2 * arch/powerpc/platforms/pseries/xics.c
4 * Copyright 2000 IBM Corporation.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/types.h>
13 #include <linux/threads.h>
14 #include <linux/kernel.h>
15 #include <linux/irq.h>
16 #include <linux/smp.h>
17 #include <linux/interrupt.h>
18 #include <linux/init.h>
19 #include <linux/radix-tree.h>
20 #include <linux/cpu.h>
21 #include <linux/msi.h>
23 #include <linux/percpu.h>
25 #include <asm/firmware.h>
27 #include <asm/pgtable.h>
30 #include <asm/hvcall.h>
31 #include <asm/machdep.h>
34 #include "plpar_wrappers.h"
36 static struct irq_host *xics_host;
39 #define XICS_IRQ_SPURIOUS 0
41 /* Want a priority other than 0. Various HW issues require this. */
42 #define DEFAULT_PRIORITY 5
45 * Mark IPIs as higher priority so we can take them inside interrupts that
46 * arent marked IRQF_DISABLED
48 #define IPI_PRIORITY 4
50 /* The least favored priority */
51 #define LOWEST_PRIORITY 0xFF
53 /* The number of priorities defined above */
54 #define MAX_NUM_PRIORITIES 3
56 static unsigned int default_server = 0xFF;
57 static unsigned int default_distrib_server = 0;
58 static unsigned int interrupt_server_size = 8;
60 /* RTAS service tokens */
61 static int ibm_get_xive;
62 static int ibm_set_xive;
63 static int ibm_int_on;
64 static int ibm_int_off;
67 unsigned char stack[MAX_NUM_PRIORITIES];
71 static DEFINE_PER_CPU(struct xics_cppr, xics_cppr);
73 /* Direct hardware low level accessors */
75 /* The part of the interrupt presentation layer that we care about */
92 static struct xics_ipl __iomem *xics_per_cpu[NR_CPUS];
94 static inline unsigned int direct_xirr_info_get(void)
96 int cpu = smp_processor_id();
98 return in_be32(&xics_per_cpu[cpu]->xirr.word);
101 static inline void direct_xirr_info_set(unsigned int value)
103 int cpu = smp_processor_id();
105 out_be32(&xics_per_cpu[cpu]->xirr.word, value);
108 static inline void direct_cppr_info(u8 value)
110 int cpu = smp_processor_id();
112 out_8(&xics_per_cpu[cpu]->xirr.bytes[0], value);
115 static inline void direct_qirr_info(int n_cpu, u8 value)
117 out_8(&xics_per_cpu[n_cpu]->qirr.bytes[0], value);
121 /* LPAR low level accessors */
123 static inline unsigned int lpar_xirr_info_get(void)
125 unsigned long lpar_rc;
126 unsigned long return_value;
128 lpar_rc = plpar_xirr(&return_value);
129 if (lpar_rc != H_SUCCESS)
130 panic(" bad return code xirr - rc = %lx \n", lpar_rc);
131 return (unsigned int)return_value;
134 static inline void lpar_xirr_info_set(unsigned int value)
136 unsigned long lpar_rc;
138 lpar_rc = plpar_eoi(value);
139 if (lpar_rc != H_SUCCESS)
140 panic("bad return code EOI - rc = %ld, value=%x\n", lpar_rc,
144 static inline void lpar_cppr_info(u8 value)
146 unsigned long lpar_rc;
148 lpar_rc = plpar_cppr(value);
149 if (lpar_rc != H_SUCCESS)
150 panic("bad return code cppr - rc = %lx\n", lpar_rc);
153 static inline void lpar_qirr_info(int n_cpu , u8 value)
155 unsigned long lpar_rc;
