xref: /illumos-gate/usr/src/uts/common/xen/io/xdf.h (revision 7c80a9608efb5c2bb78fb923e352a01088239788)
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 /*
23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * Copyright (c) 2014 by Delphix. All rights reserved.
29  */
30 
31 #ifndef _SYS_XDF_H
32 #define	_SYS_XDF_H
33 
34 #include <sys/ddi.h>
35 #include <sys/sunddi.h>
36 #include <sys/cmlb.h>
37 #include <sys/dkio.h>
38 
39 #include <sys/gnttab.h>
40 #include <xen/sys/xendev.h>
41 
42 #ifdef __cplusplus
43 extern "C" {
44 #endif
45 
46 
47 /*
48  * VBDs have standard 512 byte blocks
49  * A single blkif_request can transfer up to 11 pages of data, 1 page/segment
50  */
51 #define	XB_BSIZE	DEV_BSIZE
52 #define	XB_BMASK	(XB_BSIZE - 1)
53 #define	XB_BSHIFT	9
54 #define	XB_DTOB(bn, vdp)	((bn) * (vdp)->xdf_xdev_secsize)
55 
56 #define	XB_MAX_SEGLEN	(8 * XB_BSIZE)
57 #define	XB_SEGOFFSET	(XB_MAX_SEGLEN - 1)
58 #define	XB_MAX_XFER	(XB_MAX_SEGLEN * BLKIF_MAX_SEGMENTS_PER_REQUEST)
59 #define	XB_MAXPHYS	(XB_MAX_XFER * BLKIF_RING_SIZE)
60 
61 /* Number of sectors per segement */
62 #define	XB_NUM_SECTORS_PER_SEG	(PAGESIZE / XB_BSIZE)
63 /* sectors are number 0 through XB_NUM_SECTORS_PER_SEG - 1 */
64 #define	XB_LAST_SECTOR_IN_SEG	(XB_NUM_SECTORS_PER_SEG - 1)
65 
66 
67 /*
68  * Slice for absolute disk transaction.
69  *
70  * Hack Alert.  XB_SLICE_NONE is a magic value that can be written into the
71  * b_private field of buf structures passed to xdf_strategy().  When present
72  * it indicates that the I/O is using an absolute offset.  (ie, the I/O is
73  * not bound to any one partition.)  This magic value is currently used by
74  * the pv_cmdk driver.  This hack is shamelessly stolen from the sun4v vdc
75  * driver, another virtual disk device driver.  (Although in the case of
76  * vdc the hack is less egregious since it is self contained within the
77  * vdc driver, where as here it is used as an interface between the pv_cmdk
78  * driver and the xdf driver.)
79  */
80 #define	XB_SLICE_NONE		0xFF
81 
82 /*
83  * blkif status
84  */
85 typedef enum xdf_state {
86 	/*
87 	 * initial state
88 	 */
89 	XD_UNKNOWN = 0,
90 	/*
91 	 * ring and evtchn alloced, xenbus state changed to
92 	 * XenbusStateInitialised, wait for backend to connect
93 	 */
94 	XD_INIT = 1,
95 	/*
96 	 * backend and frontend xenbus state has changed to
97 	 * XenbusStateConnected.  IO is now allowed, but we are not still
98 	 * fully initialized.
99 	 */
100 	XD_CONNECTED = 2,
101 	/*
102 	 * We're fully initialized and allowing regular IO.
103 	 */
104 	XD_READY = 3,
105 	/*
106 	 * vbd interface close request received from backend, no more I/O
107 	 * requestis allowed to be put into ring buffer, while interrupt handler
108 	 * is allowed to run to finish any outstanding I/O request, disconnect
109 	 * process is kicked off by changing xenbus state to XenbusStateClosed
110 	 */
111 	XD_CLOSING = 4,
112 	/*
113 	 * disconnection process finished, both backend and frontend's
114 	 * xenbus state has been changed to XenbusStateClosed, can be detached
115 	 */
116 	XD_CLOSED = 5,
117 	/*
118 	 * We're either being suspended or resuming from a suspend.  If we're
119 	 * in the process of suspending, we block all new IO, but but allow
120 	 * existing IO to drain.
121 	 */
122 	XD_SUSPEND = 6
123 } xdf_state_t;
124 
125 /*
126  * 16 partitions + fdisk
127  */
128 #define	XDF_PSHIFT	6
129 #define	XDF_PMASK	((1 << XDF_PSHIFT) - 1)
130 #define	XDF_PEXT	(1 << XDF_PSHIFT)
131 #define	XDF_MINOR(i, m) (((i) << XDF_PSHIFT) | (m))
132 #define	XDF_INST(m)	((m) >> XDF_PSHIFT)
133 #define	XDF_PART(m)	((m) & XDF_PMASK)
134 
135 /*
136  * one blkif_request_t will have one corresponding ge_slot_t
137  * where we save those grant table refs used in this blkif_request_t
138  *
139  * the id of this ge_slot_t will also be put into 'id' field in
140  * each blkif_request_t when sent out to the ring buffer.
