Home | History | Annotate | Download | only in os
      1      0    stevel /*
      2      0    stevel  * CDDL HEADER START
      3      0    stevel  *
      4      0    stevel  * The contents of this file are subject to the terms of the
      5   1991     heppo  * Common Development and Distribution License (the "License").
      6   1991     heppo  * You may not use this file except in compliance with the License.
      7      0    stevel  *
      8      0    stevel  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9      0    stevel  * or http://www.opensolaris.org/os/licensing.
     10      0    stevel  * See the License for the specific language governing permissions
     11      0    stevel  * and limitations under the License.
     12      0    stevel  *
     13      0    stevel  * When distributing Covered Code, include this CDDL HEADER in each
     14      0    stevel  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15      0    stevel  * If applicable, add the following below this CDDL HEADER, with the
     16      0    stevel  * fields enclosed by brackets "[]" replaced with your own identifying
     17      0    stevel  * information: Portions Copyright [yyyy] [name of copyright owner]
     18      0    stevel  *
     19      0    stevel  * CDDL HEADER END
     20      0    stevel  */
     21      0    stevel /*
     22   8542      Haik  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
     23      0    stevel  * Use is subject to license terms.
     24      0    stevel  */
     25      0    stevel 
     26      0    stevel #include <sys/types.h>
     27      0    stevel #include <sys/systm.h>
     28      0    stevel #include <sys/archsystm.h>
     29      0    stevel #include <sys/t_lock.h>
     30      0    stevel #include <sys/uadmin.h>
     31      0    stevel #include <sys/panic.h>
     32      0    stevel #include <sys/reboot.h>
     33      0    stevel #include <sys/autoconf.h>
     34      0    stevel #include <sys/machsystm.h>
     35      0    stevel #include <sys/promif.h>
     36      0    stevel #include <sys/membar.h>
     37      0    stevel #include <vm/hat_sfmmu.h>
     38      0    stevel #include <sys/cpu_module.h>
     39      0    stevel #include <sys/cpu_sgnblk_defs.h>
     40      0    stevel #include <sys/intreg.h>
     41      0    stevel #include <sys/consdev.h>
     42      0    stevel #include <sys/kdi_impl.h>
     43   1077   svemuri #include <sys/traptrace.h>
     44      0    stevel #include <sys/hypervisor_api.h>
     45      0    stevel #include <sys/vmsystm.h>
     46      0    stevel #include <sys/dtrace.h>
     47      0    stevel #include <sys/xc_impl.h>
     48    136  achartre #include <sys/callb.h>
     49   1991     heppo #include <sys/mdesc.h>
     50   1991     heppo #include <sys/mach_descrip.h>
     51   2036   wentaoy #include <sys/wdt.h>
     52   3266  jb145095 #include <sys/soft_state.h>
     53   3266  jb145095 #include <sys/promimpl.h>
     54   3266  jb145095 #include <sys/hsvc.h>
     55   4776   jm22469 #include <sys/ldoms.h>
     56   8542      Haik #include <sys/kldc.h>
     57  11066    rafael #include <sys/clock_impl.h>
     58  11172      Haik #include <sys/suspend.h>
     59  10843      Dave #include <sys/dumphdr.h>
     60      0    stevel 
     61      0    stevel /*
     62      0    stevel  * hvdump_buf_va is a pointer to the currently-configured hvdump_buf.
     63      0    stevel  * A value of NULL indicates that this area is not configured.
     64      0    stevel  * hvdump_buf_sz is tunable but will be clamped to HVDUMP_SIZE_MAX.
     65      0    stevel  */
     66      0    stevel 
     67      0    stevel caddr_t hvdump_buf_va;
     68      0    stevel uint64_t hvdump_buf_sz = HVDUMP_SIZE_DEFAULT;
     69      0    stevel static uint64_t hvdump_buf_pa;
     70      0    stevel 
     71   1077   svemuri u_longlong_t panic_tick;
     72      0    stevel 
     73   1077   svemuri extern u_longlong_t gettick();
     74      0    stevel static void reboot_machine(char *);
     75      0    stevel static void update_hvdump_buffer(void);
     76      0    stevel 
     77      0    stevel /*
     78      0    stevel  * For xt_sync synchronization.
     79      0    stevel  */
     80      0    stevel extern uint64_t xc_tick_limit;
     81      0    stevel extern uint64_t xc_tick_jump_limit;
     82   6138   svemuri extern uint64_t xc_sync_tick_limit;
     83      0    stevel 
     84      0    stevel /*
     85      0    stevel  * We keep our own copies, used for cache flushing, because we can be called
     86      0    stevel  * before cpu_fiximpl().
     87      0    stevel  */
     88      0    stevel static int kdi_dcache_size;
     89      0    stevel static int kdi_dcache_linesize;
     90      0    stevel static int kdi_icache_size;
     91      0    stevel static int kdi_icache_linesize;
     92      0    stevel 
     93      0    stevel /*
     94      0    stevel  * Assembly support for generic modules in sun4v/ml/mach_xc.s
     95      0    stevel  */
     96      0    stevel extern void init_mondo_nocheck(xcfunc_t *func, uint64_t arg1, uint64_t arg2);
     97      0    stevel extern void kdi_flush_idcache(int, int, int, int);
     98      0    stevel extern uint64_t get_cpuaddr(uint64_t, uint64_t);
     99      0    stevel 
    100   4776   jm22469 
    101   4776   jm22469 #define	BOOT_CMD_MAX_LEN	256
    102   4776   jm22469 #define	BOOT_CMD_BASE		"boot "
    103   4776   jm22469 
    104   4776   jm22469 /*
    105   4776   jm22469  * In an LDoms system we do not save the user's boot args in NVRAM
    106   4776   jm22469  * as is done on legacy systems.  Instead, we format and send a
    107   4776   jm22469  * 'reboot-command' variable to the variable service.  The contents
    108   4776   jm22469  * of the variable are retrieved by OBP and used verbatim for
    109   4776   jm22469  * the next boot.
    110   4776   jm22469  */
    111   4776   jm22469 static void
    112   5974   jm22469 store_boot_cmd(char *args, boolean_t add_boot_str)
    113   4776   jm22469 {
    114   4776   jm22469 	static char	cmd_buf[BOOT_CMD_MAX_LEN];
    115   5974   jm22469 	size_t		len = 1;
    116   4776   jm22469 	pnode_t		node;
    117   5974   jm22469 	size_t		base_len = 0;
    118   4776   jm22469 	size_t		args_len;
    119   4776   jm22469 	size_t		args_max;
    120   4776   jm22469 
    121   5974   jm22469 	if (add_boot_str) {
    122   5974   jm22469 		(void) strcpy(cmd_buf, BOOT_CMD_BASE);
    123   4776   jm22469 
    124   5974   jm22469 		base_len = strlen(BOOT_CMD_BASE);
    125   5974   jm22469 		len = base_len + 1;
    126   5974   jm22469 	}
    127   4776   jm22469 
    128   4776   jm22469 	if (args != NULL) {
    129   4776   jm22469 		args_len = strlen(args);
    130   4776   jm22469 		args_max = BOOT_CMD_MAX_LEN - len;
    131   4776   jm22469 
    132   4776   jm22469 		if (args_len > args_max) {
    133   4776   jm22469 			cmn_err(CE_WARN, "Reboot command too long (%ld), "
    134   4776   jm22469 			    "truncating command arguments", len + args_len);
    135   4776   jm22469 
    136   4776   jm22469 			args_len = args_max;
    137   4776   jm22469 		}
    138   4776   jm22469 
    139   4776   jm22469 		len += args_len;
    140   4776   jm22469 		(void) strncpy(&cmd_buf[base_len], args, args_len);
    141   4776   jm22469 	}
    142   4776   jm22469 
    143   4776   jm22469 	node = prom_optionsnode();
    144   4776   jm22469 	if ((node == OBP_NONODE) || (node == OBP_BADNODE) ||
    145   4776   jm22469 	    prom_setprop(node, "reboot-command", cmd_buf, len) == -1)
    146   4776   jm22469 		cmn_err(CE_WARN, "Unable to store boot command for "
    147   4776   jm22469 		    "use on reboot");
    148   4776   jm22469 }
    149   4776   jm22469 
    150   4776   jm22469 
    151      0    stevel /*
    152      0    stevel  * Machine dependent code to reboot.
    153   4776   jm22469  *
    154   4776   jm22469  * "bootstr", when non-null, points to a string to be used as the
    155   4776   jm22469  * argument string when rebooting.
    156    136  achartre  *
    157    136  achartre  * "invoke_cb" is a boolean. It is set to true when mdboot() can safely
    158    136  achartre  * invoke CB_CL_MDBOOT callbacks before shutting the system down, i.e. when
    159    136  achartre  * we are in a normal shutdown sequence (interrupts are not blocked, the
    160    136  achartre  * system is not panic'ing or being suspended).
    161      0    stevel  */
    162      0    stevel /*ARGSUSED*/
    163      0    stevel void
    164    136  achartre mdboot(int cmd, int fcn, char *bootstr, boolean_t invoke_cb)
    165      0    stevel {
    166      0    stevel 	extern void pm_cfb_check_and_powerup(void);
    167   5630     jbeck 
    168   5630     jbeck 	/*
    169   5630     jbeck 	 * XXX - rconsvp is set to NULL to ensure that output messages
    170   5630     jbeck 	 * are sent to the underlying "hardware" device using the
    171   5630     jbeck 	 * monitor's printf routine since we are in the process of
    172   5630     jbeck 	 * either rebooting or halting the machine.
    173   5630     jbeck 	 */
    174   5630     jbeck 	rconsvp = NULL;
    175   4776   jm22469 
    176   4776   jm22469 	switch (fcn) {
    177   4776   jm22469 	case AD_HALT:
    178   5974   jm22469 		/*
    179   5974   jm22469 		 * LDoms: By storing a no-op command
    180   5974   jm22469 		 * in the 'reboot-command' variable we cause OBP
    181   5974   jm22469 		 * to ignore the setting of 'auto-boot?' after
    182   5974   jm22469 		 * it completes the reset.  This causes the system
    183   5974   jm22469 		 * to stop at the ok prompt.
