aboutsummaryrefslogtreecommitdiffstats
path: root/arch/sparc64/kernel
diff options
context:
space:
mode:
authorDavid S. Miller <davem@sunset.davemloft.net>2007-10-11 06:16:13 -0400
committerDavid S. Miller <davem@sunset.davemloft.net>2007-10-14 00:53:13 -0400
commit759f89e03c9e5656ff18c02e21b439506f7c0cdc (patch)
tree6e7703c0672210f2c0a1168de161d15c7470081d /arch/sparc64/kernel
parenta2cd15586e630b0870bf34783568d83901890743 (diff)
[SPARC64]: Consolidate MSI support code.
This also makes us use the MSI queues correctly. Each MSI queue is serviced by a normal sun4u/sun4v INO interrupt handler. This handler runs the MSI queue and dispatches the virtual interrupts indicated by arriving MSIs in that MSI queue. All of the common logic is placed in pci_msi.c, with callbacks to handle the PCI controller specific aspects of the operations. This common infrastructure will make it much easier to add MSG support. Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc64/kernel')
-rw-r--r--arch/sparc64/kernel/Makefile1
-rw-r--r--arch/sparc64/kernel/irq.c230
-rw-r--r--arch/sparc64/kernel/pci_fire.c380
-rw-r--r--arch/sparc64/kernel/pci_impl.h31
-rw-r--r--arch/sparc64/kernel/pci_msi.c433
-rw-r--r--arch/sparc64/kernel/pci_sun4v.c400
6 files changed, 700 insertions, 775 deletions
diff --git a/arch/sparc64/kernel/Makefile b/arch/sparc64/kernel/Makefile
index 40d2f3aae91e..112c46e66578 100644
--- a/arch/sparc64/kernel/Makefile
+++ b/arch/sparc64/kernel/Makefile
@@ -18,6 +18,7 @@ obj-$(CONFIG_STACKTRACE) += stacktrace.o
18obj-$(CONFIG_PCI) += ebus.o isa.o pci_common.o \ 18obj-$(CONFIG_PCI) += ebus.o isa.o pci_common.o \
19 pci_psycho.o pci_sabre.o pci_schizo.o \ 19 pci_psycho.o pci_sabre.o pci_schizo.o \
20 pci_sun4v.o pci_sun4v_asm.o pci_fire.o 20 pci_sun4v.o pci_sun4v_asm.o pci_fire.o
21obj-$(CONFIG_PCI_MSI) += pci_msi.o
21obj-$(CONFIG_SMP) += smp.o trampoline.o hvtramp.o 22obj-$(CONFIG_SMP) += smp.o trampoline.o hvtramp.o
22obj-$(CONFIG_SPARC32_COMPAT) += sys32.o sys_sparc32.o signal32.o 23obj-$(CONFIG_SPARC32_COMPAT) += sys32.o sys_sparc32.o signal32.o
23obj-$(CONFIG_BINFMT_ELF32) += binfmt_elf32.o 24obj-$(CONFIG_BINFMT_ELF32) += binfmt_elf32.o
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index 7f5a4c77e3e4..045ab27d4271 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -21,7 +21,6 @@
21#include <linux/seq_file.h> 21#include <linux/seq_file.h>
22#include <linux/bootmem.h> 22#include <linux/bootmem.h>
23#include <linux/irq.h> 23#include <linux/irq.h>
24#include <linux/msi.h>
25 24
26#include <asm/ptrace.h> 25#include <asm/ptrace.h>
27#include <asm/processor.h> 26#include <asm/processor.h>
@@ -92,39 +91,46 @@ static struct {
92 unsigned int dev_handle; 91 unsigned int dev_handle;
93 unsigned int dev_ino; 92 unsigned int dev_ino;
94} virt_to_real_irq_table[NR_IRQS]; 93} virt_to_real_irq_table[NR_IRQS];
94static DEFINE_SPINLOCK(virt_irq_alloc_lock);
95 95
96static unsigned char virt_irq_alloc(unsigned int real_irq) 96unsigned char virt_irq_alloc(unsigned int real_irq)
97{ 97{
98 unsigned long flags;
98 unsigned char ent; 99 unsigned char ent;
99 100
100 BUILD_BUG_ON(NR_IRQS >= 256); 101 BUILD_BUG_ON(NR_IRQS >= 256);
101 102
103 spin_lock_irqsave(&virt_irq_alloc_lock, flags);
104
102 for (ent = 1; ent < NR_IRQS; ent++) { 105 for (ent = 1; ent < NR_IRQS; ent++) {
103 if (!virt_to_real_irq_table[ent].irq) 106 if (!virt_to_real_irq_table[ent].irq)
104 break; 107 break;
105 } 108 }
106 if (ent >= NR_IRQS) { 109 if (ent >= NR_IRQS) {
107 printk(KERN_ERR "IRQ: Out of virtual IRQs.\n"); 110 printk(KERN_ERR "IRQ: Out of virtual IRQs.\n");
108 return 0; 111 ent = 0;
112 } else {
113 virt_to_real_irq_table[ent].irq = real_irq;
109 } 114 }
110 115
111 virt_to_real_irq_table[ent].irq = real_irq; 116 spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
112 117
113 return ent; 118 return ent;
114} 119}
115 120
116#ifdef CONFIG_PCI_MSI 121#ifdef CONFIG_PCI_MSI
117static void virt_irq_free(unsigned int virt_irq) 122void virt_irq_free(unsigned int virt_irq)
118{ 123{
119 unsigned int real_irq; 124 unsigned long flags;
120 125
121 if (virt_irq >= NR_IRQS) 126 if (virt_irq >= NR_IRQS)
122 return; 127 return;
123 128
124 real_irq = virt_to_real_irq_table[virt_irq].irq; 129 spin_lock_irqsave(&virt_irq_alloc_lock, flags);
130
125 virt_to_real_irq_table[virt_irq].irq = 0; 131 virt_to_real_irq_table[virt_irq].irq = 0;
126 132
127 __bucket(real_irq)->virt_irq = 0; 133 spin_unlock_irqrestore(&virt_irq_alloc_lock, flags);
128} 134}
129#endif 135#endif
130 136
@@ -217,27 +223,8 @@ struct irq_handler_data {
217 void (*pre_handler)(unsigned int, void *, void *); 223 void (*pre_handler)(unsigned int, void *, void *);
218 void *pre_handler_arg1; 224 void *pre_handler_arg1;
219 void *pre_handler_arg2; 225 void *pre_handler_arg2;
220
221 u32 msi;
222}; 226};
223 227
224void sparc64_set_msi(unsigned int virt_irq, u32 msi)
225{
226 struct irq_handler_data *data = get_irq_chip_data(virt_irq);
227
228 if (data)
229 data->msi = msi;
230}
231
232u32 sparc64_get_msi(unsigned int virt_irq)
233{
234 struct irq_handler_data *data = get_irq_chip_data(virt_irq);
235
236 if (data)
237 return data->msi;
238 return 0xffffffff;
239}
240
241static inline struct ino_bucket *virt_irq_to_bucket(unsigned int virt_irq) 228static inline struct ino_bucket *virt_irq_to_bucket(unsigned int virt_irq)
242{ 229{
243 unsigned int real_irq = virt_to_real_irq(virt_irq); 230 unsigned int real_irq = virt_to_real_irq(virt_irq);
@@ -405,32 +392,6 @@ static void sun4v_irq_disable(unsigned int virt_irq)
405 } 392 }
406} 393}
407 394
408#ifdef CONFIG_PCI_MSI
409static void sun4u_msi_enable(unsigned int virt_irq)
410{
411 sun4u_irq_enable(virt_irq);
412 unmask_msi_irq(virt_irq);
413}
414
415static void sun4u_msi_disable(unsigned int virt_irq)
416{
417 mask_msi_irq(virt_irq);
418 sun4u_irq_disable(virt_irq);
419}
420
421static void sun4v_msi_enable(unsigned int virt_irq)
422{
423 sun4v_irq_enable(virt_irq);
424 unmask_msi_irq(virt_irq);
425}
426
427static void sun4v_msi_disable(unsigned int virt_irq)
428{
429 mask_msi_irq(virt_irq);
430 sun4v_irq_disable(virt_irq);
431}
432#endif
433
434static void sun4v_irq_end(unsigned int virt_irq) 395static void sun4v_irq_end(unsigned int virt_irq)
435{ 396{
436 struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq); 397 struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
@@ -585,39 +546,6 @@ static struct irq_chip sun4v_irq = {
585 .set_affinity = sun4v_set_affinity, 546 .set_affinity = sun4v_set_affinity,
586}; 547};
587 548
588static struct irq_chip sun4v_irq_ack = {
589 .typename = "sun4v+ack",
