It seems that the CPUID lies about the monitor-line size, or at least our

interpretation of it isn't quite right.  So instead of allocating memory
and slicing it based on the parameters returned by CPUID, simply use a member
in struct cpu_info like basically all other OSes out there do.  Our struct
cpu_info is large enough to never cause any overlap.  This makes the
mwait-based idle loop actually work.  We still execute the CPUID instruction
to make sure monitor/mwait is properly supported by the hardware we're
running on.

ok sthen@, deraadt@, guenther@

diff --git a/sys/arch/amd64/amd64/cpu.c b/sys/arch/amd64/amd64/cpu.c
index db7efb8..4337d0e 100644 (file)
--- a/sys/arch/amd64/amd64/cpu.c
+++ b/sys/arch/amd64/amd64/cpu.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: cpu.c,v 1.81 2015/03/25 21:05:18 kettenis Exp $       */
+/*     $OpenBSD: cpu.c,v 1.82 2015/04/18 22:16:21 kettenis Exp $       */
 /* $NetBSD: cpu.c,v 1.1 2003/04/26 18:39:26 fvdl Exp $ */
 
 /*-
@@ -243,7 +243,6 @@ void
 cpu_idle_mwait_cycle(void)
 {
        struct cpu_info *ci = curcpu();
-       volatile int *state = &ci->ci_mwait[0];
 
        if ((read_rflags() & PSL_I) == 0)
                panic("idle with interrupts blocked!");
@@ -262,18 +261,18 @@ cpu_idle_mwait_cycle(void)
         * something to the queue and called cpu_unidle() between
         * the check in sched_idle() and here.
         */
-       atomic_setbits_int(state, MWAIT_IDLING);
+       atomic_setbits_int(&ci->ci_mwait, MWAIT_IDLING);
        if (ci->ci_schedstate.spc_whichqs == 0) {
-               monitor(state, 0, 0);
-               if ((*state & MWAIT_IDLING) == MWAIT_IDLING)
+               monitor(&ci->ci_mwait, 0, 0);
+               if ((ci->ci_mwait & MWAIT_IDLING) == MWAIT_IDLING)
                        mwait(0, 0);
        }
 
        /* done idling; let cpu_kick() know that an IPI is required */
-       atomic_clearbits_int(state, MWAIT_IDLING);
+       atomic_clearbits_int(&ci->ci_mwait, MWAIT_IDLING);
 }
 
-unsigned int mwait_size;
+u_int cpu_mwait_size;
 
 void
 cpu_init_mwait(struct cpu_softc *sc)
@@ -306,40 +305,15 @@ cpu_init_mwait(struct cpu_softc *sc)
            (largest & (sizeof(int)-1)))
                printf(" (bogus)");
        else
-               mwait_size = largest;
+               cpu_mwait_size = largest;
        printf("\n");
-       /* XXX disable mwait: ACPI says not to use it on too many systems */
-       mwait_size = 0;
 }
 
 void
 cpu_enable_mwait(void)
 {
-       unsigned long area;
-       struct cpu_info *ci;
-       CPU_INFO_ITERATOR cii;
-
-       if (mwait_size == 0)
-               return;
-
-       /*
-        * Allocate the area, with a bit extra so that we can align
-        * to a multiple of mwait_size
-        */
-       area = (unsigned long)malloc((ncpus * mwait_size) + mwait_size
-           - sizeof(int), M_DEVBUF, M_NOWAIT|M_ZERO);
-       if (area == 0) {
-               printf("cpu0: mwait failed\n");
-       } else {
-               /* round to a multiple of mwait_size  */
-               area = ((area + mwait_size - sizeof(int)) / mwait_size)
-                   * mwait_size;
-               CPU_INFO_FOREACH(cii, ci) {
-                       ci->ci_mwait = (int *)area;
-                       area += mwait_size;
-               }
+       if (cpu_mwait_size > 0)
                cpu_idle_cycle_fcn = &cpu_idle_mwait_cycle;
-       }
 }
 #endif /* MULTIPROCESSOR */
 

diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index a519b91..f18e5bc 100644 (file)
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: machdep.c,v 1.208 2015/03/25 21:05:18 kettenis Exp $  */
+/*     $OpenBSD: machdep.c,v 1.209 2015/04/18 22:16:21 kettenis Exp $  */
 /*     $NetBSD: machdep.c,v 1.3 2003/05/07 22:58:18 fvdl Exp $ */
 