    184   5974   jm22469 		 */
    185   5974   jm22469 		if (domaining_enabled() && invoke_cb)
    186   5974   jm22469 			store_boot_cmd("noop", B_FALSE);
    187   5974   jm22469 		break;
    188   5974   jm22469 
    189   4776   jm22469 	case AD_POWEROFF:
    190   4776   jm22469 		break;
    191   5974   jm22469 
    192   4776   jm22469 	default:
    193   4776   jm22469 		if (bootstr == NULL) {
    194   4776   jm22469 			switch (fcn) {
    195   4776   jm22469 
    196   4776   jm22469 			case AD_BOOT:
    197   4776   jm22469 				bootstr = "";
    198   4776   jm22469 				break;
    199   4776   jm22469 
    200   4776   jm22469 			case AD_IBOOT:
    201   4776   jm22469 				bootstr = "-a";
    202   4776   jm22469 				break;
    203   4776   jm22469 
    204   4776   jm22469 			case AD_SBOOT:
    205   4776   jm22469 				bootstr = "-s";
    206   4776   jm22469 				break;
    207   4776   jm22469 
    208   4776   jm22469 			case AD_SIBOOT:
    209   4776   jm22469 				bootstr = "-sa";
    210   4776   jm22469 				break;
    211   4776   jm22469 			default:
    212   4776   jm22469 				cmn_err(CE_WARN,
    213   4776   jm22469 				    "mdboot: invalid function %d", fcn);
    214   4776   jm22469 				bootstr = "";
    215   4776   jm22469 				break;
    216   4776   jm22469 			}
    217   4776   jm22469 		}
    218   5177   jm22469 
    219   5177   jm22469 		/*
    220   5177   jm22469 		 * If LDoms is running, we must save the boot string
    221   5177   jm22469 		 * before we enter restricted mode.  This is possible
    222   5177   jm22469 		 * only if we are not being called from panic.
    223   5177   jm22469 		 */
    224   5177   jm22469 		if (domaining_enabled() && invoke_cb)
    225   5974   jm22469 			store_boot_cmd(bootstr, B_TRUE);
    226   4776   jm22469 	}
    227      0    stevel 
    228      0    stevel 	/*
    229      0    stevel 	 * At a high interrupt level we can't:
    230      0    stevel 	 *	1) bring up the console
    231      0    stevel 	 * or
    232      0    stevel 	 *	2) wait for pending interrupts prior to redistribution
    233      0    stevel 	 *	   to the current CPU
    234      0    stevel 	 *
    235      0    stevel 	 * so we do them now.
    236      0    stevel 	 */
    237      0    stevel 	pm_cfb_check_and_powerup();
    238      0    stevel 
    239      0    stevel 	/* make sure there are no more changes to the device tree */
    240      0    stevel 	devtree_freeze();
    241    136  achartre 
    242    136  achartre 	if (invoke_cb)
    243    136  achartre 		(void) callb_execute_class(CB_CL_MDBOOT, NULL);
    244    917     elowe 
    245    917     elowe 	/*
    246    917     elowe 	 * Clear any unresolved UEs from memory.
    247    917     elowe 	 */
    248   3253       mec 	page_retire_mdboot();
    249      0    stevel 
    250      0    stevel 	/*
    251      0    stevel 	 * stop other cpus which also raise our priority. since there is only
    252      0    stevel 	 * one active cpu after this, and our priority will be too high
    253      0    stevel 	 * for us to be preempted, we're essentially single threaded
    254      0    stevel 	 * from here on out.
    255      0    stevel 	 */
    256      0    stevel 	stop_other_cpus();
    257      0    stevel 
    258      0    stevel 	/*
    259      0    stevel 	 * try and reset leaf devices.  reset_leaves() should only
    260      0    stevel 	 * be called when there are no other threads that could be
    261      0    stevel 	 * accessing devices
    262      0    stevel 	 */
    263      0    stevel 	reset_leaves();
    264   2036   wentaoy 
    265   2036   wentaoy 	watchdog_clear();
    266      0    stevel 
    267      0    stevel 	if (fcn == AD_HALT) {
    268   3266  jb145095 		mach_set_soft_state(SIS_TRANSITION,
    269   4776   jm22469 		    &SOLARIS_SOFT_STATE_HALT_MSG);
    270      0    stevel 		halt((char *)NULL);
    271      0    stevel 	} else if (fcn == AD_POWEROFF) {
    272   3266  jb145095 		mach_set_soft_state(SIS_TRANSITION,
    273   4776   jm22469 		    &SOLARIS_SOFT_STATE_POWER_MSG);
    274      0    stevel 		power_down(NULL);
    275      0    stevel 	} else {
    276   4776   jm22469 		mach_set_soft_state(SIS_TRANSITION,
    277   4776   jm22469 		    &SOLARIS_SOFT_STATE_REBOOT_MSG);
    278      0    stevel 		reboot_machine(bootstr);
    279      0    stevel 	}
    280      0    stevel 	/* MAYBE REACHED */
    281      0    stevel }
    282      0    stevel 
    283      0    stevel /* mdpreboot - may be called prior to mdboot while root fs still mounted */
    284      0    stevel /*ARGSUSED*/
    285      0    stevel void
    286      0    stevel mdpreboot(int cmd, int fcn, char *bootstr)
    287      0    stevel {
    288      0    stevel }
    289      0    stevel 
    290      0    stevel /*
    291      0    stevel  * Halt the machine and then reboot with the device
    292      0    stevel  * and arguments specified in bootstr.
    293      0    stevel  */
    294      0    stevel static void
    295      0    stevel reboot_machine(char *bootstr)
    296      0    stevel {
    297      0    stevel 	flush_windows();
    298      0    stevel 	stop_other_cpus();		/* send stop signal to other CPUs */
    299      0    stevel 	prom_printf("rebooting...\n");
    300      0    stevel 	/*
    301      0    stevel 	 * For platforms that use CPU signatures, we
    302      0    stevel 	 * need to set the signature block to OS and
    303      0    stevel 	 * the state to exiting for all the processors.
    304      0    stevel 	 */
    305      0    stevel 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_REBOOT, -1);
    306      0    stevel 	prom_reboot(bootstr);
    307      0    stevel 	/*NOTREACHED*/
    308      0    stevel }
    309      0    stevel 
    310      0    stevel /*
    311      0    stevel  * We use the x-trap mechanism and idle_stop_xcall() to stop the other CPUs.
    312      0    stevel  * Once in panic_idle() they raise spl, record their location, and spin.
    313      0    stevel  */
    314      0    stevel static void
    315      0    stevel panic_idle(void)
    316      0    stevel {
    317      0    stevel 	(void) spl7();
    318      0    stevel 
    319      0    stevel 	debug_flush_windows();
    320      0    stevel 	(void) setjmp(&curthread->t_pcb);
    321      0    stevel 
    322      0    stevel 	CPU->cpu_m.in_prom = 1;
    323      0    stevel 	membar_stld();
    324  10843      Dave 
    325  10843      Dave 	dumpsys_helper();
    326      0    stevel 
    327   4776   jm22469 	for (;;)
    328   5974   jm22469 		;
    329      0    stevel }
    330      0    stevel 
    331      0    stevel /*
    332      0    stevel  * Force the other CPUs to trap into panic_idle(), and then remove them
    333      0    stevel  * from the cpu_ready_set so they will no longer receive cross-calls.
    334      0    stevel  */
    335      0    stevel /*ARGSUSED*/
    336      0    stevel void
    337      0    stevel panic_stopcpus(cpu_t *cp, kthread_t *t, int spl)
    338      0    stevel {
    339      0    stevel 	cpuset_t cps;
    340      0    stevel 	int i;
    341      0    stevel 
    342      0    stevel 	(void) splzs();
    343      0    stevel 	CPUSET_ALL_BUT(cps, cp->cpu_id);
    344      0    stevel 	xt_some(cps, (xcfunc_t *)idle_stop_xcall, (uint64_t)&panic_idle, NULL);
    345      0    stevel 
    346      0    stevel 	for (i = 0; i < NCPU; i++) {
    347      0    stevel 		if (i != cp->cpu_id && CPU_XCALL_READY(i)) {
    348      0    stevel 			int ntries = 0x10000;
    349      0    stevel 
    350      0    stevel 			while (!cpu[i]->cpu_m.in_prom && ntries) {
    351      0    stevel 				DELAY(50);
    352      0    stevel 				ntries--;
    353      0    stevel 			}
    354      0    stevel 
    355      0    stevel 			if (!cpu[i]->cpu_m.in_prom)
    356      0    stevel 				printf("panic: failed to stop cpu%d\n", i);
    357      0    stevel 
    358      0    stevel 			cpu[i]->cpu_flags &= ~CPU_READY;
    359      0    stevel 			cpu[i]->cpu_flags |= CPU_QUIESCED;
    360      0    stevel 			CPUSET_DEL(cpu_ready_set, cpu[i]->cpu_id);
    361      0    stevel 		}
    362      0    stevel 	}
    363      0    stevel }
    364      0    stevel 
    365      0    stevel /*
    366      0    stevel  * Platform callback following each entry to panicsys().  If we've panicked at
    367      0    stevel  * level 14, we examine t_panic_trap to see if a fatal trap occurred.  If so,
    368      0    stevel  * we disable further %tick_cmpr interrupts.  If not, an explicit call to panic
    369      0    stevel  * was made and so we re-enqueue an interrupt request structure to allow
    370      0    stevel  * further level 14 interrupts to be processed once we lower PIL.  This allows
    371      0    stevel  * us to handle panics from the deadman() CY_HIGH_LEVEL cyclic.
    372      0    stevel  */
    373      0    stevel void
    374      0    stevel panic_enter_hw(int spl)
    375      0    stevel {
    376   1077   svemuri 	if (!panic_tick) {
    377   1077   svemuri 		panic_tick = gettick();
    378   1077   svemuri 		if (mach_htraptrace_enable) {
    379   1077   svemuri 			uint64_t prev_freeze;
    380   1077   svemuri 
    381   1077   svemuri 			/*  there are no possible error codes for this hcall */
    382   1077   svemuri 			(void) hv_ttrace_freeze((uint64_t)TRAP_TFREEZE_ALL,
    383   1077   svemuri 			    &prev_freeze);
    384   1077   svemuri 		}
    385    526      arao #ifdef TRAPTRACE
    386    526      arao 		TRAPTRACE_FREEZE;
    387   1077   svemuri #endif
    388    526      arao 	}
    389   3266  jb145095 
    390   3266  jb145095 	mach_set_soft_state(SIS_TRANSITION, &SOLARIS_SOFT_STATE_PANIC_MSG);
    391   3266  jb145095 
    392      0    stevel 	if (spl == ipltospl(PIL_14)) {
    393      0    stevel 		uint_t opstate = disable_vec_intr();
    394      0    stevel 
    395      0    stevel 		if (curthread->t_panic_trap != NULL) {
    396      0    stevel 			tickcmpr_disable();
    397      0    stevel 			intr_dequeue_req(PIL_14, cbe_level14_inum);
    398      0    stevel 		} else {
    399      0    stevel 			if (!tickcmpr_disabled())
    400      0    stevel 				intr_enqueue_req(PIL_14, cbe_level14_inum);
    401      0    stevel 			/*
    402      0    stevel 			 * Clear SOFTINT<14>, SOFTINT<0> (TICK_INT)
    403      0    stevel 			 * and SOFTINT<16> (STICK_INT) to indicate
    404      0    stevel 			 * that the current level 14 has been serviced.
