[PATCH] ppc64: kexec support for ppc64

This patch implements the kexec support for ppc64 platforms.

A couple of notes:

1)  We copy the pages in virtual mode, using the full base kernel
    and a statically allocated stack.   At kexec_prepare time we
    scan the pages and if any overlap our (0, _end[]) range we
    return -ETXTBSY.

    On PowerPC 64 systems running in LPAR (logical partitioning)
    mode, only a small region of memory, referred to as the RMO,
    can be accessed in real mode.  Since Linux runs with only one
    zone of memory in the memory allocator, and it can be orders of
    magnitude more memory than the RMO, looping until we allocate
    pages in the source region is not feasible.  Copying in virtual
    means we don't have to write a hash table generation and call
    hypervisor to insert translations, instead we rely on the pinned
    kernel linear mapping.  The kernel already has move to linked
    location built in, so there is no requirement to load it at 0.

    If we want to load something other than a kernel, then a stub
    can be written to copy a linear chunk in real mode.

2)  The start entry point gets passed parameters from the kernel.
    Slaves are started at a fixed address after copying code from
    the entry point.

    All CPUs get passed their firmware assigned physical id in r3
    (most calling conventions use this register for the first
    argument).

    This is used to distinguish each CPU from all other CPUs.
    Since firmware is not around, there is no other way to obtain
    this information other than to pass it somewhere.

    A single CPU, referred to here as the master and the one executing
    the kexec call, branches to start with the address of start in r4.
    While this can be calculated, we have to load it through a gpr to
    branch to this point so defining the register this is contained
    in is free.  A stack of unspecified size is available at r1
    (also common calling convention).

    All remaining running CPUs are sent to start at absolute address
    0x60 after copying the first 0x100 bytes from start to address 0.
    This convention was chosen because it matches what the kernel
    has been doing itself.  (only gpr3 is defined).

    Note: This is not quite the convention of the kexec bootblock v2
    in the kernel.  A stub has been written to convert between them,
    and we may adjust the kernel in the future to allow this directly
    without any stub.

3)  Destination pages can be placed anywhere, even where they
    would not be accessible in real mode.  This will allow us to
    place ram disks above the RMO if we choose.

Signed-off-by: Milton Miller <[email protected]>
Signed-off-by: R Sharada <[email protected]>
Signed-off-by: Paul Mackerras <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

arch/ppc64/Kconfig

@@ -142,6 +142,23 @@ config PPC_SPLPAR
 	  processors, that is, which share physical processors between
 	  two or more partitions.
 
+config KEXEC
+	bool "kexec system call (EXPERIMENTAL)"
+	depends on PPC_MULTIPLATFORM && EXPERIMENTAL
+	help
+	  kexec is a system call that implements the ability to shutdown your
+	  current kernel, and to start another kernel.  It is like a reboot
+	  but it is indepedent of the system firmware.  And like a reboot
+	  you can start any kernel with it, not just Linux.
+
+	  The name comes from the similiarity to the exec system call.
+
+	  It is an ongoing process to be certain the hardware in a machine
+	  is properly shutdown, so do not be surprised if this code does not
+	  initially work for you.  It may help to enable device hotplugging
+	  support.  As of this writing the exact hardware interface is
+	  strongly in flux, so no good recommendation can be made.
+
 config IBMVIO
 	depends on PPC_PSERIES || PPC_ISERIES
 	bool

arch/ppc64/kernel/Makefile

@@ -36,6 +36,7 @@ obj-$(CONFIG_PPC_PSERIES) += pSeries_pci.o pSeries_lpar.o pSeries_hvCall.o \
 obj-$(CONFIG_PPC_BPA) += bpa_setup.o bpa_iommu.o bpa_nvram.o \
 			 bpa_iic.o spider-pic.o
 
+obj-$(CONFIG_KEXEC)		+= machine_kexec.o
 obj-$(CONFIG_EEH)		+= eeh.o
 obj-$(CONFIG_PROC_FS)		+= proc_ppc64.o
 obj-$(CONFIG_RTAS_FLASH)	+= rtas_flash.o

arch/ppc64/kernel/head.S

@@ -1194,7 +1194,7 @@ _GLOBAL(pSeries_secondary_smp_init)
 	bl	.__restore_cpu_setup
 
 	/* Set up a paca value for this processor. Since we have the
-	 * physical cpu id in r3, we need to search the pacas to find
+	 * physical cpu id in r24, we need to search the pacas to find
 	 * which logical id maps to our physical one.
 	 */
 	LOADADDR(r13, paca) 		/* Get base vaddr of paca array	 */
@@ -1207,8 +1207,8 @@ _GLOBAL(pSeries_secondary_smp_init)
 	cmpwi	r5,NR_CPUS
 	blt	1b
 
-99:	HMT_LOW				/* Couldn't find our CPU id      */
-	b	99b
+	mr	r3,r24			/* not found, copy phys to r3	 */
+	b	.kexec_wait		/* next kernel might do better	 */
 
