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>
23 #include <asm/firmware.h>
25 #include <asm/pgtable.h>
28 #include <asm/hvcall.h>
29 #include <asm/machdep.h>
32 #include "plpar_wrappers.h"
34 static struct irq_host *xics_host;
37 #define XICS_IRQ_SPURIOUS 0
39 /* Want a priority other than 0. Various HW issues require this. */
40 #define DEFAULT_PRIORITY 5
43 * Mark IPIs as higher priority so we can take them inside interrupts that
44 * arent marked IRQF_DISABLED
46 #define IPI_PRIORITY 4
48 static unsigned int default_server = 0xFF;
49 static unsigned int default_distrib_server = 0;
50 static unsigned int interrupt_server_size = 8;
52 /* RTAS service tokens */
53 static int ibm_get_xive;
54 static int ibm_set_xive;
55 static int ibm_int_on;
56 static int ibm_int_off;
59 /* Direct hardware low level accessors */
61 /* The part of the interrupt presentation layer that we care about */
78 static struct xics_ipl __iomem *xics_per_cpu[NR_CPUS];
80 static inline unsigned int direct_xirr_info_get(void)
82 int cpu = smp_processor_id();
84 return in_be32(&xics_per_cpu[cpu]->xirr.word);
87 static inline void direct_xirr_info_set(unsigned int value)
89 int cpu = smp_processor_id();
91 out_be32(&xics_per_cpu[cpu]->xirr.word, value);
94 static inline void direct_cppr_info(u8 value)
96 int cpu = smp_processor_id();
98 out_8(&xics_per_cpu[cpu]->xirr.bytes[0], value);
101 static inline void direct_qirr_info(int n_cpu, u8 value)
103 out_8(&xics_per_cpu[n_cpu]->qirr.bytes[0], value);
107 /* LPAR low level accessors */
109 static inline unsigned int lpar_xirr_info_get(void)
111 unsigned long lpar_rc;
112 unsigned long return_value;
114 lpar_rc = plpar_xirr(&return_value);
115 if (lpar_rc != H_SUCCESS)
116 panic(" bad return code xirr - rc = %lx \n", lpar_rc);
117 return (unsigned int)return_value;
120 static inline void lpar_xirr_info_set(unsigned int value)
122 unsigned long lpar_rc;
124 lpar_rc = plpar_eoi(value);
125 if (lpar_rc != H_SUCCESS)
126 panic("bad return code EOI - rc = %ld, value=%x\n", lpar_rc,
130 static inline void lpar_cppr_info(u8 value)
132 unsigned long lpar_rc;
134 lpar_rc = plpar_cppr(value);
135 if (lpar_rc != H_SUCCESS)
136 panic("bad return code cppr - rc = %lx\n", lpar_rc);
139 static inline void lpar_qirr_info(int n_cpu , u8 value)
141 unsigned long lpar_rc;
143 lpar_rc = plpar_ipi(get_hard_smp_processor_id(n_cpu), value);
144 if (lpar_rc != H_SUCCESS)
145 panic("bad return code qirr - rc = %lx\n", lpar_rc);
149 /* Interface to generic irq subsystem */
152 static int get_irq_server(unsigned int virq, unsigned int strict_check)
155 /* For the moment only implement delivery to all cpus or one cpu */
157 cpumask_t tmp = CPU_MASK_NONE;
159 cpumask_copy(&cpumask, irq_desc[virq].affinity);
160 if (!distribute_irqs)
161 return default_server;
163 if (!cpus_equal(cpumask, CPU_MASK_ALL)) {
164 cpus_and(tmp, cpu_online_map, cpumask);
166 server = first_cpu(tmp);
168 if (server < NR_CPUS)
169 return get_hard_smp_processor_id(server);
175 if (cpus_equal(cpu_online_map, cpu_present_map))
176 return default_distrib_server;
178 return default_server;
181 static int get_irq_server(unsigned int virq, unsigned int strict_check)
183 return default_server;
187 static void xics_unmask_irq(unsigned int virq)
193 pr_debug("xics: unmask virq %d\n", virq);
195 irq = (unsigned int)irq_map[virq].hwirq;