    405      0    stevel 			 */
    406      0    stevel 			wr_clr_softint((1 << PIL_14) |
    407   4776   jm22469 			    TICK_INT_MASK | STICK_INT_MASK);
    408      0    stevel 		}
    409      0    stevel 
    410      0    stevel 		enable_vec_intr(opstate);
    411      0    stevel 	}
    412      0    stevel }
    413      0    stevel 
    414      0    stevel /*
    415      0    stevel  * Miscellaneous hardware-specific code to execute after panicstr is set
    416      0    stevel  * by the panic code: we also print and record PTL1 panic information here.
    417      0    stevel  */
    418      0    stevel /*ARGSUSED*/
    419      0    stevel void
    420      0    stevel panic_quiesce_hw(panic_data_t *pdp)
    421      0    stevel {
    422      0    stevel 	extern uint_t getpstate(void);
    423      0    stevel 	extern void setpstate(uint_t);
    424      0    stevel 
    425      0    stevel 	/*
    426      0    stevel 	 * Turn off TRAPTRACE and save the current %tick value in panic_tick.
    427      0    stevel 	 */
    428   1077   svemuri 	if (!panic_tick) {
    429      0    stevel 		panic_tick = gettick();
    430   1077   svemuri 		if (mach_htraptrace_enable) {
    431   1077   svemuri 			uint64_t prev_freeze;
    432   1077   svemuri 
    433   1077   svemuri 			/*  there are no possible error codes for this hcall */
    434   1077   svemuri 			(void) hv_ttrace_freeze((uint64_t)TRAP_TFREEZE_ALL,
    435   1077   svemuri 			    &prev_freeze);
    436   1077   svemuri 		}
    437   1077   svemuri #ifdef TRAPTRACE
    438   1077   svemuri 		TRAPTRACE_FREEZE;
    439      0    stevel #endif
    440   1077   svemuri 	}
    441      0    stevel 	/*
    442      0    stevel 	 * For Platforms that use CPU signatures, we
    443      0    stevel 	 * need to set the signature block to OS, the state to
    444      0    stevel 	 * exiting, and the substate to panic for all the processors.
    445      0    stevel 	 */
    446      0    stevel 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_PANIC, -1);
    447      0    stevel 
    448      0    stevel 	update_hvdump_buffer();
    449      0    stevel 
    450      0    stevel 	/*
    451      0    stevel 	 * Disable further ECC errors from the bus nexus.
    452      0    stevel 	 */
    453      0    stevel 	(void) bus_func_invoke(BF_TYPE_ERRDIS);
    454      0    stevel 
    455      0    stevel 	/*
    456      0    stevel 	 * Redirect all interrupts to the current CPU.
    457      0    stevel 	 */
    458      0    stevel 	intr_redist_all_cpus_shutdown();
    459      0    stevel 
    460      0    stevel 	/*
    461      0    stevel 	 * This call exists solely to support dumps to network
    462      0    stevel 	 * devices after sync from OBP.
    463      0    stevel 	 *
    464      0    stevel 	 * If we came here via the sync callback, then on some
    465      0    stevel 	 * platforms, interrupts may have arrived while we were
    466      0    stevel 	 * stopped in OBP.  OBP will arrange for those interrupts to
    467      0    stevel 	 * be redelivered if you say "go", but not if you invoke a
    468      0    stevel 	 * client callback like 'sync'.	 For some dump devices
    469      0    stevel 	 * (network swap devices), we need interrupts to be
    470      0    stevel 	 * delivered in order to dump, so we have to call the bus
    471      0    stevel 	 * nexus driver to reset the interrupt state machines.
    472      0    stevel 	 */
    473      0    stevel 	(void) bus_func_invoke(BF_TYPE_RESINTR);
    474      0    stevel 
    475      0    stevel 	setpstate(getpstate() | PSTATE_IE);
    476      0    stevel }
    477      0    stevel 
    478      0    stevel /*
    479      0    stevel  * Platforms that use CPU signatures need to set the signature block to OS and
    480      0    stevel  * the state to exiting for all CPUs. PANIC_CONT indicates that we're about to
    481      0    stevel  * write the crash dump, which tells the SSP/SMS to begin a timeout routine to
    482      0    stevel  * reboot the machine if the dump never completes.
    483      0    stevel  */
    484      0    stevel /*ARGSUSED*/
    485      0    stevel void
    486      0    stevel panic_dump_hw(int spl)
    487      0    stevel {
    488      0    stevel 	CPU_SIGNATURE(OS_SIG, SIGST_EXIT, SIGSUBST_DUMP, -1);
    489      0    stevel }
    490      0    stevel 
    491      0    stevel /*
    492      0    stevel  * for ptl1_panic
    493      0    stevel  */
    494      0    stevel void
    495      0    stevel ptl1_init_cpu(struct cpu *cpu)
    496      0    stevel {
    497      0    stevel 	ptl1_state_t *pstate = &cpu->cpu_m.ptl1_state;
    498      0    stevel 
    499      0    stevel 	/*CONSTCOND*/
    500      0    stevel 	if (sizeof (struct cpu) + PTL1_SSIZE > CPU_ALLOC_SIZE) {
    501      0    stevel 		panic("ptl1_init_cpu: not enough space left for ptl1_panic "
    502    911   iskreen 		    "stack, sizeof (struct cpu) = %lu",
    503    911   iskreen 		    (unsigned long)sizeof (struct cpu));
    504      0    stevel 	}
    505      0    stevel 
    506      0    stevel 	pstate->ptl1_stktop = (uintptr_t)cpu + CPU_ALLOC_SIZE;
    507      0    stevel 	cpu_pa[cpu->cpu_id] = va_to_pa(cpu);
    508      0    stevel }
    509      0    stevel 
    510      0    stevel void
    511      0    stevel ptl1_panic_handler(ptl1_state_t *pstate)
    512      0    stevel {
    513      0    stevel 	static const char *ptl1_reasons[] = {
    514      0    stevel #ifdef	PTL1_PANIC_DEBUG
    515      0    stevel 		"trap for debug purpose",	/* PTL1_BAD_DEBUG */
    516      0    stevel #else
    517      0    stevel 		"unknown trap",			/* PTL1_BAD_DEBUG */
    518      0    stevel #endif
    519      0    stevel 		"register window trap",		/* PTL1_BAD_WTRAP */
    520      0    stevel 		"kernel MMU miss",		/* PTL1_BAD_KMISS */
    521      0    stevel 		"kernel protection fault",	/* PTL1_BAD_KPROT_FAULT */
    522      0    stevel 		"ISM MMU miss",			/* PTL1_BAD_ISM */
    523      0    stevel 		"kernel MMU trap",		/* PTL1_BAD_MMUTRAP */
    524      0    stevel 		"kernel trap handler state",	/* PTL1_BAD_TRAP */
    525      0    stevel 		"floating point trap",		/* PTL1_BAD_FPTRAP */
    526      0    stevel #ifdef	DEBUG
    527   2973   govinda 		"pointer to intr_vec",		/* PTL1_BAD_INTR_VEC */
    528      0    stevel #else
    529   2973   govinda 		"unknown trap",			/* PTL1_BAD_INTR_VEC */
    530      0    stevel #endif
    531      0    stevel #ifdef	TRAPTRACE
    532      0    stevel 		"TRACE_PTR state",		/* PTL1_BAD_TRACE_PTR */
    533      0    stevel #else
    534      0    stevel 		"unknown trap",			/* PTL1_BAD_TRACE_PTR */
    535      0    stevel #endif
    536      0    stevel 		"stack overflow",		/* PTL1_BAD_STACK */
    537      0    stevel 		"DTrace flags",			/* PTL1_BAD_DTRACE_FLAGS */
    538      0    stevel 		"attempt to steal locked ctx",  /* PTL1_BAD_CTX_STEAL */
    539      0    stevel 		"CPU ECC error loop",		/* PTL1_BAD_ECC */
    540      0    stevel 		"unexpected error from hypervisor call", /* PTL1_BAD_HCALL */
    541    526      arao 		"unexpected global level(%gl)", /* PTL1_BAD_GL */
    542   1991     heppo 		"Watchdog Reset", 		/* PTL1_BAD_WATCHDOG */
    543   1991     heppo 		"unexpected RED mode trap", 	/* PTL1_BAD_RED */
    544   1991     heppo 		"return value EINVAL from hcall: "\
    545   1991     heppo 		    "UNMAP_PERM_ADDR",	/* PTL1_BAD_HCALL_UNMAP_PERM_EINVAL */
    546   1991     heppo 		"return value ENOMAP from hcall: "\
    547   1991     heppo 		    "UNMAP_PERM_ADDR", /* PTL1_BAD_HCALL_UNMAP_PERM_ENOMAP */
    548   6044   paulsan 		"error raising a TSB exception", /* PTL1_BAD_RAISE_TSBEXCP */
    549   6044   paulsan 		"missing shared TSB"	/* PTL1_NO_SCDTSB8K */
    550      0    stevel 	};
    551      0    stevel 
    552    357   svemuri 	uint_t reason = pstate->ptl1_regs.ptl1_gregs[0].ptl1_g1;
    553      0    stevel 	uint_t tl = pstate->ptl1_regs.ptl1_trap_regs[0].ptl1_tl;
    554   5084   johnlev 	struct panic_trap_info ti = { 0 };
    555      0    stevel 
    556      0    stevel 	/*
    557      0    stevel 	 * Use trap_info for a place holder to call panic_savetrap() and
    558      0    stevel 	 * panic_showtrap() to save and print out ptl1_panic information.
    559      0    stevel 	 */
    560      0    stevel 	if (curthread->t_panic_trap == NULL)
    561      0    stevel 		curthread->t_panic_trap = &ti;
    562      0    stevel 
    563      0    stevel 	if (reason < sizeof (ptl1_reasons) / sizeof (ptl1_reasons[0]))
    564      0    stevel 		panic("bad %s at TL %u", ptl1_reasons[reason], tl);
    565      0    stevel 	else
    566      0    stevel 		panic("ptl1_panic reason 0x%x at TL %u", reason, tl);
    567      0    stevel }
    568      0    stevel 
    569      0    stevel void
    570      0    stevel clear_watchdog_on_exit(void)
    571      0    stevel {
    572   3349   wentaoy 	if (watchdog_enabled && watchdog_activated) {
    573   3349   wentaoy 		prom_printf("Debugging requested; hardware watchdog "
    574   4776   jm22469 		    "suspended.\n");
    575   3349   wentaoy 		(void) watchdog_suspend();
    576   3349   wentaoy 	}
    577      0    stevel }
    578      0    stevel 
    579   2036   wentaoy /*
    580   2036   wentaoy  * Restore the watchdog timer when returning from a debugger
    581   2036   wentaoy  * after a panic or L1-A and resume watchdog pat.
