kernel-ark/arch/powerpc/platforms/pseries/xics.c

880 lines
20 KiB
C
Raw Normal View History

/*
* arch/powerpc/platforms/pseries/xics.c
*
* Copyright 2000 IBM Corporation.
*
* 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.
*/
2006-07-03 11:36:01 +00:00
#undef DEBUG
#include <linux/types.h>
#include <linux/threads.h>
#include <linux/kernel.h>
#include <linux/irq.h>
#include <linux/smp.h>
#include <linux/interrupt.h>
#include <linux/signal.h>
#include <linux/init.h>
#include <linux/gfp.h>
#include <linux/radix-tree.h>
#include <linux/cpu.h>
2006-07-03 11:36:01 +00:00
#include <asm/firmware.h>
#include <asm/prom.h>
#include <asm/io.h>
#include <asm/pgtable.h>
#include <asm/smp.h>
#include <asm/rtas.h>
#include <asm/hvcall.h>
#include <asm/machdep.h>
#include <asm/i8259.h>
#include "xics.h"
#define XICS_IPI 2
#define XICS_IRQ_SPURIOUS 0
/* Want a priority other than 0. Various HW issues require this. */
#define DEFAULT_PRIORITY 5
/*
* Mark IPIs as higher priority so we can take them inside interrupts that
* arent marked IRQF_DISABLED
*/
#define IPI_PRIORITY 4
struct xics_ipl {
union {
u32 word;
u8 bytes[4];
} xirr_poll;
union {
u32 word;
u8 bytes[4];
} xirr;
u32 dummy;
union {
u32 word;
u8 bytes[4];
} qirr;
};
static struct xics_ipl __iomem *xics_per_cpu[NR_CPUS];
static unsigned int default_server = 0xFF;
static unsigned int default_distrib_server = 0;
static unsigned int interrupt_server_size = 8;
2006-07-03 11:36:01 +00:00
static struct irq_host *xics_host;
/*
* XICS only has a single IPI, so encode the messages per CPU
*/
struct xics_ipi_struct xics_ipi_message[NR_CPUS] __cacheline_aligned;
/* RTAS service tokens */
static int ibm_get_xive;
static int ibm_set_xive;
static int ibm_int_on;
static int ibm_int_off;
/* Direct HW low level accessors */
2006-07-03 11:36:01 +00:00
static inline unsigned int direct_xirr_info_get(int n_cpu)
{
return in_be32(&xics_per_cpu[n_cpu]->xirr.word);
}
static inline void direct_xirr_info_set(int n_cpu, int value)
{
out_be32(&xics_per_cpu[n_cpu]->xirr.word, value);
}
static inline void direct_cppr_info(int n_cpu, u8 value)
{
out_8(&xics_per_cpu[n_cpu]->xirr.bytes[0], value);
}
static inline void direct_qirr_info(int n_cpu, u8 value)
{
out_8(&xics_per_cpu[n_cpu]->qirr.bytes[0], value);
}
/* LPAR low level accessors */
static inline long plpar_eoi(unsigned long xirr)
{
return plpar_hcall_norets(H_EOI, xirr);
}
static inline long plpar_cppr(unsigned long cppr)
{
return plpar_hcall_norets(H_CPPR, cppr);
}
static inline long plpar_ipi(unsigned long servernum, unsigned long mfrr)
{
return plpar_hcall_norets(H_IPI, servernum, mfrr);
}
static inline long plpar_xirr(unsigned long *xirr_ret)
{
unsigned long dummy;
return plpar_hcall(H_XIRR, 0, 0, 0, 0, xirr_ret, &dummy, &dummy);
}
2006-07-03 11:36:01 +00:00
static inline unsigned int lpar_xirr_info_get(int n_cpu)
{
unsigned long lpar_rc;
unsigned long return_value;
lpar_rc = plpar_xirr(&return_value);
if (lpar_rc != H_SUCCESS)
panic(" bad return code xirr - rc = %lx \n", lpar_rc);
2006-07-03 11:36:01 +00:00
return (unsigned int)return_value;
}
static inline void lpar_xirr_info_set(int n_cpu, int value)
{
unsigned long lpar_rc;
unsigned long val64 = value & 0xffffffff;
lpar_rc = plpar_eoi(val64);
if (lpar_rc != H_SUCCESS)
panic("bad return code EOI - rc = %ld, value=%lx\n", lpar_rc,
val64);
}
static inline void lpar_cppr_info(int n_cpu, u8 value)
{
unsigned long lpar_rc;
lpar_rc = plpar_cppr(value);
if (lpar_rc != H_SUCCESS)
panic("bad return code cppr - rc = %lx\n", lpar_rc);
}
static inline void lpar_qirr_info(int n_cpu , u8 value)
