git.oblomov.eu Git - linux-2.6/blob - arch/ppc64/kernel/mf.c

   1 /*
   2   * mf.c
   3   * Copyright (C) 2001 Troy D. Armstrong  IBM Corporation
   4   * Copyright (C) 2004-2005 Stephen Rothwell  IBM Corporation
   5   *
   6   * This modules exists as an interface between a Linux secondary partition
   7   * running on an iSeries and the primary partition's Virtual Service
   8   * Processor (VSP) object.  The VSP has final authority over powering on/off
   9   * all partitions in the iSeries.  It also provides miscellaneous low-level
  10   * machine facility type operations.
  11   *
  12   *
  13   * This program is free software; you can redistribute it and/or modify
  14   * it under the terms of the GNU General Public License as published by
  15   * the Free Software Foundation; either version 2 of the License, or
  16   * (at your option) any later version.
  17   *
  18   * This program is distributed in the hope that it will be useful,
  19   * but WITHOUT ANY WARRANTY; without even the implied warranty of
  20   * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  21   * GNU General Public License for more details.
  22   *
  23   * You should have received a copy of the GNU General Public License
  24   * along with this program; if not, write to the Free Software
  25   * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  26   */
  27
  28 #include <linux/types.h>
  29 #include <linux/errno.h>
  30 #include <linux/kernel.h>
  31 #include <linux/init.h>
  32 #include <linux/completion.h>
  33 #include <linux/delay.h>
  34 #include <linux/dma-mapping.h>
  35 #include <linux/bcd.h>
  36
  37 #include <asm/time.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/paca.h>
  40 #include <asm/iSeries/vio.h>
  41 #include <asm/iSeries/mf.h>
  42 #include <asm/iSeries/HvLpConfig.h>
  43 #include <asm/iSeries/ItSpCommArea.h>
  44 #include <asm/iSeries/ItLpQueue.h>
  45
  46 /*
  47  * This is the structure layout for the Machine Facilites LPAR event
  48  * flows.
  49  */
  50 struct vsp_cmd_data {
  51         u64 token;
  52         u16 cmd;
  53         HvLpIndex lp_index;
  54         u8 result_code;
  55         u32 reserved;
  56         union {
  57                 u64 state;      /* GetStateOut */
  58                 u64 ipl_type;   /* GetIplTypeOut, Function02SelectIplTypeIn */
  59                 u64 ipl_mode;   /* GetIplModeOut, Function02SelectIplModeIn */
  60                 u64 page[4];    /* GetSrcHistoryIn */
  61                 u64 flag;       /* GetAutoIplWhenPrimaryIplsOut,
  62                                    SetAutoIplWhenPrimaryIplsIn,
  63                                    WhiteButtonPowerOffIn,
  64                                    Function08FastPowerOffIn,
  65                                    IsSpcnRackPowerIncompleteOut */
  66                 struct {
  67                         u64 token;
  68                         u64 address_type;
  69                         u64 side;
  70                         u32 length;
  71                         u32 offset;
  72                 } kern;         /* SetKernelImageIn, GetKernelImageIn,
  73                                    SetKernelCmdLineIn, GetKernelCmdLineIn */
  74                 u32 length_out; /* GetKernelImageOut, GetKernelCmdLineOut */
  75                 u8 reserved[80];
  76         } sub_data;
  77 };
  78
  79 struct vsp_rsp_data {
  80         struct completion com;
  81         struct vsp_cmd_data *response;
  82 };
  83
  84 struct alloc_data {
  85         u16 size;
  86         u16 type;
  87         u32 count;
  88         u16 reserved1;
  89         u8 reserved2;
  90         HvLpIndex target_lp;
  91 };
  92
  93 struct ce_msg_data;
  94
  95 typedef void (*ce_msg_comp_hdlr)(void *token, struct ce_msg_data *vsp_cmd_rsp);
  96
  97 struct ce_msg_comp_data {
  98         ce_msg_comp_hdlr handler;
  99         void *token;
 100 };
 101
 102 struct ce_msg_data {
 103         u8 ce_msg[12];
 104         char reserved[4];
 105         struct ce_msg_comp_data *completion;
 106 };
 107
 108 struct io_mf_lp_event {
 109         struct HvLpEvent hp_lp_event;
 110         u16 subtype_result_code;
 111         u16 reserved1;
 112         u32 reserved2;
 113         union {
 114                 struct alloc_data alloc;
 115                 struct ce_msg_data ce_msg;
 116                 struct vsp_cmd_data vsp_cmd;
 117         } data;
 118 };
 119
 120 #define subtype_data(a, b, c, d)        \
 121                 (((a) << 24) + ((b) << 16) + ((c) << 8) + (d))
 122
 123 /*
 124  * All outgoing event traffic is kept on a FIFO queue.  The first
 125  * pointer points to the one that is outstanding, and all new
 126  * requests get stuck on the end.  Also, we keep a certain number of
 127  * preallocated pending events so that we can operate very early in
 128  * the boot up sequence (before kmalloc is ready).
 129  */
 130 struct pending_event {
 131         struct pending_event *next;
 132         struct io_mf_lp_event event;
 133         MFCompleteHandler hdlr;
 134         char dma_data[72];
 135         unsigned dma_data_length;
 136         unsigned remote_address;
 137 };
 138 static spinlock_t pending_event_spinlock;
 139 static struct pending_event *pending_event_head;
 140 static struct pending_event *pending_event_tail;
 141 static struct pending_event *pending_event_avail;
 142 static struct pending_event pending_event_prealloc[16];
 143
 144 /*
 145  * Put a pending event onto the available queue, so it can get reused.
 146  * Attention! You must have the pending_event_spinlock before calling!
 147  */
 148 static void free_pending_event(struct pending_event *ev)
 149 {
 150         if (ev != NULL) {
 151                 ev->next = pending_event_avail;
 152                 pending_event_avail = ev;
 153         }
 154 }
 155
 156 /*
 157  * Enqueue the outbound event onto the stack.  If the queue was
 158  * empty to begin with, we must also issue it via the Hypervisor
 159  * interface.  There is a section of code below that will touch
 160  * the first stack pointer without the protection of the pending_event_spinlock.