157 lpar_rc = plpar_ipi(get_hard_smp_processor_id(n_cpu), value);
158 if (lpar_rc != H_SUCCESS)
159 panic("bad return code qirr - rc = %lx\n", lpar_rc);
163 /* Interface to generic irq subsystem */
166 static int get_irq_server(unsigned int virq, cpumask_t cpumask,
167 unsigned int strict_check)
170 /* For the moment only implement delivery to all cpus or one cpu */
171 cpumask_t tmp = CPU_MASK_NONE;
173 if (!distribute_irqs)
174 return default_server;
176 if (!cpus_equal(cpumask, CPU_MASK_ALL)) {
177 cpus_and(tmp, cpu_online_map, cpumask);
179 server = first_cpu(tmp);
181 if (server < NR_CPUS)
182 return get_hard_smp_processor_id(server);
188 if (cpus_equal(cpu_online_map, cpu_present_map))
189 return default_distrib_server;
191 return default_server;
194 #define get_irq_server(virq, cpumask, strict_check) (default_server)
197 static void xics_unmask_irq(unsigned int virq)
203 pr_devel("xics: unmask virq %d\n", virq);
205 irq = (unsigned int)irq_map[virq].hwirq;
206 pr_devel(" -> map to hwirq 0x%x\n", irq);
207 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
210 server = get_irq_server(virq, *(irq_to_desc(virq)->affinity), 0);
212 call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq, server,
214 if (call_status != 0) {
216 "%s: ibm_set_xive irq %u server %x returned %d\n",
217 __func__, irq, server, call_status);
221 /* Now unmask the interrupt (often a no-op) */
222 call_status = rtas_call(ibm_int_on, 1, 1, NULL, irq);
223 if (call_status != 0) {
224 printk(KERN_ERR "%s: ibm_int_on irq=%u returned %d\n",
225 __func__, irq, call_status);
230 static unsigned int xics_startup(unsigned int virq)
233 * The generic MSI code returns with the interrupt disabled on the
234 * card, using the MSI mask bits. Firmware doesn't appear to unmask
235 * at that level, so we do it here by hand.
237 if (irq_to_desc(virq)->msi_desc)
238 unmask_msi_irq(virq);
241 xics_unmask_irq(virq);
245 static void xics_mask_real_irq(unsigned int irq)
252 call_status = rtas_call(ibm_int_off, 1, 1, NULL, irq);
253 if (call_status != 0) {
254 printk(KERN_ERR "%s: ibm_int_off irq=%u returned %d\n",
255 __func__, irq, call_status);
259 /* Have to set XIVE to 0xff to be able to remove a slot */
260 call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq,
261 default_server, 0xff);
262 if (call_status != 0) {
263 printk(KERN_ERR "%s: ibm_set_xive(0xff) irq=%u returned %d\n",
264 __func__, irq, call_status);
269 static void xics_mask_irq(unsigned int virq)
273 pr_devel("xics: mask virq %d\n", virq);
275 irq = (unsigned int)irq_map[virq].hwirq;
276 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
278 xics_mask_real_irq(irq);
281 static void xics_mask_unknown_vec(unsigned int vec)
283 printk(KERN_ERR "Interrupt %u (real) is invalid, disabling it.\n", vec);
284 xics_mask_real_irq(vec);
287 static inline unsigned int xics_xirr_vector(unsigned int xirr)
290 * The top byte is the old cppr, to be restored on EOI.
291 * The remaining 24 bits are the vector.
