Backout the acpicpu_setpdc code and the code to allow est to use acpicpu

on amd64.
Evaluating the _PDC seems to lead to instability (PR5878 and a report by
jmc@) on lower end machines its too close to release and I don't have
enough nickels.

ok deraadt@

1 files changed, 167 insertions, 101 deletions

diff --git a/sys/arch/amd64/amd64/est.c b/sys/arch/amd64/amd64/est.c
index c1fcd87d753..f31c0a4f72d 100644
--- a/sys/arch/amd64/amd64/est.c
+++ b/sys/arch/amd64/amd64/est.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: est.c,v 1.6 2008/06/15 05:24:07 gwk Exp $ */
+/*	$OpenBSD: est.c,v 1.7 2008/08/06 05:24:44 gwk Exp $ */
 /*
  * Copyright (c) 2003 Michael Eriksson.
  * All rights reserved.
@@ -56,122 +56,148 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/sysctl.h>
-#include <sys/malloc.h>
 
 #include <machine/cpu.h>
 #include <machine/cpufunc.h>
 #include <machine/specialreg.h>
-#include <machine/bus.h>
 
-#include "acpicpu.h"
+#define CPUVENDOR_INTEL 0
+#define CPUVENDOR_VIA 8
 
-#if NACPICPU > 0
-#include <dev/acpi/acpidev.h>
-#include <dev/acpi/acpivar.h>
-#endif
 
-struct est_op {
-	uint16_t ctrl;
-	uint16_t mhz;
-};
+/* Convert MHz and mV into IDs for passing to the MSR. */
+#define ID16(MHz, mV, bus_clk) \
+	((((MHz * 100 + 50) / bus_clk) << 8) | ((mV ? mV - 700 : 0) >> 4))
+
+/* Possible bus speeds (multiplied by 100 for rounding) */
+#define BUS100 10000
+#define BUS133 13333
+#define BUS166 16667
+#define BUS200 20000
+#define BUS266 26667
+#define BUS333 33333
+
+#define MSR2MHZ(msr, bus) \
+	(((((int) (msr) >> 8) & 0xff) * (bus) + 50) / 100)
+#define MSR2MV(msr) \
+	(((int) (msr) & 0xff) * 16 + 700)
 
 struct fqlist {
+	int vendor: 5;
+	unsigned bus_clk : 1;
 	unsigned n : 5;
-	struct est_op *table;
+	const u_int16_t *table;
 };
 
+static const struct fqlist *est_fqlist;
 
-static struct fqlist *est_fqlist;
+static u_int16_t fake_table[3];
+static struct fqlist fake_fqlist;
 
 extern int setperf_prio;
 extern int perflevel;
 
-#if NACPICPU > 0
-struct fqlist * est_acpi_init(void);
-void est_acpi_pss_changed(struct acpicpu_pss *, int);
-
-struct fqlist *
-est_acpi_init()
-{
-	struct acpicpu_pss *pss;
-	struct fqlist *acpilist;
-	int nstates, i;
-
-	if ((nstates = acpicpu_fetch_pss(&pss)) == 0)
-		goto nolist;
+int bus_clock;
 
-	if ((acpilist = malloc(sizeof(struct fqlist), M_DEVBUF, M_NOWAIT))
-	    == NULL)
-		goto nolist;
+void p4_get_bus_clock(struct cpu_info *);
+void p3_get_bus_clock(struct cpu_info *);
 