{
unsigned long lpar_rc;
lpar_rc = plpar_ipi(get_hard_smp_processor_id(n_cpu), value);
if (lpar_rc != H_SUCCESS)
panic("bad return code qirr - rc = %lx\n", lpar_rc);
}
/* High level handlers and init code */
#ifdef CONFIG_SMP
2006-07-03 11:36:01 +00:00
static int get_irq_server(unsigned int virq)
{
unsigned int server;
/* For the moment only implement delivery to all cpus or one cpu */
2006-07-03 11:36:01 +00:00
cpumask_t cpumask = irq_desc[virq].affinity;
cpumask_t tmp = CPU_MASK_NONE;
if (!distribute_irqs)
return default_server;
if (cpus_equal(cpumask, CPU_MASK_ALL)) {
server = default_distrib_server;
} else {
cpus_and(tmp, cpu_online_map, cpumask);
if (cpus_empty(tmp))
server = default_distrib_server;
else
server = get_hard_smp_processor_id(first_cpu(tmp));
}
return server;
}
#else
2006-07-03 11:36:01 +00:00
static int get_irq_server(unsigned int virq)
{
return default_server;
}
#endif
static void xics_unmask_irq(unsigned int virq)
{
unsigned int irq;
int call_status;
unsigned int server;
2006-07-03 11:36:01 +00:00
pr_debug("xics: unmask virq %d\n", virq);
irq = (unsigned int)irq_map[virq].hwirq;
pr_debug(" -> map to hwirq 0x%x\n", irq);
if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
return;
server = get_irq_server(virq);
call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq, server,
DEFAULT_PRIORITY);
if (call_status != 0) {
printk(KERN_ERR "xics_enable_irq: irq=%u: ibm_set_xive "
"returned %d\n", irq, call_status);
printk("set_xive %x, server %x\n", ibm_set_xive, server);
return;
}
/* Now unmask the interrupt (often a no-op) */
call_status = rtas_call(ibm_int_on, 1, 1, NULL, irq);
if (call_status != 0) {
printk(KERN_ERR "xics_enable_irq: irq=%u: ibm_int_on "
"returned %d\n", irq, call_status);
return;
}
}
static void xics_mask_real_irq(unsigned int irq)
{
int call_status;
unsigned int server;
if (irq == XICS_IPI)
return;
call_status = rtas_call(ibm_int_off, 1, 1, NULL, irq);
if (call_status != 0) {
printk(KERN_ERR "xics_disable_real_irq: irq=%u: "
"ibm_int_off returned %d\n", irq, call_status);
return;
}
server = get_irq_server(irq);
/* Have to set XIVE to 0xff to be able to remove a slot */
call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq, server, 0xff);
if (call_status != 0) {
printk(KERN_ERR "xics_disable_irq: irq=%u: ibm_set_xive(0xff)"
" returned %d\n", irq, call_status);
return;
}
}
static void xics_mask_irq(unsigned int virq)
{
unsigned int irq;
2006-07-03 11:36:01 +00:00
pr_debug("xics: mask virq %d\n", virq);
irq = (unsigned int)irq_map[virq].hwirq;
if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
return;
xics_mask_real_irq(irq);
}
2006-07-03 11:36:01 +00:00
static unsigned int xics_startup(unsigned int virq)
{
unsigned int irq;
2006-07-03 11:36:01 +00:00
/* force a reverse mapping of the interrupt so it gets in the cache */
irq = (unsigned int)irq_map[virq].hwirq;
irq_radix_revmap(xics_host, irq);
2006-07-03 11:36:01 +00:00
/* unmask it */
xics_unmask_irq(virq);
return 0;
}
2006-07-03 11:36:01 +00:00
static void xics_eoi_direct(unsigned int virq)
{
int cpu = smp_processor_id();
2006-07-03 11:36:01 +00:00
unsigned int irq = (unsigned int)irq_map[virq].hwirq;
iosync();
2006-07-03 11:36:01 +00:00
direct_xirr_info_set(cpu, (0xff << 24) | irq);
}
2006-07-03 11:36:01 +00:00
static void xics_eoi_lpar(unsigned int virq)
{
int cpu = smp_processor_id();
2006-07-03 11:36:01 +00:00
unsigned int irq = (unsigned int)irq_map[virq].hwirq;
iosync();
2006-07-03 11:36:01 +00:00
lpar_xirr_info_set(cpu, (0xff << 24) | irq);
}
2006-07-03 11:36:01 +00:00
