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 2008 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 #ifdef __cplusplus 32 extern "C" { 33 #endif 34 35 36 /* 37 * VBDs have standard 512 byte blocks 38 * A single blkif_request can transfer up to 11 pages of data, 1 page/segment 39 */ 40 #define XB_BSIZE DEV_BSIZE 41 #define XB_BMASK (XB_BSIZE - 1) 42 #define XB_BSHIFT 9 43 #define XB_DTOB(bn) ((bn) << XB_BSHIFT) 44 45 #define XB_MAX_SEGLEN (8 * XB_BSIZE) 46 #define XB_SEGOFFSET (XB_MAX_SEGLEN - 1) 47 #define XB_MAX_XFER (XB_MAX_SEGLEN * BLKIF_MAX_SEGMENTS_PER_REQUEST) 48 #define XB_MAXPHYS (XB_MAX_XFER * BLKIF_RING_SIZE) 49 50 51 /* 52 * Slice for absolute disk transaction. 53 * 54 * Hack Alert. XB_SLICE_NONE is a magic value that can be written into the 55 * b_private field of buf structures passed to xdf_strategy(). When present 56 * it indicates that the I/O is using an absolute offset. (ie, the I/O is 57 * not bound to any one partition.) This magic value is currently used by 58 * the pv_cmdk driver. This hack is shamelessly stolen from the sun4v vdc 59 * driver, another virtual disk device driver. (Although in the case of 60 * vdc the hack is less egregious since it is self contained within the 61 * vdc driver, where as here it is used as an interface between the pv_cmdk 62 * driver and the xdf driver.) 63 */ 64 #define XB_SLICE_NONE 0xFF 65 66 /* 67 * blkif status 68 */ 69 enum xdf_state { 70 /* 71 * initial state 72 */ 73 XD_UNKNOWN, 74 /* 75 * ring and evtchn alloced, xenbus state changed to 76 * XenbusStateInitialised, wait for backend to connect 77 */ 78 XD_INIT, 79 /* 80 * backend's xenbus state has changed to XenbusStateConnected, 81 * this is the only state allowing I/Os 82 */ 83 XD_READY, 84 /* 85 * vbd interface close request received from backend, no more I/O 86 * requestis allowed to be put into ring buffer, while interrupt handler 87 * is allowed to run to finish any outstanding I/O request, disconnect 88 * process is kicked off by changing xenbus state to XenbusStateClosed 89 */ 90 XD_CLOSING, 91 /* 92 * disconnection process finished, both backend and frontend's 93 * xenbus state has been changed to XenbusStateClosed, can be detached 94 */ 95 XD_CLOSED, 96 /* 97 * disconnection process finished, frontend is suspended 98 */ 99 XD_SUSPEND 100 }; 101 102 /* 103 * 16 partitions + fdisk 104 */ 105 #define XDF_PSHIFT 6 106 #define XDF_PMASK ((1 << XDF_PSHIFT) - 1) 107 #define XDF_PEXT (1 << XDF_PSHIFT) 108 #define XDF_MINOR(i, m) (((i) << XDF_PSHIFT) | (m)) 109 #define XDF_INST(m) ((m) >> XDF_PSHIFT) 110 #define XDF_PART(m) ((m) & XDF_PMASK) 111 112 /* 113 * one blkif_request_t will have one corresponding ge_slot_t 114 * where we save those grant table refs used in this blkif_request_t 115 * 116 * the id of this ge_slot_t will also be put into 'id' field in 117 * each blkif_request_t when sent out to the ring buffer. 