590 .enable = sun4v_irq_enable,
591 .disable = sun4v_irq_disable,
592 .ack = run_pre_handler,
593 .end = sun4v_irq_end,
594 .set_affinity = sun4v_set_affinity,
595};
596
597#ifdef CONFIG_PCI_MSI
598static struct irq_chip sun4u_msi = {
599 .typename = "sun4u+msi",
600 .mask = mask_msi_irq,
601 .unmask = unmask_msi_irq,
602 .enable = sun4u_msi_enable,
603 .disable = sun4u_msi_disable,
604 .ack = run_pre_handler,
605 .end = sun4u_irq_end,
606 .set_affinity = sun4u_set_affinity,
607};
608
609static struct irq_chip sun4v_msi = {
610 .typename = "sun4v+msi",
611 .mask = mask_msi_irq,
612 .unmask = unmask_msi_irq,
613 .enable = sun4v_msi_enable,
614 .disable = sun4v_msi_disable,
615 .ack = run_pre_handler,
616 .end = sun4v_irq_end,
617 .set_affinity = sun4v_set_affinity,
618};
619#endif
620
621static struct irq_chip sun4v_virq = { 549static struct irq_chip sun4v_virq = {
622 .typename = "vsun4v", 550 .typename = "vsun4v",
623 .enable = sun4v_virq_enable, 551 .enable = sun4v_virq_enable,
@@ -626,42 +554,27 @@ static struct irq_chip sun4v_virq = {
626 .set_affinity = sun4v_virt_set_affinity, 554 .set_affinity = sun4v_virt_set_affinity,
627}; 555};
628 556
629static struct irq_chip sun4v_virq_ack = {
630 .typename = "vsun4v+ack",
631 .enable = sun4v_virq_enable,
632 .disable = sun4v_virq_disable,
633 .ack = run_pre_handler,
634 .end = sun4v_virq_end,
635 .set_affinity = sun4v_virt_set_affinity,
636};
637
638void irq_install_pre_handler(int virt_irq, 557void irq_install_pre_handler(int virt_irq,
639 void (*func)(unsigned int, void *, void *), 558 void (*func)(unsigned int, void *, void *),
640 void *arg1, void *arg2) 559 void *arg1, void *arg2)
641{ 560{
642 struct irq_handler_data *data = get_irq_chip_data(virt_irq); 561 struct irq_handler_data *data = get_irq_chip_data(virt_irq);
643 struct irq_chip *chip; 562 struct irq_chip *chip = get_irq_chip(virt_irq);
563
564 if (WARN_ON(chip == &sun4v_irq || chip == &sun4v_virq)) {
565 printk(KERN_ERR "IRQ: Trying to install pre-handler on "
566 "sun4v irq %u\n", virt_irq);
567 return;
568 }
644 569
645 data->pre_handler = func; 570 data->pre_handler = func;
646 data->pre_handler_arg1 = arg1; 571 data->pre_handler_arg1 = arg1;
647 data->pre_handler_arg2 = arg2; 572 data->pre_handler_arg2 = arg2;
648 573
649 chip = get_irq_chip(virt_irq); 574 if (chip == &sun4u_irq_ack)
650 if (chip == &sun4u_irq_ack ||
651 chip == &sun4v_irq_ack ||
652 chip == &sun4v_virq_ack
653#ifdef CONFIG_PCI_MSI
654 || chip == &sun4u_msi
655 || chip == &sun4v_msi
656#endif
657 )
658 return; 575 return;
659 576
660 chip = (chip == &sun4u_irq ? 577 set_irq_chip(virt_irq, &sun4u_irq_ack);
661 &sun4u_irq_ack :
662 (chip == &sun4v_irq ?
663 &sun4v_irq_ack : &sun4v_virq_ack));
664 set_irq_chip(virt_irq, chip);
665} 578}
666 579
667unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap) 580unsigned int build_irq(int inofixup, unsigned long iclr, unsigned long imap)
@@ -765,103 +678,6 @@ unsigned int sun4v_build_virq(u32 devhandle, unsigned int devino)
765 return virq; 678 return virq;
766} 679}
767 680
768#ifdef CONFIG_PCI_MSI
769unsigned int sun4v_build_msi(u32 devhandle, unsigned int *virt_irq_p,
770 unsigned int msi_start, unsigned int msi_end)
771{
772 struct ino_bucket *bucket;
773 struct irq_handler_data *data;
774 unsigned long sysino;
775 unsigned int devino;
776
777 BUG_ON(tlb_type != hypervisor);
778
779 /* Find a free devino in the given range. */
780 for (devino = msi_start; devino < msi_end; devino++) {
781 sysino = sun4v_devino_to_sysino(devhandle, devino);
782 bucket = &ivector_table[sysino];
783 if (!bucket->virt_irq)
784 break;
785 }
786 if (devino >= msi_end)
787 return -ENOSPC;
788
789 sysino = sun4v_devino_to_sysino(devhandle, devino);
790 bucket = &ivector_table[sysino];
791 bucket->virt_irq = virt_irq_alloc(__irq(bucket));
792 *virt_irq_p = bucket->virt_irq;
793 set_irq_chip(bucket->virt_irq, &sun4v_msi);
794
795 data = get_irq_chip_data(bucket->virt_irq);
796 if (unlikely(data))
797 return devino;
798
799 data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
800 if (unlikely(!data)) {
801 virt_irq_free(*virt_irq_p);
802 return -ENOMEM;
803 }
804 set_irq_chip_data(bucket->virt_irq, data);
805
806 data->imap = ~0UL;
807 data->iclr = ~0UL;
808
809 return devino;
810}
811
812void sun4v_destroy_msi(unsigned int virt_irq)
813{
814 virt_irq_free(virt_irq);
815}
816
817unsigned int sun4u_build_msi(u32 portid, unsigned int *virt_irq_p,
818 unsigned int msi_start, unsigned int msi_end,
819 unsigned long imap_base, unsigned long iclr_base)
820{
821 struct ino_bucket *bucket;
822 struct irq_handler_data *data;
823 unsigned long sysino;
824 unsigned int devino;
825
826 /* Find a free devino in the given range. */
827 for (devino = msi_start; devino < msi_end; devino++) {
828 sysino = (portid << 6) | devino;
829 bucket = &ivector_table[sysino];
830 if (!bucket->virt_irq)
831 break;
832 }
833 if (devino >= msi_end)
834 return -ENOSPC;
835
836 sysino = (portid << 6) | devino;
837 bucket = &ivector_table[sysino];
838 bucket->virt_irq = virt_irq_alloc(__irq(bucket));
839 *virt_irq_p = bucket->virt_irq;
840 set_irq_chip(bucket->virt_irq, &sun4u_msi);
841
842 data = get_irq_chip_data(bucket->virt_irq);
843 if (unlikely(data))
844 return devino;
845
846 data = kzalloc(sizeof(struct irq_handler_data), GFP_ATOMIC);
847 if (unlikely(!data)) {
848 virt_irq_free(*virt_irq_p);
849 return -ENOMEM;
850 }
851 set_irq_chip_data(bucket->virt_irq, data);
852
853 data->imap = (imap_base + (devino * 0x8UL));
854 data->iclr = (iclr_base + (devino * 0x8UL));
855
856 return devino;
857}
858
859void sun4u_destroy_msi(unsigned int virt_irq)
860{
861 virt_irq_free(virt_irq);
862}
863#endif
864
865void ack_bad_irq(unsigned int virt_irq) 681void ack_bad_irq(unsigned int virt_irq)
866{ 682{
867 struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq); 683 struct ino_bucket *bucket = virt_irq_to_bucket(virt_irq);
diff --git a/arch/sparc64/kernel/pci_fire.c b/arch/sparc64/kernel/pci_fire.c
index 090f26579678..bcf6a5d425ab 100644
--- a/arch/sparc64/kernel/pci_fire.c
+++ b/arch/sparc64/kernel/pci_fire.c
@@ -161,90 +161,92 @@ struct pci_msiq_entry {
161#define MSI_64BIT_ADDR 0x034008UL 161#define MSI_64BIT_ADDR 0x034008UL
162#define MSI_64BIT_ADDR_VAL 0xffffffffffff0000UL 162#define MSI_64BIT_ADDR_VAL 0xffffffffffff0000UL
163 163
164/* For now this just runs as a pre-handler for the real interrupt handler. 164static int pci_fire_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
165 * So we just walk through the queue and ACK all the entries, update the 165 unsigned long *head)
