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