static inline unsigned int xics_remap_irq(unsigned int vec)
{
2006-07-03 11:36:01 +00:00
unsigned int irq;
vec &= 0x00ffffff;
if (vec == XICS_IRQ_SPURIOUS)
return NO_IRQ;
2006-07-03 11:36:01 +00:00
irq = irq_radix_revmap(xics_host, vec);
if (likely(irq != NO_IRQ))
2006-07-03 11:36:01 +00:00
return irq;
printk(KERN_ERR "Interrupt %u (real) is invalid,"
" disabling it.\n", vec);
xics_mask_real_irq(vec);
return NO_IRQ;
}
2006-07-03 11:36:01 +00:00
static unsigned int xics_get_irq_direct(struct pt_regs *regs)
{
unsigned int cpu = smp_processor_id();
return xics_remap_irq(direct_xirr_info_get(cpu));
}
2006-07-03 11:36:01 +00:00
static unsigned int xics_get_irq_lpar(struct pt_regs *regs)
{
unsigned int cpu = smp_processor_id();
return xics_remap_irq(lpar_xirr_info_get(cpu));
}
#ifdef CONFIG_SMP
static irqreturn_t xics_ipi_dispatch(int cpu, struct pt_regs *regs)
{
WARN_ON(cpu_is_offline(cpu));
while (xics_ipi_message[cpu].value) {
if (test_and_clear_bit(PPC_MSG_CALL_FUNCTION,
&xics_ipi_message[cpu].value)) {
mb();
smp_message_recv(PPC_MSG_CALL_FUNCTION, regs);
}
if (test_and_clear_bit(PPC_MSG_RESCHEDULE,
&xics_ipi_message[cpu].value)) {
mb();
smp_message_recv(PPC_MSG_RESCHEDULE, regs);
}
#if 0
if (test_and_clear_bit(PPC_MSG_MIGRATE_TASK,
&xics_ipi_message[cpu].value)) {
mb();
smp_message_recv(PPC_MSG_MIGRATE_TASK, regs);
}
#endif
#if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC)
if (test_and_clear_bit(PPC_MSG_DEBUGGER_BREAK,
&xics_ipi_message[cpu].value)) {
mb();
smp_message_recv(PPC_MSG_DEBUGGER_BREAK, regs);
}
#endif
}
return IRQ_HANDLED;
}
static irqreturn_t xics_ipi_action_direct(int irq, void *dev_id, struct pt_regs *regs)
{
int cpu = smp_processor_id();
direct_qirr_info(cpu, 0xff);
return xics_ipi_dispatch(cpu, regs);
}
static irqreturn_t xics_ipi_action_lpar(int irq, void *dev_id, struct pt_regs *regs)
{
int cpu = smp_processor_id();
lpar_qirr_info(cpu, 0xff);
return xics_ipi_dispatch(cpu, regs);
}
void xics_cause_IPI(int cpu)
{
if (firmware_has_feature(FW_FEATURE_LPAR))
lpar_qirr_info(cpu, IPI_PRIORITY);
else
direct_qirr_info(cpu, IPI_PRIORITY);
}
#endif /* CONFIG_SMP */
static void xics_set_cpu_priority(int cpu, unsigned char cppr)
{
if (firmware_has_feature(FW_FEATURE_LPAR))
lpar_cppr_info(cpu, cppr);
else
direct_cppr_info(cpu, cppr);
iosync();
}
static void xics_set_affinity(unsigned int virq, cpumask_t cpumask)
{
unsigned int irq;
int status;
int xics_status[2];
unsigned long newmask;
cpumask_t tmp = CPU_MASK_NONE;
2006-07-03 11:36:01 +00:00
irq = (unsigned int)irq_map[virq].hwirq;
if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
return;
status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
if (status) {
printk(KERN_ERR "xics_set_affinity: irq=%u ibm,get-xive "
"returns %d\n", irq, status);
return;
}
/* For the moment only implement delivery to all cpus or one cpu */
if (cpus_equal(cpumask, CPU_MASK_ALL)) {
newmask = default_distrib_server;
} else {
cpus_and(tmp, cpu_online_map, cpumask);
if (cpus_empty(tmp))
return;
newmask = get_hard_smp_processor_id(first_cpu(tmp));
}
status = rtas_call(ibm_set_xive, 3, 1, NULL,
irq, newmask, xics_status[1]);
if (status) {
printk(KERN_ERR "xics_set_affinity: irq=%u ibm,set-xive "
"returns %d\n", irq, status);
return;
}
}
2006-07-03 11:36:01 +00:00
void xics_setup_cpu(void)
{
int cpu = smp_processor_id();
xics_set_cpu_priority(cpu, 0xff);
/*
* Put the calling processor into the GIQ. This is really only
* necessary from a secondary thread as the OF start-cpu interface
* performs this function for us on primary threads.