196 pr_debug(" -> map to hwirq 0x%x\n", irq);
197 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
200 server = get_irq_server(virq, 0);
202 call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq, server,
204 if (call_status != 0) {
206 "%s: ibm_set_xive irq %u server %x returned %d\n",
207 __func__, irq, server, call_status);
211 /* Now unmask the interrupt (often a no-op) */
212 call_status = rtas_call(ibm_int_on, 1, 1, NULL, irq);
213 if (call_status != 0) {
214 printk(KERN_ERR "%s: ibm_int_on irq=%u returned %d\n",
215 __func__, irq, call_status);
220 static unsigned int xics_startup(unsigned int virq)
223 xics_unmask_irq(virq);
227 static void xics_mask_real_irq(unsigned int irq)
234 call_status = rtas_call(ibm_int_off, 1, 1, NULL, irq);
235 if (call_status != 0) {
236 printk(KERN_ERR "%s: ibm_int_off irq=%u returned %d\n",
237 __func__, irq, call_status);
241 /* Have to set XIVE to 0xff to be able to remove a slot */
242 call_status = rtas_call(ibm_set_xive, 3, 1, NULL, irq,
243 default_server, 0xff);
244 if (call_status != 0) {
245 printk(KERN_ERR "%s: ibm_set_xive(0xff) irq=%u returned %d\n",
246 __func__, irq, call_status);
251 static void xics_mask_irq(unsigned int virq)
255 pr_debug("xics: mask virq %d\n", virq);
257 irq = (unsigned int)irq_map[virq].hwirq;
258 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
260 xics_mask_real_irq(irq);
263 static void xics_mask_unknown_vec(unsigned int vec)
265 printk(KERN_ERR "Interrupt %u (real) is invalid, disabling it.\n", vec);
266 xics_mask_real_irq(vec);
269 static inline unsigned int xics_xirr_vector(unsigned int xirr)
272 * The top byte is the old cppr, to be restored on EOI.
273 * The remaining 24 bits are the vector.
275 return xirr & 0x00ffffff;
278 static unsigned int xics_get_irq_direct(void)
280 unsigned int xirr = direct_xirr_info_get();
281 unsigned int vec = xics_xirr_vector(xirr);
284 if (vec == XICS_IRQ_SPURIOUS)
287 irq = irq_radix_revmap_lookup(xics_host, vec);
288 if (likely(irq != NO_IRQ))
291 /* We don't have a linux mapping, so have rtas mask it. */
292 xics_mask_unknown_vec(vec);
294 /* We might learn about it later, so EOI it */
295 direct_xirr_info_set(xirr);
299 static unsigned int xics_get_irq_lpar(void)
301 unsigned int xirr = lpar_xirr_info_get();
302 unsigned int vec = xics_xirr_vector(xirr);
305 if (vec == XICS_IRQ_SPURIOUS)
308 irq = irq_radix_revmap_lookup(xics_host, vec);
309 if (likely(irq != NO_IRQ))
312 /* We don't have a linux mapping, so have RTAS mask it. */
313 xics_mask_unknown_vec(vec);
315 /* We might learn about it later, so EOI it */
316 lpar_xirr_info_set(xirr);
320 static void xics_eoi_direct(unsigned int virq)
322 unsigned int irq = (unsigned int)irq_map[virq].hwirq;
325 direct_xirr_info_set((0xff << 24) | irq);
328 static void xics_eoi_lpar(unsigned int virq)
330 unsigned int irq = (unsigned int)irq_map[virq].hwirq;
333 lpar_xirr_info_set((0xff << 24) | irq);
336 static void xics_set_affinity(unsigned int virq, const struct cpumask *cpumask)
343 irq = (unsigned int)irq_map[virq].hwirq;
344 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
347 status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
350 printk(KERN_ERR "%s: ibm,get-xive irq=%u returns %d\n",
351 __func__, irq, status);
356 * For the moment only implement delivery to all cpus or one cpu.