166 * head pointer, and return. 166{
167 * 167 *head = fire_read(pbm->pbm_regs + EVENT_QUEUE_HEAD(msiqid));
168 * In the longer term it would be nice to do something more integrated 168 return 0;
169 * wherein we can pass in some of this MSI info to the drivers. This 169}
170 * would be most useful for PCIe fabric error messages, although we could 170
171 * invoke those directly from the loop here in order to pass the info around. 171static int pci_fire_dequeue_msi(struct pci_pbm_info *pbm, unsigned long msiqid,
172 */ 172 unsigned long *head, unsigned long *msi)
173static void pci_msi_prehandler(unsigned int ino, void *data1, void *data2)
174{ 173{
175 unsigned long msiqid, orig_head, head, type_fmt, type; 174 unsigned long type_fmt, type, msi_num;
176 struct pci_pbm_info *pbm = data1;
177 struct pci_msiq_entry *base, *ep; 175 struct pci_msiq_entry *base, *ep;
178 176
179 msiqid = (unsigned long) data2; 177 base = (pbm->msi_queues + ((msiqid - pbm->msiq_first) * 8192));
178 ep = &base[*head];
180 179
181 head = fire_read(pbm->pbm_regs + EVENT_QUEUE_HEAD(msiqid)); 180 if ((ep->word0 & MSIQ_WORD0_FMT_TYPE) == 0)
181 return 0;
182 182
183 orig_head = head; 183 type_fmt = ((ep->word0 & MSIQ_WORD0_FMT_TYPE) >>
184 base = (pbm->msi_queues + ((msiqid - pbm->msiq_first) * 8192)); 184 MSIQ_WORD0_FMT_TYPE_SHIFT);
185 ep = &base[head]; 185 type = (type_fmt >> 3);
186 while ((ep->word0 & MSIQ_WORD0_FMT_TYPE) != 0) { 186 if (unlikely(type != MSIQ_TYPE_MSI32 &&
187 unsigned long msi_num; 187 type != MSIQ_TYPE_MSI64))
188 188 return -EINVAL;
189 type_fmt = ((ep->word0 & MSIQ_WORD0_FMT_TYPE) >>
190 MSIQ_WORD0_FMT_TYPE_SHIFT);
191 type = (type_fmt >>3);
192 if (unlikely(type != MSIQ_TYPE_MSI32 &&
193 type != MSIQ_TYPE_MSI64))
194 goto bad_type;
195
196 msi_num = ((ep->word0 & MSIQ_WORD0_DATA0) >>
197 MSIQ_WORD0_DATA0_SHIFT);
198
199 fire_write(pbm->pbm_regs + MSI_CLEAR(msi_num),
200 MSI_CLEAR_EQWR_N);
201
202 /* Clear the entry. */
203 ep->word0 &= ~MSIQ_WORD0_FMT_TYPE;
204
205 /* Go to next entry in ring. */
206 head++;
207 if (head >= pbm->msiq_ent_count)
208 head = 0;
209 ep = &base[head];
210 }
211 189
212 if (likely(head != orig_head)) { 190 *msi = msi_num = ((ep->word0 & MSIQ_WORD0_DATA0) >>
213 /* ACK entries by updating head pointer. */ 191 MSIQ_WORD0_DATA0_SHIFT);
214 fire_write(pbm->pbm_regs +
215 EVENT_QUEUE_HEAD(msiqid),
216 head);
217 }
218 return;
219 192
220bad_type: 193 fire_write(pbm->pbm_regs + MSI_CLEAR(msi_num),
221 printk(KERN_EMERG "MSI: Entry has bad type %lx\n", type); 194 MSI_CLEAR_EQWR_N);
222 return; 195
196 /* Clear the entry. */
197 ep->word0 &= ~MSIQ_WORD0_FMT_TYPE;
198
199 /* Go to next entry in ring. */
200 (*head)++;
201 if (*head >= pbm->msiq_ent_count)
202 *head = 0;
203
204 return 1;
223} 205}
224 206
225static int msi_bitmap_alloc(struct pci_pbm_info *pbm) 207static int pci_fire_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
208 unsigned long head)
226{ 209{
227 unsigned long size, bits_per_ulong; 210 fire_write(pbm->pbm_regs + EVENT_QUEUE_HEAD(msiqid), head);
211 return 0;
212}
228 213
229 bits_per_ulong = sizeof(unsigned long) * 8; 214static int pci_fire_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
230 size = (pbm->msi_num + (bits_per_ulong - 1)) & ~(bits_per_ulong - 1); 215 unsigned long msi, int is_msi64)
231 size /= 8; 216{
232 BUG_ON(size % sizeof(unsigned long)); 217 u64 val;
233 218
234 pbm->msi_bitmap = kzalloc(size, GFP_KERNEL); 219 val = fire_read(pbm->pbm_regs + MSI_MAP(msi));
235 if (!pbm->msi_bitmap) 220 val &= ~(MSI_MAP_EQNUM);
236 return -ENOMEM; 221 val |= msiqid;
222 fire_write(pbm->pbm_regs + MSI_MAP(msi), val);
223
224 fire_write(pbm->pbm_regs + MSI_CLEAR(msi),
225 MSI_CLEAR_EQWR_N);
226
227 val = fire_read(pbm->pbm_regs + MSI_MAP(msi));
228 val |= MSI_MAP_VALID;
229 fire_write(pbm->pbm_regs + MSI_MAP(msi), val);
237 230
238 return 0; 231 return 0;
239} 232}
240 233
241static void msi_bitmap_free(struct pci_pbm_info *pbm) 234static int pci_fire_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
242{ 235{
243 kfree(pbm->msi_bitmap); 236 unsigned long msiqid;
244 pbm->msi_bitmap = NULL; 237 u64 val;
238
239 val = fire_read(pbm->pbm_regs + MSI_MAP(msi));
240 msiqid = (val & MSI_MAP_EQNUM);
241
242 val &= ~MSI_MAP_VALID;
243
244 fire_write(pbm->pbm_regs + MSI_MAP(msi), val);
245
246 return 0;
245} 247}
246 248
247static int msi_queue_alloc(struct pci_pbm_info *pbm) 249static int pci_fire_msiq_alloc(struct pci_pbm_info *pbm)
248{ 250{
249 unsigned long pages, order, i; 251 unsigned long pages, order, i;
250 252
@@ -279,241 +281,65 @@ static int msi_queue_alloc(struct pci_pbm_info *pbm)
279 return 0; 281 return 0;
280} 282}
281 283
282static int alloc_msi(struct pci_pbm_info *pbm) 284static void pci_fire_msiq_free(struct pci_pbm_info *pbm)
283{ 285{
284 int i; 286 unsigned long pages, order;
285 287
286 for (i = 0; i < pbm->msi_num; i++) { 288 order = get_order(512 * 1024);
287 if (!test_and_set_bit(i, pbm->msi_bitmap)) 289 pages = (unsigned long) pbm->msi_queues;
288 return i + pbm->msi_first;
289 }
290 290
291 return -ENOENT; 291 free_pages(pages, order);
292}
293 292
294static void free_msi(struct pci_pbm_info *pbm, int msi_num) 293 pbm->msi_queues = NULL;
295{
296 msi_num -= pbm->msi_first;
297 clear_bit(msi_num, pbm->msi_bitmap);
298} 294}
299 295
300static int pci_setup_msi_irq(unsigned int *virt_irq_p, 296static int pci_fire_msiq_build_irq(struct pci_pbm_info *pbm,
301 struct pci_dev *pdev, 297 unsigned long msiqid,
302 struct msi_desc *entry) 298 unsigned long devino)
303{ 299{
304 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller; 300 unsigned long cregs = (unsigned long) pbm->pbm_regs;
305 unsigned long devino, msiqid, cregs, imap_off; 301 unsigned long imap_reg, iclr_reg, int_ctrlr;
306 struct msi_msg msg; 302 unsigned int virt_irq;
307 int msi_num, err; 303 int fixup;
308 u64 val; 304 u64 val;
309 305
310 *virt_irq_p = 0; 306 imap_reg = cregs + (0x001000UL + (devino * 0x08UL));
311 307 iclr_reg = cregs + (0x001400UL + (devino * 0x08UL));
312 msi_num = alloc_msi(pbm);
313 if (msi_num < 0)
314 return msi_num;
315
316 cregs = (unsigned long) pbm->pbm_regs;
317 308
318 err = sun4u_build_msi(pbm->portid, virt_irq_p, 309 /* XXX iterate amongst the 4 IRQ controllers XXX */
319 pbm->msiq_first_devino, 310 int_ctrlr = (1UL << 6);
320 (pbm->msiq_first_devino +
321 pbm->msiq_num),
322 cregs + 0x001000UL,
323 cregs + 0x001400UL);
324 if (err < 0)
325 goto out_err;
326 devino = err;
327 311
328 imap_off = 0x001000UL + (devino * 0x8UL); 312 val = fire_read(imap_reg);
313 val |= (1UL << 63) | int_ctrlr;