 161  * This is OK, because we know that nobody else will be modifying
 162  * the first pointer when we do this.
 163  */
 164 static int signal_event(struct pending_event *ev)
 165 {
 166         int rc = 0;
 167         unsigned long flags;
 168         int go = 1;
 169         struct pending_event *ev1;
 170         HvLpEvent_Rc hv_rc;
 171
 172         /* enqueue the event */
 173         if (ev != NULL) {
 174                 ev->next = NULL;
 175                 spin_lock_irqsave(&pending_event_spinlock, flags);
 176                 if (pending_event_head == NULL)
 177                         pending_event_head = ev;
 178                 else {
 179                         go = 0;
 180                         pending_event_tail->next = ev;
 181                 }
 182                 pending_event_tail = ev;
 183                 spin_unlock_irqrestore(&pending_event_spinlock, flags);
 184         }
 185
 186         /* send the event */
 187         while (go) {
 188                 go = 0;
 189
 190                 /* any DMA data to send beforehand? */
 191                 if (pending_event_head->dma_data_length > 0)
 192                         HvCallEvent_dmaToSp(pending_event_head->dma_data,
 193                                         pending_event_head->remote_address,
 194                                         pending_event_head->dma_data_length,
 195                                         HvLpDma_Direction_LocalToRemote);
 196
 197                 hv_rc = HvCallEvent_signalLpEvent(
 198                                 &pending_event_head->event.hp_lp_event);
 199                 if (hv_rc != HvLpEvent_Rc_Good) {
 200                         printk(KERN_ERR "mf.c: HvCallEvent_signalLpEvent() "
 201                                         "failed with %d\n", (int)hv_rc);
 202
 203                         spin_lock_irqsave(&pending_event_spinlock, flags);
 204                         ev1 = pending_event_head;
 205                         pending_event_head = pending_event_head->next;
 206                         if (pending_event_head != NULL)
 207                                 go = 1;
 208                         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 209
 210                         if (ev1 == ev)
 211                                 rc = -EIO;
 212                         else if (ev1->hdlr != NULL)
 213                                 (*ev1->hdlr)((void *)ev1->event.hp_lp_event.xCorrelationToken, -EIO);
 214
 215                         spin_lock_irqsave(&pending_event_spinlock, flags);
 216                         free_pending_event(ev1);
 217                         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 218                 }
 219         }
 220
 221         return rc;
 222 }
 223
 224 /*
 225  * Allocate a new pending_event structure, and initialize it.
 226  */
 227 static struct pending_event *new_pending_event(void)
 228 {
 229         struct pending_event *ev = NULL;
 230         HvLpIndex primary_lp = HvLpConfig_getPrimaryLpIndex();
 231         unsigned long flags;
 232         struct HvLpEvent *hev;
 233
 234         spin_lock_irqsave(&pending_event_spinlock, flags);
 235         if (pending_event_avail != NULL) {
 236                 ev = pending_event_avail;
 237                 pending_event_avail = pending_event_avail->next;
 238         }
 239         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 240         if (ev == NULL) {
 241                 ev = kmalloc(sizeof(struct pending_event), GFP_ATOMIC);
 242                 if (ev == NULL) {
 243                         printk(KERN_ERR "mf.c: unable to kmalloc %ld bytes\n",
 244                                         sizeof(struct pending_event));
 245                         return NULL;
 246                 }
 247         }
 248         memset(ev, 0, sizeof(struct pending_event));
 249         hev = &ev->event.hp_lp_event;
 250         hev->xFlags.xValid = 1;
 251         hev->xFlags.xAckType = HvLpEvent_AckType_ImmediateAck;
 252         hev->xFlags.xAckInd = HvLpEvent_AckInd_DoAck;
 253         hev->xFlags.xFunction = HvLpEvent_Function_Int;
 254         hev->xType = HvLpEvent_Type_MachineFac;
 255         hev->xSourceLp = HvLpConfig_getLpIndex();
 256         hev->xTargetLp = primary_lp;
 257         hev->xSizeMinus1 = sizeof(ev->event) - 1;
 258         hev->xRc = HvLpEvent_Rc_Good;
 259         hev->xSourceInstanceId = HvCallEvent_getSourceLpInstanceId(primary_lp,
 260                         HvLpEvent_Type_MachineFac);
 261         hev->xTargetInstanceId = HvCallEvent_getTargetLpInstanceId(primary_lp,
 262                         HvLpEvent_Type_MachineFac);
 263
 264         return ev;
 265 }
 266
 267 static int signal_vsp_instruction(struct vsp_cmd_data *vsp_cmd)
 268 {
 269         struct pending_event *ev = new_pending_event();
 270         int rc;
 271         struct vsp_rsp_data response;
 272
 273         if (ev == NULL)
 274                 return -ENOMEM;
 275
 276         init_completion(&response.com);
 277         response.response = vsp_cmd;
 278         ev->event.hp_lp_event.xSubtype = 6;
 279         ev->event.hp_lp_event.x.xSubtypeData =
 280                 subtype_data('M', 'F',  'V',  'I');
 281         ev->event.data.vsp_cmd.token = (u64)&response;
 282         ev->event.data.vsp_cmd.cmd = vsp_cmd->cmd;
 283         ev->event.data.vsp_cmd.lp_index = HvLpConfig_getLpIndex();
 284         ev->event.data.vsp_cmd.result_code = 0xFF;
 285         ev->event.data.vsp_cmd.reserved = 0;
 286         memcpy(&(ev->event.data.vsp_cmd.sub_data),
 287                         &(vsp_cmd->sub_data), sizeof(vsp_cmd->sub_data));
 288         mb();
 289
 290         rc = signal_event(ev);
 291         if (rc == 0)
 292                 wait_for_completion(&response.com);
 293         return rc;
 294 }
 295
 296
 297 /*
