patch-2.4.10 linux/arch/ppc/kernel/pmac_smp.c

• Lines: 490
• Date: Sat Sep 8 12:38:42 2001
• Orig file: v2.4.9/linux/arch/ppc/kernel/pmac_smp.c
• Orig date: Wed Dec 31 16:00:00 1969

diff -u --recursive --new-file v2.4.9/linux/arch/ppc/kernel/pmac_smp.c linux/arch/ppc/kernel/pmac_smp.c
@@ -0,0 +1,489 @@
+/*
+ * BK Id: %F% %I% %G% %U% %#%
+ */
+/*
+ * SMP support for power macintosh.
+ *
+ * We support both the old "powersurge" SMP architecture
+ * and the current Core99 (G4 PowerMac) machines.
+ *
+ * Support Macintosh G4 SMP by Troy Benjegerdes (hozer@drgw.net)
+ * and Ben Herrenschmidt <benh@kernel.crashing.org>.
+ *
+ * Support for DayStar quad CPU cards
+ * Copyright (C) XLR8, Inc. 1994-2000
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/interrupt.h>
+#include <linux/kernel_stat.h>
+#include <linux/delay.h>
+#define __KERNEL_SYSCALLS__
+#include <linux/unistd.h>
+#include <linux/init.h>
+#include <linux/spinlock.h>
+
+#include <asm/ptrace.h>
+#include <asm/atomic.h>
+#include <asm/irq.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/hardirq.h>
+#include <asm/softirq.h>
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/smp.h>
+#include <asm/residual.h>
+#include <asm/feature.h>
+#include <asm/time.h>
+#include <asm/gemini.h>
+
+#include "open_pic.h"
+
+/*
+ * Powersurge (old powermac SMP) support.
+ */
+
+extern void __secondary_start_psurge(void);
+extern void __secondary_start_psurge2(void);	/* Temporary horrible hack */
+extern void __secondary_start_psurge3(void);	/* Temporary horrible hack */
+
+/* Addresses for powersurge registers */
+#define HAMMERHEAD_BASE		0xf8000000
+#define HHEAD_CONFIG		0x90
+#define HHEAD_SEC_INTR		0xc0
+
+/* register for interrupting the primary processor on the powersurge */
+/* N.B. this is actually the ethernet ROM! */
+#define PSURGE_PRI_INTR		0xf3019000
+
+/* register for storing the start address for the secondary processor */
+/* N.B. this is the PCI config space address register for the 1st bridge */
+#define PSURGE_START		0xf2800000
+
+/* Daystar/XLR8 4-CPU card */
+#define PSURGE_QUAD_REG_ADDR	0xf8800000
+
+#define PSURGE_QUAD_IRQ_SET	0
+#define PSURGE_QUAD_IRQ_CLR	1
+#define PSURGE_QUAD_IRQ_PRIMARY	2
+#define PSURGE_QUAD_CKSTOP_CTL	3
+#define PSURGE_QUAD_PRIMARY_ARB	4
+#define PSURGE_QUAD_BOARD_ID	6
+#define PSURGE_QUAD_WHICH_CPU	7
+#define PSURGE_QUAD_CKSTOP_RDBK	8
+#define PSURGE_QUAD_RESET_CTL	11
+
+#define PSURGE_QUAD_OUT(r, v)	(out_8(quad_base + ((r) << 4) + 4, (v)))
+#define PSURGE_QUAD_IN(r)	(in_8(quad_base + ((r) << 4) + 4) & 0x0f)
+#define PSURGE_QUAD_BIS(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
+#define PSURGE_QUAD_BIC(r, v)	(PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
+
+/* virtual addresses for the above */
+static volatile u8 *hhead_base;
+static volatile u8 *quad_base;
+static volatile u32 *psurge_pri_intr;
+static volatile u8 *psurge_sec_intr;
+static volatile u32 *psurge_start;
+
+/* what sort of powersurge board we have */
+static int psurge_type;
+
+/* values for psurge_type */
+#define PSURGE_DUAL		0
+#define PSURGE_QUAD_OKEE	1
+#define PSURGE_QUAD_COTTON	2
+#define PSURGE_QUAD_ICEGRASS	3
+
+/* l2 cache stuff for dual G4 macs */
+extern void core99_init_l2(void);
+
+/*
+ * Set and clear IPIs for powersurge.
+ */
+static inline void psurge_set_ipi(int cpu)
+{
+	if (cpu == 0)
+		in_be32(psurge_pri_intr);
+	else if (psurge_type == PSURGE_DUAL)
+		out_8(psurge_sec_intr, 0);
+	else
+		PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
+}
+
+static inline void psurge_clr_ipi(int cpu)
+{
+	if (cpu > 0) {
+		if (psurge_type == PSURGE_DUAL)
+			out_8(psurge_sec_intr, ~0);
+		else
+			PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
+	}
+}
+
+/*
+ * On powersurge (old SMP powermac architecture) we don't have
+ * separate IPIs for separate messages like openpic does.  Instead
+ * we have a bitmap for each processor, where a 1 bit means that
+ * the corresponding message is pending for that processor.
