]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/ppc/cpu.c
xnu-1228.3.13.tar.gz
[apple/xnu.git] / osfmk / ppc / cpu.c
index cd42fdd121ae938e4270402c0a54c1f13694b58d..3a77bccb6384ade11a6d321bc492e317f30a2381 100644 (file)
@@ -1,23 +1,29 @@
 /*
- * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
  * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ * 
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
  * 
- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
 #include <mach/mach_types.h>
 #include <kern/thread.h>
 #include <kern/sched_prim.h>
 #include <kern/processor.h>
+#include <kern/pms.h>
 
 #include <vm/pmap.h>
 #include <IOKit/IOHibernatePrivate.h>
 
 #include <ppc/proc_reg.h>
 #include <ppc/misc_protos.h>
+#include <ppc/fpu_protos.h>
 #include <ppc/machine_routines.h>
 #include <ppc/cpu_internal.h>
 #include <ppc/exception.h>
@@ -48,7 +56,6 @@
 #include <ppc/Diagnostics.h>
 #include <ppc/trap.h>
 #include <ppc/machine_cpu.h>
-#include <ppc/pms.h>
 #include <ppc/rtclock.h>
 
 decl_mutex_data(static,ppt_lock);
@@ -65,15 +72,15 @@ static unsigned int rht_state = 0;
 decl_simple_lock_data(static,SignalReadyLock);
 
 struct SIGtimebase {
-       boolean_t       avail;
-       boolean_t       ready;
-       boolean_t       done;
+       volatile boolean_t      avail;
+       volatile boolean_t      ready;
+       volatile boolean_t      done;
        uint64_t        abstime;
 };
 
-perfCallback           perfCpuSigHook = 0;                     /* Pointer to CHUD cpu signal hook routine */
+perfCallback           perfCpuSigHook;                 /* Pointer to CHUD cpu signal hook routine */
 
-extern int                     debugger_sync;
+extern uint32_t                        debugger_sync;
 
 /*
  * Forward definitions
@@ -120,8 +127,9 @@ cpu_init(
                mttbu(proc_info->save_tbu);
                mttb(proc_info->save_tbl);
        }
-       
-       setTimerReq();                          /* Now that the time base is sort of correct, request the next timer pop */
+
+       proc_info->rtcPop = EndOfAllTime;                       /* forget any existing decrementer setting */
+       etimer_resync_deadlines();                              /* Now that the time base is sort of correct, request the next timer pop */
 
        proc_info->cpu_type = CPU_TYPE_POWERPC;
        proc_info->cpu_subtype = (cpu_subtype_t)proc_info->pf.rptdProc;
@@ -186,7 +194,7 @@ cpu_machine_init(
        proc_info->cpu_flags |= BootDone|SignalReady;
        if (proc_info != mproc_info) {
                if (proc_info->ppXFlags & SignalReadyWait) {
-                       hw_atomic_and(&proc_info->ppXFlags, ~SignalReadyWait);
+                       (void)hw_atomic_and(&proc_info->ppXFlags, ~SignalReadyWait);
                        thread_wakeup(&proc_info->cpu_flags);
                }
                simple_unlock(&SignalReadyLock);
@@ -203,9 +211,9 @@ struct per_proc_info *
 cpu_per_proc_alloc(
                void)
 {
-       struct per_proc_info    *proc_info=0;
-       void                    *interrupt_stack=0;
-       void                    *debugger_stack=0;
+       struct per_proc_info    *proc_info = NULL;
+       void                    *interrupt_stack = NULL;
+       void                    *debugger_stack = NULL;
 
        if ((proc_info = (struct per_proc_info*)kalloc(sizeof(struct per_proc_info))) == (struct per_proc_info*)0)
                return (struct per_proc_info *)NULL;
@@ -222,7 +230,10 @@ cpu_per_proc_alloc(
 
        bzero((void *)proc_info, sizeof(struct per_proc_info));
 