 298  * Send a 12-byte CE message to the primary partition VSP object
 299  */
 300 static int signal_ce_msg(char *ce_msg, struct ce_msg_comp_data *completion)
 301 {
 302         struct pending_event *ev = new_pending_event();
 303
 304         if (ev == NULL)
 305                 return -ENOMEM;
 306
 307         ev->event.hp_lp_event.xSubtype = 0;
 308         ev->event.hp_lp_event.x.xSubtypeData =
 309                 subtype_data('M',  'F',  'C',  'E');
 310         memcpy(ev->event.data.ce_msg.ce_msg, ce_msg, 12);
 311         ev->event.data.ce_msg.completion = completion;
 312         return signal_event(ev);
 313 }
 314
 315 /*
 316  * Send a 12-byte CE message (with no data) to the primary partition VSP object
 317  */
 318 static int signal_ce_msg_simple(u8 ce_op, struct ce_msg_comp_data *completion)
 319 {
 320         u8 ce_msg[12];
 321
 322         memset(ce_msg, 0, sizeof(ce_msg));
 323         ce_msg[3] = ce_op;
 324         return signal_ce_msg(ce_msg, completion);
 325 }
 326
 327 /*
 328  * Send a 12-byte CE message and DMA data to the primary partition VSP object
 329  */
 330 static int dma_and_signal_ce_msg(char *ce_msg,
 331                 struct ce_msg_comp_data *completion, void *dma_data,
 332                 unsigned dma_data_length, unsigned remote_address)
 333 {
 334         struct pending_event *ev = new_pending_event();
 335
 336         if (ev == NULL)
 337                 return -ENOMEM;
 338
 339         ev->event.hp_lp_event.xSubtype = 0;
 340         ev->event.hp_lp_event.x.xSubtypeData =
 341                 subtype_data('M', 'F', 'C', 'E');
 342         memcpy(ev->event.data.ce_msg.ce_msg, ce_msg, 12);
 343         ev->event.data.ce_msg.completion = completion;
 344         memcpy(ev->dma_data, dma_data, dma_data_length);
 345         ev->dma_data_length = dma_data_length;
 346         ev->remote_address = remote_address;
 347         return signal_event(ev);
 348 }
 349
 350 /*
 351  * Initiate a nice (hopefully) shutdown of Linux.  We simply are
 352  * going to try and send the init process a SIGINT signal.  If
 353  * this fails (why?), we'll simply force it off in a not-so-nice
 354  * manner.
 355  */
 356 static int shutdown(void)
 357 {
 358         int rc = kill_proc(1, SIGINT, 1);
 359
 360         if (rc) {
 361                 printk(KERN_ALERT "mf.c: SIGINT to init failed (%d), "
 362                                 "hard shutdown commencing\n", rc);
 363                 mf_power_off();
 364         } else
 365                 printk(KERN_INFO "mf.c: init has been successfully notified "
 366                                 "to proceed with shutdown\n");
 367         return rc;
 368 }
 369
 370 /*
 371  * The primary partition VSP object is sending us a new
 372  * event flow.  Handle it...
 373  */
 374 static void handle_int(struct io_mf_lp_event *event)
 375 {
 376         struct ce_msg_data *ce_msg_data;
 377         struct ce_msg_data *pce_msg_data;
 378         unsigned long flags;
 379         struct pending_event *pev;
 380
 381         /* ack the interrupt */
 382         event->hp_lp_event.xRc = HvLpEvent_Rc_Good;
 383         HvCallEvent_ackLpEvent(&event->hp_lp_event);
 384
 385         /* process interrupt */
 386         switch (event->hp_lp_event.xSubtype) {
 387         case 0: /* CE message */
 388                 ce_msg_data = &event->data.ce_msg;
 389                 switch (ce_msg_data->ce_msg[3]) {
 390                 case 0x5B:      /* power control notification */
 391                         if ((ce_msg_data->ce_msg[5] & 0x20) != 0) {
 392                                 printk(KERN_INFO "mf.c: Commencing partition shutdown\n");
 393                                 if (shutdown() == 0)
 394                                         signal_ce_msg_simple(0xDB, NULL);
 395                         }
 396                         break;
 397                 case 0xC0:      /* get time */
 398                         spin_lock_irqsave(&pending_event_spinlock, flags);
 399                         pev = pending_event_head;
 400                         if (pev != NULL)
 401                                 pending_event_head = pending_event_head->next;
 402                         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 403                         if (pev == NULL)
 404                                 break;
 405                         pce_msg_data = &pev->event.data.ce_msg;
 406                         if (pce_msg_data->ce_msg[3] != 0x40)
 407                                 break;
 408                         if (pce_msg_data->completion != NULL) {
 409                                 ce_msg_comp_hdlr handler =
 410                                         pce_msg_data->completion->handler;
 411                                 void *token = pce_msg_data->completion->token;
 412
 413                                 if (handler != NULL)
 414                                         (*handler)(token, ce_msg_data);
 415                         }
 416                         spin_lock_irqsave(&pending_event_spinlock, flags);
 417                         free_pending_event(pev);
 418                         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 419                         /* send next waiting event */
 420                         if (pending_event_head != NULL)
 421                                 signal_event(NULL);
 422                         break;
 423                 }
 424                 break;
 425         case 1: /* IT sys shutdown */
 426                 printk(KERN_INFO "mf.c: Commencing system shutdown\n");
 427                 shutdown();
 428                 break;
 429         }
 430 }
 431
 432 /*
 433  * The primary partition VSP object is acknowledging the receipt
 434  * of a flow we sent to them.  If there are other flows queued
 435  * up, we must send another one now...