*
* XXX: undo of teardown on kexec needs this too, as may hotplug
*/
rtas_set_indicator(GLOBAL_INTERRUPT_QUEUE,
(1UL << interrupt_server_size) - 1 - default_distrib_server, 1);
}
static struct irq_chip xics_pic_direct = {
.typename = " XICS ",
.startup = xics_startup,
.mask = xics_mask_irq,
.unmask = xics_unmask_irq,
.eoi = xics_eoi_direct,
.set_affinity = xics_set_affinity
};
static struct irq_chip xics_pic_lpar = {
.typename = " XICS ",
.startup = xics_startup,
.mask = xics_mask_irq,
.unmask = xics_unmask_irq,
.eoi = xics_eoi_lpar,
.set_affinity = xics_set_affinity
};
2006-07-03 11:36:01 +00:00
static int xics_host_match(struct irq_host *h, struct device_node *node)
{
2006-07-03 11:36:01 +00:00
/* IBM machines have interrupt parents of various funky types for things
* like vdevices, events, etc... The trick we use here is to match
* everything here except the legacy 8259 which is compatible "chrp,iic"
*/
return !device_is_compatible(node, "chrp,iic");
}
2006-07-03 11:36:01 +00:00
static int xics_host_map_direct(struct irq_host *h, unsigned int virq,
irq_hw_number_t hw, unsigned int flags)
{
unsigned int sense = flags & IRQ_TYPE_SENSE_MASK;
2006-07-03 11:36:01 +00:00
pr_debug("xics: map_direct virq %d, hwirq 0x%lx, flags: 0x%x\n",
virq, hw, flags);
if (sense && sense != IRQ_TYPE_LEVEL_LOW)
printk(KERN_WARNING "xics: using unsupported sense 0x%x"
" for irq %d (h: 0x%lx)\n", flags, virq, hw);
get_irq_desc(virq)->status |= IRQ_LEVEL;
set_irq_chip_and_handler(virq, &xics_pic_direct, handle_fasteoi_irq);
return 0;
}
static int xics_host_map_lpar(struct irq_host *h, unsigned int virq,
irq_hw_number_t hw, unsigned int flags)
{
unsigned int sense = flags & IRQ_TYPE_SENSE_MASK;
pr_debug("xics: map_lpar virq %d, hwirq 0x%lx, flags: 0x%x\n",
virq, hw, flags);
if (sense && sense != IRQ_TYPE_LEVEL_LOW)
printk(KERN_WARNING "xics: using unsupported sense 0x%x"
" for irq %d (h: 0x%lx)\n", flags, virq, hw);
get_irq_desc(virq)->status |= IRQ_LEVEL;
set_irq_chip_and_handler(virq, &xics_pic_lpar, handle_fasteoi_irq);
return 0;
}
static int xics_host_xlate(struct irq_host *h, struct device_node *ct,
u32 *intspec, unsigned int intsize,
irq_hw_number_t *out_hwirq, unsigned int *out_flags)
{
/* Current xics implementation translates everything
* to level. It is not technically right for MSIs but this
* is irrelevant at this point. We might get smarter in the future
*/
2006-07-03 11:36:01 +00:00
*out_hwirq = intspec[0];
*out_flags = IRQ_TYPE_LEVEL_LOW;
return 0;
}
static struct irq_host_ops xics_host_direct_ops = {
.match = xics_host_match,
.map = xics_host_map_direct,
.xlate = xics_host_xlate,
};
static struct irq_host_ops xics_host_lpar_ops = {
.match = xics_host_match,
.map = xics_host_map_lpar,
.xlate = xics_host_xlate,
};
static void __init xics_init_host(void)
{
struct irq_host_ops *ops;
if (firmware_has_feature(FW_FEATURE_LPAR))
ops = &xics_host_lpar_ops;
else
ops = &xics_host_direct_ops;
xics_host = irq_alloc_host(IRQ_HOST_MAP_TREE, 0, ops,
XICS_IRQ_SPURIOUS);
BUG_ON(xics_host == NULL);
irq_set_default_host(xics_host);
}
2006-07-03 11:36:01 +00:00
static void __init xics_map_one_cpu(int hw_id, unsigned long addr,
unsigned long size)
{
2006-07-03 11:36:01 +00:00
#ifdef CONFIG_SMP
int i;
2006-07-03 11:36:01 +00:00
/* This may look gross but it's good enough for now, we don't quite
* have a hard -> linux processor id matching.