141  */
142 typedef struct ge_slot {
143 	list_node_t	gs_vreq_link;
144 	struct v_req	*gs_vreq;
145 	domid_t		gs_oeid;
146 	int		gs_isread;
147 	grant_ref_t	gs_ghead;
148 	int		gs_ngrefs;
149 	grant_ref_t	gs_ge[BLKIF_MAX_SEGMENTS_PER_REQUEST];
150 } ge_slot_t;
151 
152 /*
153  * vbd I/O request
154  *
155  * An instance of this structure is bound to each buf passed to
156  * the driver's strategy by setting the pointer into bp->av_back.
157  * The id of this vreq will also be put into 'id' field in each
158  * blkif_request_t when sent out to the ring buffer for one DMA
159  * window of this buf.
160  *
161  * Vreq mainly contains DMA information for this buf. In one vreq/buf,
162  * there could be more than one DMA window, each of which will be
163  * mapped to one blkif_request_t/ge_slot_t. Ge_slot_t contains all grant
164  * table entry information for this buf. The ge_slot_t for current DMA
165  * window is pointed to by v_gs in vreq.
166  *
167  * So, grant table entries will only be alloc'ed when the DMA window is
168  * about to be transferred via blkif_request_t to the ring buffer. And
169  * they will be freed right after the blkif_response_t is seen. By this
170  * means, we can make use of grant table entries more efficiently.
171  */
172 typedef struct v_req {
173 	list_node_t	v_link;
174 	list_t		v_gs;
175 	int		v_status;
176 	buf_t		*v_buf;
177 	uint_t		v_ndmacs;
178 	uint_t		v_dmaw;
179 	uint_t		v_ndmaws;
180 	uint_t		v_nslots;
181 	uint64_t	v_blkno;
182 	ddi_dma_handle_t v_memdmahdl;
183 	ddi_acc_handle_t v_align;
184 	ddi_dma_handle_t v_dmahdl;
185 	ddi_dma_cookie_t v_dmac;
186 	caddr_t		v_abuf;
187 	uint8_t		v_flush_diskcache;
188 	boolean_t	v_runq;
189 } v_req_t;
190 
191 /*
192  * Status set and checked in vreq->v_status by vreq_setup()
193  *
194  * These flags will help us to continue the vreq setup work from last failure
195  * point, instead of starting from scratch after each failure.
196  */
197 #define	VREQ_INIT		0x0
198 #define	VREQ_INIT_DONE		0x1
199 #define	VREQ_DMAHDL_ALLOCED	0x2
200 #define	VREQ_MEMDMAHDL_ALLOCED	0x3
201 #define	VREQ_DMAMEM_ALLOCED	0x4
202 #define	VREQ_DMABUF_BOUND	0x5
203 #define	VREQ_GS_ALLOCED		0x6
204 #define	VREQ_DMAWIN_DONE	0x7
205 
206 /*
207  * virtual block device per-instance softstate
208  */
209 typedef struct xdf {
210 	dev_info_t	*xdf_dip;
211 	char		*xdf_addr;
212 	ddi_iblock_cookie_t xdf_ibc; /* mutex iblock cookie */
213 	domid_t		xdf_peer; /* otherend's dom ID */
214 	xendev_ring_t	*xdf_xb_ring; /* I/O ring buffer */
215 	ddi_acc_handle_t xdf_xb_ring_hdl; /* access handler for ring buffer */
216 	list_t		xdf_vreq_act; /* active vreq list */
217 	buf_t		*xdf_f_act; /* active buf list head */
218 	buf_t		*xdf_l_act; /* active buf list tail */
219 	buf_t		*xdf_i_act; /* active buf list index */
220 	xdf_state_t	xdf_state; /* status of this virtual disk */
221 	boolean_t	xdf_suspending;
222 	ulong_t		xdf_vd_open[OTYPCNT];
223 	ulong_t		xdf_vd_lyropen[XDF_PEXT];
224 	ulong_t		xdf_connect_req;
225 	kthread_t	*xdf_connect_thread;
226 	ulong_t		xdf_vd_exclopen;
227 	kmutex_t	xdf_iostat_lk; /* muxes lock for the iostat ptr */
228 	kmutex_t	xdf_dev_lk; /* mutex lock for I/O path */
229 	kmutex_t	xdf_cb_lk; /* mutex lock for event handling path */
230 	kcondvar_t	xdf_dev_cv; /* cv used in I/O path */
231 	uint_t		xdf_dinfo; /* disk info from backend xenstore */
232 	diskaddr_t	xdf_xdev_nblocks; /* total size in block */
233 	uint_t		xdf_xdev_secsize; /* disk blksize from backend */