 2:	mtspr	SPRG3,r13		/* Save vaddr of paca in SPRG3	 */
 	/* From now on, r24 is expected to be logica cpuid */

arch/ppc64/kernel/machine_kexec.c

@@ -0,0 +1,301 @@
+/*
+ * machine_kexec.c - handle transition of Linux booting another kernel
+ *
+ * Copyright (C) 2004-2005, IBM Corp.
+ *
+ * Created by: Milton D Miller II
+ *
+ * This source code is licensed under the GNU General Public License,
+ * Version 2.  See the file COPYING for more details.
+ */
+
+
+#include <linux/cpumask.h>
+#include <linux/kexec.h>
+#include <linux/smp.h>
+#include <linux/thread_info.h>
+#include <linux/errno.h>
+
+#include <asm/page.h>
+#include <asm/current.h>
+#include <asm/machdep.h>
+#include <asm/cacheflush.h>
+#include <asm/paca.h>
+#include <asm/mmu.h>
+#include <asm/sections.h>	/* _end */
+#include <asm/prom.h>
+
+#define HASH_GROUP_SIZE 0x80	/* size of each hash group, asm/mmu.h */
+
+/* Have this around till we move it into crash specific file */
+note_buf_t crash_notes[NR_CPUS];
+
+/* Dummy for now. Not sure if we need to have a crash shutdown in here
+ * and if what it will achieve. Letting it be now to compile the code
+ * in generic kexec environment
+ */
+void machine_crash_shutdown(void)
+{
+	/* do nothing right now */
+	/* smp_relase_cpus() if we want smp on panic kernel */
+	/* cpu_irq_down to isolate us until we are ready */
+}
+
+int machine_kexec_prepare(struct kimage *image)
+{
+	int i;
+	unsigned long begin, end;	/* limits of segment */
+	unsigned long low, high;	/* limits of blocked memory range */
+	struct device_node *node;
+	unsigned long *basep;
+	unsigned int *sizep;
+
+	if (!ppc_md.hpte_clear_all)
+		return -ENOENT;
+
+	/*
+	 * Since we use the kernel fault handlers and paging code to
+	 * handle the virtual mode, we must make sure no destination
+	 * overlaps kernel static data or bss.
+	 */
+	for(i = 0; i < image->nr_segments; i++)
+		if (image->segment[i].mem < __pa(_end))
+			return -ETXTBSY;
+
+	/*
+	 * For non-LPAR, we absolutely can not overwrite the mmu hash
+	 * table, since we are still using the bolted entries in it to
+	 * do the copy.  Check that here.
+	 *
+	 * It is safe if the end is below the start of the blocked
+	 * region (end <= low), or if the beginning is after the
+	 * end of the blocked region (begin >= high).  Use the
+	 * boolean identity !(a || b)  === (!a && !b).
+	 */
+	if (htab_address) {
+		low = __pa(htab_address);
+		high = low + (htab_hash_mask + 1) * HASH_GROUP_SIZE;
+
+		for(i = 0; i < image->nr_segments; i++) {
+			begin = image->segment[i].mem;
+			end = begin + image->segment[i].memsz;
+
+			if ((begin < high) && (end > low))
+				return -ETXTBSY;
+		}
+	}
+
+	/* We also should not overwrite the tce tables */
+	for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
+			node = of_find_node_by_type(node, "pci")) {
+		basep = (unsigned long *)get_property(node, "linux,tce-base",
+							NULL);
+		sizep = (unsigned int *)get_property(node, "linux,tce-size",
+							NULL);
+		if (basep == NULL || sizep == NULL)
+			continue;
+
+		low = *basep;
+		high = low + (*sizep);
+
+		for(i = 0; i < image->nr_segments; i++) {
+			begin = image->segment[i].mem;
+			end = begin + image->segment[i].memsz;
+
+			if ((begin < high) && (end > low))
+				return -ETXTBSY;
+		}
+	}
+
+	return 0;
+}
+
+void machine_kexec_cleanup(struct kimage *image)
+{
+	/* we do nothing in prepare that needs to be undone */
+}
+
+#define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
+
+static void copy_segments(unsigned long ind)
+{
+	unsigned long entry;
+	unsigned long *ptr;
+	void *dest;
+	void *addr;
+
+	/*
+	 * We rely on kexec_load to create a lists that properly
+	 * initializes these pointers before they are used.
+	 * We will still crash if the list is wrong, but at least
+	 * the compiler will be quiet.
+	 */
+	ptr = NULL;
+	dest = NULL;
+
+	for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
+		addr = __va(entry & PAGE_MASK);
+
+		switch (entry & IND_FLAGS) {
+		case IND_DESTINATION:
+			dest = addr;
+			break;
+		case IND_INDIRECTION:
+			ptr = addr;
+			break;
+		case IND_SOURCE:
+			copy_page(dest, addr);
+			dest += PAGE_SIZE;
+		}
+	}
+}
+
+void kexec_copy_flush(struct kimage *image)
+{
+	long i, nr_segments = image->nr_segments;
+	struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
+
+	/* save the ranges on the stack to efficiently flush the icache */
+	memcpy(ranges, image->segment, sizeof(ranges));
+
+	/*
+	 * After this call we may not use anything allocated in dynamic
+	 * memory, including *image.
+	 *
+	 * Only globals and the stack are allowed.
+	 */
+	copy_segments(image->head);
+
+	/*
+	 * we need to clear the icache for all dest pages sometime,
+	 * including ones that were in place on the original copy
+	 */
+	for (i = 0; i < nr_segments; i++)
+		flush_icache_range(ranges[i].mem + KERNELBASE,
+				ranges[i].mem + KERNELBASE +
+				ranges[i].memsz);
+}
+
+#ifdef CONFIG_SMP
+
+/* FIXME: we should schedule this function to be called on all cpus based
+ * on calling the interrupts, but we would like to call it off irq level
+ * so that the interrupt controller is clean.
+ */
+void kexec_smp_down(void *arg)
+{
+	if (ppc_md.cpu_irq_down)
+		ppc_md.cpu_irq_down();
+
+	local_irq_disable();
+	kexec_smp_wait();
+	/* NOTREACHED */
+}
+
+static void kexec_prepare_cpus(void)
+{
+	int my_cpu, i, notified=-1;
+
+	smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
+	my_cpu = get_cpu();
+
+	/* check the others cpus are now down (via paca hw cpu id == -1) */
+	for (i=0; i < NR_CPUS; i++) {
+		if (i == my_cpu)
+			continue;
+
+		while (paca[i].hw_cpu_id != -1) {
+			if (!cpu_possible(i)) {
+				printk("kexec: cpu %d hw_cpu_id %d is not"
+						" possible, ignoring\n",
+						i, paca[i].hw_cpu_id);
+				break;
+			}
+			if (!cpu_online(i)) {
+				/* Fixme: this can be spinning in
+				 * pSeries_secondary_wait with a paca
+				 * waiting for it to go online.
+				 */
+				printk("kexec: cpu %d hw_cpu_id %d is not"
+						" online, ignoring\n",
+						i, paca[i].hw_cpu_id);
+				break;
+			}
+			if (i != notified) {
+				printk( "kexec: waiting for cpu %d (physical"
+						" %d) to go down\n",
+						i, paca[i].hw_cpu_id);
+				notified = i;
+			}
+		}
+	}
+
+	/* after we tell the others to go down */
+	if (ppc_md.cpu_irq_down)
+		ppc_md.cpu_irq_down();
+
+	put_cpu();
+
+	local_irq_disable();
+}
+
+#else /* ! SMP */
+
+static void kexec_prepare_cpus(void)
+{
+	/*
+	 * move the secondarys to us so that we can copy
+	 * the new kernel 0-0x100 safely
+	 *
+	 * do this if kexec in setup.c ?
+	 */
+	smp_relase_cpus();
+	if (ppc_md.cpu_irq_down)
+		ppc_md.cpu_irq_down();
+	local_irq_disable();
+}
+
+#endif /* SMP */
+
+/*
+ * kexec thread structure and stack.
+ *
+ * We need to make sure that this is 16384-byte aligned due to the
+ * way process stacks are handled.  It also must be statically allocated
+ * or allocated as part of the kimage, because everything else may be
+ * overwritten when we copy the kexec image.  We piggyback on the
+ * "init_task" linker section here to statically allocate a stack.
+ *
+ * We could use a smaller stack if we don't care about anything using
+ * current, but that audit has not been performed.
+ */
+union thread_union kexec_stack
+	__attribute__((__section__(".data.init_task"))) = { };
+
+/* Our assembly helper, in kexec_stub.S */
+extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
+	void *image, void *control, void (*clear_all)(void)) ATTRIB_NORET;
+
+/* too late to fail here */
+void machine_kexec(struct kimage *image)
+{
+
+	/* prepare control code if any */
+
+	/* shutdown other cpus into our wait loop and quiesce interrupts */
+	kexec_prepare_cpus();
+
+	/* switch to a staticly allocated stack.  Based on irq stack code.
+	 * XXX: the task struct will likely be invalid once we do the copy!
+	 */
+	kexec_stack.thread_info.task = current_thread_info()->task;
+	kexec_stack.thread_info.flags = 0;
+
+	/* Some things are best done in assembly.  Finding globals with
+	 * a toc is easier in C, so pass in what we can.
+	 */
+	kexec_sequence(&kexec_stack, image->start, image,
+			page_address(image->control_code_page),
+			ppc_md.hpte_clear_all);
+	/* NOTREACHED */
+}