-       proc_info->pp2ndPage = (addr64_t)pmap_find_phys(kernel_pmap, (addr64_t)proc_info + 0x1000) << PAGE_SHIFT;       /* Set physical address of the second page */
+       /* Set physical address of the second page */
+       proc_info->pp2ndPage = (addr64_t)pmap_find_phys(kernel_pmap,
+                               ((addr64_t)(unsigned int)proc_info) + 0x1000)
+                              << PAGE_SHIFT;
        proc_info->next_savearea = (uint64_t)save_get_init();
        proc_info->pf = BootProcInfo.pf;
        proc_info->istackptr = (vm_offset_t)interrupt_stack + INTSTACK_SIZE - FM_SIZE;
@@ -271,7 +282,7 @@ cpu_per_proc_register(
        cpu = real_ncpus;
        proc_info->cpu_number = cpu;
        PerProcTable[cpu].ppe_vaddr = proc_info;
-       PerProcTable[cpu].ppe_paddr = (addr64_t)pmap_find_phys(kernel_pmap, (addr64_t)proc_info) << PAGE_SHIFT;
+       PerProcTable[cpu].ppe_paddr = (addr64_t)pmap_find_phys(kernel_pmap, (addr64_t)(unsigned int)proc_info) << PAGE_SHIFT;
        eieio();
        real_ncpus++;
        mutex_unlock(&ppt_lock);
@@ -305,8 +316,8 @@ cpu_start(
                proc_info->pending_ast = AST_NONE;
                proc_info->istackptr = proc_info->intstack_top_ss;
                proc_info->rtcPop = EndOfAllTime;
-               proc_info->FPU_owner = 0;
-               proc_info->VMX_owner = 0;
+               proc_info->FPU_owner = NULL;
+               proc_info->VMX_owner = NULL;
                proc_info->pms.pmsStamp = 0;                                                                    /* Dummy transition time */
                proc_info->pms.pmsPop = EndOfAllTime;                                                   /* Set the pop way into the future */
                proc_info->pms.pmsState = pmsParked;                                                    /* Park the stepper */
@@ -358,7 +369,7 @@ cpu_start(
                } else {
                        simple_lock(&SignalReadyLock);
                        if (!((*(volatile short *)&proc_info->cpu_flags) & SignalReady)) {
-                               hw_atomic_or(&proc_info->ppXFlags, SignalReadyWait);
+                               (void)hw_atomic_or(&proc_info->ppXFlags, SignalReadyWait);
                                thread_sleep_simple_lock((event_t)&proc_info->cpu_flags,
                                                          &SignalReadyLock, THREAD_UNINT);
                        }
@@ -417,12 +428,13 @@ cpu_sleep(
        proc_info->running = FALSE;
 
        fowner = proc_info->FPU_owner;                                  /* Cache this */
-       if(fowner) fpu_save(fowner);                                    /* If anyone owns FPU, save it */
-       proc_info->FPU_owner = 0;                                               /* Set no fpu owner now */
+       if(fowner) /* If anyone owns FPU, save it */
+               fpu_save(fowner);
+       proc_info->FPU_owner = NULL;                                            /* Set no fpu owner now */
 
        fowner = proc_info->VMX_owner;                                  /* Cache this */
        if(fowner) vec_save(fowner);                                    /* If anyone owns vectors, save it */
-       proc_info->VMX_owner = 0;                                               /* Set no vector owner now */
+       proc_info->VMX_owner = NULL;                                            /* Set no vector owner now */
 
        if (proc_info->cpu_number == master_cpu)  {
                proc_info->cpu_flags &= BootDone;
@@ -558,11 +570,9 @@ cpu_signal(
  *     the lock and signal the other guy.
  */
 void 
-cpu_signal_handler(
-       void)
+cpu_signal_handler(void)
 {
-
-       unsigned int holdStat, holdParm0, holdParm1, holdParm2, mtype;
+       unsigned int holdStat, holdParm0, holdParm1, holdParm2;
        unsigned int *parmAddr;
        struct per_proc_info    *proc_info;
        int cpu;
@@ -654,7 +664,7 @@ cpu_signal_handler(
 
                                        proc_info->hwCtr.numSIGPdebug++;        /* Count this one */
                                        proc_info->debugger_is_slave++;         /* Bump up the count to show we're here */
-                                       hw_atomic_sub(&debugger_sync, 1);       /* Show we've received the 'rupt */
+                                       (void)hw_atomic_sub(&debugger_sync, 1); /* Show we've received the 'rupt */
                                        __asm__ volatile("tw 4,r3,r3"); /* Enter the debugger */
                                        return;                                                 /* All done now... */
                                        
@@ -664,7 +674,7 @@ cpu_signal_handler(
                                        
                                case SIGPcall:                                          /* Call function on CPU */
                                        proc_info->hwCtr.numSIGPcall++; /* Count this one */
-                                       xfunc = holdParm1;                              /* Do this since I can't seem to figure C out */
+                                       xfunc = (broadcastFunc)holdParm1;                               /* Do this since I can't seem to figure C out */
                                        xfunc(holdParm2);                               /* Call the passed function */
                                        return;                                                 /* Done... */
                                        
@@ -705,7 +715,7 @@ cpu_sync_timebase(
                                                        (unsigned int)&syncClkSpot) != KERN_SUCCESS)
                continue;
 
-       while (*(volatile int *)&(syncClkSpot.avail) == FALSE)
+       while (syncClkSpot.avail == FALSE)
                continue;
 
        isync();
@@ -723,11 +733,10 @@ cpu_sync_timebase(
 
        syncClkSpot.ready = TRUE;
 