*/
for_each_possible_cpu(i) {
if (!cpu_present(i))
continue;
if (hw_id == get_hard_smp_processor_id(i)) {
xics_per_cpu[i] = ioremap(addr, size);
return;
}
}
#else
if (hw_id != 0)
return;
xics_per_cpu[0] = ioremap(addr, size);
#endif /* CONFIG_SMP */
}
2006-07-03 11:36:01 +00:00
static void __init xics_init_one_node(struct device_node *np,
unsigned int *indx)
{
unsigned int ilen;
u32 *ireg;
2006-07-03 11:36:01 +00:00
/* This code does the theorically broken assumption that the interrupt
* server numbers are the same as the hard CPU numbers.
* This happens to be the case so far but we are playing with fire...
* should be fixed one of these days. -BenH.
*/
ireg = (u32 *)get_property(np, "ibm,interrupt-server-ranges", NULL);
2006-07-03 11:36:01 +00:00
/* Do that ever happen ? we'll know soon enough... but even good'old
* f80 does have that property ..
*/
WARN_ON(ireg == NULL);
if (ireg) {
/*
* set node starting index for this node
*/
2006-07-03 11:36:01 +00:00
*indx = *ireg;
}
2006-07-03 11:36:01 +00:00
ireg = (u32 *)get_property(np, "reg", &ilen);
if (!ireg)
panic("xics_init_IRQ: can't find interrupt reg property");
2006-07-03 11:36:01 +00:00
while (ilen >= (4 * sizeof(u32))) {
unsigned long addr, size;
/* XXX Use proper OF parsing code here !!! */
addr = (unsigned long)*ireg++ << 32;
ilen -= sizeof(u32);
addr |= *ireg++;
ilen -= sizeof(u32);
size = (unsigned long)*ireg++ << 32;
ilen -= sizeof(u32);
size |= *ireg++;
ilen -= sizeof(u32);
xics_map_one_cpu(*indx, addr, size);
(*indx)++;
}
}
static void __init xics_setup_8259_cascade(void)
{
struct device_node *np, *old, *found = NULL;
int cascade, naddr;
u32 *addrp;
unsigned long intack = 0;
for_each_node_by_type(np, "interrupt-controller")
if (device_is_compatible(np, "chrp,iic")) {
found = np;
break;
}
if (found == NULL) {
printk(KERN_DEBUG "xics: no ISA interrupt controller\n");
return;
}
2006-07-03 11:36:01 +00:00
cascade = irq_of_parse_and_map(found, 0);
if (cascade == NO_IRQ) {
printk(KERN_ERR "xics: failed to map cascade interrupt");
return;
}
pr_debug("xics: cascade mapped to irq %d\n", cascade);
for (old = of_node_get(found); old != NULL ; old = np) {
np = of_get_parent(old);
of_node_put(old);
if (np == NULL)
break;
if (strcmp(np->name, "pci") != 0)
continue;
addrp = (u32 *)get_property(np, "8259-interrupt-acknowledge", NULL);
if (addrp == NULL)
continue;
naddr = prom_n_addr_cells(np);
intack = addrp[naddr-1];
if (naddr > 1)
intack |= ((unsigned long)addrp[naddr-2]) << 32;
}
if (intack)
printk(KERN_DEBUG "xics: PCI 8259 intack at 0x%016lx\n", intack);
i8259_init(found, intack);
of_node_put(found);
set_irq_chained_handler(cascade, pseries_8259_cascade);
}
2006-07-03 11:36:01 +00:00
void __init xics_init_IRQ(void)
{
int i;
struct device_node *np;
