xref: /titanic_50/usr/src/uts/common/xen/io/xdf.h (revision 2df1fe9ca32bb227b9158c67f5c00b54c20b10fd)
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 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 
28 #ifndef _SYS_XDF_H
29 #define	_SYS_XDF_H
30 
31 #pragma ident	"%Z%%M%	%I%	%E% SMI"
32 
33 #ifdef __cplusplus
34 extern "C" {
35 #endif
36 
37 
38 #include <sys/types.h>
39 #include <sys/conf.h>
40 #include <sys/ddi.h>
41 #include <sys/dditypes.h>
42 #include <sys/sunddi.h>
43 #include <sys/list.h>
44 #include <sys/dkio.h>
45 #include <sys/vtoc.h>
46 #include <sys/modctl.h>
47 #include <sys/bootconf.h>
48 #include <sys/promif.h>
49 #include <sys/open.h>
50 #include <sys/sysmacros.h>
51 #include <sys/kstat.h>
52 #include <sys/gnttab.h>
53 #include <xen/sys/xenbus_impl.h>
54 #include <xen/sys/xendev.h>
55 #include <sys/cmlb.h>
56 #include <sys/scsi/generic/inquiry.h>
57 
58 #define	BLKIF_RING_SIZE	__RING_SIZE((blkif_sring_t *)NULL, PAGESIZE)
59 
60 /*
61  * VBDs have standard 512 byte blocks
62  * A single blkif_request can transfer up to 11 pages of data, 1 page/segment
63  */
64 #define	XB_BSIZE	DEV_BSIZE
65 #define	XB_BMASK	(XB_BSIZE - 1)
66 #define	XB_BSHIFT	9
67 #define	XB_DTOB(bn)	((bn) << XB_BSHIFT)
68 
69 #define	XB_MAX_SEGLEN	(8 * XB_BSIZE)
70 #define	XB_SEGOFFSET	(XB_MAX_SEGLEN - 1)
71 #define	XB_MAX_XFER	(XB_MAX_SEGLEN * BLKIF_MAX_SEGMENTS_PER_REQUEST)
72 #define	XB_MAXPHYS	(XB_MAX_XFER * BLKIF_RING_SIZE)
73 
74 /*
75  * blkif status
76  */
77 enum xdf_state {
78 	/*
79 	 * initial state
80 	 */
81 	XD_UNKNOWN,
82 	/*
83 	 * ring and evtchn alloced, xenbus state changed to
84 	 * XenbusStateInitialised, wait for backend to connect
85 	 */
86 	XD_INIT,
87 	/*
88 	 * backend's xenbus state has changed to XenbusStateConnected,
89 	 * this is the only state allowing I/Os
90 	 */
91 	XD_READY,
92 	/*
93 	 * vbd interface close request received from backend, no more I/O
94 	 * requestis allowed to be put into ring buffer, while interrupt handler
95 	 * is allowed to run to finish any outstanding I/O request, disconnect
96 	 * process is kicked off by changing xenbus state to XenbusStateClosed
97 	 */
98 	XD_CLOSING,
99 	/*
100 	 * disconnection process finished, both backend and frontend's
101 	 * xenbus state has been changed to XenbusStateClosed, can be detached
102 	 */
103 	XD_CLOSED,
104 	/*
105 	 * disconnection process finished, frontend is suspended
106 	 */
107 	XD_SUSPEND
108 };
109 
110 /*
111  * 16 paritions + fdisk
112  */
113 #define	XDF_PSHIFT	6
114 #define	XDF_PMASK	((1 << XDF_PSHIFT) - 1)
115 #define	XDF_PEXT	(1 << XDF_PSHIFT)
116 #define	XDF_MINOR(i, m) (((i) << XDF_PSHIFT) | (m))
117 #define	XDF_INST(m)	((m) >> XDF_PSHIFT)
118 #define	XDF_PART(m)	((m) & XDF_PMASK)
119 
120 /*
121  * one blkif_request_t will have one corresponding ge_slot_t
122  * where we save those grant table refs used in this blkif_request_t
123  *
124  * the id of this ge_slot_t will also be put into 'id' field in
125  * each blkif_request_t when sent out to the ring buffer.