234 	cmlb_geom_t	xdf_pgeom;
235 	boolean_t	xdf_pgeom_set;
236 	boolean_t	xdf_pgeom_fixed;
237 	kstat_t		*xdf_xdev_iostat;
238 	cmlb_handle_t	xdf_vd_lbl;
239 	ddi_softintr_t	xdf_softintr_id;
240 	timeout_id_t	xdf_timeout_id;
241 	struct gnttab_free_callback xdf_gnt_callback;
242 	boolean_t	xdf_feature_barrier;
243 	boolean_t	xdf_flush_supported;
244 	boolean_t	xdf_media_req_supported;
245 	boolean_t	xdf_wce;
246 	boolean_t	xdf_cmbl_reattach;
247 	char		*xdf_flush_mem;
248 	char		*xdf_cache_flush_block;
249 	int		xdf_evtchn;
250 	enum dkio_state	xdf_mstate;
251 	kcondvar_t	xdf_mstate_cv;
252 	kcondvar_t	xdf_hp_status_cv;
253 	struct buf	*xdf_ready_bp;
254 	ddi_taskq_t	*xdf_ready_tq;
255 	kthread_t	*xdf_ready_tq_thread;
256 	struct buf	*xdf_ready_tq_bp;
257 	ddi_devid_t	xdf_tgt_devid;
258 #ifdef	DEBUG
259 	int		xdf_dmacallback_num;
260 	kthread_t	*xdf_oe_change_thread;
261 #endif
262 } xdf_t;
263 
264 /*
265  * VBD I/O requests must be aligned on a 512-byte boundary and specify
266  * a transfer size which is a mutiple of 512-bytes
267  */
268 #define	ALIGNED_XFER(bp) \
269 	((((uintptr_t)((bp)->b_un.b_addr) & XB_BMASK) == 0) && \
270 	(((bp)->b_bcount & XB_BMASK) == 0))
271 
272 #define	U_INVAL(u)	(((u)->uio_loffset & (offset_t)(XB_BMASK)) || \
273 	((u)->uio_iov->iov_len & (offset_t)(XB_BMASK)))
274 
275 /* wrap pa_to_ma() for xdf to run in dom0 */
276 #define	PATOMA(addr)	(DOMAIN_IS_INITDOMAIN(xen_info) ? addr : pa_to_ma(addr))
277 
278 #define	XD_IS_RO(vbd)	VOID2BOOLEAN((vbd)->xdf_dinfo & VDISK_READONLY)
279 #define	XD_IS_CD(vbd)	VOID2BOOLEAN((vbd)->xdf_dinfo & VDISK_CDROM)
280 #define	XD_IS_RM(vbd)	VOID2BOOLEAN((vbd)->xdf_dinfo & VDISK_REMOVABLE)
281 #define	IS_READ(bp)	VOID2BOOLEAN((bp)->b_flags & B_READ)
282 #define	IS_ERROR(bp)	VOID2BOOLEAN((bp)->b_flags & B_ERROR)
283 
284 #define	XDF_UPDATE_IO_STAT(vdp, bp)					\
285 	{								\
286 		kstat_io_t *kip = KSTAT_IO_PTR((vdp)->xdf_xdev_iostat);	\
287 		size_t n_done = (bp)->b_bcount - (bp)->b_resid;		\
288 		if ((bp)->b_flags & B_READ) {				\
289 			kip->reads++;					\
290 			kip->nread += n_done;				\
291 		} else {                                                \
292 			kip->writes++;					\
293 			kip->nwritten += n_done;			\
294 		}							\
295 	}
296 
297 #ifdef DEBUG
298 #define	DPRINTF(flag, args)	{if (xdf_debug & (flag)) prom_printf args; }
299 #define	SETDMACBON(vbd)		{(vbd)->xdf_dmacallback_num++; }
300 #define	SETDMACBOFF(vbd)	{(vbd)->xdf_dmacallback_num--; }
301 #define	ISDMACBON(vbd)		((vbd)->xdf_dmacallback_num > 0)
302 #else
303 #define	DPRINTF(flag, args)
304 #define	SETDMACBON(vbd)
305 #define	SETDMACBOFF(vbd)
306 #define	ISDMACBON(vbd)
307 #endif /* DEBUG */
308 
309 #define	DDI_DBG		0x1
310 #define	DMA_DBG		0x2
311 #define	INTR_DBG	0x8
312 #define	IO_DBG		0x10
313 #define	IOCTL_DBG	0x20
314 #define	SUSRES_DBG	0x40
315 #define	LBL_DBG		0x80
316 
317 #ifdef XPV_HVM_DRIVER
318 extern int xdf_lb_getinfo(dev_info_t *, int, void *, void *);
319 extern int xdf_lb_rdwr(dev_info_t *, uchar_t, void *, diskaddr_t, size_t,
320     void *);
321 extern void xdfmin(struct buf *bp);
322 extern dev_info_t *xdf_hvm_hold(const char *);
323 extern boolean_t xdf_hvm_connect(dev_info_t *);
324 extern int xdf_hvm_setpgeom(dev_info_t *, cmlb_geom_t *);
325 extern boolean_t xdf_is_cd(dev_info_t *);
326 extern boolean_t xdf_is_rm(dev_info_t *);
327 extern boolean_t xdf_media_req_supported(dev_info_t *);
328 #endif /* XPV_HVM_DRIVER */
329 
330 #ifdef __cplusplus
331 }
332 #endif
333 
334 #endif	/* _SYS_XDF_H */
335