293 return xirr & 0x00ffffff;
296 static void push_cppr(unsigned int vec)
298 struct xics_cppr *os_cppr = &__get_cpu_var(xics_cppr);
300 if (WARN_ON(os_cppr->index >= MAX_NUM_PRIORITIES - 1))
304 os_cppr->stack[++os_cppr->index] = IPI_PRIORITY;
306 os_cppr->stack[++os_cppr->index] = DEFAULT_PRIORITY;
309 static unsigned int xics_get_irq_direct(void)
311 unsigned int xirr = direct_xirr_info_get();
312 unsigned int vec = xics_xirr_vector(xirr);
315 if (vec == XICS_IRQ_SPURIOUS)
318 irq = irq_radix_revmap_lookup(xics_host, vec);
319 if (likely(irq != NO_IRQ)) {
324 /* We don't have a linux mapping, so have rtas mask it. */
325 xics_mask_unknown_vec(vec);
327 /* We might learn about it later, so EOI it */
328 direct_xirr_info_set(xirr);
332 static unsigned int xics_get_irq_lpar(void)
334 unsigned int xirr = lpar_xirr_info_get();
335 unsigned int vec = xics_xirr_vector(xirr);
338 if (vec == XICS_IRQ_SPURIOUS)
341 irq = irq_radix_revmap_lookup(xics_host, vec);
342 if (likely(irq != NO_IRQ)) {
347 /* We don't have a linux mapping, so have RTAS mask it. */
348 xics_mask_unknown_vec(vec);
350 /* We might learn about it later, so EOI it */
351 lpar_xirr_info_set(xirr);
355 static unsigned char pop_cppr(void)
357 struct xics_cppr *os_cppr = &__get_cpu_var(xics_cppr);
359 if (WARN_ON(os_cppr->index < 1))
360 return LOWEST_PRIORITY;
362 return os_cppr->stack[--os_cppr->index];
365 static void xics_eoi_direct(unsigned int virq)
367 unsigned int irq = (unsigned int)irq_map[virq].hwirq;
370 direct_xirr_info_set((pop_cppr() << 24) | irq);
373 static void xics_eoi_lpar(unsigned int virq)
375 unsigned int irq = (unsigned int)irq_map[virq].hwirq;
378 lpar_xirr_info_set((pop_cppr() << 24) | irq);
381 static int xics_set_affinity(unsigned int virq, const struct cpumask *cpumask)
388 irq = (unsigned int)irq_map[virq].hwirq;
389 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
392 status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
395 printk(KERN_ERR "%s: ibm,get-xive irq=%u returns %d\n",
396 __func__, irq, status);
401 * For the moment only implement delivery to all cpus or one cpu.
402 * Get current irq_server for the given irq
404 irq_server = get_irq_server(virq, *cpumask, 1);
405 if (irq_server == -1) {
407 cpumask_scnprintf(cpulist, sizeof(cpulist), cpumask);
409 "%s: No online cpus in the mask %s for irq %d\n",
410 __func__, cpulist, virq);
414 status = rtas_call(ibm_set_xive, 3, 1, NULL,
415 irq, irq_server, xics_status[1]);
418 printk(KERN_ERR "%s: ibm,set-xive irq=%u returns %d\n",
419 __func__, irq, status);
426 static struct irq_chip xics_pic_direct = {
428 .startup = xics_startup,
429 .mask = xics_mask_irq,
430 .unmask = xics_unmask_irq,
431 .eoi = xics_eoi_direct,
432 .set_affinity = xics_set_affinity
435 static struct irq_chip xics_pic_lpar = {
437 .startup = xics_startup,
438 .mask = xics_mask_irq,
439 .unmask = xics_unmask_irq,
440 .eoi = xics_eoi_lpar,
441 .set_affinity = xics_set_affinity
445 /* Interface to arch irq controller subsystem layer */
447 /* Points to the irq_chip we're actually using */
448 static struct irq_chip *xics_irq_chip;
450 static int xics_host_match(struct irq_host *h, struct device_node *node)
452 /* IBM machines have interrupt parents of various funky types for things
453 * like vdevices, events, etc... The trick we use here is to match
454 * everything here except the legacy 8259 which is compatible "chrp,iic"
456 return !of_device_is_compatible(node, "chrp,iic");
459 static int xics_host_map(struct irq_host *h, unsigned int virq,
462 pr_devel("xics: map virq %d, hwirq 0x%lx\n", virq, hw);
464 /* Insert the interrupt mapping into the radix tree for fast lookup */
465 irq_radix_revmap_insert(xics_host, virq, hw);
467 irq_to_desc(virq)->status |= IRQ_LEVEL;
468 set_irq_chip_and_handler(virq, xics_irq_chip, handle_fasteoi_irq);
472 static int xics_host_xlate(struct irq_host *h, struct device_node *ct,
473 const u32 *intspec, unsigned int intsize,
474 irq_hw_number_t *out_hwirq, unsigned int *out_flags)