126  */
127 typedef struct ge_slot {
128 	list_node_t	link;
129 	domid_t		oeid;
130 	struct v_req	*vreq;
131 	int		isread;
132 	grant_ref_t	ghead;
133 	int		ngrefs;
134 	grant_ref_t	ge[BLKIF_MAX_SEGMENTS_PER_REQUEST];
135 } ge_slot_t;
136 
137 /*
138  * vbd I/O request
139  *
140  * An instance of this structure is bound to each buf passed to
141  * the driver's strategy by setting the pointer into bp->av_back.
142  * The id of this vreq will also be put into 'id' field in each
143  * blkif_request_t when sent out to the ring buffer for one DMA
144  * window of this buf.
145  *
146  * Vreq mainly contains DMA information for this buf. In one vreq/buf,
147  * there could be more than one DMA window, each of which will be
148  * mapped to one blkif_request_t/ge_slot_t. Ge_slot_t contains all grant
149  * table entry information for this buf. The ge_slot_t for current DMA
150  * window is pointed to by v_gs in vreq.
151  *
152  * So, grant table entries will only be alloc'ed when the DMA window is
153  * about to be transferred via blkif_request_t to the ring buffer. And
154  * they will be freed right after the blkif_response_t is seen. By this
155  * means, we can make use of grant table entries more efficiently.
156  */
157 typedef struct v_req {
158 	list_node_t	v_link;
159 	int		v_status;
160 	buf_t		*v_buf;
161 	ddi_dma_handle_t v_dmahdl;
162 	ddi_dma_cookie_t v_dmac;
163 	uint_t		v_ndmacs;
164 	uint_t		v_dmaw;
165 	uint_t		v_ndmaws;
166 	uint_t		v_nslots;
167 	ge_slot_t	*v_gs;
168 	uint64_t	v_blkno;
169 	ddi_acc_handle_t v_align;
170 	caddr_t		v_abuf;
171 	ddi_dma_handle_t v_memdmahdl;
172 	uint8_t		v_flush_diskcache;
173 } v_req_t;
174 
175 /*
176  * Status set and checked in vreq->v_status by vreq_setup()
177  *
178  * These flags will help us to continue the vreq setup work from last failure
179  * point, instead of starting from scrath after each failure.
180  */
181 #define	VREQ_INIT		0x0
182 #define	VREQ_INIT_DONE		0x1
183 #define	VREQ_DMAHDL_ALLOCED	0x2
184 #define	VREQ_MEMDMAHDL_ALLOCED	0x3
185 #define	VREQ_DMAMEM_ALLOCED	0x4
186 #define	VREQ_DMABUF_BOUND	0x5
187 #define	VREQ_GS_ALLOCED		0x6
188 #define	VREQ_DMAWIN_DONE	0x7
189 
190 /*
191  * virtual block device per-instance softstate
192  */
193 typedef struct xdf {
194 	dev_info_t	*xdf_dip;
195 	domid_t		xdf_peer; /* otherend's dom ID */
196 	xendev_ring_t	*xdf_xb_ring; /* I/O ring buffer */
197 	ddi_acc_handle_t xdf_xb_ring_hdl; /* access handler for ring buffer */
198 	list_t		xdf_vreq_act; /* active vreq list */
199 	list_t		xdf_gs_act; /* active grant table slot list */
200 	buf_t		*xdf_f_act; /* active buf list head */
201 	buf_t		*xdf_l_act; /* active buf list tail */
202 	enum xdf_state	xdf_status; /* status of this virtual disk */
203 	ulong_t		xdf_vd_open[OTYPCNT];
204 	ulong_t		xdf_vd_lyropen[XDF_PEXT];