314 fire_write(imap_reg, val);
329 315
330 val = fire_read(pbm->pbm_regs + imap_off); 316 fixup = ((pbm->portid << 6) | devino) - int_ctrlr;
331 val |= (1UL << 63) | (1UL << 6);
332 fire_write(pbm->pbm_regs + imap_off, val);
333 317
334 msiqid = ((devino - pbm->msiq_first_devino) + 318 virt_irq = build_irq(fixup, iclr_reg, imap_reg);
335 pbm->msiq_first); 319 if (!virt_irq)
320 return -ENOMEM;
336 321
337 fire_write(pbm->pbm_regs + 322 fire_write(pbm->pbm_regs +
338 EVENT_QUEUE_CONTROL_SET(msiqid), 323 EVENT_QUEUE_CONTROL_SET(msiqid),
339 EVENT_QUEUE_CONTROL_SET_EN); 324 EVENT_QUEUE_CONTROL_SET_EN);
340 325
341 val = fire_read(pbm->pbm_regs + MSI_MAP(msi_num)); 326 return virt_irq;
342 val &= ~(MSI_MAP_EQNUM);
343 val |= msiqid;
344 fire_write(pbm->pbm_regs + MSI_MAP(msi_num), val);
345
346 fire_write(pbm->pbm_regs + MSI_CLEAR(msi_num),
347 MSI_CLEAR_EQWR_N);
348
349 val = fire_read(pbm->pbm_regs + MSI_MAP(msi_num));
350 val |= MSI_MAP_VALID;
351 fire_write(pbm->pbm_regs + MSI_MAP(msi_num), val);
352
353 sparc64_set_msi(*virt_irq_p, msi_num);
354
355 if (entry->msi_attrib.is_64) {
356 msg.address_hi = pbm->msi64_start >> 32;
357 msg.address_lo = pbm->msi64_start & 0xffffffff;
358 } else {
359 msg.address_hi = 0;
360 msg.address_lo = pbm->msi32_start;
361 }
362 msg.data = msi_num;
363
364 set_irq_msi(*virt_irq_p, entry);
365 write_msi_msg(*virt_irq_p, &msg);
366
367 irq_install_pre_handler(*virt_irq_p,
368 pci_msi_prehandler,
369 pbm, (void *) msiqid);
370
371 return 0;
372
373out_err:
374 free_msi(pbm, msi_num);
375 return err;
376} 327}
377 328
378static void pci_teardown_msi_irq(unsigned int virt_irq, 329static const struct sparc64_msiq_ops pci_fire_msiq_ops = {
379 struct pci_dev *pdev) 330 .get_head = pci_fire_get_head,
380{ 331 .dequeue_msi = pci_fire_dequeue_msi,
381 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller; 332 .set_head = pci_fire_set_head,
382 unsigned long msiqid, msi_num; 333 .msi_setup = pci_fire_msi_setup,
383 u64 val; 334 .msi_teardown = pci_fire_msi_teardown,
384 335 .msiq_alloc = pci_fire_msiq_alloc,
385 msi_num = sparc64_get_msi(virt_irq); 336 .msiq_free = pci_fire_msiq_free,
386 337 .msiq_build_irq = pci_fire_msiq_build_irq,
387 val = fire_read(pbm->pbm_regs + MSI_MAP(msi_num)); 338};
388
389 msiqid = (val & MSI_MAP_EQNUM);
390
391 val &= ~MSI_MAP_VALID;
392 fire_write(pbm->pbm_regs + MSI_MAP(msi_num), val);
393
394 fire_write(pbm->pbm_regs + EVENT_QUEUE_CONTROL_CLEAR(msiqid),
395 EVENT_QUEUE_CONTROL_CLEAR_DIS);
396
397 free_msi(pbm, msi_num);
398
399 /* The sun4u_destroy_msi() will liberate the devino and thus the MSIQ
400 * allocation.
401 */
402 sun4u_destroy_msi(virt_irq);
403}
404 339
405static void pci_fire_msi_init(struct pci_pbm_info *pbm) 340static void pci_fire_msi_init(struct pci_pbm_info *pbm)
406{ 341{
407 const u32 *val; 342 sparc64_pbm_msi_init(pbm, &pci_fire_msiq_ops);
408 int len;
409
410 val = of_get_property(pbm->prom_node, "#msi-eqs", &len);
411 if (!val || len != 4)
412 goto no_msi;
413 pbm->msiq_num = *val;
414 if (pbm->msiq_num) {
415 const struct msiq_prop {
416 u32 first_msiq;
417 u32 num_msiq;
418 u32 first_devino;
419 } *mqp;
420 const struct msi_range_prop {
421 u32 first_msi;
422 u32 num_msi;
423 } *mrng;
424 const struct addr_range_prop {
425 u32 msi32_high;
426 u32 msi32_low;
427 u32 msi32_len;
428 u32 msi64_high;
429 u32 msi64_low;
430 u32 msi64_len;
431 } *arng;
432
433 val = of_get_property(pbm->prom_node, "msi-eq-size", &len);
434 if (!val || len != 4)
435 goto no_msi;
436
437 pbm->msiq_ent_count = *val;
438
439 mqp = of_get_property(pbm->prom_node,
440 "msi-eq-to-devino", &len);
441 if (!mqp)
442 mqp = of_get_property(pbm->prom_node,
443 "msi-eq-devino", &len);
444 if (!mqp || len != sizeof(struct msiq_prop))
445 goto no_msi;
446
447 pbm->msiq_first = mqp->first_msiq;
448 pbm->msiq_first_devino = mqp->first_devino;
449
450 val = of_get_property(pbm->prom_node, "#msi", &len);
451 if (!val || len != 4)
452 goto no_msi;
453 pbm->msi_num = *val;
454
455 mrng = of_get_property(pbm->prom_node, "msi-ranges", &len);
456 if (!mrng || len != sizeof(struct msi_range_prop))
457 goto no_msi;
458 pbm->msi_first = mrng->first_msi;
459
460 val = of_get_property(pbm->prom_node, "msi-data-mask", &len);
461 if (!val || len != 4)
462 goto no_msi;
463 pbm->msi_data_mask = *val;
464
465 val = of_get_property(pbm->prom_node, "msix-data-width", &len);
466 if (!val || len != 4)
467 goto no_msi;
468 pbm->msix_data_width = *val;
469
470 arng = of_get_property(pbm->prom_node, "msi-address-ranges",
471 &len);
472 if (!arng || len != sizeof(struct addr_range_prop))
473 goto no_msi;
474 pbm->msi32_start = ((u64)arng->msi32_high << 32) |
475 (u64) arng->msi32_low;
476 pbm->msi64_start = ((u64)arng->msi64_high << 32) |
477 (u64) arng->msi64_low;
478 pbm->msi32_len = arng->msi32_len;
479 pbm->msi64_len = arng->msi64_len;
480
481 if (msi_bitmap_alloc(pbm))
482 goto no_msi;
483
484 if (msi_queue_alloc(pbm)) {
485 msi_bitmap_free(pbm);
486 goto no_msi;
487 }
488
489 printk(KERN_INFO "%s: MSI Queue first[%u] num[%u] count[%u] "
490 "devino[0x%x]\n",
491 pbm->name,
492 pbm->msiq_first, pbm->msiq_num,
493 pbm->msiq_ent_count,
494 pbm->msiq_first_devino);
495 printk(KERN_INFO "%s: MSI first[%u] num[%u] mask[0x%x] "
496 "width[%u]\n",
497 pbm->name,
498 pbm->msi_first, pbm->msi_num, pbm->msi_data_mask,
499 pbm->msix_data_width);
500 printk(KERN_INFO "%s: MSI addr32[0x%lx:0x%x] "
501 "addr64[0x%lx:0x%x]\n",
502 pbm->name,
503 pbm->msi32_start, pbm->msi32_len,
504 pbm->msi64_start, pbm->msi64_len);
505 printk(KERN_INFO "%s: MSI queues at RA [%016lx]\n",
506 pbm->name,
507 __pa(pbm->msi_queues));
508 }
509 pbm->setup_msi_irq = pci_setup_msi_irq;
510 pbm->teardown_msi_irq = pci_teardown_msi_irq;
511
512 return;
513
514no_msi:
515 pbm->msiq_num = 0;
516 printk(KERN_INFO "%s: No MSI support.\n", pbm->name);
517} 343}
518#else /* CONFIG_PCI_MSI */ 344#else /* CONFIG_PCI_MSI */
519static void pci_fire_msi_init(struct pci_pbm_info *pbm) 345static void pci_fire_msi_init(struct pci_pbm_info *pbm)
diff --git a/arch/sparc64/kernel/pci_impl.h b/arch/sparc64/kernel/pci_impl.h
index f660c2b685eb..ccbb188f2e61 100644
--- a/arch/sparc64/kernel/pci_impl.h
+++ b/arch/sparc64/kernel/pci_impl.h
@@ -29,6 +29,33 @@
29#define PCI_STC_FLUSHFLAG_SET(STC) \ 29#define PCI_STC_FLUSHFLAG_SET(STC) \
30 (*((STC)->strbuf_flushflag) != 0UL) 30 (*((STC)->strbuf_flushflag) != 0UL)
31 31
32#ifdef CONFIG_PCI_MSI
33struct pci_pbm_info;
34struct sparc64_msiq_ops {
35 int (*get_head)(struct pci_pbm_info *pbm, unsigned long msiqid,
36 unsigned long *head);