    582   2036   wentaoy  */
    583      0    stevel void
    584   2036   wentaoy restore_watchdog_on_entry()
    585      0    stevel {
    586   2036   wentaoy 	watchdog_resume();
    587      0    stevel }
    588      0    stevel 
    589      0    stevel int
    590      0    stevel kdi_watchdog_disable(void)
    591      0    stevel {
    592   2036   wentaoy 	watchdog_suspend();
    593   2036   wentaoy 
    594   2036   wentaoy 	return (0);
    595      0    stevel }
    596      0    stevel 
    597      0    stevel void
    598      0    stevel kdi_watchdog_restore(void)
    599      0    stevel {
    600   2036   wentaoy 	watchdog_resume();
    601      0    stevel }
    602      0    stevel 
    603      0    stevel void
    604      0    stevel mach_dump_buffer_init(void)
    605      0    stevel {
    606      0    stevel 	uint64_t  ret, minsize = 0;
    607      0    stevel 
    608      0    stevel 	if (hvdump_buf_sz > HVDUMP_SIZE_MAX)
    609      0    stevel 		hvdump_buf_sz = HVDUMP_SIZE_MAX;
    610      0    stevel 
    611    288      arao 	hvdump_buf_va = contig_mem_alloc_align(hvdump_buf_sz, PAGESIZE);
    612      0    stevel 	if (hvdump_buf_va == NULL)
    613      0    stevel 		return;
    614      0    stevel 
    615      0    stevel 	hvdump_buf_pa = va_to_pa(hvdump_buf_va);
    616      0    stevel 
    617      0    stevel 	ret = hv_dump_buf_update(hvdump_buf_pa, hvdump_buf_sz,
    618      0    stevel 	    &minsize);
    619      0    stevel 
    620      0    stevel 	if (ret != H_EOK) {
    621      0    stevel 		contig_mem_free(hvdump_buf_va, hvdump_buf_sz);
    622      0    stevel 		hvdump_buf_va = NULL;
    623      0    stevel 		cmn_err(CE_NOTE, "!Error in setting up hvstate"
    624      0    stevel 		    "dump buffer. Error = 0x%lx, size = 0x%lx,"
    625      0    stevel 		    "buf_pa = 0x%lx", ret, hvdump_buf_sz,
    626      0    stevel 		    hvdump_buf_pa);
    627      0    stevel 
    628      0    stevel 		if (ret == H_EINVAL) {
    629      0    stevel 			cmn_err(CE_NOTE, "!Buffer size too small."
    630      0    stevel 			    "Available buffer size = 0x%lx,"
    631      0    stevel 			    "Minimum buffer size required = 0x%lx",
    632      0    stevel 			    hvdump_buf_sz, minsize);
    633      0    stevel 		}
    634      0    stevel 	}
    635      0    stevel }
    636      0    stevel 
    637      0    stevel 
    638      0    stevel static void
    639      0    stevel update_hvdump_buffer(void)
    640      0    stevel {
    641      0    stevel 	uint64_t ret, dummy_val;
    642      0    stevel 
    643      0    stevel 	if (hvdump_buf_va == NULL)
    644      0    stevel 		return;
    645      0    stevel 
    646      0    stevel 	ret = hv_dump_buf_update(hvdump_buf_pa, hvdump_buf_sz,
    647      0    stevel 	    &dummy_val);
    648      0    stevel 	if (ret != H_EOK) {
    649      0    stevel 		cmn_err(CE_NOTE, "!Cannot update hvstate dump"
    650      0    stevel 		    "buffer. Error = 0x%lx", ret);
    651      0    stevel 	}
    652      0    stevel }
    653      0    stevel 
    654      0    stevel 
    655      0    stevel static int
    656    789    ahrens getintprop(pnode_t node, char *name, int deflt)
    657      0    stevel {
    658      0    stevel 	int	value;
    659      0    stevel 
    660      0    stevel 	switch (prom_getproplen(node, name)) {
    661      0    stevel 	case 0:
    662      0    stevel 		value = 1;	/* boolean properties */
    663      0    stevel 		break;
    664      0    stevel 
    665      0    stevel 	case sizeof (int):
    666      0    stevel 		(void) prom_getprop(node, name, (caddr_t)&value);
    667      0    stevel 		break;
    668      0    stevel 
    669      0    stevel 	default:
    670      0    stevel 		value = deflt;
    671      0    stevel 		break;
    672      0    stevel 	}
    673      0    stevel 
    674      0    stevel 	return (value);
    675      0    stevel }
    676      0    stevel 
    677      0    stevel /*
    678      0    stevel  * Called by setcpudelay
    679      0    stevel  */
    680      0    stevel void
    681      0    stevel cpu_init_tick_freq(void)
    682      0    stevel {
    683   1991     heppo 	md_t *mdp;
    684   1991     heppo 	mde_cookie_t rootnode;
    685   1991     heppo 	int		listsz;
    686   1991     heppo 	mde_cookie_t	*listp = NULL;
    687   1991     heppo 	int	num_nodes;
    688   1991     heppo 	uint64_t stick_prop;
    689   1991     heppo 
    690   1991     heppo 	if (broken_md_flag) {
    691   1991     heppo 		sys_tick_freq = cpunodes[CPU->cpu_id].clock_freq;
    692   1991     heppo 		return;
    693   1991     heppo 	}
    694   1991     heppo 
    695   1991     heppo 	if ((mdp = md_get_handle()) == NULL)
    696   1991     heppo 		panic("stick_frequency property not found in MD");
    697   1991     heppo 
    698   1991     heppo 	rootnode = md_root_node(mdp);
    699   1991     heppo 	ASSERT(rootnode != MDE_INVAL_ELEM_COOKIE);
    700   1991     heppo 
    701   1991     heppo 	num_nodes = md_node_count(mdp);
    702   1991     heppo 
    703   1991     heppo 	ASSERT(num_nodes > 0);
    704   1991     heppo 	listsz = num_nodes * sizeof (mde_cookie_t);
    705   1991     heppo 	listp = (mde_cookie_t *)prom_alloc((caddr_t)0, listsz, 0);
    706   1991     heppo 
    707   1991     heppo 	if (listp == NULL)
    708   1991     heppo 		panic("cannot allocate list for MD properties");
    709   1991     heppo 
    710   1991     heppo 	num_nodes = md_scan_dag(mdp, rootnode, md_find_name(mdp, "platform"),
    711   1991     heppo 	    md_find_name(mdp, "fwd"), listp);
    712   1991     heppo 
    713   1991     heppo 	ASSERT(num_nodes == 1);
    714   1991     heppo 
    715   1991     heppo 	if (md_get_prop_val(mdp, *listp, "stick-frequency", &stick_prop) != 0)
    716   1991     heppo 		panic("stick_frequency property not found in MD");
    717   1991     heppo 
    718   1991     heppo 	sys_tick_freq = stick_prop;
    719   1991     heppo 
    720   1991     heppo 	prom_free((caddr_t)listp, listsz);
    721   1991     heppo 	(void) md_fini_handle(mdp);
    722      0    stevel }
    723      0    stevel 
    724      0    stevel int shipit(int n, uint64_t cpu_list_ra);
    725      0    stevel 
    726      0    stevel #ifdef DEBUG
    727      0    stevel #define	SEND_MONDO_STATS	1
    728      0    stevel #endif
    729      0    stevel 
    730      0    stevel #ifdef SEND_MONDO_STATS
    731      0    stevel uint32_t x_one_stimes[64];
    732      0    stevel uint32_t x_one_ltimes[16];
    733      0    stevel uint32_t x_set_stimes[64];
    734      0    stevel uint32_t x_set_ltimes[16];
    735      0    stevel uint32_t x_set_cpus[NCPU];
    736      0    stevel #endif
    737      0    stevel 
    738      0    stevel void
    739      0    stevel send_one_mondo(int cpuid)
    740      0    stevel {
    741      0    stevel 	int retries, stat;
    742      0    stevel 	uint64_t starttick, endtick, tick, lasttick;
    743      0    stevel 	struct machcpu	*mcpup = &(CPU->cpu_m);
    744      0    stevel 
    745      0    stevel 	CPU_STATS_ADDQ(CPU, sys, xcalls, 1);
    746      0    stevel 	starttick = lasttick = gettick();
    747      0    stevel 	mcpup->cpu_list[0] = (uint16_t)cpuid;
    748      0    stevel 	stat = shipit(1, mcpup->cpu_list_ra);
    749      0    stevel 	endtick = starttick + xc_tick_limit;
    750      0    stevel 	retries = 0;
    751   1310  ha137994 	while (stat != H_EOK) {
    752   1310  ha137994 		if (stat != H_EWOULDBLOCK) {
    753   1310  ha137994 			if (panic_quiesce)
    754   1310  ha137994 				return;
    755   1310  ha137994 			if (stat == H_ECPUERROR)
    756   1310  ha137994 				cmn_err(CE_PANIC, "send_one_mondo: "
    757   1310  ha137994 				    "cpuid: 0x%x has been marked in "
    758   1310  ha137994 				    "error", cpuid);
    759   1310  ha137994 			else
    760   1310  ha137994 				cmn_err(CE_PANIC, "send_one_mondo: "
    761   1310  ha137994 				    "unexpected hypervisor error 0x%x "
    762   1310  ha137994 				    "while sending a mondo to cpuid: "
    763   1310  ha137994 				    "0x%x", stat, cpuid);
    764   1310  ha137994 		}
    765      0    stevel 		tick = gettick();
    766      0    stevel 		/*
    767      0    stevel 		 * If there is a big jump between the current tick
    768      0    stevel 		 * count and lasttick, we have probably hit a break
    769      0    stevel 		 * point.  Adjust endtick accordingly to avoid panic.