 436  */
 437 static void handle_ack(struct io_mf_lp_event *event)
 438 {
 439         unsigned long flags;
 440         struct pending_event *two = NULL;
 441         unsigned long free_it = 0;
 442         struct ce_msg_data *ce_msg_data;
 443         struct ce_msg_data *pce_msg_data;
 444         struct vsp_rsp_data *rsp;
 445
 446         /* handle current event */
 447         if (pending_event_head == NULL) {
 448                 printk(KERN_ERR "mf.c: stack empty for receiving ack\n");
 449                 return;
 450         }
 451
 452         switch (event->hp_lp_event.xSubtype) {
 453         case 0:     /* CE msg */
 454                 ce_msg_data = &event->data.ce_msg;
 455                 if (ce_msg_data->ce_msg[3] != 0x40) {
 456                         free_it = 1;
 457                         break;
 458                 }
 459                 if (ce_msg_data->ce_msg[2] == 0)
 460                         break;
 461                 free_it = 1;
 462                 pce_msg_data = &pending_event_head->event.data.ce_msg;
 463                 if (pce_msg_data->completion != NULL) {
 464                         ce_msg_comp_hdlr handler =
 465                                 pce_msg_data->completion->handler;
 466                         void *token = pce_msg_data->completion->token;
 467
 468                         if (handler != NULL)
 469                                 (*handler)(token, ce_msg_data);
 470                 }
 471                 break;
 472         case 4: /* allocate */
 473         case 5: /* deallocate */
 474                 if (pending_event_head->hdlr != NULL)
 475                         (*pending_event_head->hdlr)((void *)event->hp_lp_event.xCorrelationToken, event->data.alloc.count);
 476                 free_it = 1;
 477                 break;
 478         case 6:
 479                 free_it = 1;
 480                 rsp = (struct vsp_rsp_data *)event->data.vsp_cmd.token;
 481                 if (rsp == NULL) {
 482                         printk(KERN_ERR "mf.c: no rsp\n");
 483                         break;
 484                 }
 485                 if (rsp->response != NULL)
 486                         memcpy(rsp->response, &event->data.vsp_cmd,
 487                                         sizeof(event->data.vsp_cmd));
 488                 complete(&rsp->com);
 489                 break;
 490         }
 491
 492         /* remove from queue */
 493         spin_lock_irqsave(&pending_event_spinlock, flags);
 494         if ((pending_event_head != NULL) && (free_it == 1)) {
 495                 struct pending_event *oldHead = pending_event_head;
 496
 497                 pending_event_head = pending_event_head->next;
 498                 two = pending_event_head;
 499                 free_pending_event(oldHead);
 500         }
 501         spin_unlock_irqrestore(&pending_event_spinlock, flags);
 502
 503         /* send next waiting event */
 504         if (two != NULL)
 505                 signal_event(NULL);
 506 }
 507
 508 /*
 509  * This is the generic event handler we are registering with
 510  * the Hypervisor.  Ensure the flows are for us, and then
 511  * parse it enough to know if it is an interrupt or an
 512  * acknowledge.
 513  */
 514 static void hv_handler(struct HvLpEvent *event, struct pt_regs *regs)
 515 {
 516         if ((event != NULL) && (event->xType == HvLpEvent_Type_MachineFac)) {
 517                 switch(event->xFlags.xFunction) {
 518                 case HvLpEvent_Function_Ack:
 519                         handle_ack((struct io_mf_lp_event *)event);
 520                         break;
 521                 case HvLpEvent_Function_Int:
 522                         handle_int((struct io_mf_lp_event *)event);
 523                         break;
 524                 default:
 525                         printk(KERN_ERR "mf.c: non ack/int event received\n");
 526                         break;
 527                 }
 528         } else
 529                 printk(KERN_ERR "mf.c: alien event received\n");
 530 }
 531
 532 /*
 533  * Global kernel interface to allocate and seed events into the
 534  * Hypervisor.
 535  */
 536 void mf_allocate_lp_events(HvLpIndex target_lp, HvLpEvent_Type type,
 537                 unsigned size, unsigned count, MFCompleteHandler hdlr,
 538                 void *user_token)
 539 {
 540         struct pending_event *ev = new_pending_event();
 541         int rc;
 542
 543         if (ev == NULL) {
 544                 rc = -ENOMEM;
 545         } else {
 546                 ev->event.hp_lp_event.xSubtype = 4;
 547                 ev->event.hp_lp_event.xCorrelationToken = (u64)user_token;
 548                 ev->event.hp_lp_event.x.xSubtypeData =
 549                         subtype_data('M', 'F', 'M', 'A');
 550                 ev->event.data.alloc.target_lp = target_lp;
 551                 ev->event.data.alloc.type = type;
 552                 ev->event.data.alloc.size = size;
 553                 ev->event.data.alloc.count = count;
 554                 ev->hdlr = hdlr;
 555                 rc = signal_event(ev);
 556         }
 557         if ((rc != 0) && (hdlr != NULL))
 558                 (*hdlr)(user_token, rc);
 559 }
 560 EXPORT_SYMBOL(mf_allocate_lp_events);
 561
 562 /*
 563  * Global kernel interface to unseed and deallocate events already in
 564  * Hypervisor.
 565  */
 566 void mf_deallocate_lp_events(HvLpIndex target_lp, HvLpEvent_Type type,
 567                 unsigned count, MFCompleteHandler hdlr, void *user_token)
 568 {
 569         struct pending_event *ev = new_pending_event();
 570         int rc;
 571
 572         if (ev == NULL)
 573                 rc = -ENOMEM;
 574         else {
 575                 ev->event.hp_lp_event.xSubtype = 5;
 576                 ev->event.hp_lp_event.xCorrelationToken = (u64)user_token;
 577                 ev->event.hp_lp_event.x.xSubtypeData =
 578                         subtype_data('M', 'F', 'M', 'D');
 579                 ev->event.data.alloc.target_lp = target_lp;
 580                 ev->event.data.alloc.type = type;
 581                 ev->event.data.alloc.count = count;
 582                 ev->hdlr = hdlr;
 583                 rc = signal_event(ev);
 584         }
 585         if ((rc != 0) && (hdlr != NULL))
 586                 (*hdlr)(user_token, rc);
 587 }
 588 EXPORT_SYMBOL(mf_deallocate_lp_events);
 589
 590 /*
 591  * Global kernel interface to tell the VSP object in the primary
 592  * partition to power this partition off.
 593  */
 594 void mf_power_off(void)
 595 {
 596         printk(KERN_INFO "mf.c: Down it goes...\n");
 597         signal_ce_msg_simple(0x4d, NULL);
 598         for (;;)
 599                 ;
 600 }
 601
 602 /*
 603  * Global kernel interface to tell the VSP object in the primary
 604  * partition to reboot this partition.
 605  */
 606 void mf_reboot(void)
 607 {
 608         printk(KERN_INFO "mf.c: Preparing to bounce...\n");
 609         signal_ce_msg_simple(0x4e, NULL);
 610         for (;;)
 611                 ;
 612 }
 613
 614 /*
 615  * Display a single word SRC onto the VSP control panel.