357 * Get current irq_server for the given irq
359 irq_server = get_irq_server(virq, 1);
360 if (irq_server == -1) {
362 cpumask_scnprintf(cpulist, sizeof(cpulist), cpumask);
364 "%s: No online cpus in the mask %s for irq %d\n",
365 __func__, cpulist, virq);
369 status = rtas_call(ibm_set_xive, 3, 1, NULL,
370 irq, irq_server, xics_status[1]);
373 printk(KERN_ERR "%s: ibm,set-xive irq=%u returns %d\n",
374 __func__, irq, status);
379 static struct irq_chip xics_pic_direct = {
380 .typename = " XICS ",
381 .startup = xics_startup,
382 .mask = xics_mask_irq,
383 .unmask = xics_unmask_irq,
384 .eoi = xics_eoi_direct,
385 .set_affinity = xics_set_affinity
388 static struct irq_chip xics_pic_lpar = {
389 .typename = " XICS ",
390 .startup = xics_startup,
391 .mask = xics_mask_irq,
392 .unmask = xics_unmask_irq,
393 .eoi = xics_eoi_lpar,
394 .set_affinity = xics_set_affinity
398 /* Interface to arch irq controller subsystem layer */
400 /* Points to the irq_chip we're actually using */
401 static struct irq_chip *xics_irq_chip;
403 static int xics_host_match(struct irq_host *h, struct device_node *node)
405 /* IBM machines have interrupt parents of various funky types for things
406 * like vdevices, events, etc... The trick we use here is to match
407 * everything here except the legacy 8259 which is compatible "chrp,iic"
409 return !of_device_is_compatible(node, "chrp,iic");
412 static int xics_host_map(struct irq_host *h, unsigned int virq,
415 pr_debug("xics: map virq %d, hwirq 0x%lx\n", virq, hw);
417 /* Insert the interrupt mapping into the radix tree for fast lookup */
418 irq_radix_revmap_insert(xics_host, virq, hw);
420 get_irq_desc(virq)->status |= IRQ_LEVEL;
421 set_irq_chip_and_handler(virq, xics_irq_chip, handle_fasteoi_irq);
425 static int xics_host_xlate(struct irq_host *h, struct device_node *ct,
426 u32 *intspec, unsigned int intsize,
427 irq_hw_number_t *out_hwirq, unsigned int *out_flags)
430 /* Current xics implementation translates everything
431 * to level. It is not technically right for MSIs but this
432 * is irrelevant at this point. We might get smarter in the future
434 *out_hwirq = intspec[0];
435 *out_flags = IRQ_TYPE_LEVEL_LOW;
440 static struct irq_host_ops xics_host_ops = {
441 .match = xics_host_match,
442 .map = xics_host_map,
443 .xlate = xics_host_xlate,
446 static void __init xics_init_host(void)
448 if (firmware_has_feature(FW_FEATURE_LPAR))
449 xics_irq_chip = &xics_pic_lpar;
451 xics_irq_chip = &xics_pic_direct;
453 xics_host = irq_alloc_host(NULL, IRQ_HOST_MAP_TREE, 0, &xics_host_ops,
455 BUG_ON(xics_host == NULL);
456 irq_set_default_host(xics_host);
460 /* Inter-processor interrupt support */
464 * XICS only has a single IPI, so encode the messages per CPU
466 struct xics_ipi_struct {
468 } ____cacheline_aligned;
470 static struct xics_ipi_struct xics_ipi_message[NR_CPUS] __cacheline_aligned;
472 static inline void smp_xics_do_message(int cpu, int msg)
474 set_bit(msg, &xics_ipi_message[cpu].value);
476 if (firmware_has_feature(FW_FEATURE_LPAR))
477 lpar_qirr_info(cpu, IPI_PRIORITY);
479 direct_qirr_info(cpu, IPI_PRIORITY);
482 void smp_xics_message_pass(int target, int msg)
486 if (target < NR_CPUS) {
487 smp_xics_do_message(target, msg);
489 for_each_online_cpu(i) {
490 if (target == MSG_ALL_BUT_SELF
491 && i == smp_processor_id())
493 smp_xics_do_message(i, msg);
498 static irqreturn_t xics_ipi_dispatch(int cpu)
500 WARN_ON(cpu_is_offline(cpu));
502 mb(); /* order mmio clearing qirr */
503 while (xics_ipi_message[cpu].value) {
504 if (test_and_clear_bit(PPC_MSG_CALL_FUNCTION,
505 &xics_ipi_message[cpu].value)) {
506 smp_message_recv(PPC_MSG_CALL_FUNCTION);
508 if (test_and_clear_bit(PPC_MSG_RESCHEDULE,
509 &xics_ipi_message[cpu].value)) {
510 smp_message_recv(PPC_MSG_RESCHEDULE);