u32 *ireg, ilen, indx = 0;
int found = 0;
ppc64_boot_msg(0x20, "XICS Init");
ibm_get_xive = rtas_token("ibm,get-xive");
ibm_set_xive = rtas_token("ibm,set-xive");
ibm_int_on = rtas_token("ibm,int-on");
ibm_int_off = rtas_token("ibm,int-off");
for_each_node_by_type(np, "PowerPC-External-Interrupt-Presentation") {
found = 1;
if (firmware_has_feature(FW_FEATURE_LPAR))
break;
xics_init_one_node(np, &indx);
}
if (found == 0)
return;
xics_init_host();
/* Find the server numbers for the boot cpu. */
for (np = of_find_node_by_type(NULL, "cpu");
np;
np = of_find_node_by_type(np, "cpu")) {
2006-07-03 11:36:01 +00:00
ireg = (u32 *)get_property(np, "reg", &ilen);
if (ireg && ireg[0] == get_hard_smp_processor_id(boot_cpuid)) {
2006-07-03 11:36:01 +00:00
ireg = (u32 *)get_property(np,
"ibm,ppc-interrupt-gserver#s",
&ilen);
i = ilen / sizeof(int);
if (ireg && i > 0) {
default_server = ireg[0];
2006-07-03 11:36:01 +00:00
/* take last element */
default_distrib_server = ireg[i-1];
}
2006-07-03 11:36:01 +00:00
ireg = (u32 *)get_property(np,
"ibm,interrupt-server#-size", NULL);
if (ireg)
interrupt_server_size = *ireg;
break;
}
}
of_node_put(np);
2006-07-03 11:36:01 +00:00
if (firmware_has_feature(FW_FEATURE_LPAR))
ppc_md.get_irq = xics_get_irq_lpar;
else
ppc_md.get_irq = xics_get_irq_direct;
xics_setup_cpu();
2006-07-03 11:36:01 +00:00
xics_setup_8259_cascade();
2006-07-03 11:36:01 +00:00
ppc64_boot_msg(0x21, "XICS Done");
}
#ifdef CONFIG_SMP
void xics_request_IPIs(void)
{
2006-07-03 11:36:01 +00:00
unsigned int ipi;
ipi = irq_create_mapping(xics_host, XICS_IPI, 0);
BUG_ON(ipi == NO_IRQ);
/*
* IPIs are marked IRQF_DISABLED as they must run with irqs
* disabled
*/
2006-07-03 11:36:01 +00:00
set_irq_handler(ipi, handle_percpu_irq);
if (firmware_has_feature(FW_FEATURE_LPAR))
2006-07-03 11:36:01 +00:00
request_irq(ipi, xics_ipi_action_lpar, IRQF_DISABLED,
"IPI", NULL);
else
2006-07-03 11:36:01 +00:00
request_irq(ipi, xics_ipi_action_direct, IRQF_DISABLED,
"IPI", NULL);
}
#endif /* CONFIG_SMP */
void xics_teardown_cpu(int secondary)
[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 <miltonm@bga.com> Signed-off-by: R Sharada <sharada@in.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-25 21:58:10 +00:00
{
int cpu = smp_processor_id();
2006-07-03 11:36:01 +00:00
unsigned int ipi;
struct irq_desc *desc;
[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 <miltonm@bga.com> Signed-off-by: R Sharada <sharada@in.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-25 21:58:10 +00:00
2006-07-03 11:36:01 +00:00
xics_set_cpu_priority(cpu, 0);
/*
* we need to EOI the IPI if we got here from kexec down IPI
*
* probably need to check all the other interrupts too
* should we be flagging idle loop instead?
* or creating some task to be scheduled?