 616  */
 617 void mf_display_src(u32 word)
 618 {
 619         u8 ce[12];
 620
 621         memset(ce, 0, sizeof(ce));
 622         ce[3] = 0x4a;
 623         ce[7] = 0x01;
 624         ce[8] = word >> 24;
 625         ce[9] = word >> 16;
 626         ce[10] = word >> 8;
 627         ce[11] = word;
 628         signal_ce_msg(ce, NULL);
 629 }
 630
 631 /*
 632  * Display a single word SRC of the form "PROGXXXX" on the VSP control panel.
 633  */
 634 void mf_display_progress(u16 value)
 635 {
 636         u8 ce[12];
 637         u8 src[72];
 638
 639         memcpy(ce, "\x00\x00\x04\x4A\x00\x00\x00\x48\x00\x00\x00\x00", 12);
 640         memcpy(src, "\x01\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x00"
 641                 "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"
 642                 "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"
 643                 "\x00\x00\x00\x00PROGxxxx                        ",
 644                 72);
 645         src[6] = value >> 8;
 646         src[7] = value & 255;
 647         src[44] = "0123456789ABCDEF"[(value >> 12) & 15];
 648         src[45] = "0123456789ABCDEF"[(value >> 8) & 15];
 649         src[46] = "0123456789ABCDEF"[(value >> 4) & 15];
 650         src[47] = "0123456789ABCDEF"[value & 15];
 651         dma_and_signal_ce_msg(ce, NULL, src, sizeof(src), 9 * 64 * 1024);
 652 }
 653
 654 /*
 655  * Clear the VSP control panel.  Used to "erase" an SRC that was
 656  * previously displayed.
 657  */
 658 void mf_clear_src(void)
 659 {
 660         signal_ce_msg_simple(0x4b, NULL);
 661 }
 662
 663 /*
 664  * Initialization code here.
 665  */
 666 void mf_init(void)
 667 {
 668         int i;
 669
 670         /* initialize */
 671         spin_lock_init(&pending_event_spinlock);
 672         for (i = 0;
 673              i < sizeof(pending_event_prealloc) / sizeof(*pending_event_prealloc);
 674              ++i)
 675                 free_pending_event(&pending_event_prealloc[i]);
 676         HvLpEvent_registerHandler(HvLpEvent_Type_MachineFac, &hv_handler);
 677
 678         /* virtual continue ack */
 679         signal_ce_msg_simple(0x57, NULL);
 680
 681         /* initialization complete */
 682         printk(KERN_NOTICE "mf.c: iSeries Linux LPAR Machine Facilities "
 683                         "initialized\n");
 684 }
 685
 686 struct rtc_time_data {
 687         struct completion com;
 688         struct ce_msg_data ce_msg;
 689         int rc;
 690 };
 691
 692 static void get_rtc_time_complete(void *token, struct ce_msg_data *ce_msg)
 693 {
 694         struct rtc_time_data *rtc = token;
 695
 696         memcpy(&rtc->ce_msg, ce_msg, sizeof(rtc->ce_msg));
 697         rtc->rc = 0;
 698         complete(&rtc->com);
 699 }
 700
 701 static int rtc_set_tm(int rc, u8 *ce_msg, struct rtc_time *tm)
 702 {
 703         tm->tm_wday = 0;
 704         tm->tm_yday = 0;
 705         tm->tm_isdst = 0;
 706         if (rc) {
 707                 tm->tm_sec = 0;
 708                 tm->tm_min = 0;
 709                 tm->tm_hour = 0;
 710                 tm->tm_mday = 15;
 711                 tm->tm_mon = 5;
 712                 tm->tm_year = 52;
 713                 return rc;
 714         }
 715
 716         if ((ce_msg[2] == 0xa9) ||
 717             (ce_msg[2] == 0xaf)) {
 718                 /* TOD clock is not set */
 719                 tm->tm_sec = 1;
 720                 tm->tm_min = 1;
 721                 tm->tm_hour = 1;
 722                 tm->tm_mday = 10;
 723                 tm->tm_mon = 8;
 724                 tm->tm_year = 71;
 725                 mf_set_rtc(tm);
 726         }
 727         {
 728                 u8 year = ce_msg[5];
 729                 u8 sec = ce_msg[6];
 730                 u8 min = ce_msg[7];
 731                 u8 hour = ce_msg[8];
 732                 u8 day = ce_msg[10];
 733                 u8 mon = ce_msg[11];
 734
 735                 BCD_TO_BIN(sec);
 736                 BCD_TO_BIN(min);
 737                 BCD_TO_BIN(hour);
 738                 BCD_TO_BIN(day);
 739                 BCD_TO_BIN(mon);
 740                 BCD_TO_BIN(year);
 741
 742                 if (year <= 69)
 743                         year += 100;
 744
 745                 tm->tm_sec = sec;
 746                 tm->tm_min = min;
 747                 tm->tm_hour = hour;
 748                 tm->tm_mday = day;
 749                 tm->tm_mon = mon;
 750                 tm->tm_year = year;
 751         }
 752
 753         return 0;
 754 }
 755
 756 int mf_get_rtc(struct rtc_time *tm)
 757 {
 758         struct ce_msg_comp_data ce_complete;
 759         struct rtc_time_data rtc_data;
 760         int rc;
 761
 762         memset(&ce_complete, 0, sizeof(ce_complete));
 763         memset(&rtc_data, 0, sizeof(rtc_data));
 764         init_completion(&rtc_data.com);
 765         ce_complete.handler = &get_rtc_time_complete;
 766         ce_complete.token = &rtc_data;
 767         rc = signal_ce_msg_simple(0x40, &ce_complete);
 768         if (rc)
 769                 return rc;
 770         wait_for_completion(&rtc_data.com);
 771         return rtc_set_tm(rtc_data.rc, rtc_data.ce_msg.ce_msg, tm);
 772 }
 773
 774 struct boot_rtc_time_data {
 775         int busy;
 776         struct ce_msg_data ce_msg;
 777         int rc;
 778 };
 779
 780 static void get_boot_rtc_time_complete(void *token, struct ce_msg_data *ce_msg)
 781 {
 782         struct boot_rtc_time_data *rtc = token;
 783
 784         memcpy(&rtc->ce_msg, ce_msg, sizeof(rtc->ce_msg));
 785         rtc->rc = 0;
 786         rtc->busy = 0;
 787 }
 788
 789 int mf_get_boot_rtc(struct rtc_time *tm)
 790 {
 791         struct ce_msg_comp_data ce_complete;