    770      0    stevel 		 */
    771      0    stevel 		if (tick > (lasttick + xc_tick_jump_limit))
    772      0    stevel 			endtick += (tick - lasttick);
    773      0    stevel 		lasttick = tick;
    774      0    stevel 		if (tick > endtick) {
    775      0    stevel 			if (panic_quiesce)
    776      0    stevel 				return;
    777      0    stevel 			cmn_err(CE_PANIC, "send mondo timeout "
    778      0    stevel 			    "(target 0x%x) [retries: 0x%x hvstat: 0x%x]",
    779      0    stevel 			    cpuid, retries, stat);
    780      0    stevel 		}
    781      0    stevel 		drv_usecwait(1);
    782      0    stevel 		stat = shipit(1, mcpup->cpu_list_ra);
    783      0    stevel 		retries++;
    784      0    stevel 	}
    785      0    stevel #ifdef SEND_MONDO_STATS
    786      0    stevel 	{
    787   1310  ha137994 		uint64_t n = gettick() - starttick;
    788      0    stevel 		if (n < 8192)
    789      0    stevel 			x_one_stimes[n >> 7]++;
    790   1310  ha137994 		else if (n < 15*8192)
    791   1310  ha137994 			x_one_ltimes[n >> 13]++;
    792      0    stevel 		else
    793      0    stevel 			x_one_ltimes[0xf]++;
    794      0    stevel 	}
    795      0    stevel #endif
    796      0    stevel }
    797      0    stevel 
    798      0    stevel void
    799      0    stevel send_mondo_set(cpuset_t set)
    800      0    stevel {
    801      0    stevel 	uint64_t starttick, endtick, tick, lasttick;
    802   2492  ha137994 	uint_t largestid, smallestid;
    803   2492  ha137994 	int i, j;
    804   2492  ha137994 	int ncpuids = 0;
    805      0    stevel 	int shipped = 0;
    806   1310  ha137994 	int retries = 0;
    807      0    stevel 	struct machcpu	*mcpup = &(CPU->cpu_m);
    808      0    stevel 
    809      0    stevel 	ASSERT(!CPUSET_ISNULL(set));
    810   2492  ha137994 	CPUSET_BOUNDS(set, smallestid, largestid);
    811   2492  ha137994 	if (smallestid == CPUSET_NOTINSET) {
    812   2492  ha137994 		return;
    813   2492  ha137994 	}
    814   2492  ha137994 
    815      0    stevel 	starttick = lasttick = gettick();
    816      0    stevel 	endtick = starttick + xc_tick_limit;
    817      0    stevel 
    818   2492  ha137994 	/*
    819   2492  ha137994 	 * Assemble CPU list for HV argument. We already know
    820   2492  ha137994 	 * smallestid and largestid are members of set.
    821   2492  ha137994 	 */
    822   2492  ha137994 	mcpup->cpu_list[ncpuids++] = (uint16_t)smallestid;
    823   2492  ha137994 	if (largestid != smallestid) {
    824   2492  ha137994 		for (i = smallestid+1; i <= largestid-1; i++) {
    825   1310  ha137994 			if (CPU_IN_SET(set, i)) {
    826   2492  ha137994 				mcpup->cpu_list[ncpuids++] = (uint16_t)i;
    827      0    stevel 			}
    828      0    stevel 		}
    829   2492  ha137994 		mcpup->cpu_list[ncpuids++] = (uint16_t)largestid;
    830   2492  ha137994 	}
    831   2492  ha137994 
    832   2492  ha137994 	do {
    833   2492  ha137994 		int stat;
    834      0    stevel 
    835   1310  ha137994 		stat = shipit(ncpuids, mcpup->cpu_list_ra);
    836   1310  ha137994 		if (stat == H_EOK) {
    837   1310  ha137994 			shipped += ncpuids;
    838   1310  ha137994 			break;
    839   1310  ha137994 		}
    840   1310  ha137994 
    841   1310  ha137994 		/*
    842   1310  ha137994 		 * Either not all CPU mondos were sent, or an
    843   1310  ha137994 		 * error occurred. CPUs that were sent mondos
    844   1310  ha137994 		 * have their CPU IDs overwritten in cpu_list.
    845   2492  ha137994 		 * Reset cpu_list so that it only holds those
    846   2492  ha137994 		 * CPU IDs that still need to be sent.
    847   1310  ha137994 		 */
    848   2492  ha137994 		for (i = 0, j = 0; i < ncpuids; i++) {
    849   1310  ha137994 			if (mcpup->cpu_list[i] == HV_SEND_MONDO_ENTRYDONE) {
    850   1310  ha137994 				shipped++;
    851   1310  ha137994 			} else {
    852   2492  ha137994 				mcpup->cpu_list[j++] = mcpup->cpu_list[i];
    853   1310  ha137994 			}
    854   1310  ha137994 		}
    855   2492  ha137994 		ncpuids = j;
    856   1310  ha137994 
    857   1310  ha137994 		/*
    858   1310  ha137994 		 * Now handle possible errors returned
    859   1310  ha137994 		 * from hypervisor.
    860   1310  ha137994 		 */
    861   1310  ha137994 		if (stat == H_ECPUERROR) {
    862   2492  ha137994 			int errorcpus;
    863   2492  ha137994 
    864   2492  ha137994 			if (!panic_quiesce)
    865   2492  ha137994 				cmn_err(CE_CONT, "send_mondo_set: cpuid(s) ");
    866   1310  ha137994 
    867   1310  ha137994 			/*
    868   2492  ha137994 			 * Remove any CPUs in the error state from
    869   2492  ha137994 			 * cpu_list. At this point cpu_list only
    870   2492  ha137994 			 * contains the CPU IDs for mondos not
    871   2492  ha137994 			 * succesfully sent.
    872   1310  ha137994 			 */
    873   2492  ha137994 			for (i = 0, errorcpus = 0; i < ncpuids; i++) {
    874   2492  ha137994 				uint64_t state = CPU_STATE_INVALID;
    875   2492  ha137994 				uint16_t id = mcpup->cpu_list[i];
    876   2492  ha137994 
    877   2492  ha137994 				(void) hv_cpu_state(id, &state);
    878   2492  ha137994 				if (state == CPU_STATE_ERROR) {
    879   2492  ha137994 					if (!panic_quiesce)
    880   2492  ha137994 						cmn_err(CE_CONT, "0x%x ", id);
    881   2492  ha137994 					errorcpus++;
    882   2492  ha137994 				} else if (errorcpus > 0) {
    883   2492  ha137994 					mcpup->cpu_list[i - errorcpus] =
    884   2492  ha137994 					    mcpup->cpu_list[i];
    885   1310  ha137994 				}
    886   1310  ha137994 			}
    887   2492  ha137994 			ncpuids -= errorcpus;
    888   1310  ha137994 
    889   1310  ha137994 			if (!panic_quiesce) {
    890   2492  ha137994 				if (errorcpus == 0) {
    891   2492  ha137994 					cmn_err(CE_CONT, "<none> have been "
    892   2492  ha137994 					    "marked in error\n");
    893   1310  ha137994 					cmn_err(CE_PANIC, "send_mondo_set: "
    894   1310  ha137994 					    "hypervisor returned "
    895   1310  ha137994 					    "H_ECPUERROR but no CPU in "
    896   1310  ha137994 					    "cpu_list in error state");
    897   2492  ha137994 				} else {
    898   2492  ha137994 					cmn_err(CE_CONT, "have been marked in "
    899   2492  ha137994 					    "error\n");
    900   2492  ha137994 					cmn_err(CE_PANIC, "send_mondo_set: "
    901   2492  ha137994 					    "CPU(s) in error state");
    902   1310  ha137994 				}
    903   1310  ha137994 			}
    904   1310  ha137994 		} else if (stat != H_EWOULDBLOCK) {
    905   1310  ha137994 			if (panic_quiesce)
    906   1310  ha137994 				return;
    907   1310  ha137994 			/*
    908   1310  ha137994 			 * For all other errors, panic.
    909   1310  ha137994 			 */
    910   1310  ha137994 			cmn_err(CE_CONT, "send_mondo_set: unexpected "
    911   1310  ha137994 			    "hypervisor error 0x%x while sending a "
    912   1310  ha137994 			    "mondo to cpuid(s):", stat);
    913   2492  ha137994 			for (i = 0; i < ncpuids; i++) {
    914   2492  ha137994 				cmn_err(CE_CONT, " 0x%x", mcpup->cpu_list[i]);
    915   1310  ha137994 			}
    916   1310  ha137994 			cmn_err(CE_CONT, "\n");
    917   1310  ha137994 			cmn_err(CE_PANIC, "send_mondo_set: unexpected "
    918   1310  ha137994 			    "hypervisor error");
    919   1310  ha137994 		}
    920   1310  ha137994 
    921      0    stevel 		tick = gettick();
    922      0    stevel 		/*
    923      0    stevel 		 * If there is a big jump between the current tick
    924      0    stevel 		 * count and lasttick, we have probably hit a break
    925      0    stevel 		 * point.  Adjust endtick accordingly to avoid panic.
