrier();
422                 if (n > dst->u.rw.nr_segments)
423                         n = dst->u.rw.nr_segments;
424                 for (i = 0; i < n; i++)
425                         dst->u.rw.seg[i] = src->u.rw.seg[i];
426                 break;
427         case BLKIF_OP_DISCARD:
428                 dst->u.discard.flag = src->u.discard.flag;
429                 dst->u.discard.id = src->u.discard.id;
430                 dst->u.discard.sector_number = src->u.discard.sector_number;
431                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
432                 break;
433         case BLKIF_OP_INDIRECT:
434                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
435                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
436                 dst->u.indirect.handle = src->u.indirect.handle;
437                 dst->u.indirect.id = src->u.indirect.id;
438                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
439                 barrier();
440                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
441                 for (i = 0; i < j; i++)
442                         dst->u.indirect.indirect_grefs[i] =
443                                 src->u.indirect.indirect_grefs[i];
444                 break;
445         default:
446                 /*
447                  * Don't know how to translate this op. Only get the
448                  * ID so failure can be reported to the frontend.
449                  */
450                 dst->u.other.id = src->u.other.id;
451                 break;
452         }
453 }
454
455 static inline void blkif_get_x86_64_req(struct blkif_request *dst,
456                                         struct blkif_x86_64_request *src)
457 {
458         int i, n = BLKIF_MAX_SEGMENTS_PER_REQUEST, j;
459         dst->operation = READ_ONCE(src->operation);
460         switch (dst->operation) {
461         case BLKIF_OP_READ:
462         case BLKIF_OP_WRITE:
463         case BLKIF_OP_WRITE_BARRIER:
464         case BLKIF_OP_FLUSH_DISKCACHE:
465                 dst->u.rw.nr_segments = src->u.rw.nr_segments;
466                 dst->u.rw.handle = src->u.rw.handle;
467                 dst->u.rw.id = src->u.rw.id;
468                 dst->u.rw.sector_number = src->u.rw.sector_number;
469                 barrier();
470                 if (n > dst->u.rw.nr_segments)
471                         n = dst->u.rw.nr_segments;
472                 for (i = 0; i < n; i++)
473                         dst->u.rw.seg[i] = src->u.rw.seg[i];
474                 break;
475         case BLKIF_OP_DISCARD:
476                 dst->u.discard.flag = src->u.discard.flag;
477                 dst->u.discard.id = src->u.discard.id;
478                 dst->u.discard.sector_number = src->u.discard.sector_number;
479                 dst->u.discard.nr_sectors = src->u.discard.nr_sectors;
480                 break;
481         case BLKIF_OP_INDIRECT:
482                 dst->u.indirect.indirect_op = src->u.indirect.indirect_op;
483                 dst->u.indirect.nr_segments = src->u.indirect.nr_segments;
484                 dst->u.indirect.handle = src->u.indirect.handle;
485                 dst->u.indirect.id = src->u.indirect.id;
486                 dst->u.indirect.sector_number = src->u.indirect.sector_number;
487                 barrier();
488                 j = min(MAX_INDIRECT_PAGES, INDIRECT_PAGES(dst->u.indirect.nr_segments));
489                 for (i = 0; i < j; i++)
490                         dst->u.indirect.indirect_grefs[i] =
491                                 src->u.indirect.indirect_grefs[i];
492                 break;
493         default:
494                 /*
495                  * Don't know how to translate this op. Only get the
496                  * ID so failure can be reported to the frontend.
497                  */
498                 dst->u.other.id = src->u.other.id;
499                 break;
500         }
501 }
502
503 #endif /* __XEN_BLKIF__BACKEND__COMMON_H__ */