 792         struct boot_rtc_time_data rtc_data;
 793         int rc;
 794
 795         memset(&ce_complete, 0, sizeof(ce_complete));
 796         memset(&rtc_data, 0, sizeof(rtc_data));
 797         rtc_data.busy = 1;
 798         ce_complete.handler = &get_boot_rtc_time_complete;
 799         ce_complete.token = &rtc_data;
 800         rc = signal_ce_msg_simple(0x40, &ce_complete);
 801         if (rc)
 802                 return rc;
 803         /* We need to poll here as we are not yet taking interrupts */
 804         while (rtc_data.busy) {
 805                 extern unsigned long lpevent_count;
 806                 struct ItLpQueue *lpq = get_paca()->lpqueue_ptr;
 807                 if (lpq && ItLpQueue_isLpIntPending(lpq))
 808                         lpevent_count += ItLpQueue_process(lpq, NULL);
 809         }
 810         return rtc_set_tm(rtc_data.rc, rtc_data.ce_msg.ce_msg, tm);
 811 }
 812
 813 int mf_set_rtc(struct rtc_time *tm)
 814 {
 815         char ce_time[12];
 816         u8 day, mon, hour, min, sec, y1, y2;
 817         unsigned year;
 818
 819         year = 1900 + tm->tm_year;
 820         y1 = year / 100;
 821         y2 = year % 100;
 822
 823         sec = tm->tm_sec;
 824         min = tm->tm_min;
 825         hour = tm->tm_hour;
 826         day = tm->tm_mday;
 827         mon = tm->tm_mon + 1;
 828
 829         BIN_TO_BCD(sec);
 830         BIN_TO_BCD(min);
 831         BIN_TO_BCD(hour);
 832         BIN_TO_BCD(mon);
 833         BIN_TO_BCD(day);
 834         BIN_TO_BCD(y1);
 835         BIN_TO_BCD(y2);
 836
 837         memset(ce_time, 0, sizeof(ce_time));
 838         ce_time[3] = 0x41;
 839         ce_time[4] = y1;
 840         ce_time[5] = y2;
 841         ce_time[6] = sec;
 842         ce_time[7] = min;
 843         ce_time[8] = hour;
 844         ce_time[10] = day;
 845         ce_time[11] = mon;
 846
 847         return signal_ce_msg(ce_time, NULL);
 848 }
 849
 850 #ifdef CONFIG_PROC_FS
 851
 852 static int proc_mf_dump_cmdline(char *page, char **start, off_t off,
 853                 int count, int *eof, void *data)
 854 {
 855         int len;
 856         char *p;
 857         struct vsp_cmd_data vsp_cmd;
 858         int rc;
 859         dma_addr_t dma_addr;
 860
 861         /* The HV appears to return no more than 256 bytes of command line */
 862         if (off >= 256)
 863                 return 0;
 864         if ((off + count) > 256)
 865                 count = 256 - off;
 866
 867         dma_addr = dma_map_single(iSeries_vio_dev, page, off + count,
 868                         DMA_FROM_DEVICE);
 869         if (dma_mapping_error(dma_addr))
 870                 return -ENOMEM;
 871         memset(page, 0, off + count);
 872         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
 873         vsp_cmd.cmd = 33;
 874         vsp_cmd.sub_data.kern.token = dma_addr;
 875         vsp_cmd.sub_data.kern.address_type = HvLpDma_AddressType_TceIndex;
 876         vsp_cmd.sub_data.kern.side = (u64)data;
 877         vsp_cmd.sub_data.kern.length = off + count;
 878         mb();
 879         rc = signal_vsp_instruction(&vsp_cmd);
 880         dma_unmap_single(iSeries_vio_dev, dma_addr, off + count,
 881                         DMA_FROM_DEVICE);
 882         if (rc)
 883                 return rc;
 884         if (vsp_cmd.result_code != 0)
 885                 return -ENOMEM;
 886         p = page;
 887         len = 0;
 888         while (len < (off + count)) {
 889                 if ((*p == '\0') || (*p == '\n')) {
 890                         if (*p == '\0')
 891                                 *p = '\n';
 892                         p++;
 893                         len++;
 894                         *eof = 1;
 895                         break;
 896                 }
 897                 p++;
 898                 len++;
 899         }
 900
 901         if (len < off) {
 902                 *eof = 1;
 903                 len = 0;
 904         }
 905         return len;
 906 }
 907
 908 #if 0
 909 static int mf_getVmlinuxChunk(char *buffer, int *size, int offset, u64 side)
 910 {
 911         struct vsp_cmd_data vsp_cmd;
 912         int rc;
 913         int len = *size;
 914         dma_addr_t dma_addr;
 915
 916         dma_addr = dma_map_single(iSeries_vio_dev, buffer, len,
 917                         DMA_FROM_DEVICE);
 918         memset(buffer, 0, len);
 919         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
 920         vsp_cmd.cmd = 32;
 921         vsp_cmd.sub_data.kern.token = dma_addr;
 922         vsp_cmd.sub_data.kern.address_type = HvLpDma_AddressType_TceIndex;
 923         vsp_cmd.sub_data.kern.side = side;
 924         vsp_cmd.sub_data.kern.offset = offset;
 925         vsp_cmd.sub_data.kern.length = len;
 926         mb();
 927         rc = signal_vsp_instruction(&vsp_cmd);
 928         if (rc == 0) {
 929                 if (vsp_cmd.result_code == 0)
 930                         *size = vsp_cmd.sub_data.length_out;
 931                 else
 932                         rc = -ENOMEM;
 933         }
 934
 935         dma_unmap_single(iSeries_vio_dev, dma_addr, len, DMA_FROM_DEVICE);
 936
 937         return rc;
 938 }
 939
 940 static int proc_mf_dump_vmlinux(char *page, char **start, off_t off,
 941                 int count, int *eof, void *data)
 942 {
 943         int sizeToGet = count;
 944
 945         if (!capable(CAP_SYS_ADMIN))
 946                 return -EACCES;
 947
 948         if (mf_getVmlinuxChunk(page, &sizeToGet, off, (u64)data) == 0) {
 949                 if (sizeToGet != 0) {
 950                         *start = page + off;
 951                         return sizeToGet;
 952                 }
 953                 *eof = 1;