    926      0    stevel 		 */
    927      0    stevel 		if (tick > (lasttick + xc_tick_jump_limit))
    928      0    stevel 			endtick += (tick - lasttick);
    929      0    stevel 		lasttick = tick;
    930      0    stevel 		if (tick > endtick) {
    931      0    stevel 			if (panic_quiesce)
    932      0    stevel 				return;
    933      0    stevel 			cmn_err(CE_CONT, "send mondo timeout "
    934      0    stevel 			    "[retries: 0x%x]  cpuids: ", retries);
    935   2492  ha137994 			for (i = 0; i < ncpuids; i++)
    936   2492  ha137994 				cmn_err(CE_CONT, " 0x%x", mcpup->cpu_list[i]);
    937      0    stevel 			cmn_err(CE_CONT, "\n");
    938      0    stevel 			cmn_err(CE_PANIC, "send_mondo_set: timeout");
    939      0    stevel 		}
    940      0    stevel 
    941      0    stevel 		while (gettick() < (tick + sys_clock_mhz))
    942      0    stevel 			;
    943      0    stevel 		retries++;
    944   2492  ha137994 	} while (ncpuids > 0);
    945   1310  ha137994 
    946   1310  ha137994 	CPU_STATS_ADDQ(CPU, sys, xcalls, shipped);
    947      0    stevel 
    948      0    stevel #ifdef SEND_MONDO_STATS
    949      0    stevel 	{
    950   1310  ha137994 		uint64_t n = gettick() - starttick;
    951      0    stevel 		if (n < 8192)
    952      0    stevel 			x_set_stimes[n >> 7]++;
    953   1310  ha137994 		else if (n < 15*8192)
    954   1310  ha137994 			x_set_ltimes[n >> 13]++;
    955      0    stevel 		else
    956      0    stevel 			x_set_ltimes[0xf]++;
    957      0    stevel 	}
    958      0    stevel 	x_set_cpus[shipped]++;
    959      0    stevel #endif
    960      0    stevel }
    961      0    stevel 
    962      0    stevel void
    963      0    stevel syncfpu(void)
    964      0    stevel {
    965      0    stevel }
    966      0    stevel 
    967      0    stevel void
    968      0    stevel sticksync_slave(void)
    969  11172      Haik {
    970  11172      Haik 	suspend_sync_tick_stick_npt();
    971  11172      Haik }
    972      0    stevel 
    973      0    stevel void
    974      0    stevel sticksync_master(void)
    975      0    stevel {}
    976      0    stevel 
    977      0    stevel void
    978      0    stevel cpu_init_cache_scrub(void)
    979   3266  jb145095 {
    980   3266  jb145095 	mach_set_soft_state(SIS_NORMAL, &SOLARIS_SOFT_STATE_RUN_MSG);
    981   3266  jb145095 }
    982      0    stevel 
    983      0    stevel int
    984      0    stevel dtrace_blksuword32_err(uintptr_t addr, uint32_t *data)
    985      0    stevel {
    986      0    stevel 	int ret, watched;
    987      0    stevel 
    988      0    stevel 	watched = watch_disable_addr((void *)addr, 4, S_WRITE);
    989      0    stevel 	ret = dtrace_blksuword32(addr, data, 0);
    990      0    stevel 	if (watched)
    991      0    stevel 		watch_enable_addr((void *)addr, 4, S_WRITE);
    992      0    stevel 
    993      0    stevel 	return (ret);
    994      0    stevel }
    995      0    stevel 
    996      0    stevel int
    997      0    stevel dtrace_blksuword32(uintptr_t addr, uint32_t *data, int tryagain)
    998      0    stevel {
    999      0    stevel 	if (suword32((void *)addr, *data) == -1)
   1000      0    stevel 		return (tryagain ? dtrace_blksuword32_err(addr, data) : -1);
   1001      0    stevel 	dtrace_flush_sec(addr);
   1002      0    stevel 
   1003      0    stevel 	return (0);
   1004      0    stevel }
   1005      0    stevel 
   1006      0    stevel /*ARGSUSED*/
   1007      0    stevel void
   1008      0    stevel cpu_faulted_enter(struct cpu *cp)
   1009      0    stevel {
   1010      0    stevel }
   1011      0    stevel 
   1012      0    stevel /*ARGSUSED*/
   1013      0    stevel void
   1014      0    stevel cpu_faulted_exit(struct cpu *cp)
   1015      0    stevel {
   1016      0    stevel }
   1017      0    stevel 
   1018      0    stevel static int
   1019      0    stevel kdi_cpu_ready_iter(int (*cb)(int, void *), void *arg)
   1020      0    stevel {
   1021      0    stevel 	int rc, i;
   1022      0    stevel 
   1023      0    stevel 	for (rc = 0, i = 0; i < NCPU; i++) {
   1024      0    stevel 		if (CPU_IN_SET(cpu_ready_set, i))
   1025      0    stevel 			rc += cb(i, arg);
   1026      0    stevel 	}
   1027      0    stevel 
   1028      0    stevel 	return (rc);
   1029      0    stevel }
   1030      0    stevel 
   1031      0    stevel /*
   1032      0    stevel  * Sends a cross-call to a specified processor.  The caller assumes
   1033      0    stevel  * responsibility for repetition of cross-calls, as appropriate (MARSA for
   1034      0    stevel  * debugging).
   1035      0    stevel  */
   1036      0    stevel static int
   1037      0    stevel kdi_xc_one(int cpuid, void (*func)(uintptr_t, uintptr_t), uintptr_t arg1,
   1038      0    stevel     uintptr_t arg2)
   1039      0    stevel {
   1040      0    stevel 	int stat;
   1041      0    stevel 	struct machcpu	*mcpup;
   1042      0    stevel 	uint64_t cpuaddr_reg = 0, cpuaddr_scr = 0;
   1043      0    stevel 
   1044      0    stevel 	mcpup = &(((cpu_t *)get_cpuaddr(cpuaddr_reg, cpuaddr_scr))->cpu_m);
   1045      0    stevel 
   1046      0    stevel 	/*
   1047      0    stevel 	 * if (idsr_busy())
   1048      0    stevel 	 *	return (KDI_XC_RES_ERR);
   1049      0    stevel 	 */
   1050      0    stevel 
   1051      0    stevel 	init_mondo_nocheck((xcfunc_t *)func, arg1, arg2);
   1052      0    stevel 
   1053      0    stevel 	mcpup->cpu_list[0] = (uint16_t)cpuid;
   1054      0    stevel 	stat = shipit(1, mcpup->cpu_list_ra);
   1055      0    stevel 
   1056      0    stevel 	if (stat == 0)
   1057      0    stevel 		return (KDI_XC_RES_OK);
   1058      0    stevel 	else
   1059      0    stevel 		return (KDI_XC_RES_NACK);
   1060      0    stevel }
   1061      0    stevel 
   1062      0    stevel static void
   1063      0    stevel kdi_tickwait(clock_t nticks)
   1064      0    stevel {
   1065      0    stevel 	clock_t endtick = gettick() + nticks;
   1066      0    stevel 
   1067   4776   jm22469 	while (gettick() < endtick)
   1068   5974   jm22469 		;
   1069      0    stevel }
   1070      0    stevel 
   1071      0    stevel static void
   1072      0    stevel kdi_cpu_init(int dcache_size, int dcache_linesize, int icache_size,
   1073      0    stevel     int icache_linesize)
   1074      0    stevel {
   1075      0    stevel 	kdi_dcache_size = dcache_size;
   1076      0    stevel 	kdi_dcache_linesize = dcache_linesize;
   1077      0    stevel 	kdi_icache_size = icache_size;
   1078      0    stevel 	kdi_icache_linesize = icache_linesize;
   1079      0    stevel }
   1080      0    stevel 
   1081      0    stevel /* used directly by kdi_read/write_phys */
   1082      0    stevel void
   1083      0    stevel kdi_flush_caches(void)
   1084      0    stevel {
   1085  10271     Jason 	/* Not required on sun4v architecture. */
   1086      0    stevel }
   1087      0    stevel 
   1088      0    stevel /*ARGSUSED*/
   1089      0    stevel int
   1090      0    stevel kdi_get_stick(uint64_t *stickp)
   1091      0    stevel {
   1092      0    stevel 	return (-1);
   1093      0    stevel }
   1094      0    stevel 
   1095      0    stevel void
   1096      0    stevel cpu_kdi_init(kdi_t *kdi)
   1097      0    stevel {
   1098      0    stevel 	kdi->kdi_flush_caches = kdi_flush_caches;
   1099      0    stevel 	kdi->mkdi_cpu_init = kdi_cpu_init;
   1100      0    stevel 	kdi->mkdi_cpu_ready_iter = kdi_cpu_ready_iter;
   1101      0    stevel 	kdi->mkdi_xc_one = kdi_xc_one;
   1102      0    stevel 	kdi->mkdi_tickwait = kdi_tickwait;
   1103      0    stevel 	kdi->mkdi_get_stick = kdi_get_stick;
   1104      0    stevel }
   1105      0    stevel 
   1106   3266  jb145095 uint64_t	soft_state_message_ra[SOLARIS_SOFT_STATE_MSG_CNT];
   1107   3266  jb145095 static uint64_t	soft_state_saved_state = (uint64_t)-1;
   1108   3266  jb145095 static int	soft_state_initialized = 0;
   1109   3266  jb145095 static uint64_t soft_state_sup_minor;		/* Supported minor number */
   1110   3266  jb145095 static hsvc_info_t soft_state_hsvc = {
   1111   3266  jb145095 			HSVC_REV_1, NULL, HSVC_GROUP_SOFT_STATE, 1, 0, NULL };
   1112   3266  jb145095 
   1113   3266  jb145095 
   1114   2036   wentaoy static void
   1115   2036   wentaoy sun4v_system_claim(void)
   1116   2036   wentaoy {
   1117  11066    rafael 	lbolt_debug_entry();
   1118  11066    rafael 
   1119   2036   wentaoy 	watchdog_suspend();
   1120   8542      Haik 	kldc_debug_enter();
   1121   3266  jb145095 	/*
   1122   3266  jb145095 	 * For "mdb -K", set soft state to debugging
   1123   3266  jb145095 	 */
   1124   3266  jb145095 	if (soft_state_saved_state == -1) {
   1125   3266  jb145095 		mach_get_soft_state(&soft_state_saved_state,
   1126   4776   jm22469 		    &SOLARIS_SOFT_STATE_SAVED_MSG);
   1127   3266  jb145095 	}
   1128   3266  jb145095 	/*
   1129   3266  jb145095 	 * check again as the read above may or may not have worked and if
   1130   3266  jb145095 	 * it didn't then soft state will still be -1
   1131   3266  jb145095 	 */
   1132   3266  jb145095 	if (soft_state_saved_state != -1) {
   1133   3266  jb145095 		mach_set_soft_state(SIS_TRANSITION,
   1134   4776   jm22469 		    &SOLARIS_SOFT_STATE_DEBUG_MSG);
   1135   3266  jb145095 	}
   1136   2036   wentaoy }
   1137   2036   wentaoy 
   1138   2036   wentaoy static void
   1139   2036   wentaoy sun4v_system_release(void)
   1140   2036   wentaoy {
   1141   2036   wentaoy 	watchdog_resume();
   1142   3266  jb145095 	/*
   1143   3266  jb145095 	 * For "mdb -K", set soft_state state back to original state on exit
   1144   3266  jb145095 	 */
   1145   3266  jb145095 	if (soft_state_saved_state != -1) {
   1146   3266  jb145095 		mach_set_soft_state(soft_state_saved_state,
   1147   4776   jm22469 		    &SOLARIS_SOFT_STATE_SAVED_MSG);
   1148   3266  jb145095 		soft_state_saved_state = -1;
   1149   3266  jb145095 	}
   1150  11066    rafael 
   1151  11066    rafael 	lbolt_debug_return();
   1152   2036   wentaoy }
   1153   2036   wentaoy 
   1154   2036   wentaoy void
   1155   2036   wentaoy plat_kdi_init(kdi_t *kdi)
   1156   2036   wentaoy {
   1157   2036   wentaoy 	kdi->pkdi_system_claim = sun4v_system_claim;
   1158   2036   wentaoy 	kdi->pkdi_system_release = sun4v_system_release;
   1159   2036   wentaoy }
   1160   2036   wentaoy 
   1161      0    stevel /*
   1162      0    stevel  * Routine to return memory information associated
   1163      0    stevel  * with a physical address and syndrome.
   1164      0    stevel  */
   1165      0    stevel /* ARGSUSED */
   1166      0    stevel int
   1167      0    stevel cpu_get_mem_info(uint64_t synd, uint64_t afar,
   1168      0    stevel     uint64_t *mem_sizep, uint64_t *seg_sizep, uint64_t *bank_sizep,
   1169      0    stevel     int *segsp, int *banksp, int *mcidp)
   1170      0    stevel {
   1171      0    stevel 	return (ENOTSUP);
   1172      0    stevel }
   1173      0    stevel 
   1174      0    stevel /*
   1175      0    stevel  * This routine returns the size of the kernel's FRU name buffer.