477 /* Current xics implementation translates everything
478 * to level. It is not technically right for MSIs but this
479 * is irrelevant at this point. We might get smarter in the future
481 *out_hwirq = intspec[0];
482 *out_flags = IRQ_TYPE_LEVEL_LOW;
487 static struct irq_host_ops xics_host_ops = {
488 .match = xics_host_match,
489 .map = xics_host_map,
490 .xlate = xics_host_xlate,
493 static void __init xics_init_host(void)
495 if (firmware_has_feature(FW_FEATURE_LPAR))
496 xics_irq_chip = &xics_pic_lpar;
498 xics_irq_chip = &xics_pic_direct;
500 xics_host = irq_alloc_host(NULL, IRQ_HOST_MAP_TREE, 0, &xics_host_ops,
502 BUG_ON(xics_host == NULL);
503 irq_set_default_host(xics_host);
507 /* Inter-processor interrupt support */
511 * XICS only has a single IPI, so encode the messages per CPU
513 struct xics_ipi_struct {
515 } ____cacheline_aligned;
517 static struct xics_ipi_struct xics_ipi_message[NR_CPUS] __cacheline_aligned;
519 static inline void smp_xics_do_message(int cpu, int msg)
521 set_bit(msg, &xics_ipi_message[cpu].value);
523 if (firmware_has_feature(FW_FEATURE_LPAR))
524 lpar_qirr_info(cpu, IPI_PRIORITY);
526 direct_qirr_info(cpu, IPI_PRIORITY);
529 void smp_xics_message_pass(int target, int msg)
533 if (target < NR_CPUS) {
534 smp_xics_do_message(target, msg);
536 for_each_online_cpu(i) {
537 if (target == MSG_ALL_BUT_SELF
538 && i == smp_processor_id())
540 smp_xics_do_message(i, msg);
545 static irqreturn_t xics_ipi_dispatch(int cpu)
547 WARN_ON(cpu_is_offline(cpu));
549 mb(); /* order mmio clearing qirr */
550 while (xics_ipi_message[cpu].value) {
551 if (test_and_clear_bit(PPC_MSG_CALL_FUNCTION,
552 &xics_ipi_message[cpu].value)) {
553 smp_message_recv(PPC_MSG_CALL_FUNCTION);
555 if (test_and_clear_bit(PPC_MSG_RESCHEDULE,
556 &xics_ipi_message[cpu].value)) {
557 smp_message_recv(PPC_MSG_RESCHEDULE);
559 if (test_and_clear_bit(PPC_MSG_CALL_FUNC_SINGLE,
560 &xics_ipi_message[cpu].value)) {
561 smp_message_recv(PPC_MSG_CALL_FUNC_SINGLE);
563 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC)
564 if (test_and_clear_bit(PPC_MSG_DEBUGGER_BREAK,
565 &xics_ipi_message[cpu].value)) {
566 smp_message_recv(PPC_MSG_DEBUGGER_BREAK);
573 static irqreturn_t xics_ipi_action_direct(int irq, void *dev_id)
575 int cpu = smp_processor_id();
577 direct_qirr_info(cpu, 0xff);
579 return xics_ipi_dispatch(cpu);
582 static irqreturn_t xics_ipi_action_lpar(int irq, void *dev_id)
584 int cpu = smp_processor_id();
586 lpar_qirr_info(cpu, 0xff);
588 return xics_ipi_dispatch(cpu);
591 static void xics_request_ipi(void)
596 ipi = irq_create_mapping(xics_host, XICS_IPI);
597 BUG_ON(ipi == NO_IRQ);
600 * IPIs are marked IRQF_DISABLED as they must run with irqs
603 set_irq_handler(ipi, handle_percpu_irq);
604 if (firmware_has_feature(FW_FEATURE_LPAR))
605 rc = request_irq(ipi, xics_ipi_action_lpar,
606 IRQF_DISABLED|IRQF_PERCPU, "IPI", NULL);
608 rc = request_irq(ipi, xics_ipi_action_direct,
609 IRQF_DISABLED|IRQF_PERCPU, "IPI", NULL);
613 int __init smp_xics_probe(void)
617 return cpus_weight(cpu_possible_map);
620 #endif /* CONFIG_SMP */
625 static void xics_update_irq_servers(void)
628 struct device_node *np;
633 /* Find the server numbers for the boot cpu. */
634 np = of_get_cpu_node(boot_cpuid, NULL);
637 ireg = of_get_property(np, "ibm,ppc-interrupt-gserver#s", &ilen);
643 i = ilen / sizeof(int);
644 hcpuid = get_hard_smp_processor_id(boot_cpuid);
646 /* Global interrupt distribution server is specified in the last
647 * entry of "ibm,ppc-interrupt-gserver#s" property. Get the last
648 * entry fom this property for current boot cpu id and use it as
649 * default distribution server
651 for (j = 0; j < i; j += 2) {
652 if (ireg[j] == hcpuid) {
653 default_server = hcpuid;
654 default_distrib_server = ireg[j+1];
661 static void __init xics_map_one_cpu(int hw_id, unsigned long addr,
666 /* This may look gross but it's good enough for now, we don't quite
667 * have a hard -> linux processor id matching.