37 int (*dequeue_msi)(struct pci_pbm_info *pbm, unsigned long msiqid,
38 unsigned long *head, unsigned long *msi);
39 int (*set_head)(struct pci_pbm_info *pbm, unsigned long msiqid,
40 unsigned long head);
41 int (*msi_setup)(struct pci_pbm_info *pbm, unsigned long msiqid,
42 unsigned long msi, int is_msi64);
43 int (*msi_teardown)(struct pci_pbm_info *pbm, unsigned long msi);
44 int (*msiq_alloc)(struct pci_pbm_info *pbm);
45 void (*msiq_free)(struct pci_pbm_info *pbm);
46 int (*msiq_build_irq)(struct pci_pbm_info *pbm, unsigned long msiqid,
47 unsigned long devino);
48};
49
50extern void sparc64_pbm_msi_init(struct pci_pbm_info *pbm,
51 const struct sparc64_msiq_ops *ops);
52
53struct sparc64_msiq_cookie {
54 struct pci_pbm_info *pbm;
55 unsigned long msiqid;
56};
57#endif
58
32struct pci_controller_info; 59struct pci_controller_info;
33 60
34struct pci_pbm_info { 61struct pci_pbm_info {
@@ -90,6 +117,8 @@ struct pci_pbm_info {
90 u32 msiq_ent_count; 117 u32 msiq_ent_count;
91 u32 msiq_first; 118 u32 msiq_first;
92 u32 msiq_first_devino; 119 u32 msiq_first_devino;
120 u32 msiq_rotor;
121 struct sparc64_msiq_cookie *msiq_irq_cookies;
93 u32 msi_num; 122 u32 msi_num;
94 u32 msi_first; 123 u32 msi_first;
95 u32 msi_data_mask; 124 u32 msi_data_mask;
@@ -100,9 +129,11 @@ struct pci_pbm_info {
100 u32 msi64_len; 129 u32 msi64_len;
101 void *msi_queues; 130 void *msi_queues;
102 unsigned long *msi_bitmap; 131 unsigned long *msi_bitmap;
132 unsigned int *msi_irq_table;
103 int (*setup_msi_irq)(unsigned int *virt_irq_p, struct pci_dev *pdev, 133 int (*setup_msi_irq)(unsigned int *virt_irq_p, struct pci_dev *pdev,
104 struct msi_desc *entry); 134 struct msi_desc *entry);
105 void (*teardown_msi_irq)(unsigned int virt_irq, struct pci_dev *pdev); 135 void (*teardown_msi_irq)(unsigned int virt_irq, struct pci_dev *pdev);
136 const struct sparc64_msiq_ops *msi_ops;
106#endif /* !(CONFIG_PCI_MSI) */ 137#endif /* !(CONFIG_PCI_MSI) */
107 138
108 /* This PBM's streaming buffer. */ 139 /* This PBM's streaming buffer. */
diff --git a/arch/sparc64/kernel/pci_msi.c b/arch/sparc64/kernel/pci_msi.c
new file mode 100644
index 000000000000..0fa33b1f4274
--- /dev/null
+++ b/arch/sparc64/kernel/pci_msi.c
@@ -0,0 +1,433 @@
1/* pci_msi.c: Sparc64 MSI support common layer.
2 *
3 * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
4 */
5#include <linux/kernel.h>
6#include <linux/interrupt.h>
7#include <linux/irq.h>
8
9#include "pci_impl.h"
10
11static irqreturn_t sparc64_msiq_interrupt(int irq, void *cookie)
12{
13 struct sparc64_msiq_cookie *msiq_cookie = cookie;
14 struct pci_pbm_info *pbm = msiq_cookie->pbm;
15 unsigned long msiqid = msiq_cookie->msiqid;
16 const struct sparc64_msiq_ops *ops;
17 unsigned long orig_head, head;
18 int err;
19
20 ops = pbm->msi_ops;
21
22 err = ops->get_head(pbm, msiqid, &head);
23 if (unlikely(err < 0))
24 goto err_get_head;
25
26 orig_head = head;
27 for (;;) {
28 unsigned long msi;
29
30 err = ops->dequeue_msi(pbm, msiqid, &head, &msi);
31 if (likely(err > 0))
32 __do_IRQ(pbm->msi_irq_table[msi - pbm->msi_first]);
33
34 if (unlikely(err < 0))
35 goto err_dequeue;
36
37 if (err == 0)
38 break;
39 }
40 if (likely(head != orig_head)) {
41 err = ops->set_head(pbm, msiqid, head);
42 if (unlikely(err < 0))
43 goto err_set_head;
44 }
45 return IRQ_HANDLED;
46
47err_get_head:
48 printk(KERN_EMERG "MSI: Get head on msiqid[%lu] gives error %d\n",
49 msiqid, err);
50 goto err_out;
51
52err_dequeue:
53 printk(KERN_EMERG "MSI: Dequeue head[%lu] from msiqid[%lu] "
54 "gives error %d\n",
55 head, msiqid, err);
56 goto err_out;
57
58err_set_head:
59 printk(KERN_EMERG "MSI: Set head[%lu] on msiqid[%lu] "
60 "gives error %d\n",
61 head, msiqid, err);
62 goto err_out;
63
64err_out:
65 return IRQ_NONE;
66}
67
68static u32 pick_msiq(struct pci_pbm_info *pbm)
69{
70 static DEFINE_SPINLOCK(rotor_lock);
71 unsigned long flags;
72 u32 ret, rotor;
73
74 spin_lock_irqsave(&rotor_lock, flags);
75
76 rotor = pbm->msiq_rotor;
77 ret = pbm->msiq_first + rotor;
78
79 if (++rotor >= pbm->msiq_num)
80 rotor = 0;
81 pbm->msiq_rotor = rotor;
82
83 spin_unlock_irqrestore(&rotor_lock, flags);
84
85 return ret;
86}
87
88
89static int alloc_msi(struct pci_pbm_info *pbm)
90{
91 int i;
92
93 for (i = 0; i < pbm->msi_num; i++) {
94 if (!test_and_set_bit(i, pbm->msi_bitmap))
95 return i + pbm->msi_first;
96 }
97
98 return -ENOENT;
99}
100
101static void free_msi(struct pci_pbm_info *pbm, int msi_num)
102{
103 msi_num -= pbm->msi_first;
104 clear_bit(msi_num, pbm->msi_bitmap);
105}
106
107static struct irq_chip msi_irq = {
108 .typename = "PCI-MSI",
109 .mask = mask_msi_irq,
110 .unmask = unmask_msi_irq,
111 .enable = unmask_msi_irq,
112 .disable = mask_msi_irq,
113 /* XXX affinity XXX */
114};
115
116int sparc64_setup_msi_irq(unsigned int *virt_irq_p,
117 struct pci_dev *pdev,
118 struct msi_desc *entry)
119{
120 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
121 const struct sparc64_msiq_ops *ops = pbm->msi_ops;
122 struct msi_msg msg;
123 int msi, err;
124 u32 msiqid;
125
126 *virt_irq_p = virt_irq_alloc(~0);
127 err = -ENOMEM;
128 if (!*virt_irq_p)
129 goto out_err;
130
131 set_irq_chip(*virt_irq_p, &msi_irq);
132
133 err = alloc_msi(pbm);
134 if (unlikely(err < 0))
135 goto out_virt_irq_free;
136
137 msi = err;
138
139 msiqid = pick_msiq(pbm);
140
141 err = ops->msi_setup(pbm, msiqid, msi,
142 (entry->msi_attrib.is_64 ? 1 : 0));
143 if (err)
144 goto out_msi_free;
145
146 pbm->msi_irq_table[msi - pbm->msi_first] = *virt_irq_p;
147
148 if (entry->msi_attrib.is_64) {
149 msg.address_hi = pbm->msi64_start >> 32;
150 msg.address_lo = pbm->msi64_start & 0xffffffff;
151 } else {
152 msg.address_hi = 0;
153 msg.address_lo = pbm->msi32_start;
154 }
155 msg.data = msi;
156
157 set_irq_msi(*virt_irq_p, entry);
158 write_msi_msg(*virt_irq_p, &msg);
159
160 return 0;
161
162out_msi_free:
163 free_msi(pbm, msi);
164
165out_virt_irq_free:
166 set_irq_chip(*virt_irq_p, NULL);
167 virt_irq_free(*virt_irq_p);
168 *virt_irq_p = 0;
169
170out_err:
171 return err;
172}
173
174void sparc64_teardown_msi_irq(unsigned int virt_irq,
175 struct pci_dev *pdev)
176{
177 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
178 const struct sparc64_msiq_ops *ops = pbm->msi_ops;
179 unsigned int msi_num;
180 int i, err;
181
182 for (i = 0; i < pbm->msi_num; i++) {
183 if (pbm->msi_irq_table[i] == virt_irq)
184 break;
185 }
186 if (i >= pbm->msi_num) {
187 printk(KERN_ERR "%s: teardown: No MSI for irq %u\n",
188 pbm->name, virt_irq);
189 return;
190 }
191
192 msi_num = pbm->msi_first + i;
193 pbm->msi_irq_table[i] = ~0U;
194