 954                 return 0;
 955         }
 956         *eof = 1;
 957         return 0;
 958 }
 959 #endif
 960
 961 static int proc_mf_dump_side(char *page, char **start, off_t off,
 962                 int count, int *eof, void *data)
 963 {
 964         int len;
 965         char mf_current_side = ' ';
 966         struct vsp_cmd_data vsp_cmd;
 967
 968         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
 969         vsp_cmd.cmd = 2;
 970         vsp_cmd.sub_data.ipl_type = 0;
 971         mb();
 972
 973         if (signal_vsp_instruction(&vsp_cmd) == 0) {
 974                 if (vsp_cmd.result_code == 0) {
 975                         switch (vsp_cmd.sub_data.ipl_type) {
 976                         case 0: mf_current_side = 'A';
 977                                 break;
 978                         case 1: mf_current_side = 'B';
 979                                 break;
 980                         case 2: mf_current_side = 'C';
 981                                 break;
 982                         default:        mf_current_side = 'D';
 983                                 break;
 984                         }
 985                 }
 986         }
 987
 988         len = sprintf(page, "%c\n", mf_current_side);
 989
 990         if (len <= (off + count))
 991                 *eof = 1;
 992         *start = page + off;
 993         len -= off;
 994         if (len > count)
 995                 len = count;
 996         if (len < 0)
 997                 len = 0;
 998         return len;
 999 }
1000
1001 static int proc_mf_change_side(struct file *file, const char __user *buffer,
1002                 unsigned long count, void *data)
1003 {
1004         char side;
1005         u64 newSide;
1006         struct vsp_cmd_data vsp_cmd;
1007
1008         if (!capable(CAP_SYS_ADMIN))
1009                 return -EACCES;
1010
1011         if (count == 0)
1012                 return 0;
1013
1014         if (get_user(side, buffer))
1015                 return -EFAULT;
1016
1017         switch (side) {
1018         case 'A':       newSide = 0;
1019                         break;
1020         case 'B':       newSide = 1;
1021                         break;
1022         case 'C':       newSide = 2;
1023                         break;
1024         case 'D':       newSide = 3;
1025                         break;
1026         default:
1027                 printk(KERN_ERR "mf_proc.c: proc_mf_change_side: invalid side\n");
1028                 return -EINVAL;
1029         }
1030
1031         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
1032         vsp_cmd.sub_data.ipl_type = newSide;
1033         vsp_cmd.cmd = 10;
1034
1035         (void)signal_vsp_instruction(&vsp_cmd);
1036
1037         return count;
1038 }
1039
1040 #if 0
1041 static void mf_getSrcHistory(char *buffer, int size)
1042 {
1043         struct IplTypeReturnStuff return_stuff;
1044         struct pending_event *ev = new_pending_event();
1045         int rc = 0;
1046         char *pages[4];
1047
1048         pages[0] = kmalloc(4096, GFP_ATOMIC);
1049         pages[1] = kmalloc(4096, GFP_ATOMIC);
1050         pages[2] = kmalloc(4096, GFP_ATOMIC);
1051         pages[3] = kmalloc(4096, GFP_ATOMIC);
1052         if ((ev == NULL) || (pages[0] == NULL) || (pages[1] == NULL)
1053                          || (pages[2] == NULL) || (pages[3] == NULL))
1054                 return -ENOMEM;
1055
1056         return_stuff.xType = 0;
1057         return_stuff.xRc = 0;
1058         return_stuff.xDone = 0;
1059         ev->event.hp_lp_event.xSubtype = 6;
1060         ev->event.hp_lp_event.x.xSubtypeData =
1061                 subtype_data('M', 'F', 'V', 'I');
1062         ev->event.data.vsp_cmd.xEvent = &return_stuff;
1063         ev->event.data.vsp_cmd.cmd = 4;
1064         ev->event.data.vsp_cmd.lp_index = HvLpConfig_getLpIndex();
1065         ev->event.data.vsp_cmd.result_code = 0xFF;
1066         ev->event.data.vsp_cmd.reserved = 0;
1067         ev->event.data.vsp_cmd.sub_data.page[0] = ISERIES_HV_ADDR(pages[0]);
1068         ev->event.data.vsp_cmd.sub_data.page[1] = ISERIES_HV_ADDR(pages[1]);
1069         ev->event.data.vsp_cmd.sub_data.page[2] = ISERIES_HV_ADDR(pages[2]);
1070         ev->event.data.vsp_cmd.sub_data.page[3] = ISERIES_HV_ADDR(pages[3]);
1071         mb();
1072         if (signal_event(ev) != 0)
1073                 return;
1074
1075         while (return_stuff.xDone != 1)
1076                 udelay(10);
1077         if (return_stuff.xRc == 0)
1078                 memcpy(buffer, pages[0], size);
1079         kfree(pages[0]);
1080         kfree(pages[1]);
1081         kfree(pages[2]);
1082         kfree(pages[3]);
1083 }
1084 #endif
1085
1086 static int proc_mf_dump_src(char *page, char **start, off_t off,
1087                 int count, int *eof, void *data)
1088 {
1089 #if 0
1090         int len;
1091
1092         mf_getSrcHistory(page, count);
1093         len = count;
1094         len -= off;
1095         if (len < count) {
1096                 *eof = 1;
1097                 if (len <= 0)
1098                         return 0;
1099         } else
1100                 len = count;
1101         *start = page + off;
1102         return len;
1103 #else
1104         return 0;
1105 #endif
1106 }
1107
1108 static int proc_mf_change_src(struct file *file, const char __user *buffer,
1109                 unsigned long count, void *data)
1110 {
1111         char stkbuf[10];
1112
1113         if (!capable(CAP_SYS_ADMIN))
1114                 return -EACCES;
1115
1116         if ((count < 4) && (count != 1)) {
1117                 printk(KERN_ERR "mf_proc: invalid src\n");
1118                 return -EINVAL;
1119         }
1120