*/
2006-07-03 11:36:01 +00:00
ipi = irq_find_mapping(xics_host, XICS_IPI);
if (ipi == XICS_IRQ_SPURIOUS)
return;
desc = get_irq_desc(ipi);
if (desc->chip && desc->chip->eoi)
desc->chip->eoi(XICS_IPI);
[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 <miltonm@bga.com> Signed-off-by: R Sharada <sharada@in.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-25 21:58:10 +00:00
/*
* Some machines need to have at least one cpu in the GIQ,
* so leave the master cpu in the group.
[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 <miltonm@bga.com> Signed-off-by: R Sharada <sharada@in.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-25 21:58:10 +00:00
*/
if (secondary)
rtas_set_indicator(GLOBAL_INTERRUPT_QUEUE,
2006-07-03 11:36:01 +00:00
(1UL << interrupt_server_size) - 1 -
default_distrib_server, 0);
[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 <miltonm@bga.com> Signed-off-by: R Sharada <sharada@in.ibm.com> Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-25 21:58:10 +00:00
}
#ifdef CONFIG_HOTPLUG_CPU
/* Interrupts are disabled. */
void xics_migrate_irqs_away(void)
{
int status;
unsigned int irq, virq, cpu = smp_processor_id();
/* Reject any interrupt that was queued to us... */
xics_set_cpu_priority(cpu, 0);
/* remove ourselves from the global interrupt queue */
status = rtas_set_indicator(GLOBAL_INTERRUPT_QUEUE,
(1UL << interrupt_server_size) - 1 - default_distrib_server, 0);
WARN_ON(status < 0);
/* Allow IPIs again... */
xics_set_cpu_priority(cpu, DEFAULT_PRIORITY);
for_each_irq(virq) {
struct irq_desc *desc;
int xics_status[2];
unsigned long flags;
/* We cant set affinity on ISA interrupts */
2006-07-03 11:36:01 +00:00
if (virq < NUM_ISA_INTERRUPTS)
continue;
2006-07-03 11:36:01 +00:00
if (irq_map[virq].host != xics_host)
continue;
irq = (unsigned int)irq_map[virq].hwirq;
/* We need to get IPIs still. */
2006-07-03 11:36:01 +00:00
if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
continue;
2006-07-03 11:36:01 +00:00
desc = get_irq_desc(virq);
/* We only need to migrate enabled IRQS */
[PATCH] genirq: rename desc->handler to desc->chip This patch-queue improves the generic IRQ layer to be truly generic, by adding various abstractions and features to it, without impacting existing functionality. While the queue can be best described as "fix and improve everything in the generic IRQ layer that we could think of", and thus it consists of many smaller features and lots of cleanups, the one feature that stands out most is the new 'irq chip' abstraction. The irq-chip abstraction is about describing and coding and IRQ controller driver by mapping its raw hardware capabilities [and quirks, if needed] in a straightforward way, without having to think about "IRQ flow" (level/edge/etc.) type of details. This stands in contrast with the current 'irq-type' model of genirq architectures, which 'mixes' raw hardware capabilities with 'flow' details. The patchset supports both types of irq controller designs at once, and converts i386 and x86_64 to the new irq-chip design. As a bonus side-effect of the irq-chip approach, chained interrupt controllers (master/slave PIC constructs, etc.) are now supported by design as well. The end result of this patchset intends to be simpler architecture-level code and more consolidation between architectures. We reused many bits of code and many concepts from Russell King's ARM IRQ layer, the merging of which was one of the motivations for this patchset. This patch: rename desc->handler to desc->chip. Originally i did not want to do this, because it's a big patch. But having both "desc->handler", "desc->handle_irq" and "action->handler" caused a large degree of confusion and made the code appear alot less clean than it truly is. I have also attempted a dual approach as well by introducing a desc->chip alias - but that just wasnt robust enough and broke frequently. So lets get over with this quickly. The conversion was done automatically via scripts and converts all the code in the kernel. This renaming patch is the first one amongst the patches, so that the remaining patches can stay flexible and can be merged and split up without having some big monolithic patch act as a merge barrier. [akpm@osdl.org: build fix] [akpm@osdl.org: another build fix] Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-29 09:24:36 +00:00
if (desc == NULL || desc->chip == NULL
|| desc->action == NULL