512 if (test_and_clear_bit(PPC_MSG_CALL_FUNC_SINGLE,
513 &xics_ipi_message[cpu].value)) {
514 smp_message_recv(PPC_MSG_CALL_FUNC_SINGLE);
516 #if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC)
517 if (test_and_clear_bit(PPC_MSG_DEBUGGER_BREAK,
518 &xics_ipi_message[cpu].value)) {
519 smp_message_recv(PPC_MSG_DEBUGGER_BREAK);
526 static irqreturn_t xics_ipi_action_direct(int irq, void *dev_id)
528 int cpu = smp_processor_id();
530 direct_qirr_info(cpu, 0xff);
532 return xics_ipi_dispatch(cpu);
535 static irqreturn_t xics_ipi_action_lpar(int irq, void *dev_id)
537 int cpu = smp_processor_id();
539 lpar_qirr_info(cpu, 0xff);
541 return xics_ipi_dispatch(cpu);
544 static void xics_request_ipi(void)
549 ipi = irq_create_mapping(xics_host, XICS_IPI);
550 BUG_ON(ipi == NO_IRQ);
553 * IPIs are marked IRQF_DISABLED as they must run with irqs
556 set_irq_handler(ipi, handle_percpu_irq);
557 if (firmware_has_feature(FW_FEATURE_LPAR))
558 rc = request_irq(ipi, xics_ipi_action_lpar,
559 IRQF_DISABLED|IRQF_PERCPU, "IPI", NULL);
561 rc = request_irq(ipi, xics_ipi_action_direct,
562 IRQF_DISABLED|IRQF_PERCPU, "IPI", NULL);
566 int __init smp_xics_probe(void)
570 return cpus_weight(cpu_possible_map);
573 #endif /* CONFIG_SMP */
578 static void xics_update_irq_servers(void)
581 struct device_node *np;
586 /* Find the server numbers for the boot cpu. */
587 np = of_get_cpu_node(boot_cpuid, NULL);
590 ireg = of_get_property(np, "ibm,ppc-interrupt-gserver#s", &ilen);
596 i = ilen / sizeof(int);
597 hcpuid = get_hard_smp_processor_id(boot_cpuid);
599 /* Global interrupt distribution server is specified in the last
600 * entry of "ibm,ppc-interrupt-gserver#s" property. Get the last
601 * entry fom this property for current boot cpu id and use it as
602 * default distribution server
604 for (j = 0; j < i; j += 2) {
605 if (ireg[j] == hcpuid) {
606 default_server = hcpuid;
607 default_distrib_server = ireg[j+1];
614 static void __init xics_map_one_cpu(int hw_id, unsigned long addr,
619 /* This may look gross but it's good enough for now, we don't quite
620 * have a hard -> linux processor id matching.
622 for_each_possible_cpu(i) {
625 if (hw_id == get_hard_smp_processor_id(i)) {
626 xics_per_cpu[i] = ioremap(addr, size);
632 static void __init xics_init_one_node(struct device_node *np,
638 /* This code does the theorically broken assumption that the interrupt
639 * server numbers are the same as the hard CPU numbers.
640 * This happens to be the case so far but we are playing with fire...
641 * should be fixed one of these days. -BenH.
643 ireg = of_get_property(np, "ibm,interrupt-server-ranges", NULL);
645 /* Do that ever happen ? we'll know soon enough... but even good'old
646 * f80 does have that property ..
648 WARN_ON(ireg == NULL);
651 * set node starting index for this node
655 ireg = of_get_property(np, "reg", &ilen);
657 panic("xics_init_IRQ: can't find interrupt reg property");
659 while (ilen >= (4 * sizeof(u32))) {
660 unsigned long addr, size;
662 /* XXX Use proper OF parsing code here !!! */
663 addr = (unsigned long)*ireg++ << 32;
667 size = (unsigned long)*ireg++ << 32;
671 xics_map_one_cpu(*indx, addr, size);
676 void __init xics_init_IRQ(void)
678 struct device_node *np;
683 ppc64_boot_msg(0x20, "XICS Init");
685 ibm_get_xive = rtas_token("ibm,get-xive");
686 ibm_set_xive = rtas_token("ibm,set-xive");
687 ibm_int_on = rtas_token("ibm,int-on");
688 ibm_int_off = rtas_token("ibm,int-off");
690 for_each_node_by_type(np, "PowerPC-External-Interrupt-Presentation") {
692 if (firmware_has_feature(FW_FEATURE_LPAR)) {
696 xics_init_one_node(np, &indx);
701 /* get the bit size of server numbers */
704 for_each_compatible_node(np, NULL, "ibm,ppc-xics") {
705 isize = of_get_property(np, "ibm,interrupt-server#-size", NULL);