118 */ 119 typedef struct ge_slot { 120 list_node_t link; 121 domid_t oeid; 122 struct v_req *vreq; 123 int isread; 124 grant_ref_t ghead; 125 int ngrefs; 126 grant_ref_t ge[BLKIF_MAX_SEGMENTS_PER_REQUEST]; 127 } ge_slot_t; 128 129 /* 130 * vbd I/O request 131 * 132 * An instance of this structure is bound to each buf passed to 133 * the driver's strategy by setting the pointer into bp->av_back. 134 * The id of this vreq will also be put into 'id' field in each 135 * blkif_request_t when sent out to the ring buffer for one DMA 136 * window of this buf. 137 * 138 * Vreq mainly contains DMA information for this buf. In one vreq/buf, 139 * there could be more than one DMA window, each of which will be 140 * mapped to one blkif_request_t/ge_slot_t. Ge_slot_t contains all grant 141 * table entry information for this buf. The ge_slot_t for current DMA 142 * window is pointed to by v_gs in vreq. 143 * 144 * So, grant table entries will only be alloc'ed when the DMA window is 145 * about to be transferred via blkif_request_t to the ring buffer. And 146 * they will be freed right after the blkif_response_t is seen. By this 147 * means, we can make use of grant table entries more efficiently. 148 */ 149 typedef struct v_req { 150 list_node_t v_link; 151 int v_status; 152 buf_t *v_buf; 153 ddi_dma_handle_t v_dmahdl; 154 ddi_dma_cookie_t v_dmac; 155 uint_t v_ndmacs; 156 uint_t v_dmaw; 157 uint_t v_ndmaws; 158 uint_t v_nslots; 159 ge_slot_t *v_gs; 160 uint64_t v_blkno; 161 ddi_acc_handle_t v_align; 162 caddr_t v_abuf; 163 ddi_dma_handle_t v_memdmahdl; 164 uint8_t v_flush_diskcache; 165 } v_req_t; 166 167 /* 168 * Status set and checked in vreq->v_status by vreq_setup() 169 * 170 * These flags will help us to continue the vreq setup work from last failure 171 * point, instead of starting from scratch after each failure. 172 */ 173 #define VREQ_INIT 0x0 174 #define VREQ_INIT_DONE 0x1 175 #define VREQ_DMAHDL_ALLOCED 0x2 176 #define VREQ_MEMDMAHDL_ALLOCED 0x3 177 #define VREQ_DMAMEM_ALLOCED 0x4 178 #define VREQ_DMABUF_BOUND 0x5 179 #define VREQ_GS_ALLOCED 0x6 180 #define VREQ_DMAWIN_DONE 0x7 181 182 /* 183 * virtual block device per-instance softstate 184 */ 185 typedef struct xdf { 186 dev_info_t *xdf_dip; 187 ddi_iblock_cookie_t xdf_ibc; /* mutex iblock cookie */ 188 domid_t xdf_peer; /* otherend's dom ID */ 189 xendev_ring_t *xdf_xb_ring; /* I/O ring buffer */ 190 ddi_acc_handle_t xdf_xb_ring_hdl; /* access handler for ring buffer */ 191 list_t xdf_vreq_act; /* active vreq list */ 192 list_t xdf_gs_act; /* active grant table slot list */ 193 buf_t *xdf_f_act; /* active buf list head */ 194 buf_t *xdf_l_act; /* active buf list tail */ 195 enum xdf_state xdf_status; /* status of this virtual disk */ 196 ulong_t xdf_vd_open[OTYPCNT]; 197 ulong_t xdf_vd_lyropen[XDF_PEXT]; 198 ulong_t xdf_vd_exclopen; 199 kmutex_t xdf_iostat_lk; /* muxes lock for the iostat ptr */ 200 kmutex_t xdf_dev_lk; /* mutex lock for I/O path */ 201 kmutex_t xdf_cb_lk; /* mutex lock