-	if ((acpilist->table = malloc(sizeof(struct est_op) * nstates, M_DEVBUF,
-	   M_NOWAIT)) == NULL)
-		goto notable;
-
-	acpilist->n = nstates;
-
-	for (i = 0; i < nstates; i++) {
-		acpilist->table[i].mhz = pss[i].pss_core_freq;
-		acpilist->table[i].ctrl = pss[i].pss_ctrl;
+void
+p4_get_bus_clock(struct cpu_info *ci)
+{
+	u_int64_t msr;
+	int model, bus;
+
+	model = (ci->ci_signature >> 4) & 15;
+	msr = rdmsr(MSR_EBC_FREQUENCY_ID);
+	if (model < 2) {
+		bus = (msr >> 21) & 0x7;
+		switch (bus) {
+		case 0:
+			bus_clock = BUS100;
+			break;
+		case 1:
+			bus_clock = BUS133;
+			break;
+		default:
+			printf("%s: unknown Pentium 4 (model %d) "
+			    "EBC_FREQUENCY_ID value %d\n",
+			    ci->ci_dev->dv_xname, model, bus);
+			break;
+		}
+	} else {
+		bus = (msr >> 16) & 0x7;
+		switch (bus) {
+		case 0:
+			bus_clock = (model == 2) ? BUS100 : BUS266;
+			break;
+		case 1:
+			bus_clock = BUS133;
+			break;
+		case 2:
+			bus_clock = BUS200;
+			break;
+		case 3:
+			bus_clock = BUS166;
+			break;
+		default:
+			printf("%s: unknown Pentium 4 (model %d) "
+			    "EBC_FREQUENCY_ID value %d\n",
+			    ci->ci_dev->dv_xname, model, bus);
+			break;
+		}
 	}
-
-	acpicpu_set_notify(est_acpi_pss_changed);
-
-	return acpilist;
-
-notable:
-	free(acpilist, M_DEVBUF);
-	acpilist = NULL;
-nolist:
-	return NULL;
 }
 
 void
-est_acpi_pss_changed(struct acpicpu_pss *pss, int npss)
+p3_get_bus_clock(struct cpu_info *ci)
 {
-	struct fqlist *acpilist;
-	int needtran = 1, nstates, i;
 	u_int64_t msr;
-	u_int16_t cur;
-
-	msr = rdmsr(MSR_PERF_STATUS);
-	cur = msr & 0xffff;
-
-	if ((acpilist = malloc(sizeof(struct fqlist), M_DEVBUF, M_NOWAIT))
-	    == NULL) {
-		printf("est_acpi_pss_changed: cannot allocate memory for new est state");
-		return;
-	}
-
-	if ((acpilist->table = malloc(sizeof(struct est_op) * nstates, M_DEVBUF,
-	   M_NOWAIT)) == NULL) {
-		printf("est_acpi_pss_changed: cannot allocate memory for new operating points");
-		free(acpilist, M_DEVBUF);
-		return;
-	}
-
-	for (i = 0; i < nstates; i++) {
-		acpilist->table[i].mhz = pss[i].pss_core_freq;
-		acpilist->table[i].ctrl = pss[i].pss_ctrl;
-		if (pss[i].pss_ctrl == cur)
-			needtran = 0;
-	}
-
-	free(est_fqlist->table, M_DEVBUF);
-	free(est_fqlist, M_DEVBUF);
-	est_fqlist = acpilist;
-
-	if (needtran) {
-		est_setperf(perflevel);
+	int model, bus;
+
+	model = (ci->ci_signature >> 4) & 15;
+	switch (model) {
+	case 0xe: /* Core Duo/Solo */
+	case 0xf: /* Core Xeon */
+		msr = rdmsr(MSR_FSB_FREQ);
+		bus = (msr >> 0) & 0x7;
+		switch (bus) {
+		case 5:
+			bus_clock = BUS100;
+			break;
+		case 1:
+			bus_clock = BUS133;
+			break;
+		case 3:
+			bus_clock = BUS166;
+			break;
+		case 2:
+			bus_clock = BUS200;
+			break;
+		case 0:
+			bus_clock = BUS266;
+			break;
+		case 4:
+			bus_clock = BUS333;
+			break;
+		default:
+			printf("%s: unknown Core FSB_FREQ value %d",
+			    ci->ci_dev->dv_xname, bus);
+			break;
+		}
+		break;
+	default: 
+		printf("%s: unknown i686 model %d, can't get bus clock",
+		    ci->ci_dev->dv_xname, model);
 	}
 }
-#endif
 
 void
 est_init(struct cpu_info *ci)
 {
 	const char *cpu_device = ci->ci_dev->dv_xname;
-	int i, low, high;
+	int vendor = -1;
+	int i, mhz, mv, low, high, family;
 	u_int64_t msr;
 	u_int16_t idhi, idlo, cur;
 	u_int8_t crhi, crlo, crcur;
@@ -179,6 +205,17 @@ est_init(struct cpu_info *ci)
 	if (setperf_prio > 3)
 		return;
 