711 interrupt_server_size = *isize;
713 } else if (*isize != interrupt_server_size) {
714 printk(KERN_WARNING "XICS: "
715 "mismatched ibm,interrupt-server#-size\n");
716 interrupt_server_size = max(*isize,
717 interrupt_server_size);
721 xics_update_irq_servers();
724 if (firmware_has_feature(FW_FEATURE_LPAR))
725 ppc_md.get_irq = xics_get_irq_lpar;
727 ppc_md.get_irq = xics_get_irq_direct;
731 ppc64_boot_msg(0x21, "XICS Done");
734 /* Cpu startup, shutdown, and hotplug */
736 static void xics_set_cpu_priority(unsigned char cppr)
738 if (firmware_has_feature(FW_FEATURE_LPAR))
739 lpar_cppr_info(cppr);
741 direct_cppr_info(cppr);
745 /* Have the calling processor join or leave the specified global queue */
746 static void xics_set_cpu_giq(unsigned int gserver, unsigned int join)
751 if (!rtas_indicator_present(GLOBAL_INTERRUPT_QUEUE, NULL))
754 index = (1UL << interrupt_server_size) - 1 - gserver;
756 status = rtas_set_indicator_fast(GLOBAL_INTERRUPT_QUEUE, index, join);
758 WARN(status < 0, "set-indicator(%d, %d, %u) returned %d\n",
759 GLOBAL_INTERRUPT_QUEUE, index, join, status);
762 void xics_setup_cpu(void)
764 xics_set_cpu_priority(0xff);
766 xics_set_cpu_giq(default_distrib_server, 1);
769 void xics_teardown_cpu(void)
771 int cpu = smp_processor_id();
773 xics_set_cpu_priority(0);
775 /* Clear any pending IPI request */
776 if (firmware_has_feature(FW_FEATURE_LPAR))
777 lpar_qirr_info(cpu, 0xff);
779 direct_qirr_info(cpu, 0xff);
782 void xics_kexec_teardown_cpu(int secondary)
787 * we take the ipi irq but and never return so we
788 * need to EOI the IPI, but want to leave our priority 0
790 * should we check all the other interrupts too?
791 * should we be flagging idle loop instead?
792 * or creating some task to be scheduled?
795 if (firmware_has_feature(FW_FEATURE_LPAR))
796 lpar_xirr_info_set((0x00 << 24) | XICS_IPI);
798 direct_xirr_info_set((0x00 << 24) | XICS_IPI);
801 * Some machines need to have at least one cpu in the GIQ,
802 * so leave the master cpu in the group.
805 xics_set_cpu_giq(default_distrib_server, 0);
808 #ifdef CONFIG_HOTPLUG_CPU
810 /* Interrupts are disabled. */
811 void xics_migrate_irqs_away(void)
813 int cpu = smp_processor_id(), hw_cpu = hard_smp_processor_id();
814 unsigned int irq, virq;
816 /* If we used to be the default server, move to the new "boot_cpuid" */
817 if (hw_cpu == default_server)
818 xics_update_irq_servers();
820 /* Reject any interrupt that was queued to us... */
821 xics_set_cpu_priority(0);
823 /* Remove ourselves from the global interrupt queue */
824 xics_set_cpu_giq(default_distrib_server, 0);
826 /* Allow IPIs again... */
827 xics_set_cpu_priority(DEFAULT_PRIORITY);
830 struct irq_desc *desc;
835 /* We cant set affinity on ISA interrupts */
836 if (virq < NUM_ISA_INTERRUPTS)
838 if (irq_map[virq].host != xics_host)
840 irq = (unsigned int)irq_map[virq].hwirq;
841 /* We need to get IPIs still. */
842 if (irq == XICS_IPI || irq == XICS_IRQ_SPURIOUS)
844 desc = get_irq_desc(virq);
846 /* We only need to migrate enabled IRQS */
847 if (desc == NULL || desc->chip == NULL
848 || desc->action == NULL
849 || desc->chip->set_affinity == NULL)
852 spin_lock_irqsave(&desc->lock, flags);
854 status = rtas_call(ibm_get_xive, 1, 3, xics_status, irq);
856 printk(KERN_ERR "%s: ibm,get-xive irq=%u returns %d\n",
857 __func__, irq, status);
862 * We only support delivery to all cpus or to one cpu.
863 * The irq has to be migrated only in the single cpu
866 if (xics_status[0] != hw_cpu)
869 printk(KERN_WARNING "IRQ %u affinity broken off cpu %u\n",
872 /* Reset affinity to all cpus */
873 cpumask_setall(irq_desc[virq].affinity);
874 desc->chip->set_affinity(virq, cpu_all_mask);
876 spin_unlock_irqrestore(&desc->lock, flags);