1121         if (count > (sizeof(stkbuf) - 1))
1122                 count = sizeof(stkbuf) - 1;
1123         if (copy_from_user(stkbuf, buffer, count))
1124                 return -EFAULT;
1125
1126         if ((count == 1) && (*stkbuf == '\0'))
1127                 mf_clear_src();
1128         else
1129                 mf_display_src(*(u32 *)stkbuf);
1130
1131         return count;
1132 }
1133
1134 static int proc_mf_change_cmdline(struct file *file, const char __user *buffer,
1135                 unsigned long count, void *data)
1136 {
1137         struct vsp_cmd_data vsp_cmd;
1138         dma_addr_t dma_addr;
1139         char *page;
1140         int ret = -EACCES;
1141
1142         if (!capable(CAP_SYS_ADMIN))
1143                 goto out;
1144
1145         dma_addr = 0;
1146         page = dma_alloc_coherent(iSeries_vio_dev, count, &dma_addr,
1147                         GFP_ATOMIC);
1148         ret = -ENOMEM;
1149         if (page == NULL)
1150                 goto out;
1151
1152         ret = -EFAULT;
1153         if (copy_from_user(page, buffer, count))
1154                 goto out_free;
1155
1156         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
1157         vsp_cmd.cmd = 31;
1158         vsp_cmd.sub_data.kern.token = dma_addr;
1159         vsp_cmd.sub_data.kern.address_type = HvLpDma_AddressType_TceIndex;
1160         vsp_cmd.sub_data.kern.side = (u64)data;
1161         vsp_cmd.sub_data.kern.length = count;
1162         mb();
1163         (void)signal_vsp_instruction(&vsp_cmd);
1164         ret = count;
1165
1166 out_free:
1167         dma_free_coherent(iSeries_vio_dev, count, page, dma_addr);
1168 out:
1169         return ret;
1170 }
1171
1172 static ssize_t proc_mf_change_vmlinux(struct file *file,
1173                                       const char __user *buf,
1174                                       size_t count, loff_t *ppos)
1175 {
1176         struct proc_dir_entry *dp = PDE(file->f_dentry->d_inode);
1177         ssize_t rc;
1178         dma_addr_t dma_addr;
1179         char *page;
1180         struct vsp_cmd_data vsp_cmd;
1181
1182         rc = -EACCES;
1183         if (!capable(CAP_SYS_ADMIN))
1184                 goto out;
1185
1186         dma_addr = 0;
1187         page = dma_alloc_coherent(iSeries_vio_dev, count, &dma_addr,
1188                         GFP_ATOMIC);
1189         rc = -ENOMEM;
1190         if (page == NULL) {
1191                 printk(KERN_ERR "mf.c: couldn't allocate memory to set vmlinux chunk\n");
1192                 goto out;
1193         }
1194         rc = -EFAULT;
1195         if (copy_from_user(page, buf, count))
1196                 goto out_free;
1197
1198         memset(&vsp_cmd, 0, sizeof(vsp_cmd));
1199         vsp_cmd.cmd = 30;
1200         vsp_cmd.sub_data.kern.token = dma_addr;
1201         vsp_cmd.sub_data.kern.address_type = HvLpDma_AddressType_TceIndex;
1202         vsp_cmd.sub_data.kern.side = (u64)dp->data;
1203         vsp_cmd.sub_data.kern.offset = *ppos;
1204         vsp_cmd.sub_data.kern.length = count;
1205         mb();
1206         rc = signal_vsp_instruction(&vsp_cmd);
1207         if (rc)
1208                 goto out_free;
1209         rc = -ENOMEM;
1210         if (vsp_cmd.result_code != 0)
1211                 goto out_free;
1212
1213         *ppos += count;
1214         rc = count;
1215 out_free:
1216         dma_free_coherent(iSeries_vio_dev, count, page, dma_addr);
1217 out:
1218         return rc;
1219 }
1220
1221 static struct file_operations proc_vmlinux_operations = {
1222         .write          = proc_mf_change_vmlinux,
1223 };
1224
1225 static int __init mf_proc_init(void)
1226 {
1227         struct proc_dir_entry *mf_proc_root;
1228         struct proc_dir_entry *ent;
1229         struct proc_dir_entry *mf;
1230         char name[2];
1231         int i;
1232
1233         mf_proc_root = proc_mkdir("iSeries/mf", NULL);
1234         if (!mf_proc_root)
1235                 return 1;
1236
1237         name[1] = '\0';
1238         for (i = 0; i < 4; i++) {
1239                 name[0] = 'A' + i;
1240                 mf = proc_mkdir(name, mf_proc_root);
1241                 if (!mf)
1242                         return 1;
1243
1244                 ent = create_proc_entry("cmdline", S_IFREG|S_IRUSR|S_IWUSR, mf);
1245                 if (!ent)
1246                         return 1;
1247                 ent->nlink = 1;
1248                 ent->data = (void *)(long)i;
1249                 ent->read_proc = proc_mf_dump_cmdline;
1250                 ent->write_proc = proc_mf_change_cmdline;
1251
1252                 if (i == 3)     /* no vmlinux entry for 'D' */
1253                         continue;
1254
1255                 ent = create_proc_entry("vmlinux", S_IFREG|S_IWUSR, mf);
1256                 if (!ent)
1257                         return 1;
1258                 ent->nlink = 1;
1259                 ent->data = (void *)(long)i;
1260                 ent->proc_fops = &proc_vmlinux_operations;
1261         }
1262
1263         ent = create_proc_entry("side", S_IFREG|S_IRUSR|S_IWUSR, mf_proc_root);
1264         if (!ent)
1265                 return 1;
1266         ent->nlink = 1;
1267         ent->data = (void *)0;
1268         ent->read_proc = proc_mf_dump_side;
1269         ent->write_proc = proc_mf_change_side;
1270
1271         ent = create_proc_entry("src", S_IFREG|S_IRUSR|S_IWUSR, mf_proc_root);
1272         if (!ent)
1273                 return 1;
1274         ent->nlink = 1;
1275         ent->data = (void *)0;
1276         ent->read_proc = proc_mf_dump_src;
1277         ent->write_proc = proc_mf_change_src;
1278
1279         return 0;
1280 }
1281
1282 __initcall(mf_proc_init);
1283
1284 #endif /* CONFIG_PROC_FS */