+	family = (ci->ci_signature >> 8) & 15;
+	if (family == 0xf) {
+		p4_get_bus_clock(ci);
+	} else if (family == 6) {
+		p3_get_bus_clock(ci);
+	}
+	if (bus_clock == 0) {
+		printf("%s: EST: unknown system bus clock\n", cpu_device);
+		return;
+	}
+
 	msr = rdmsr(MSR_PERF_STATUS);
 	idhi = (msr >> 32) & 0xffff;
 	idlo = (msr >> 48) & 0xffff;
@@ -186,7 +223,6 @@ est_init(struct cpu_info *ci)
 	crhi = (idhi  >> 8) & 0xff;
 	crlo = (idlo  >> 8) & 0xff;
 	crcur = (cur >> 8) & 0xff;
-
 	if (crlo == 0 || crhi == crlo) {
 		/*
 		 * Don't complain about these cases, and silently disable EST:
@@ -197,27 +233,58 @@ est_init(struct cpu_info *ci)
 		 */
 		return;
 	}
-
-
-#if NACPICPU > 0
-		est_fqlist = est_acpi_init();
-#endif
-	if (est_fqlist == NULL)
+	if (crhi == 0 || crcur == 0 || crlo > crhi ||
+	    crcur < crlo || crcur > crhi) {
+		/*
+		 * Do complain about other weirdness, because we first want to
+		 * know about it, before we decide what to do with it.
+		 */
+		printf("%s: EST: strange msr value 0x%016llx\n",
+		    cpu_device, msr);
 		return;
+	}
+	if (est_fqlist == NULL) {
+		printf("%s: unknown Enhanced SpeedStep CPU, msr 0x%016llx\n",
+		    cpu_device, msr);
 
-	printf("%s: Enhanced SpeedStep %d MHz", cpu_device, cpuspeed);
+		/*
+		 * Generate a fake table with the power states we know.
+		 */
+		fake_table[0] = idhi;
+		if (cur == idhi || cur == idlo) {
+			printf("%s: using only highest and lowest power "
+			    "states\n", cpu_device);
+
+			fake_table[1] = idlo;
+			fake_fqlist.n = 2;
+		} else {
+			printf("%s: using only highest, current and lowest "
+			    "power states\n", cpu_device);
+
+			fake_table[1] = cur;
+			fake_table[2] = idlo;
+			fake_fqlist.n = 3;
+		}
+		fake_fqlist.vendor = vendor;
+		fake_fqlist.table = fake_table;
+		est_fqlist = &fake_fqlist;
+	}
 
-	low = est_fqlist->table[est_fqlist->n - 1].mhz;
-	high = est_fqlist->table[0].mhz;
-	perflevel = (cpuspeed - low) * 100 / (high - low);
+	mhz = MSR2MHZ(cur, bus_clock);
+	mv = MSR2MV(cur);
+	printf("%s: Enhanced SpeedStep %d MHz (%d mV)", cpu_device, mhz, mv);
+
+	low = MSR2MHZ(est_fqlist->table[est_fqlist->n - 1], bus_clock);
+	high = MSR2MHZ(est_fqlist->table[0], bus_clock);
+	perflevel = (mhz - low) * 100 / (high - low);
 
 	/*
 	 * OK, tell the user the available frequencies.
 	 */
 	printf(": speeds: ");
 	for (i = 0; i < est_fqlist->n; i++)
-		printf("%d%s", est_fqlist->table[i].mhz, i < est_fqlist->n - 1 ?
-		    ", " : " MHz\n");
+		printf("%d%s", MSR2MHZ(est_fqlist->table[i], bus_clock),
+		    i < est_fqlist->n - 1 ? ", " : " MHz\n");
 
 	cpu_setperf = est_setperf;
 	setperf_prio = 3;
@@ -239,8 +306,7 @@ est_setperf(int level)
 
 	msr = rdmsr(MSR_PERF_CTL);
 	msr &= ~0xffffULL;
-	msr |= est_fqlist->table[i].ctrl;
-
+	msr |= est_fqlist->table[i];
 	wrmsr(MSR_PERF_CTL, msr);
-	cpuspeed = est_fqlist->table[i].mhz;
+	cpuspeed = MSR2MHZ(est_fqlist->table[i], bus_clock);
 }