[PATCH] genirq: rename desc->handler to desc->chip This patch-queue improves the generic IRQ layer to be truly generic, by adding various abstractions and features to it, without impacting existing functionality. While the queue can be best described as "fix and improve everything in the generic IRQ layer that we could think of", and thus it consists of many smaller features and lots of cleanups, the one feature that stands out most is the new 'irq chip' abstraction. The irq-chip abstraction is about describing and coding and IRQ controller driver by mapping its raw hardware capabilities [and quirks, if needed] in a straightforward way, without having to think about "IRQ flow" (level/edge/etc.) type of details. This stands in contrast with the current 'irq-type' model of genirq architectures, which 'mixes' raw hardware capabilities with 'flow' details. The patchset supports both types of irq controller designs at once, and converts i386 and x86_64 to the new irq-chip design. As a bonus side-effect of the irq-chip approach, chained interrupt controllers (master/slave PIC constructs, etc.) are now supported by design as well. The end result of this patchset intends to be simpler architecture-level code and more consolidation between architectures. We reused many bits of code and many concepts from Russell King's ARM IRQ layer, the merging of which was one of the motivations for this patchset. This patch: rename desc->handler to desc->chip. Originally i did not want to do this, because it's a big patch. But having both "desc->handler", "desc->handle_irq" and "action->handler" caused a large degree of confusion and made the code appear alot less clean than it truly is. I have also attempted a dual approach as well by introducing a desc->chip alias - but that just wasnt robust enough and broke frequently. So lets get over with this quickly. The conversion was done automatically via scripts and converts all the code in the kernel. This renaming patch is the first one amongst the patches, so that the remaining patches can stay flexible and can be merged and split up without having some big monolithic patch act as a merge barrier. [akpm@osdl.org: build fix] [akpm@osdl.org: another build fix] Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-29 09:24:36 +00:00
|| desc->chip->set_affinity == NULL)
continue;
spin_lock_irqsave(&desc->lock, flags);
status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
if (status) {
printk(KERN_ERR "migrate_irqs_away: irq=%u "
"ibm,get-xive returns %d\n",
virq, status);
goto unlock;
}
/*
* We only support delivery to all cpus or to one cpu.
* The irq has to be migrated only in the single cpu
* case.
*/
if (xics_status[0] != get_hard_smp_processor_id(cpu))
goto unlock;
printk(KERN_WARNING "IRQ %u affinity broken off cpu %u\n",
virq, cpu);
/* Reset affinity to all cpus */
[PATCH] genirq: rename desc->handler to desc->chip This patch-queue improves the generic IRQ layer to be truly generic, by adding various abstractions and features to it, without impacting existing functionality. While the queue can be best described as "fix and improve everything in the generic IRQ layer that we could think of", and thus it consists of many smaller features and lots of cleanups, the one feature that stands out most is the new 'irq chip' abstraction. The irq-chip abstraction is about describing and coding and IRQ controller driver by mapping its raw hardware capabilities [and quirks, if needed] in a straightforward way, without having to think about "IRQ flow" (level/edge/etc.) type of details. This stands in contrast with the current 'irq-type' model of genirq architectures, which 'mixes' raw hardware capabilities with 'flow' details. The patchset supports both types of irq controller designs at once, and converts i386 and x86_64 to the new irq-chip design. As a bonus side-effect of the irq-chip approach, chained interrupt controllers (master/slave PIC constructs, etc.) are now supported by design as well. The end result of this patchset intends to be simpler architecture-level code and more consolidation between architectures. We reused many bits of code and many concepts from Russell King's ARM IRQ layer, the merging of which was one of the motivations for this patchset. This patch: rename desc->handler to desc->chip. Originally i did not want to do this, because it's a big patch. But having both "desc->handler", "desc->handle_irq" and "action->handler" caused a large degree of confusion and made the code appear alot less clean than it truly is. I have also attempted a dual approach as well by introducing a desc->chip alias - but that just wasnt robust enough and broke frequently. So lets get over with this quickly. The conversion was done automatically via scripts and converts all the code in the kernel. This renaming patch is the first one amongst the patches, so that the remaining patches can stay flexible and can be merged and split up without having some big monolithic patch act as a merge barrier. [akpm@osdl.org: build fix] [akpm@osdl.org: another build fix] Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-29 09:24:36 +00:00
desc->chip->set_affinity(virq, CPU_MASK_ALL);
irq_desc[irq].affinity = CPU_MASK_ALL;
unlock:
spin_unlock_irqrestore(&desc->lock, flags);
}
}
#endif