for event handling path */ 202 kcondvar_t xdf_dev_cv; /* cv used in I/O path */ 203 uint_t xdf_xdev_info; /* disk info from backend xenstore */ 204 diskaddr_t xdf_xdev_nblocks; /* total size in block */ 205 cmlb_geom_t xdf_pgeom; 206 kstat_t *xdf_xdev_iostat; 207 cmlb_handle_t xdf_vd_lbl; 208 ddi_softintr_t xdf_softintr_id; 209 timeout_id_t xdf_timeout_id; 210 struct gnttab_free_callback xdf_gnt_callback; 211 int xdf_feature_barrier; 212 int xdf_flush_supported; 213 int xdf_wce; 214 char *xdf_flush_mem; 215 char *xdf_cache_flush_block; 216 int xdf_evtchn; 217 #ifdef DEBUG 218 int xdf_dmacallback_num; 219 #endif 220 } xdf_t; 221 222 #define BP2VREQ(bp) ((v_req_t *)((bp)->av_back)) 223 224 /* 225 * VBD I/O requests must be aligned on a 512-byte boundary and specify 226 * a transfer size which is a mutiple of 512-bytes 227 */ 228 #define ALIGNED_XFER(bp) \ 229 ((((uintptr_t)((bp)->b_un.b_addr) & XB_BMASK) == 0) && \ 230 (((bp)->b_bcount & XB_BMASK) == 0)) 231 232 #define U_INVAL(u) (((u)->uio_loffset & (offset_t)(XB_BMASK)) || \ 233 ((u)->uio_iov->iov_len & (offset_t)(XB_BMASK))) 234 235 /* wrap pa_to_ma() for xdf to run in dom0 */ 236 #define PATOMA(addr) (DOMAIN_IS_INITDOMAIN(xen_info) ? addr : pa_to_ma(addr)) 237 238 #define XD_IS_RO(vbd) ((vbd)->xdf_xdev_info & VDISK_READONLY) 239 #define XD_IS_CD(vbd) ((vbd)->xdf_xdev_info & VDISK_CDROM) 240 #define XD_IS_RM(vbd) ((vbd)->xdf_xdev_info & VDISK_REMOVABLE) 241 #define IS_READ(bp) ((bp)->b_flags & B_READ) 242 #define IS_ERROR(bp) ((bp)->b_flags & B_ERROR) 243 244 #define XDF_UPDATE_IO_STAT(vdp, bp) \ 245 if ((vdp)->xdf_xdev_iostat != NULL) { \ 246 kstat_io_t *kip = KSTAT_IO_PTR((vdp)->xdf_xdev_iostat); \ 247 size_t n_done = (bp)->b_bcount - (bp)->b_resid; \ 248 if ((bp)->b_flags & B_READ) { \ 249 kip->reads++; \ 250 kip->nread += n_done; \ 251 } else { \ 252 kip->writes++; \ 253 kip->nwritten += n_done; \ 254 } \ 255 } 256 257 extern int xdfdebug; 258 #ifdef DEBUG 259 #define DPRINTF(flag, args) {if (xdfdebug & (flag)) prom_printf args; } 260 #define SETDMACBON(vbd) {(vbd)->xdf_dmacallback_num++; } 261 #define SETDMACBOFF(vbd) {(vbd)->xdf_dmacallback_num--; } 262 #define ISDMACBON(vbd) ((vbd)->xdf_dmacallback_num > 0) 263 #else 264 #define DPRINTF(flag, args) 265 #define SETDMACBON(vbd) 266 #define SETDMACBOFF(vbd) 267 #define ISDMACBON(vbd) 268 #endif /* DEBUG */ 269 270 #define DDI_DBG 0x1 271 #define DMA_DBG 0x2 272 #define INTR_DBG 0x8 273 #define IO_DBG 0x10 274 #define IOCTL_DBG 0x20 275 #define SUSRES_DBG 0x40 276 #define LBL_DBG 0x80 277 278 #if defined(XPV_HVM_DRIVER) 279 extern dev_info_t *xdf_hvm_hold(char *); 280 extern int xdf_hvm_connect(dev_info_t *); 281 extern int xdf_hvm_setpgeom(dev_info_t *, cmlb_geom_t *); 282 extern int xdf_kstat_create(dev_info_t *, char *, int); 283 extern void xdf_kstat_delete(dev_info_t *); 284 #endif /* XPV_HVM_DRIVER */ 285 286 #ifdef __cplusplus 287 } 288 #endif 289 290 #endif /* _SYS_XDF_H */ 291