 /*-
@@ -671,15 +671,15 @@ cpu_kick(struct cpu_info *ci)
 {
        /* only need to kick other CPUs */
        if (ci != curcpu()) {
-               if (ci->ci_mwait != NULL) {
+               if (cpu_mwait_size > 0) {
                        /*
                         * If not idling, then send an IPI, else
                         * just clear the "keep idling" bit.
                         */
-                       if ((ci->ci_mwait[0] & MWAIT_IN_IDLE) == 0)
+                       if ((ci->ci_mwait & MWAIT_IN_IDLE) == 0)
                                x86_send_ipi(ci, X86_IPI_NOP);
                        else
-                               atomic_clearbits_int(&ci->ci_mwait[0],
+                               atomic_clearbits_int(&ci->ci_mwait,
                                    MWAIT_KEEP_IDLING);
                } else {
                        /* no mwait, so need an IPI */
@@ -704,12 +704,12 @@ signotify(struct proc *p)
 void
 cpu_unidle(struct cpu_info *ci)
 {
-       if (ci->ci_mwait != NULL) {
+       if (cpu_mwait_size > 0) {
                /*
                 * Just clear the "keep idling" bit; if it wasn't
                 * idling then we didn't need to do anything anyway.
                 */
-               atomic_clearbits_int(&ci->ci_mwait[0], MWAIT_KEEP_IDLING);
+               atomic_clearbits_int(&ci->ci_mwait, MWAIT_KEEP_IDLING);
                return;
        }
 

diff --git a/sys/arch/amd64/include/cpu.h b/sys/arch/amd64/include/cpu.h
index 780948b..7caefde 100644 (file)
--- a/sys/arch/amd64/include/cpu.h
+++ b/sys/arch/amd64/include/cpu.h
@@ -1,4 +1,4 @@
-/*     $OpenBSD: cpu.h,v 1.90 2015/01/15 15:30:17 sf Exp $     */
+/*     $OpenBSD: cpu.h,v 1.91 2015/04/18 22:16:21 kettenis Exp $       */
 /*     $NetBSD: cpu.h,v 1.1 2003/04/26 18:39:39 fvdl Exp $     */
 
 /*-
@@ -109,8 +109,7 @@ struct cpu_info {
        void (*cpu_setup)(struct cpu_info *);
        void (*ci_info)(struct cpu_info *);
 
-       u_int           *ci_mwait;
-/* bits in ci_mwait[0] */
+       volatile u_int  ci_mwait;
 #define        MWAIT_IN_IDLE           0x1     /* don't need IPI to wake */
 #define        MWAIT_KEEP_IDLING       0x2     /* cleared by other cpus to wake me */
 #define        MWAIT_IDLING    (MWAIT_IN_IDLE | MWAIT_KEEP_IDLING)
@@ -191,6 +190,8 @@ extern struct cpu_info *cpu_info[MAXCPUS];
 void cpu_boot_secondary_processors(void);
 void cpu_init_idle_pcbs(void);    
 
+extern u_int cpu_mwait_size;
+
 void cpu_kick(struct cpu_info *);
 void cpu_unidle(struct cpu_info *);
 

diff --git a/sys/arch/i386/i386/cpu.c b/sys/arch/i386/i386/cpu.c
index c3ecb15..78cf7fd 100644 (file)
--- a/sys/arch/i386/i386/cpu.c
+++ b/sys/arch/i386/i386/cpu.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: cpu.c,v 1.61 2015/02/11 05:54:48 dlg Exp $    */
+/*     $OpenBSD: cpu.c,v 1.62 2015/04/18 22:16:21 kettenis Exp $       */
 /* $NetBSD: cpu.c,v 1.1.2.7 2000/06/26 02:04:05 sommerfeld Exp $ */
 
 /*-
@@ -789,7 +789,6 @@ void
 cpu_idle_mwait_cycle(void)
 {
        struct cpu_info *ci = curcpu();
-       volatile int *state = &ci->ci_mwait[0];
 
        if ((read_eflags() & PSL_I) == 0)
                panic("idle with interrupts blocked!");
@@ -808,18 +807,18 @@ cpu_idle_mwait_cycle(void)
         * something to the queue and called cpu_unidle() between
         * the check in sched_idle() and here.
         */
-       atomic_setbits_int(state, MWAIT_IDLING);
+       atomic_setbits_int(&ci->ci_mwait, MWAIT_IDLING);
        if (ci->ci_schedstate.spc_whichqs == 0) {
-               monitor(state, 0, 0);
-               if ((*state & MWAIT_IDLING) == MWAIT_IDLING)
+               monitor(&ci->ci_mwait, 0, 0);
+               if ((ci->ci_mwait & MWAIT_IDLING) == MWAIT_IDLING)
                        mwait(0, 0);
        }
 
        /* done idling; let cpu_kick() know that an IPI is required */
-       atomic_clearbits_int(state, MWAIT_IDLING);
+       atomic_clearbits_int(&ci->ci_mwait, MWAIT_IDLING);
 }
 