669 for_each_possible_cpu(i) {
672 if (hw_id == get_hard_smp_processor_id(i)) {
673 xics_per_cpu[i] = ioremap(addr, size);
679 static void __init xics_init_one_node(struct device_node *np,
685 /* This code does the theorically broken assumption that the interrupt
686 * server numbers are the same as the hard CPU numbers.
687 * This happens to be the case so far but we are playing with fire...
688 * should be fixed one of these days. -BenH.
690 ireg = of_get_property(np, "ibm,interrupt-server-ranges", NULL);
692 /* Do that ever happen ? we'll know soon enough... but even good'old
693 * f80 does have that property ..
695 WARN_ON(ireg == NULL);
698 * set node starting index for this node
702 ireg = of_get_property(np, "reg", &ilen);
704 panic("xics_init_IRQ: can't find interrupt reg property");
706 while (ilen >= (4 * sizeof(u32))) {
707 unsigned long addr, size;
709 /* XXX Use proper OF parsing code here !!! */
710 addr = (unsigned long)*ireg++ << 32;
714 size = (unsigned long)*ireg++ << 32;
718 xics_map_one_cpu(*indx, addr, size);
723 void __init xics_init_IRQ(void)
725 struct device_node *np;
730 ppc64_boot_msg(0x20, "XICS Init");
732 ibm_get_xive = rtas_token("ibm,get-xive");
733 ibm_set_xive = rtas_token("ibm,set-xive");
734 ibm_int_on = rtas_token("ibm,int-on");
735 ibm_int_off = rtas_token("ibm,int-off");
737 for_each_node_by_type(np, "PowerPC-External-Interrupt-Presentation") {
739 if (firmware_has_feature(FW_FEATURE_LPAR)) {
743 xics_init_one_node(np, &indx);
748 /* get the bit size of server numbers */
751 for_each_compatible_node(np, NULL, "ibm,ppc-xics") {
752 isize = of_get_property(np, "ibm,interrupt-server#-size", NULL);
758 interrupt_server_size = *isize;
760 } else if (*isize != interrupt_server_size) {
761 printk(KERN_WARNING "XICS: "
762 "mismatched ibm,interrupt-server#-size\n");
763 interrupt_server_size = max(*isize,
764 interrupt_server_size);
768 xics_update_irq_servers();
771 if (firmware_has_feature(FW_FEATURE_LPAR))
772 ppc_md.get_irq = xics_get_irq_lpar;
774 ppc_md.get_irq = xics_get_irq_direct;
778 ppc64_boot_msg(0x21, "XICS Done");
781 /* Cpu startup, shutdown, and hotplug */
783 static void xics_set_cpu_priority(unsigned char cppr)
785 struct xics_cppr *os_cppr = &__get_cpu_var(xics_cppr);
788 * we only really want to set the priority when there's
789 * just one cppr value on the stack
791 WARN_ON(os_cppr->index != 0);
793 os_cppr->stack[0] = cppr;
795 if (firmware_has_feature(FW_FEATURE_LPAR))
796 lpar_cppr_info(cppr);
798 direct_cppr_info(cppr);
802 /* Have the calling processor join or leave the specified global queue */
803 static void xics_set_cpu_giq(unsigned int gserver, unsigned int join)
808 if (!rtas_indicator_present(GLOBAL_INTERRUPT_QUEUE, NULL))
811 index = (1UL << interrupt_server_size) - 1 - gserver;
813 status = rtas_set_indicator_fast(GLOBAL_INTERRUPT_QUEUE, index, join);
815 WARN(status < 0, "set-indicator(%d, %d, %u) returned %d\n",