195 err = ops->msi_teardown(pbm, msi_num);
196 if (err) {
197 printk(KERN_ERR "%s: teardown: ops->teardown() on MSI %u, "
198 "irq %u, gives error %d\n",
199 pbm->name, msi_num, virt_irq, err);
200 return;
201 }
202
203 free_msi(pbm, msi_num);
204
205 set_irq_chip(virt_irq, NULL);
206 virt_irq_free(virt_irq);
207}
208
209static int msi_bitmap_alloc(struct pci_pbm_info *pbm)
210{
211 unsigned long size, bits_per_ulong;
212
213 bits_per_ulong = sizeof(unsigned long) * 8;
214 size = (pbm->msi_num + (bits_per_ulong - 1)) & ~(bits_per_ulong - 1);
215 size /= 8;
216 BUG_ON(size % sizeof(unsigned long));
217
218 pbm->msi_bitmap = kzalloc(size, GFP_KERNEL);
219 if (!pbm->msi_bitmap)
220 return -ENOMEM;
221
222 return 0;
223}
224
225static void msi_bitmap_free(struct pci_pbm_info *pbm)
226{
227 kfree(pbm->msi_bitmap);
228 pbm->msi_bitmap = NULL;
229}
230
231static int msi_table_alloc(struct pci_pbm_info *pbm)
232{
233 int size, i;
234
235 size = pbm->msiq_num * sizeof(struct sparc64_msiq_cookie);
236 pbm->msiq_irq_cookies = kzalloc(size, GFP_KERNEL);
237 if (!pbm->msiq_irq_cookies)
238 return -ENOMEM;
239
240 for (i = 0; i < pbm->msiq_num; i++) {
241 struct sparc64_msiq_cookie *p;
242
243 p = &pbm->msiq_irq_cookies[i];
244 p->pbm = pbm;
245 p->msiqid = pbm->msiq_first + i;
246 }
247
248 size = pbm->msi_num * sizeof(unsigned int);
249 pbm->msi_irq_table = kzalloc(size, GFP_KERNEL);
250 if (!pbm->msi_irq_table) {
251 kfree(pbm->msiq_irq_cookies);
252 pbm->msiq_irq_cookies = NULL;
253 return -ENOMEM;
254 }
255
256 return 0;
257}
258
259static void msi_table_free(struct pci_pbm_info *pbm)
260{
261 kfree(pbm->msiq_irq_cookies);
262 pbm->msiq_irq_cookies = NULL;
263
264 kfree(pbm->msi_irq_table);
265 pbm->msi_irq_table = NULL;
266}
267
268static int bringup_one_msi_queue(struct pci_pbm_info *pbm,
269 const struct sparc64_msiq_ops *ops,
270 unsigned long msiqid,
271 unsigned long devino)
272{
273 int irq = ops->msiq_build_irq(pbm, msiqid, devino);
274 int err;
275
276 if (irq < 0)
277 return irq;
278
279 err = request_irq(irq, sparc64_msiq_interrupt, 0,
280 "MSIQ",
281 &pbm->msiq_irq_cookies[msiqid - pbm->msiq_first]);
282 if (err)
283 return err;
284
285 return 0;
286}
287
288static int sparc64_bringup_msi_queues(struct pci_pbm_info *pbm,
289 const struct sparc64_msiq_ops *ops)
290{
291 int i;
292
293 for (i = 0; i < pbm->msiq_num; i++) {
294 unsigned long msiqid = i + pbm->msiq_first;
295 unsigned long devino = i + pbm->msiq_first_devino;
296 int err;
297
298 err = bringup_one_msi_queue(pbm, ops, msiqid, devino);
299 if (err)
300 return err;
301 }
302
303 return 0;
304}
305
306void sparc64_pbm_msi_init(struct pci_pbm_info *pbm,
307 const struct sparc64_msiq_ops *ops)
308{
309 const u32 *val;
310 int len;
311
312 val = of_get_property(pbm->prom_node, "#msi-eqs", &len);
313 if (!val || len != 4)
314 goto no_msi;
315 pbm->msiq_num = *val;
316 if (pbm->msiq_num) {
317 const struct msiq_prop {
318 u32 first_msiq;
319 u32 num_msiq;
320 u32 first_devino;
321 } *mqp;
322 const struct msi_range_prop {
323 u32 first_msi;
324 u32 num_msi;
325 } *mrng;
326 const struct addr_range_prop {
327 u32 msi32_high;
328 u32 msi32_low;
329 u32 msi32_len;
330 u32 msi64_high;
331 u32 msi64_low;
332 u32 msi64_len;
333 } *arng;
334
335 val = of_get_property(pbm->prom_node, "msi-eq-size", &len);
336 if (!val || len != 4)
337 goto no_msi;
338
339 pbm->msiq_ent_count = *val;
340
341 mqp = of_get_property(pbm->prom_node,
342 "msi-eq-to-devino", &len);
343 if (!mqp)
344 mqp = of_get_property(pbm->prom_node,
345 "msi-eq-devino", &len);
346 if (!mqp || len != sizeof(struct msiq_prop))
347 goto no_msi;
348
349 pbm->msiq_first = mqp->first_msiq;
350 pbm->msiq_first_devino = mqp->first_devino;
351
352 val = of_get_property(pbm->prom_node, "#msi", &len);
353 if (!val || len != 4)
354 goto no_msi;
355 pbm->msi_num = *val;
356
357 mrng = of_get_property(pbm->prom_node, "msi-ranges", &len);
358 if (!mrng || len != sizeof(struct msi_range_prop))
359 goto no_msi;
360 pbm->msi_first = mrng->first_msi;
361
362 val = of_get_property(pbm->prom_node, "msi-data-mask", &len);
363 if (!val || len != 4)
364 goto no_msi;
365 pbm->msi_data_mask = *val;
366
367 val = of_get_property(pbm->prom_node, "msix-data-width", &len);
368 if (!val || len != 4)
369 goto no_msi;
370 pbm->msix_data_width = *val;
371
372 arng = of_get_property(pbm->prom_node, "msi-address-ranges",
373 &len);
374 if (!arng || len != sizeof(struct addr_range_prop))
375 goto no_msi;
376 pbm->msi32_start = ((u64)arng->msi32_high << 32) |
377 (u64) arng->msi32_low;
378 pbm->msi64_start = ((u64)arng->msi64_high << 32) |
379 (u64) arng->msi64_low;
380 pbm->msi32_len = arng->msi32_len;
381 pbm->msi64_len = arng->msi64_len;
382
383 if (msi_bitmap_alloc(pbm))
384 goto no_msi;
385
386 if (msi_table_alloc(pbm)) {
387 msi_bitmap_free(pbm);
388 goto no_msi;
389 }
390
391 if (ops->msiq_alloc(pbm)) {
392 msi_table_free(pbm);
393 msi_bitmap_free(pbm);
394 goto no_msi;
395 }
396
397 if (sparc64_bringup_msi_queues(pbm, ops)) {
398 ops->msiq_free(pbm);
399 msi_table_free(pbm);
400 msi_bitmap_free(pbm);
401 goto no_msi;
402 }
403
404 printk(KERN_INFO "%s: MSI Queue first[%u] num[%u] count[%u] "
405 "devino[0x%x]\n",
406 pbm->name,
407 pbm->msiq_first, pbm->msiq_num,
408 pbm->msiq_ent_count,
409 pbm->msiq_first_devino);
410 printk(KERN_INFO "%s: MSI first[%u] num[%u] mask[0x%x] "
411 "width[%u]\n",
412 pbm->name,
413 pbm->msi_first, pbm->msi_num, pbm->msi_data_mask,
414 pbm->msix_data_width);
415 printk(KERN_INFO "%s: MSI addr32[0x%lx:0x%x] "
416 "addr64[0x%lx:0x%x]\n",
417 pbm->name,
418 pbm->msi32_start, pbm->msi32_len,
419 pbm->msi64_start, pbm->msi64_len);
420 printk(KERN_INFO "%s: MSI queues at RA [%016lx]\n",
421 pbm->name,
422 __pa(pbm->msi_queues));
423
424 pbm->msi_ops = ops;
425 pbm->setup_msi_irq = sparc64_setup_msi_irq;
426 pbm->teardown_msi_irq = sparc64_teardown_msi_irq;
427 }
428 return;
429
430no_msi:
431 pbm->msiq_num = 0;
432 printk(KERN_INFO "%s: No MSI support.\n", pbm->name);
433}
diff --git a/arch/sparc64/kernel/pci_sun4v.c b/arch/sparc64/kernel/pci_sun4v.c
index da724b13e89e..97c45b26b780 100644
--- a/arch/sparc64/kernel/pci_sun4v.c
+++ b/arch/sparc64/kernel/pci_sun4v.c
@@ -748,111 +748,102 @@ struct pci_sun4v_msiq_entry {
748 u64 reserved2; 748 u64 reserved2;
749}; 749};
750 750
751/* For now this just runs as a pre-handler for the real interrupt handler. 751static int pci_sun4v_get_head(struct pci_pbm_info *pbm, unsigned long msiqid,
752 * So we just walk through the queue and ACK all the entries, update the 752 unsigned long *head)
753 * head pointer, and return.