-unsigned int mwait_size;
+u_int cpu_mwait_size;
 
 void
 cpu_init_mwait(struct device *dv)
@@ -851,40 +850,15 @@ cpu_init_mwait(struct device *dv)
            (largest & (sizeof(int)-1)))
                printf(" (bogus)");
        else
-               mwait_size = largest;
+               cpu_mwait_size = largest;
        printf("\n");
-       /* XXX disable mwait: ACPI says not to use it on too many systems */
-       mwait_size = 0;
 }
 
 void
 cpu_enable_mwait(void)
 {
-       unsigned long area;
-       struct cpu_info *ci;
-       CPU_INFO_ITERATOR cii;
-
-       if (mwait_size == 0)
-               return;
-
-       /*
-        * Allocate the area, with a bit extra so that we can align
-        * to a multiple of mwait_size
-        */
-       area = (unsigned long)malloc((ncpus * mwait_size) + mwait_size
-           - sizeof(int), M_DEVBUF, M_NOWAIT|M_ZERO);
-       if (area == 0) {
-               printf("cpu0: mwait failed\n");
-       } else {
-               /* round to a multiple of mwait_size  */
-               area = ((area + mwait_size - sizeof(int)) / mwait_size)
-                   * mwait_size;
-               CPU_INFO_FOREACH(cii, ci) {
-                       ci->ci_mwait = (int *)area;
-                       area += mwait_size;
-               }
+       if (cpu_mwait_size > 0)
                cpu_idle_cycle_fcn = &cpu_idle_mwait_cycle;
-       }
 }
 #endif /* MULTIPROCESSOR */
 

diff --git a/sys/arch/i386/i386/machdep.c b/sys/arch/i386/i386/machdep.c
index 5027a0c..40f0a9b 100644 (file)
--- a/sys/arch/i386/i386/machdep.c
+++ b/sys/arch/i386/i386/machdep.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: machdep.c,v 1.568 2015/04/12 18:37:53 mlarkin Exp $   */
+/*     $OpenBSD: machdep.c,v 1.569 2015/04/18 22:16:21 kettenis Exp $  */
 /*     $NetBSD: machdep.c,v 1.214 1996/11/10 03:16:17 thorpej Exp $    */
 
 /*-
@@ -2532,15 +2532,15 @@ cpu_kick(struct cpu_info *ci)
 {
        /* only need to kick other CPUs */
        if (ci != curcpu()) {
-               if (ci->ci_mwait != NULL) {
+               if (cpu_mwait_size > 0) {
                        /*
                         * If not idling, then send an IPI, else
                         * just clear the "keep idling" bit.
                         */
-                       if ((ci->ci_mwait[0] & MWAIT_IN_IDLE) == 0)
+                       if ((ci->ci_mwait & MWAIT_IN_IDLE) == 0)
                                i386_send_ipi(ci, I386_IPI_NOP);
                        else
-                               atomic_clearbits_int(&ci->ci_mwait[0],
+                               atomic_clearbits_int(&ci->ci_mwait,
                                    MWAIT_KEEP_IDLING);
                } else {
                        /* no mwait, so need an IPI */
@@ -2565,12 +2565,12 @@ signotify(struct proc *p)
 void
 cpu_unidle(struct cpu_info *ci)
 {
-       if (ci->ci_mwait != NULL) {
+       if (cpu_mwait_size > 0) {
                /*
                 * Just clear the "keep idling" bit; if it wasn't
                 * idling then we didn't need to do anything anyway.
                 */
-               atomic_clearbits_int(&ci->ci_mwait[0], MWAIT_KEEP_IDLING);
+               atomic_clearbits_int(&ci->ci_mwait, MWAIT_KEEP_IDLING);
                return;
        }
 

diff --git a/sys/arch/i386/include/cpu.h b/sys/arch/i386/include/cpu.h
index 516bbe1..14033d3 100644 (file)
--- a/sys/arch/i386/include/cpu.h
+++ b/sys/arch/i386/include/cpu.h
@@ -1,4 +1,4 @@
-/*     $OpenBSD: cpu.h,v 1.138 2015/04/12 18:37:54 mlarkin Exp $       */
+/*     $OpenBSD: cpu.h,v 1.139 2015/04/18 22:16:21 kettenis Exp $      */
 /*     $NetBSD: cpu.h,v 1.35 1996/05/05 19:29:26 christos Exp $        */
 
 /*-
@@ -126,8 +126,7 @@ struct cpu_info {
        struct cpu_functions *ci_func;  /* start/stop functions */
        void (*cpu_setup)(struct cpu_info *);   /* proc-dependant init */
 
-       u_int           *ci_mwait;
-/* bits in ci_mwait[0] */
+       volatile u_int  ci_mwait;
 #define        MWAIT_IN_IDLE           0x1     /* don't need IPI to wake */
 #define        MWAIT_KEEP_IDLING       0x2     /* cleared by other cpus to wake me */
 #define        MWAIT_IDLING            (MWAIT_IN_IDLE | MWAIT_KEEP_IDLING)
@@ -214,6 +213,8 @@ extern struct cpu_info      *cpu_info[MAXCPUS];
 extern void cpu_boot_secondary_processors(void);
 extern void cpu_init_idle_pcbs(void);
 
+extern u_int cpu_mwait_size;
+
 void cpu_kick(struct cpu_info *);
 void cpu_unidle(struct cpu_info *);