+ * Ideally each cpu's entry would be in a different cache line.
+ *  -- paulus.
+ */
+static unsigned long psurge_smp_message[NR_CPUS];
+
+void __pmac
+psurge_smp_message_recv(struct pt_regs *regs)
+{
+	int cpu = smp_processor_id();
+	int msg;
+
+	/* clear interrupt */
+	psurge_clr_ipi(cpu);
+
+	if (smp_num_cpus < 2)
+		return;
+
+	/* make sure there is a message there */
+	for (msg = 0; msg < 4; msg++)
+		if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
+			smp_message_recv(msg, regs);
+}
+
+void __pmac
+psurge_primary_intr(int irq, void *d, struct pt_regs *regs)
+{
+	psurge_smp_message_recv(regs);
+}
+
+static void __pmac
+smp_psurge_message_pass(int target, int msg, unsigned long data, int wait)
+{
+	int i;
+
+	if (smp_num_cpus < 2)
+		return;
+
+	for (i = 0; i < smp_num_cpus; i++) {
+		if (target == MSG_ALL
+		    || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
+		    || target == i) {
+			set_bit(msg, &psurge_smp_message[i]);
+			psurge_set_ipi(i);
+		}
+	}
+}
+
+/*
+ * Determine a quad card presence. We read the board ID register, we
+ * force the data bus to change to something else, and we read it again.
+ * It it's stable, then the register probably exist (ugh !)
+ */
+static int __init psurge_quad_probe(void)
+{
+	int type;
+	unsigned int i;
+
+	type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
+	if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
+	    || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
+		return PSURGE_DUAL;
+
+	/* looks OK, try a slightly more rigorous test */
+	/* bogus is not necessarily cacheline-aligned,
+	   though I don't suppose that really matters.  -- paulus */
+	for (i = 0; i < 100; i++) {
+		volatile u32 bogus[8];
+		bogus[(0+i)%8] = 0x00000000;
+		bogus[(1+i)%8] = 0x55555555;
+		bogus[(2+i)%8] = 0xFFFFFFFF;
+		bogus[(3+i)%8] = 0xAAAAAAAA;
+		bogus[(4+i)%8] = 0x33333333;
+		bogus[(5+i)%8] = 0xCCCCCCCC;
+		bogus[(6+i)%8] = 0xCCCCCCCC;
+		bogus[(7+i)%8] = 0x33333333;
+		wmb();
+		asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
+		mb();
+		if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
+			return PSURGE_DUAL;
+	}
+	return type;
+}
+
+static void __init psurge_quad_init(void)
+{
+	int procbits;
+
+	if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
+	procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
+	if (psurge_type == PSURGE_QUAD_ICEGRASS)
+		PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
+	else
+		PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
+	mdelay(33);
+	out_8(psurge_sec_intr, ~0);
+	PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
+	PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
+	if (psurge_type != PSURGE_QUAD_ICEGRASS)
+		PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
+	PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
+	mdelay(33);
+	PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
+	mdelay(33);
+	PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
+	mdelay(33);
+}
+
+static int __init smp_psurge_probe(void)
+{
+	int i, ncpus;
+
+	/* We don't do SMP on the PPC601 -- paulus */
+	if (PVR_VER(mfspr(PVR)) == 1)
+		return 1;
+
+	/*
+	 * The powersurge cpu board can be used in the generation
+	 * of powermacs that have a socket for an upgradeable cpu card,
+	 * including the 7500, 8500, 9500, 9600.
+	 * The device tree doesn't tell you if you have 2 cpus because
+	 * OF doesn't know anything about the 2nd processor.
+	 * Instead we look for magic bits in magic registers,
+	 * in the hammerhead memory controller in the case of the
+	 * dual-cpu powersurge board.  -- paulus.
+	 */
+	if (find_devices("hammerhead") == NULL)
+		return 1;
+
+	hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
+	quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
+	psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
+
+	psurge_type = psurge_quad_probe();
+	if (psurge_type != PSURGE_DUAL) {
+		psurge_quad_init();
+		/* All released cards using this HW design have 4 CPUs */
+		ncpus = 4;
+	} else {
+		iounmap((void *) quad_base);
+		if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
+			/* not a dual-cpu card */
+			iounmap((void *) hhead_base);
+			return 1;
+		}
+		ncpus = 2;
+	}
+
+	psurge_start = ioremap(PSURGE_START, 4);
+	psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
+
+	/* this is not actually strictly necessary -- paulus. */
+	for (i = 1; i < ncpus; ++i)
+		smp_hw_index[i] = i;
+
+	if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
+
+	return ncpus;
+}
+
+static void __init smp_psurge_kick_cpu(int nr)
+{
+	void (*start)(void) = __secondary_start_psurge;
+	unsigned long a;
+
+	/* may need to flush here if secondary bats aren't setup */
+	for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
+		asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
+	asm volatile("sync");
+
+	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
+
+	/* setup entry point of secondary processor */
+	switch (nr) {
+	case 2:
+		start = __secondary_start_psurge2;
+		break;
+	case 3:
+		start = __secondary_start_psurge3;
+		break;
+	}
+
+	out_be32(psurge_start, __pa(start));
+	mb();
+
+	psurge_set_ipi(nr);
+	udelay(10);
+	psurge_clr_ipi(nr);
+
+	if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
+}
+
+/*
+ * With the dual-cpu powersurge board, the decrementers and timebases
+ * of both cpus are frozen after the secondary cpu is started up,
+ * until we give the secondary cpu another interrupt.  This routine
+ * uses this to get the timebases synchronized.