754 *
755 * In the longer term it would be nice to do something more integrated
756 * wherein we can pass in some of this MSI info to the drivers. This
757 * would be most useful for PCIe fabric error messages, although we could
758 * invoke those directly from the loop here in order to pass the info around.
759 */
760static void pci_sun4v_msi_prehandler(unsigned int ino, void *data1, void *data2)
761{ 753{
762 struct pci_pbm_info *pbm = data1; 754 unsigned long err, limit;
763 struct pci_sun4v_msiq_entry *base, *ep;
764 unsigned long msiqid, orig_head, head, type, err;
765
766 msiqid = (unsigned long) data2;
767 755
768 head = 0xdeadbeef; 756 err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, head);
769 err = pci_sun4v_msiq_gethead(pbm->devhandle, msiqid, &head);
770 if (unlikely(err)) 757 if (unlikely(err))
771 goto hv_error_get; 758 return -ENXIO;
772
773 if (unlikely(head >= (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry))))
774 goto bad_offset;
775
776 head /= sizeof(struct pci_sun4v_msiq_entry);
777 orig_head = head;
778 base = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
779 (pbm->msiq_ent_count *
780 sizeof(struct pci_sun4v_msiq_entry))));
781 ep = &base[head];
782 while ((ep->version_type & MSIQ_TYPE_MASK) != 0) {
783 type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
784 if (unlikely(type != MSIQ_TYPE_MSI32 &&
785 type != MSIQ_TYPE_MSI64))
786 goto bad_type;
787
788 pci_sun4v_msi_setstate(pbm->devhandle,
789 ep->msi_data /* msi_num */,
790 HV_MSISTATE_IDLE);
791
792 /* Clear the entry. */
793 ep->version_type &= ~MSIQ_TYPE_MASK;
794
795 /* Go to next entry in ring. */
796 head++;
797 if (head >= pbm->msiq_ent_count)
798 head = 0;
799 ep = &base[head];
800 }
801 759
802 if (likely(head != orig_head)) { 760 limit = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
803 /* ACK entries by updating head pointer. */ 761 if (unlikely(*head >= limit))
804 head *= sizeof(struct pci_sun4v_msiq_entry); 762 return -EFBIG;
805 err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
806 if (unlikely(err))
807 goto hv_error_set;
808 }
809 return;
810 763
811hv_error_set: 764 return 0;
812 printk(KERN_EMERG "MSI: Hypervisor set head gives error %lu\n", err); 765}
813 goto hv_error_cont;
814 766
815hv_error_get: 767static int pci_sun4v_dequeue_msi(struct pci_pbm_info *pbm,
816 printk(KERN_EMERG "MSI: Hypervisor get head gives error %lu\n", err); 768 unsigned long msiqid, unsigned long *head,
769 unsigned long *msi)
770{
771 struct pci_sun4v_msiq_entry *ep;
772 unsigned long err, type;
817 773
818hv_error_cont: 774 /* Note: void pointer arithmetic, 'head' is a byte offset */
819 printk(KERN_EMERG "MSI: devhandle[%x] msiqid[%lx] head[%lu]\n", 775 ep = (pbm->msi_queues + ((msiqid - pbm->msiq_first) *
820 pbm->devhandle, msiqid, head); 776 (pbm->msiq_ent_count *
821 return; 777 sizeof(struct pci_sun4v_msiq_entry))) +
778 *head);
822 779
823bad_offset: 780 if ((ep->version_type & MSIQ_TYPE_MASK) == 0)
824 printk(KERN_EMERG "MSI: Hypervisor gives bad offset %lx max(%lx)\n", 781 return 0;
825 head, pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry));
826 return;
827 782
828bad_type: 783 type = (ep->version_type & MSIQ_TYPE_MASK) >> MSIQ_TYPE_SHIFT;
829 printk(KERN_EMERG "MSI: Entry has bad type %lx\n", type); 784 if (unlikely(type != MSIQ_TYPE_MSI32 &&
830 return; 785 type != MSIQ_TYPE_MSI64))
786 return -EINVAL;
787
788 *msi = ep->msi_data;
789
790 err = pci_sun4v_msi_setstate(pbm->devhandle,
791 ep->msi_data /* msi_num */,
792 HV_MSISTATE_IDLE);
793 if (unlikely(err))
794 return -ENXIO;
795
796 /* Clear the entry. */
797 ep->version_type &= ~MSIQ_TYPE_MASK;
798
799 (*head) += sizeof(struct pci_sun4v_msiq_entry);
800 if (*head >=
801 (pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry)))
802 *head = 0;
803
804 return 1;
831} 805}
832 806
833static int msi_bitmap_alloc(struct pci_pbm_info *pbm) 807static int pci_sun4v_set_head(struct pci_pbm_info *pbm, unsigned long msiqid,
808 unsigned long head)
834{ 809{
835 unsigned long size, bits_per_ulong; 810 unsigned long err;
836 811
837 bits_per_ulong = sizeof(unsigned long) * 8; 812 err = pci_sun4v_msiq_sethead(pbm->devhandle, msiqid, head);
838 size = (pbm->msi_num + (bits_per_ulong - 1)) & ~(bits_per_ulong - 1); 813 if (unlikely(err))
839 size /= 8; 814 return -EINVAL;
840 BUG_ON(size % sizeof(unsigned long));
841 815
842 pbm->msi_bitmap = kzalloc(size, GFP_KERNEL); 816 return 0;
843 if (!pbm->msi_bitmap) 817}
844 return -ENOMEM;
845 818
819static int pci_sun4v_msi_setup(struct pci_pbm_info *pbm, unsigned long msiqid,
820 unsigned long msi, int is_msi64)
821{
822 if (pci_sun4v_msi_setmsiq(pbm->devhandle, msi, msiqid,
823 (is_msi64 ?
824 HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
825 return -ENXIO;
826 if (pci_sun4v_msi_setstate(pbm->devhandle, msi, HV_MSISTATE_IDLE))
827 return -ENXIO;
828 if (pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_VALID))
829 return -ENXIO;
846 return 0; 830 return 0;
847} 831}
848 832
849static void msi_bitmap_free(struct pci_pbm_info *pbm) 833static int pci_sun4v_msi_teardown(struct pci_pbm_info *pbm, unsigned long msi)
850{ 834{
851 kfree(pbm->msi_bitmap); 835 unsigned long err, msiqid;
852 pbm->msi_bitmap = NULL; 836
837 err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi, &msiqid);
838 if (err)
839 return -ENXIO;
840
841 pci_sun4v_msi_setvalid(pbm->devhandle, msi, HV_MSIVALID_INVALID);
842
843 return 0;
853} 844}
854 845
855static int msi_queue_alloc(struct pci_pbm_info *pbm) 846static int pci_sun4v_msiq_alloc(struct pci_pbm_info *pbm)
856{ 847{
857 unsigned long q_size, alloc_size, pages, order; 848 unsigned long q_size, alloc_size, pages, order;
858 int i; 849 int i;
@@ -906,232 +897,59 @@ h_error:
906 return -EINVAL; 897 return -EINVAL;
907} 898}
908 899
909 900static void pci_sun4v_msiq_free(struct pci_pbm_info *pbm)
910static int alloc_msi(struct pci_pbm_info *pbm)
911{ 901{
902 unsigned long q_size, alloc_size, pages, order;
912 int i; 903 int i;
913 904
914 for (i = 0; i < pbm->msi_num; i++) { 905 for (i = 0; i < pbm->msiq_num; i++) {
915 if (!test_and_set_bit(i, pbm->msi_bitmap)) 906 unsigned long msiqid = pbm->msiq_first + i;
916 return i + pbm->msi_first;
917 }
918
919 return -ENOENT;
920}
921
922static void free_msi(struct pci_pbm_info *pbm, int msi_num)
923{
924 msi_num -= pbm->msi_first;
925 clear_bit(msi_num, pbm->msi_bitmap);
926}
927
928static int pci_sun4v_setup_msi_irq(unsigned int *virt_irq_p,
929 struct pci_dev *pdev,
930 struct msi_desc *entry)
931{
932 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
933 unsigned long devino, msiqid;
934 struct msi_msg msg;
935 int msi_num, err;
936
937 *virt_irq_p = 0;
938
939 msi_num = alloc_msi(pbm);
940 if (msi_num < 0)
941 return msi_num;
942
943 err = sun4v_build_msi(pbm->devhandle, virt_irq_p,
944 pbm->msiq_first_devino,
945 (pbm->msiq_first_devino +
946 pbm->msiq_num));
947 if (err < 0)
948 goto out_err;
949 devino = err;
950
951 msiqid = ((devino - pbm->msiq_first_devino) +
952 pbm->msiq_first);
953
954 err = -EINVAL;
955 if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
956 if (err)
957 goto out_err;
958
959 if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
960 goto out_err;
961
962 if (pci_sun4v_msi_setmsiq(pbm->devhandle,
963 msi_num, msiqid,
964 (entry->msi_attrib.is_64 ?