-       while (*(volatile int *)&(syncClkSpot.done) == FALSE)
+       while (syncClkSpot.done == FALSE)
                continue;
 
-       setTimerReq();                                                                  /* Start the timer */
-       
+       etimer_resync_deadlines();                                                                      /* Start the timer */
        (void)ml_set_interrupts_enabled(intr);
 }
 
@@ -760,7 +769,8 @@ cpu_timebase_signal_handler(
                                                
        timebaseAddr->avail = TRUE;
 
-       while (*(volatile int *)&(timebaseAddr->ready) == FALSE);
+       while (timebaseAddr->ready == FALSE)
+               continue;
 
        if(proc_info->time_base_enable !=  (void(*)(cpu_id_t, boolean_t ))NULL)
                proc_info->time_base_enable(proc_info->cpu_id, TRUE);
@@ -1115,34 +1125,56 @@ cpu_threadtype(void)
  *     It is not passed to the other processor and must be known by the called function.
  *     The called function must do a thread_wakeup on the synch if it decrements the
  *     synch count to 0.
+ *
+ *     We start by initializing the synchronizer to the number of possible cpus.
+ *     The we signal each popssible processor.
+ *     If the signal fails, we count it.  We also skip our own.
+ *     When we are finished signaling, we adjust the syncronizer count down buy the number of failed signals.
+ *     Because the signaled processors are also decrementing the synchronizer count, the adjustment may result in a 0
+ *     If this happens, all other processors are finished with the function.
+ *     If so, we clear the wait and continue
+ *     Otherwise, we block waiting for the other processor(s) to finish.
+ *
+ *     Meanwhile, the other processors are decrementing the synchronizer when they are done
+ *     If it goes to zero, thread_wakeup is called to run the broadcaster
+ *
+ *     Note that because we account for the broadcaster in the synchronization count, we will not get any
+ *     premature wakeup calls.
+ *
+ *     Also note that when we do the adjustment of the synchronization count, it the result is 0, it means that
+ *     all of the other processors are finished.  Otherwise, we know that there is at least one more. 
+ *     When that thread decrements the synchronizer to zero, it will do a thread_wake.
+ *     
  */
 
-
-int32_t cpu_broadcast(uint32_t *synch, broadcastFunc func, uint32_t parm) {
-
-       int sigproc, cpu, ocpu;
-
-       cpu = cpu_number();                                                                     /* Who are we? */
-       sigproc = 0;                                                                            /* Clear called processor count */
-
-       if(real_ncpus > 1) {                                                            /* Are we just a uni? */
+int32_t
+cpu_broadcast(uint32_t *synch, broadcastFunc func, uint32_t parm)
+{
+       int failsig;
+       unsigned int cpu, ocpu;
        
+       cpu = cpu_number();                                             /* Who are we? */
+       failsig = 0;                                                    /* Clear called processor count */
+       
+       if(real_ncpus > 1) {                                            /* Are we just a uni? */
+               
+               *synch = real_ncpus;                                    /* Set how many we are going to try */
                assert_wait((event_t)synch, THREAD_UNINT);              /* If more than one processor, we may have to wait */
-
+               
                for(ocpu = 0; ocpu < real_ncpus; ocpu++) {              /* Tell everyone to call */
-                       if(ocpu == cpu) continue;                                       /* If we talk to ourselves, people will wonder... */
-                       hw_atomic_add(synch, 1);                                        /* Tentatively bump synchronizer  */
-                       sigproc++;                                                                      /* Tentatively bump signal sent count */
+                       
+                       if(ocpu == cpu) continue;                       /* If we talk to ourselves, people will wonder... */
+                       
                        if(KERN_SUCCESS != cpu_signal(ocpu, SIGPcall, (uint32_t)func, parm)) {  /* Call the function on the other processor */
-                               hw_atomic_sub(synch, 1);                                /* Other guy isn't really there, ignore it  */
-                               sigproc--;                                                              /* and don't count it */
+                               failsig++;                              /* Count failed signals */
                        }
                }
-
-               if(!sigproc) clear_wait(current_thread(), THREAD_AWAKENED);     /* Clear wait if we never signalled */
-               else thread_block(THREAD_CONTINUE_NULL);                /* Wait for everyone to get into step... */
+               
+               if (hw_atomic_sub(synch, failsig + 1) == 0)
+                       clear_wait(current_thread(), THREAD_AWAKENED);  /* Clear wait if we never signalled or all of the others finished */
+               else
+                       thread_block(THREAD_CONTINUE_NULL);             /* Wait for everyone to get into step... */
        }
-
-       return sigproc;                                                                         /* Return the number of guys actually signalled */
-
+       
+       return (real_ncpus - failsig - 1);                              /* Return the number of guys actually signalled... */
 }