205 	ulong_t		xdf_vd_exclopen;
206 	kmutex_t	xdf_dev_lk; /* mutex lock for I/O path */
207 	kmutex_t	xdf_cb_lk; /* mutex lock for event handling path */
208 	kcondvar_t	xdf_dev_cv; /* cv used in I/O path */
209 	uint_t		xdf_xdev_info; /* disk info from backend xenstore */
210 	diskaddr_t	xdf_xdev_nblocks; /* total size in block */
211 	kstat_t		*xdf_xdev_iostat;
212 	cmlb_handle_t	xdf_vd_lbl;
213 	ddi_softintr_t	xdf_softintr_id;
214 	timeout_id_t	xdf_timeout_id;
215 	struct gnttab_free_callback xdf_gnt_callback;
216 	int		xdf_feature_barrier;
217 	int		xdf_flush_supported;
218 	int		xdf_wce;
219 	char		*xdf_flush_mem;
220 	char		*xdf_cache_flush_block;
221 #ifdef	DEBUG
222 	int		xdf_dmacallback_num;
223 #endif
224 } xdf_t;
225 
226 #define	BP2VREQ(bp)	((v_req_t *)((bp)->av_back))
227 
228 /*
229  * VBD I/O requests must be aligned on a 512-byte boundary and specify
230  * a transfer size which is a mutiple of 512-bytes
231  */
232 #define	ALIGNED_XFER(bp) \
233 	((((uintptr_t)((bp)->b_un.b_addr) & XB_BMASK) == 0) && \
234 	(((bp)->b_bcount & XB_BMASK) == 0))
235 
236 #define	U_INVAL(u)	(((u)->uio_loffset & (offset_t)(XB_BMASK)) || \
237 	((u)->uio_iov->iov_len & (offset_t)(XB_BMASK)))
238 
239 /* wrap pa_to_ma() for xdf to run in dom0 */
240 #define	PATOMA(addr)	(DOMAIN_IS_INITDOMAIN(xen_info) ? addr : pa_to_ma(addr))
241 
242 #define	XD_IS_RO(vbd)	((vbd)->xdf_xdev_info & VDISK_READONLY)
243 #define	XD_IS_CD(vbd)	((vbd)->xdf_xdev_info & VDISK_CDROM)
244 #define	XD_IS_RM(vbd)	((vbd)->xdf_xdev_info & VDISK_REMOVABLE)
245 #define	IS_READ(bp)	((bp)->b_flags & B_READ)
246 #define	IS_ERROR(bp)	((bp)->b_flags & B_ERROR)
247 
248 #define	XDF_UPDATE_IO_STAT(vdp, bp)					\
249 	if ((vdp)->xdf_xdev_iostat != NULL) {				\
250 		kstat_io_t *kip = KSTAT_IO_PTR((vdp)->xdf_xdev_iostat);	\
251 		size_t n_done = (bp)->b_bcount - (bp)->b_resid;		\
252 		if ((bp)->b_flags & B_READ) {				\
253 			kip->reads++;					\
254 			kip->nread += n_done;				\
255 		} else {                                                \
256 			kip->writes++;					\
257 			kip->nwritten += n_done;			\
258 		}							\
259 	}
260 
261 extern int xdfdebug;
262 #ifdef DEBUG
263 #define	DPRINTF(flag, args)	{if (xdfdebug & (flag)) prom_printf args; }
264 #define	SETDMACBON(vbd)		{(vbd)->xdf_dmacallback_num++; }
265 #define	SETDMACBOFF(vbd)	{(vbd)->xdf_dmacallback_num--; }
266 #define	ISDMACBON(vbd)		((vbd)->xdf_dmacallback_num > 0)
267 #else
268 #define	DPRINTF(flag, args)
269 #define	SETDMACBON(vbd)
270 #define	SETDMACBOFF(vbd)
271 #define	ISDMACBON(vbd)
272 #endif /* DEBUG */
273 
274 #define	DDI_DBG		0x1
275 #define	DMA_DBG		0x2
276 #define	INTR_DBG	0x8
277 #define	IO_DBG		0x10
278 #define	IOCTL_DBG	0x20
279 #define	SUSRES_DBG	0x40
280 #define	LBL_DBG		0x80
281 
282 #ifdef __cplusplus
283 }
284 #endif
285 
286 #endif	/* _SYS_XDF_H */
287