816 GLOBAL_INTERRUPT_QUEUE, index, join, status);
819 void xics_setup_cpu(void)
821 xics_set_cpu_priority(LOWEST_PRIORITY);
823 xics_set_cpu_giq(default_distrib_server, 1);
826 void xics_teardown_cpu(void)
828 struct xics_cppr *os_cppr = &__get_cpu_var(xics_cppr);
829 int cpu = smp_processor_id();
832 * we have to reset the cppr index to 0 because we're
833 * not going to return from the IPI
836 xics_set_cpu_priority(0);
838 /* Clear any pending IPI request */
839 if (firmware_has_feature(FW_FEATURE_LPAR))
840 lpar_qirr_info(cpu, 0xff);
842 direct_qirr_info(cpu, 0xff);
845 void xics_kexec_teardown_cpu(int secondary)
850 * we take the ipi irq but and never return so we
851 * need to EOI the IPI, but want to leave our priority 0
853 * should we check all the other interrupts too?
854 * should we be flagging idle loop instead?
855 * or creating some task to be scheduled?
858 if (firmware_has_feature(FW_FEATURE_LPAR))
859 lpar_xirr_info_set((0x00 << 24) | XICS_IPI);
861 direct_xirr_info_set((0x00 << 24) | XICS_IPI);
864 * Some machines need to have at least one cpu in the GIQ,
865 * so leave the master cpu in the group.
868 xics_set_cpu_giq(default_distrib_server, 0);
871 #ifdef CONFIG_HOTPLUG_CPU
873 /* Interrupts are disabled. */
874 void xics_migrate_irqs_away(void)
876 int cpu = smp_processor_id(), hw_cpu = hard_smp_processor_id();
877 unsigned int irq, virq;
879 /* If we used to be the default server, move to the new "boot_cpuid" */
880 if (hw_cpu == default_server)
881 xics_update_irq_servers();
883 /* Reject any interrupt that was queued to us... */
884 xics_set_cpu_priority(0);
886 /* Remove ourselves from the global interrupt queue */
887 xics_set_cpu_giq(default_distrib_server, 0);
889 /* Allow IPIs again... */
890 xics_set_cpu_priority(DEFAULT_PRIORITY);
893 struct irq_desc *desc;
898 /* We cant set affinity on ISA interrupts */
899 if (virq < NUM_ISA_INTERRUPTS)
901 if (irq_map[virq].host != xics_host)
903 irq = (unsigned int)irq_map[virq].hwirq;
904 /* We need to get IPIs still. */
905 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
907 desc = irq_to_desc(virq);
909 /* We only need to migrate enabled IRQS */
910 if (desc == NULL || desc->chip == NULL
911 || desc->action == NULL
912 || desc->chip->set_affinity == NULL)
915 raw_spin_lock_irqsave(&desc->lock, flags);
917 status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
919 printk(KERN_ERR "%s: ibm,get-xive irq=%u returns %d\n",
920 __func__, irq, status);
925 * We only support delivery to all cpus or to one cpu.
926 * The irq has to be migrated only in the single cpu
929 if (xics_status[0] != hw_cpu)
932 printk(KERN_WARNING "IRQ %u affinity broken off cpu %u\n",
935 /* Reset affinity to all cpus */
936 cpumask_setall(irq_to_desc(virq)->affinity);
937 desc->chip->set_affinity(virq, cpu_all_mask);
939 raw_spin_unlock_irqrestore(&desc->lock, flags);