   1176      0    stevel  */
   1177      0    stevel size_t
   1178      0    stevel cpu_get_name_bufsize()
   1179      0    stevel {
   1180      0    stevel 	return (UNUM_NAMLEN);
   1181      0    stevel }
   1182      0    stevel 
   1183      0    stevel /*
   1184      0    stevel  * This routine is a more generic interface to cpu_get_mem_unum(),
   1185      0    stevel  * that may be used by other modules (e.g. mm).
   1186      0    stevel  */
   1187      0    stevel /* ARGSUSED */
   1188      0    stevel int
   1189      0    stevel cpu_get_mem_name(uint64_t synd, uint64_t *afsr, uint64_t afar,
   1190      0    stevel     char *buf, int buflen, int *lenp)
   1191      0    stevel {
   1192      0    stevel 	return (ENOTSUP);
   1193      0    stevel }
   1194      0    stevel 
   1195   1186   ayznaga /* ARGSUSED */
   1196   1186   ayznaga int
   1197   1186   ayznaga cpu_get_mem_sid(char *unum, char *buf, int buflen, int *lenp)
   1198   1186   ayznaga {
   1199   1186   ayznaga 	return (ENOTSUP);
   1200   1186   ayznaga }
   1201   1186   ayznaga 
   1202   1186   ayznaga /* ARGSUSED */
   1203   1186   ayznaga int
   1204   1186   ayznaga cpu_get_mem_addr(char *unum, char *sid, uint64_t offset, uint64_t *addrp)
   1205   1186   ayznaga {
   1206   1186   ayznaga 	return (ENOTSUP);
   1207   1186   ayznaga }
   1208   1186   ayznaga 
   1209      0    stevel /*
   1210      0    stevel  * xt_sync - wait for previous x-traps to finish
   1211      0    stevel  */
   1212      0    stevel void
   1213      0    stevel xt_sync(cpuset_t cpuset)
   1214      0    stevel {
   1215      0    stevel 	union {
   1216      0    stevel 		uint8_t volatile byte[NCPU];
   1217      0    stevel 		uint64_t volatile xword[NCPU / 8];
   1218      0    stevel 	} cpu_sync;
   1219   5146   svemuri 	uint64_t starttick, endtick, tick, lasttick, traptrace_id;
   1220   2492  ha137994 	uint_t largestid, smallestid;
   1221   3316  rf157361 	int i, j;
   1222      0    stevel 
   1223      0    stevel 	kpreempt_disable();
   1224      0    stevel 	CPUSET_DEL(cpuset, CPU->cpu_id);
   1225      0    stevel 	CPUSET_AND(cpuset, cpu_ready_set);
   1226   2492  ha137994 
   1227   2492  ha137994 	CPUSET_BOUNDS(cpuset, smallestid, largestid);
   1228   2492  ha137994 	if (smallestid == CPUSET_NOTINSET)
   1229   2492  ha137994 		goto out;
   1230      0    stevel 
   1231      0    stevel 	/*
   1232      0    stevel 	 * Sun4v uses a queue for receiving mondos. Successful
   1233      0    stevel 	 * transmission of a mondo only indicates that the mondo
   1234      0    stevel 	 * has been written into the queue.
   1235      0    stevel 	 *
   1236      0    stevel 	 * We use an array of bytes to let each cpu to signal back
   1237      0    stevel 	 * to the cross trap sender that the cross trap has been
   1238      0    stevel 	 * executed. Set the byte to 1 before sending the cross trap
   1239      0    stevel 	 * and wait until other cpus reset it to 0.
   1240      0    stevel 	 */
   1241      0    stevel 	bzero((void *)&cpu_sync, NCPU);
   1242   2492  ha137994 	cpu_sync.byte[smallestid] = 1;
   1243   2492  ha137994 	if (largestid != smallestid) {
   1244   2492  ha137994 		for (i = (smallestid + 1); i <= (largestid - 1); i++)
   1245   2492  ha137994 			if (CPU_IN_SET(cpuset, i))
   1246   2492  ha137994 				cpu_sync.byte[i] = 1;
   1247   2492  ha137994 		cpu_sync.byte[largestid] = 1;
   1248   2492  ha137994 	}
   1249      0    stevel 
   1250   5146   svemuri 	/*
   1251   5146   svemuri 	 * To help debug xt_sync panic, each mondo is uniquely identified
   1252   5146   svemuri 	 * by passing the tick value, traptrace_id as the second mondo
   1253   5146   svemuri 	 * argument to xt_some which is logged in CPU's mondo queue,
   1254   5146   svemuri 	 * traptrace buffer and the panic message.
   1255   5146   svemuri 	 */
   1256   5146   svemuri 	traptrace_id = gettick();
   1257      0    stevel 	xt_some(cpuset, (xcfunc_t *)xt_sync_tl1,
   1258   5146   svemuri 	    (uint64_t)cpu_sync.byte, traptrace_id);
   1259      0    stevel 
   1260      0    stevel 	starttick = lasttick = gettick();
   1261   6138   svemuri 	endtick = starttick + xc_sync_tick_limit;
   1262      0    stevel 
   1263   2492  ha137994 	for (i = (smallestid / 8); i <= (largestid / 8); i++) {
   1264      0    stevel 		while (cpu_sync.xword[i] != 0) {
   1265      0    stevel 			tick = gettick();
   1266      0    stevel 			/*
   1267      0    stevel 			 * If there is a big jump between the current tick
   1268      0    stevel 			 * count and lasttick, we have probably hit a break
   1269      0    stevel 			 * point. Adjust endtick accordingly to avoid panic.
   1270      0    stevel 			 */
   1271      0    stevel 			if (tick > (lasttick + xc_tick_jump_limit)) {
   1272      0    stevel 				endtick += (tick - lasttick);
   1273      0    stevel 			}
   1274      0    stevel 			lasttick = tick;
   1275      0    stevel 			if (tick > endtick) {
   1276      0    stevel 				if (panic_quiesce)
   1277      0    stevel 					goto out;
   1278   3316  rf157361 				cmn_err(CE_CONT, "Cross trap sync timeout:  "
   1279   5146   svemuri 				    "at cpu_sync.xword[%d]: 0x%lx "
   1280   5146   svemuri 				    "cpu_sync.byte: 0x%lx "
   1281   5146   svemuri 				    "starttick: 0x%lx endtick: 0x%lx "
   1282   5146   svemuri 				    "traptrace_id = 0x%lx\n",
   1283   5146   svemuri 				    i, cpu_sync.xword[i],
   1284   5146   svemuri 				    (uint64_t)cpu_sync.byte,
   1285   5146   svemuri 				    starttick, endtick, traptrace_id);
   1286   5146   svemuri 				cmn_err(CE_CONT, "CPUIDs:");
   1287   3316  rf157361 				for (j = (i * 8); j <= largestid; j++) {
   1288   3316  rf157361 					if (cpu_sync.byte[j] != 0)
   1289   3316  rf157361 						cmn_err(CE_CONT, " 0x%x", j);
   1290   3316  rf157361 				}
   1291      0    stevel 				cmn_err(CE_PANIC, "xt_sync: timeout");
   1292      0    stevel 			}
   1293      0    stevel 		}
   1294      0    stevel 	}
   1295      0    stevel 
   1296      0    stevel out:
   1297      0    stevel 	kpreempt_enable();
   1298      0    stevel }
   1299   2957   jm22469 
   1300   6138   svemuri #define	QFACTOR		200
   1301   2957   jm22469 /*
   1302   2957   jm22469  * Recalculate the values of the cross-call timeout variables based
   1303   2957   jm22469  * on the value of the 'inter-cpu-latency' property of the platform node.
   1304   2957   jm22469  * The property sets the number of nanosec to wait for a cross-call
   1305   2957   jm22469  * to be acknowledged.  Other timeout variables are derived from it.
   1306   2957   jm22469  *
   1307   2957   jm22469  * N.B. This implementation is aware of the internals of xc_init()
   1308   2957   jm22469  * and updates many of the same variables.
   1309   2957   jm22469  */
   1310   2957   jm22469 void
   1311   2957   jm22469 recalc_xc_timeouts(void)
   1312   2957   jm22469 {
   1313   2957   jm22469 	typedef union {
   1314   2957   jm22469 		uint64_t whole;
   1315   2957   jm22469 		struct {
   1316   2957   jm22469 			uint_t high;
   1317   2957   jm22469 			uint_t low;
   1318   2957   jm22469 		} half;
   1319   2957   jm22469 	} u_number;
   1320   2957   jm22469 
   1321   2957   jm22469 	/* See x_call.c for descriptions of these extern variables. */
   1322   2957   jm22469 	extern uint64_t xc_tick_limit_scale;
   1323   2957   jm22469 	extern uint64_t xc_mondo_time_limit;
   1324   2957   jm22469 	extern uint64_t xc_func_time_limit;
   1325   2957   jm22469 	extern uint64_t xc_scale;
   1326   2957   jm22469 	extern uint64_t xc_mondo_multiplier;
   1327   2957   jm22469 	extern uint_t   nsec_shift;
   1328   2957   jm22469 
   1329   2957   jm22469 	/* Temp versions of the target variables */
   1330   2957   jm22469 	uint64_t tick_limit;
   1331   2957   jm22469 	uint64_t tick_jump_limit;
   1332   2957   jm22469 	uint64_t mondo_time_limit;
   1333   2957   jm22469 	uint64_t func_time_limit;
   1334   2957   jm22469 	uint64_t scale;
   1335   2957   jm22469 
   1336   2957   jm22469 	uint64_t latency;	/* nanoseconds */
   1337   2957   jm22469 	uint64_t maxfreq;
   1338   2957   jm22469 	uint64_t tick_limit_save = xc_tick_limit;
   1339   6138   svemuri 	uint64_t sync_tick_limit_save = xc_sync_tick_limit;
   1340   2957   jm22469 	uint_t   tick_scale;
   1341   2957   jm22469 	uint64_t top;
   1342   2957   jm22469 	uint64_t bottom;
   1343   2957   jm22469 	u_number tk;
   1344   2957   jm22469 
   1345   2957   jm22469 	md_t *mdp;
   1346   2957   jm22469 	int nrnode;
   1347   2957   jm22469 	mde_cookie_t *platlist;
   1348   2957   jm22469 
   1349   2957   jm22469 	/*
   1350   2957   jm22469 	 * Look up the 'inter-cpu-latency' (optional) property in the
   1351   2957   jm22469 	 * platform node of the MD.  The units are nanoseconds.