+ *  -- paulus.
+ */
+static void __init psurge_dual_sync_tb(int cpu_nr)
+{
+	static volatile int sec_tb_reset = 0;
+	int t;
+
+	set_dec(tb_ticks_per_jiffy);
+	set_tb(0, 0);
+	last_jiffy_stamp(cpu_nr) = 0;
+
+	if (cpu_nr > 0) {
+		mb();
+		sec_tb_reset = 1;
+		return;
+	}
+
+	/* wait for the secondary to have reset its TB before proceeding */
+	for (t = 10000000; t > 0 && !sec_tb_reset; --t)
+		;
+
+	/* now interrupt the secondary, starting both TBs */
+	psurge_set_ipi(1);
+
+	smp_tb_synchronized = 1;
+}
+
+static void __init
+smp_psurge_setup_cpu(int cpu_nr)
+{
+
+	if (cpu_nr == 0) {
+		if (smp_num_cpus < 2)
+			return;
+		/* reset the entry point so if we get another intr we won't
+		 * try to startup again */
+		out_be32(psurge_start, 0x100);
+		if (request_irq(30, psurge_primary_intr, 0, "primary IPI", 0))
+			printk(KERN_ERR "Couldn't get primary IPI interrupt");
+	}
+
+	if (psurge_type == PSURGE_DUAL)
+		psurge_dual_sync_tb(cpu_nr);
+}
+
+static int __init
+smp_core99_probe(void)
+{
+	struct device_node *cpus;
+	int i, ncpus = 1;
+
+	if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
+	cpus = find_type_devices("cpu");
+	if (cpus)
+		while ((cpus = cpus->next) != NULL)
+			++ncpus;
+	printk("smp_core99_probe: found %d cpus\n", ncpus);
+	if (ncpus > 1) {
+		openpic_request_IPIs();
+		for (i = 1; i < ncpus; ++i)
+			smp_hw_index[i] = i;
+	}
+
+	return ncpus;
+}
+
+static void __init
+smp_core99_kick_cpu(int nr)
+{
+	unsigned long save_vector, new_vector;
+	unsigned long flags;
+#if 1 /* New way... */
+	volatile unsigned long *vector
+		 = ((volatile unsigned long *)(KERNELBASE+0x100));
+	if (nr < 1 || nr > 3)
+		return;
+#else
+	volatile unsigned long *vector
+		 = ((volatile unsigned long *)(KERNELBASE+0x500));
+	if (nr != 1)
+		return;
+#endif
+	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu", 0x346);
+
+	local_irq_save(flags);
+	local_irq_disable();
+	
+	/* Save reset vector */
+	save_vector = *vector;
+	
+	/* Setup fake reset vector that does	  
+	 *   b __secondary_start_psurge - KERNELBASE
+	 */  
+	switch(nr) {
+		case 1:
+			new_vector = (unsigned long)__secondary_start_psurge;
+			break;
+		case 2:
+			new_vector = (unsigned long)__secondary_start_psurge2;
+			break;
+		case 3:
+			new_vector = (unsigned long)__secondary_start_psurge3;
+			break;
+	}
+	*vector = 0x48000002 + new_vector - KERNELBASE;
+	
+	/* flush data cache and inval instruction cache */
+	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
+	
+	/* Put some life in our friend */
+	feature_core99_kick_cpu(nr);
+	
+	/* FIXME: We wait a bit for the CPU to take the exception, I should
+	 * instead wait for the entry code to set something for me. Well,
+	 * ideally, all that crap will be done in prom.c and the CPU left
+	 * in a RAM-based wait loop like CHRP.
+	 */
+	mdelay(1);
+	
+	/* Restore our exception vector */
+	*vector = save_vector;
+	flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
+	
+	local_irq_restore(flags);
+	if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
+}
+
+static void __init
+smp_core99_setup_cpu(int cpu_nr)
+{
+	/* Setup openpic */
+	do_openpic_setup_cpu();
+
+	/* Setup L2 */
+	if (cpu_nr != 0)
+		core99_init_l2();
+	else
+		if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349);
+}
+
+/* PowerSurge-style Macs */
+struct smp_ops_t psurge_smp_ops __pmacdata = {
+	smp_psurge_message_pass,
+	smp_psurge_probe,
+	smp_psurge_kick_cpu,
+	smp_psurge_setup_cpu,
+};
+
+/* Core99 Macs (dual G4s) */
+struct smp_ops_t core99_smp_ops __pmacdata = {
+	smp_openpic_message_pass,
+	smp_core99_probe,
+	smp_core99_kick_cpu,
+	smp_core99_setup_cpu,
+};