965 HV_MSITYPE_MSI64 : HV_MSITYPE_MSI32)))
966 goto out_err;
967
968 if (pci_sun4v_msi_setstate(pbm->devhandle, msi_num, HV_MSISTATE_IDLE))
969 goto out_err;
970
971 if (pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_VALID))
972 goto out_err;
973
974 sparc64_set_msi(*virt_irq_p, msi_num);
975 907
976 if (entry->msi_attrib.is_64) { 908 (void) pci_sun4v_msiq_conf(pbm->devhandle, msiqid, 0UL, 0);
977 msg.address_hi = pbm->msi64_start >> 32;
978 msg.address_lo = pbm->msi64_start & 0xffffffff;
979 } else {
980 msg.address_hi = 0;
981 msg.address_lo = pbm->msi32_start;
982 } 909 }
983 msg.data = msi_num;
984 910
985 set_irq_msi(*virt_irq_p, entry); 911 q_size = pbm->msiq_ent_count * sizeof(struct pci_sun4v_msiq_entry);
986 write_msi_msg(*virt_irq_p, &msg); 912 alloc_size = (pbm->msiq_num * q_size);
987 913 order = get_order(alloc_size);
988 irq_install_pre_handler(*virt_irq_p,
989 pci_sun4v_msi_prehandler,
990 pbm, (void *) msiqid);
991 914
992 return 0; 915 pages = (unsigned long) pbm->msi_queues;
993 916
994out_err: 917 free_pages(pages, order);
995 free_msi(pbm, msi_num);
996 return err;
997 918
919 pbm->msi_queues = NULL;
998} 920}
999 921
1000static void pci_sun4v_teardown_msi_irq(unsigned int virt_irq, 922static int pci_sun4v_msiq_build_irq(struct pci_pbm_info *pbm,
1001 struct pci_dev *pdev) 923 unsigned long msiqid,
924 unsigned long devino)
1002{ 925{
1003 struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller; 926 unsigned int virt_irq = sun4v_build_irq(pbm->devhandle, devino);
1004 unsigned long msiqid, err;
1005 unsigned int msi_num;
1006
1007 msi_num = sparc64_get_msi(virt_irq);
1008 err = pci_sun4v_msi_getmsiq(pbm->devhandle, msi_num, &msiqid);
1009 if (err) {
1010 printk(KERN_ERR "%s: getmsiq gives error %lu\n",
1011 pbm->name, err);
1012 return;
1013 }
1014 927
1015 pci_sun4v_msi_setvalid(pbm->devhandle, msi_num, HV_MSIVALID_INVALID); 928 if (!virt_irq)
1016 pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_INVALID); 929 return -ENOMEM;
1017 930
1018 free_msi(pbm, msi_num); 931 if (pci_sun4v_msiq_setstate(pbm->devhandle, msiqid, HV_MSIQSTATE_IDLE))
932 return -EINVAL;
933 if (pci_sun4v_msiq_setvalid(pbm->devhandle, msiqid, HV_MSIQ_VALID))
934 return -EINVAL;
1019 935
1020 /* The sun4v_destroy_msi() will liberate the devino and thus the MSIQ 936 return virt_irq;
1021 * allocation.
1022 */
1023 sun4v_destroy_msi(virt_irq);
1024} 937}
1025 938
939static const struct sparc64_msiq_ops pci_sun4v_msiq_ops = {
940 .get_head = pci_sun4v_get_head,
941 .dequeue_msi = pci_sun4v_dequeue_msi,
942 .set_head = pci_sun4v_set_head,
943 .msi_setup = pci_sun4v_msi_setup,
944 .msi_teardown = pci_sun4v_msi_teardown,
945 .msiq_alloc = pci_sun4v_msiq_alloc,
946 .msiq_free = pci_sun4v_msiq_free,
947 .msiq_build_irq = pci_sun4v_msiq_build_irq,
948};
949
1026static void pci_sun4v_msi_init(struct pci_pbm_info *pbm) 950static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)
1027{ 951{
1028 const u32 *val; 952 sparc64_pbm_msi_init(pbm, &pci_sun4v_msiq_ops);
1029 int len;
1030
1031 val = of_get_property(pbm->prom_node, "#msi-eqs", &len);
1032 if (!val || len != 4)
1033 goto no_msi;
1034 pbm->msiq_num = *val;
1035 if (pbm->msiq_num) {
1036 const struct msiq_prop {
1037 u32 first_msiq;
1038 u32 num_msiq;
1039 u32 first_devino;
1040 } *mqp;
1041 const struct msi_range_prop {
1042 u32 first_msi;
1043 u32 num_msi;
1044 } *mrng;
1045 const struct addr_range_prop {
1046 u32 msi32_high;
1047 u32 msi32_low;
1048 u32 msi32_len;
1049 u32 msi64_high;
1050 u32 msi64_low;
1051 u32 msi64_len;
1052 } *arng;
1053
1054 val = of_get_property(pbm->prom_node, "msi-eq-size", &len);
1055 if (!val || len != 4)
1056 goto no_msi;
1057
1058 pbm->msiq_ent_count = *val;
1059
1060 mqp = of_get_property(pbm->prom_node,
1061 "msi-eq-to-devino", &len);
1062 if (!mqp || len != sizeof(struct msiq_prop))
1063 goto no_msi;
1064
1065 pbm->msiq_first = mqp->first_msiq;
1066 pbm->msiq_first_devino = mqp->first_devino;
1067
1068 val = of_get_property(pbm->prom_node, "#msi", &len);
1069 if (!val || len != 4)
1070 goto no_msi;
1071 pbm->msi_num = *val;
1072
1073 mrng = of_get_property(pbm->prom_node, "msi-ranges", &len);
1074 if (!mrng || len != sizeof(struct msi_range_prop))
1075 goto no_msi;
1076 pbm->msi_first = mrng->first_msi;
1077
1078 val = of_get_property(pbm->prom_node, "msi-data-mask", &len);
1079 if (!val || len != 4)
1080 goto no_msi;
1081 pbm->msi_data_mask = *val;
1082
1083 val = of_get_property(pbm->prom_node, "msix-data-width", &len);
1084 if (!val || len != 4)
1085 goto no_msi;
1086 pbm->msix_data_width = *val;
1087
1088 arng = of_get_property(pbm->prom_node, "msi-address-ranges",
1089 &len);
1090 if (!arng || len != sizeof(struct addr_range_prop))
1091 goto no_msi;
1092 pbm->msi32_start = ((u64)arng->msi32_high << 32) |
1093 (u64) arng->msi32_low;
1094 pbm->msi64_start = ((u64)arng->msi64_high << 32) |
1095 (u64) arng->msi64_low;
1096 pbm->msi32_len = arng->msi32_len;
1097 pbm->msi64_len = arng->msi64_len;
1098
1099 if (msi_bitmap_alloc(pbm))
1100 goto no_msi;
1101
1102 if (msi_queue_alloc(pbm)) {
1103 msi_bitmap_free(pbm);
1104 goto no_msi;
1105 }
1106
1107 printk(KERN_INFO "%s: MSI Queue first[%u] num[%u] count[%u] "
1108 "devino[0x%x]\n",
1109 pbm->name,
1110 pbm->msiq_first, pbm->msiq_num,
1111 pbm->msiq_ent_count,
1112 pbm->msiq_first_devino);
1113 printk(KERN_INFO "%s: MSI first[%u] num[%u] mask[0x%x] "
1114 "width[%u]\n",
1115 pbm->name,
1116 pbm->msi_first, pbm->msi_num, pbm->msi_data_mask,
1117 pbm->msix_data_width);
1118 printk(KERN_INFO "%s: MSI addr32[0x%lx:0x%x] "
1119 "addr64[0x%lx:0x%x]\n",
1120 pbm->name,
1121 pbm->msi32_start, pbm->msi32_len,
1122 pbm->msi64_start, pbm->msi64_len);
1123 printk(KERN_INFO "%s: MSI queues at RA [%p]\n",
1124 pbm->name,
1125 pbm->msi_queues);
1126 }
1127 pbm->setup_msi_irq = pci_sun4v_setup_msi_irq;
1128 pbm->teardown_msi_irq = pci_sun4v_teardown_msi_irq;
1129
1130 return;
1131
1132no_msi:
1133 pbm->msiq_num = 0;
1134 printk(KERN_INFO "%s: No MSI support.\n", pbm->name);
1135} 953}
1136#else /* CONFIG_PCI_MSI */ 954#else /* CONFIG_PCI_MSI */
1137static void pci_sun4v_msi_init(struct pci_pbm_info *pbm) 955static void pci_sun4v_msi_init(struct pci_pbm_info *pbm)