   1352   2957   jm22469 	 */
   1353   2957   jm22469 	if ((mdp = md_get_handle()) == NULL) {
   1354   2957   jm22469 		cmn_err(CE_WARN, "recalc_xc_timeouts: "
   1355   2957   jm22469 		    "Unable to initialize machine description");
   1356   2957   jm22469 		return;
   1357   2957   jm22469 	}
   1358   2957   jm22469 
   1359   2957   jm22469 	nrnode = md_alloc_scan_dag(mdp,
   1360   2957   jm22469 	    md_root_node(mdp), "platform", "fwd", &platlist);
   1361   2957   jm22469 
   1362   2957   jm22469 	ASSERT(nrnode == 1);
   1363   2957   jm22469 	if (nrnode < 1) {
   1364   2957   jm22469 		cmn_err(CE_WARN, "recalc_xc_timeouts: platform node missing");
   1365   3069   jm22469 		goto done;
   1366   2957   jm22469 	}
   1367   2957   jm22469 	if (md_get_prop_val(mdp, platlist[0],
   1368   2957   jm22469 	    "inter-cpu-latency", &latency) == -1)
   1369   3069   jm22469 		goto done;
   1370   2957   jm22469 
   1371   2957   jm22469 	/*
   1372   2957   jm22469 	 * clock.h defines an assembly-language macro
   1373   2957   jm22469 	 * (NATIVE_TIME_TO_NSEC_SCALE) to convert from %stick
   1374   2957   jm22469 	 * units to nanoseconds.  Since the inter-cpu-latency
   1375   2957   jm22469 	 * units are nanoseconds and the xc_* variables require
   1376   2957   jm22469 	 * %stick units, we need the inverse of that function.
   1377   2957   jm22469 	 * The trick is to perform the calculation without
   1378   2957   jm22469 	 * floating point, but also without integer truncation
   1379   2957   jm22469 	 * or overflow.  To understand the calculation below,
   1380   2957   jm22469 	 * please read the discussion of the macro in clock.h.
   1381   2957   jm22469 	 * Since this new code will be invoked infrequently,
   1382   2957   jm22469 	 * we can afford to implement it in C.
   1383   2957   jm22469 	 *
   1384   2957   jm22469 	 * tick_scale is the reciprocal of nsec_scale which is
   1385   2957   jm22469 	 * calculated at startup in setcpudelay().  The calc
   1386   2957   jm22469 	 * of tick_limit parallels that of NATIVE_TIME_TO_NSEC_SCALE
   1387   2957   jm22469 	 * except we use tick_scale instead of nsec_scale and
   1388   2957   jm22469 	 * C instead of assembler.
   1389   2957   jm22469 	 */
   1390   2957   jm22469 	tick_scale = (uint_t)(((u_longlong_t)sys_tick_freq
   1391   2957   jm22469 	    << (32 - nsec_shift)) / NANOSEC);
   1392   2957   jm22469 
   1393   2957   jm22469 	tk.whole = latency;
   1394   2957   jm22469 	top = ((uint64_t)tk.half.high << 4) * tick_scale;
   1395   2957   jm22469 	bottom = (((uint64_t)tk.half.low << 4) * (uint64_t)tick_scale) >> 32;
   1396   2957   jm22469 	tick_limit = top + bottom;
   1397   2957   jm22469 
   1398   2957   jm22469 	/*
   1399   2957   jm22469 	 * xc_init() calculated 'maxfreq' by looking at all the cpus,
   1400   2957   jm22469 	 * and used it to derive some of the timeout variables that we
   1401   2957   jm22469 	 * recalculate below.  We can back into the original value by
   1402   2957   jm22469 	 * using the inverse of one of those calculations.
   1403   2957   jm22469 	 */
   1404   2957   jm22469 	maxfreq = xc_mondo_time_limit / xc_scale;
   1405   2957   jm22469 
   1406   2957   jm22469 	/*
   1407   2957   jm22469 	 * Don't allow the new timeout (xc_tick_limit) to fall below
   1408   2957   jm22469 	 * the system tick frequency (stick).  Allowing the timeout
   1409   2957   jm22469 	 * to be set more tightly than this empirically determined
   1410   2957   jm22469 	 * value may cause panics.
   1411   2957   jm22469 	 */
   1412   2957   jm22469 	tick_limit = tick_limit < sys_tick_freq ? sys_tick_freq : tick_limit;
   1413   2957   jm22469 
   1414   2957   jm22469 	tick_jump_limit = tick_limit / 32;
   1415   2957   jm22469 	tick_limit *= xc_tick_limit_scale;
   1416   2957   jm22469 
   1417   2957   jm22469 	/*
   1418   2957   jm22469 	 * Recalculate xc_scale since it is used in a callback function
   1419   2957   jm22469 	 * (xc_func_timeout_adj) to adjust two of the timeouts dynamically.
   1420   2957   jm22469 	 * Make the change in xc_scale proportional to the change in
   1421   2957   jm22469 	 * xc_tick_limit.
   1422   2957   jm22469 	 */
   1423   2957   jm22469 	scale = (xc_scale * tick_limit + sys_tick_freq / 2) / tick_limit_save;
   1424   2957   jm22469 	if (scale == 0)
   1425   2957   jm22469 		scale = 1;
   1426   2957   jm22469 
   1427   2957   jm22469 	mondo_time_limit = maxfreq * scale;
   1428   2957   jm22469 	func_time_limit = mondo_time_limit * xc_mondo_multiplier;
   1429   2957   jm22469 
   1430   2957   jm22469 	/*
   1431   2957   jm22469 	 * Don't modify the timeouts if nothing has changed.  Else,
   1432   2957   jm22469 	 * stuff the variables with the freshly calculated (temp)
   1433   2957   jm22469 	 * variables.  This minimizes the window where the set of
   1434   2957   jm22469 	 * values could be inconsistent.
   1435   2957   jm22469 	 */
   1436   2957   jm22469 	if (tick_limit != xc_tick_limit) {
   1437   2957   jm22469 		xc_tick_limit = tick_limit;
   1438   2957   jm22469 		xc_tick_jump_limit = tick_jump_limit;
   1439   2957   jm22469 		xc_scale = scale;
   1440   2957   jm22469 		xc_mondo_time_limit = mondo_time_limit;
   1441   2957   jm22469 		xc_func_time_limit = func_time_limit;
   1442   2957   jm22469 	}
   1443   3069   jm22469 
   1444   3069   jm22469 done:
   1445   6138   svemuri 	/*
   1446   6138   svemuri 	 * Increase the timeout limit for xt_sync() cross calls.
   1447   6138   svemuri 	 */
   1448   6138   svemuri 	xc_sync_tick_limit = xc_tick_limit * (cpu_q_entries / QFACTOR);
   1449   6138   svemuri 	xc_sync_tick_limit = xc_sync_tick_limit < xc_tick_limit ?
   1450   6138   svemuri 	    xc_tick_limit : xc_sync_tick_limit;
   1451   6138   svemuri 
   1452   6138   svemuri 	/*
   1453   6138   svemuri 	 * Force the new values to be used for future cross calls.
   1454   6138   svemuri 	 * This is necessary only when we increase the timeouts.
   1455   6138   svemuri 	 */
   1456   6138   svemuri 	if ((xc_tick_limit > tick_limit_save) || (xc_sync_tick_limit >
   1457   6138   svemuri 	    sync_tick_limit_save)) {
   1458   6138   svemuri 		cpuset_t cpuset = cpu_ready_set;
   1459   6138   svemuri 		xt_sync(cpuset);
   1460   6138   svemuri 	}
   1461   6138   svemuri 
   1462   3069   jm22469 	if (nrnode > 0)
   1463   3069   jm22469 		md_free_scan_dag(mdp, &platlist);
   1464   3069   jm22469 	(void) md_fini_handle(mdp);
   1465   2957   jm22469 }
   1466   3266  jb145095 
   1467   3266  jb145095 void
   1468   3266  jb145095 mach_soft_state_init(void)
   1469   3266  jb145095 {
   1470   3266  jb145095 	int		i;
   1471   3266  jb145095 	uint64_t	ra;
   1472   3266  jb145095 
   1473   3266  jb145095 	/*
   1474   3266  jb145095 	 * Try to register soft_state api. If it fails, soft_state api has not
   1475   3266  jb145095 	 * been implemented in the firmware, so do not bother to setup
   1476   3266  jb145095 	 * soft_state in the kernel.
   1477   3266  jb145095 	 */
   1478   3266  jb145095 	if ((i = hsvc_register(&soft_state_hsvc, &soft_state_sup_minor)) != 0) {
   1479   3266  jb145095 		return;
   1480   3266  jb145095 	}
   1481   3266  jb145095 	for (i = 0; i < SOLARIS_SOFT_STATE_MSG_CNT; i++) {
   1482   3266  jb145095 		ASSERT(strlen((const char *)(void *)
   1483   4776   jm22469 		    soft_state_message_strings + i) < SSM_SIZE);
   1484   4776   jm22469 		if ((ra = va_to_pa(
   1485   4776   jm22469 		    (void *)(soft_state_message_strings + i))) == -1ll) {
   1486   3266  jb145095 			return;
   1487   3266  jb145095 		}
   1488   3266  jb145095 		soft_state_message_ra[i] = ra;
   1489   3266  jb145095 	}
   1490   3266  jb145095 	/*
   1491   3266  jb145095 	 * Tell OBP that we are supporting Guest State
   1492   3266  jb145095 	 */
   1493   3266  jb145095 	prom_sun4v_soft_state_supported();
   1494   3266  jb145095 	soft_state_initialized = 1;
   1495   3266  jb145095 }
   1496   3266  jb145095 
   1497   3266  jb145095 void
   1498   3266  jb145095 mach_set_soft_state(uint64_t state, uint64_t *string_ra)
   1499   3266  jb145095 {
   1500   3266  jb145095 	uint64_t	rc;
   1501   3266  jb145095 
   1502   3266  jb145095 	if (soft_state_initialized && *string_ra) {
   1503   3266  jb145095 		rc = hv_soft_state_set(state, *string_ra);
   1504   3266  jb145095 		if (rc != H_EOK) {
   1505   5146   svemuri 			cmn_err(CE_WARN,
   1506   5146   svemuri 			    "hv_soft_state_set returned %ld\n", rc);
   1507   3266  jb145095 		}
   1508   3266  jb145095 	}
   1509   3266  jb145095 }
   1510   3266  jb145095 
   1511   3266  jb145095 void
   1512   3266  jb145095 mach_get_soft_state(uint64_t *state, uint64_t *string_ra)
   1513   3266  jb145095 {
   1514   4776   jm22469 	uint64_t	rc;
   1515   3266  jb145095 
   1516   3266  jb145095 	if (soft_state_initialized && *string_ra) {
   1517   3266  jb145095 		rc = hv_soft_state_get(*string_ra, state);
   1518   3266  jb145095 		if (rc != H_EOK) {
   1519   5146   svemuri 			cmn_err(CE_WARN,
   1520   5146   svemuri 			    "hv_soft_state_get returned %ld\n", rc);
   1521   3266  jb145095 			*state = -1;
   1522   3266  jb145095 		}
   1523   3266  jb145095 	}
   1524   3266  jb145095 }
   1525