aboutsummaryrefslogtreecommitdiffstats
path: root/arch/ia64/sn
diff options
context:
space:
mode:
authorLen Brown <len.brown@intel.com>2005-09-03 02:44:09 -0400
committerLen Brown <len.brown@intel.com>2005-09-03 02:44:09 -0400
commit129521dcc94f781890f8f668219ab79f0073ff9f (patch)
tree9f70707c88da65577f38814fe37b24c4b4957d64 /arch/ia64/sn
parent824b558bbe2c298b165cdb54c33718994dda30bb (diff)
parentf505380ba7b98ec97bf25300c2a58aeae903530b (diff)
Merge linux-2.6 into linux-acpi-2.6 test
Diffstat (limited to 'arch/ia64/sn')
-rw-r--r--arch/ia64/sn/include/tio.h6
-rw-r--r--arch/ia64/sn/include/xtalk/hubdev.h11
-rw-r--r--arch/ia64/sn/kernel/bte.c83
-rw-r--r--arch/ia64/sn/kernel/huberror.c2
-rw-r--r--arch/ia64/sn/kernel/io_init.c35
-rw-r--r--arch/ia64/sn/kernel/irq.c75
-rw-r--r--arch/ia64/sn/kernel/setup.c7
-rw-r--r--arch/ia64/sn/kernel/sn2/ptc_deadlock.S13
-rw-r--r--arch/ia64/sn/kernel/sn2/sn2_smp.c256
-rw-r--r--arch/ia64/sn/kernel/sn2/sn_hwperf.c313
-rw-r--r--arch/ia64/sn/kernel/sn2/sn_proc_fs.c4
-rw-r--r--arch/ia64/sn/kernel/sn2/timer_interrupt.c22
-rw-r--r--arch/ia64/sn/pci/Makefile2
-rw-r--r--arch/ia64/sn/pci/pcibr/pcibr_dma.c60
-rw-r--r--arch/ia64/sn/pci/pcibr/pcibr_provider.c40
-rw-r--r--arch/ia64/sn/pci/tioca_provider.c7
-rw-r--r--arch/ia64/sn/pci/tioce_provider.c771
17 files changed, 1494 insertions, 213 deletions
diff --git a/arch/ia64/sn/include/tio.h b/arch/ia64/sn/include/tio.h
index 0139124dd54a..6b2e7b75eb19 100644
--- a/arch/ia64/sn/include/tio.h
+++ b/arch/ia64/sn/include/tio.h
@@ -3,7 +3,7 @@
3 * License. See the file "COPYING" in the main directory of this archive 3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details. 4 * for more details.
5 * 5 *
6 * Copyright (C) 2000-2004 Silicon Graphics, Inc. All rights reserved. 6 * Copyright (C) 2000-2005 Silicon Graphics, Inc. All rights reserved.
7 */ 7 */
8 8
9#ifndef _ASM_IA64_SN_TIO_H 9#ifndef _ASM_IA64_SN_TIO_H
@@ -26,6 +26,10 @@
26#define TIO_ITTE_VALID_MASK 0x1 26#define TIO_ITTE_VALID_MASK 0x1
27#define TIO_ITTE_VALID_SHIFT 16 27#define TIO_ITTE_VALID_SHIFT 16
28 28
29#define TIO_ITTE_WIDGET(itte) \
30 (((itte) >> TIO_ITTE_WIDGET_SHIFT) & TIO_ITTE_WIDGET_MASK)
31#define TIO_ITTE_VALID(itte) \
32 (((itte) >> TIO_ITTE_VALID_SHIFT) & TIO_ITTE_VALID_MASK)
29 33
30#define TIO_ITTE_PUT(nasid, bigwin, widget, addr, valid) \ 34#define TIO_ITTE_PUT(nasid, bigwin, widget, addr, valid) \
31 REMOTE_HUB_S((nasid), TIO_ITTE(bigwin), \ 35 REMOTE_HUB_S((nasid), TIO_ITTE(bigwin), \
diff --git a/arch/ia64/sn/include/xtalk/hubdev.h b/arch/ia64/sn/include/xtalk/hubdev.h
index 580a1c0403a7..71c2b271b4c6 100644
--- a/arch/ia64/sn/include/xtalk/hubdev.h
+++ b/arch/ia64/sn/include/xtalk/hubdev.h
@@ -3,7 +3,7 @@
3 * License. See the file "COPYING" in the main directory of this archive 3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details. 4 * for more details.
5 * 5 *
6 * Copyright (C) 1992 - 1997, 2000-2004 Silicon Graphics, Inc. All rights reserved. 6 * Copyright (C) 1992 - 1997, 2000-2005 Silicon Graphics, Inc. All rights reserved.
7 */ 7 */
8#ifndef _ASM_IA64_SN_XTALK_HUBDEV_H 8#ifndef _ASM_IA64_SN_XTALK_HUBDEV_H
9#define _ASM_IA64_SN_XTALK_HUBDEV_H 9#define _ASM_IA64_SN_XTALK_HUBDEV_H
@@ -16,6 +16,9 @@
16#define IIO_ITTE_WIDGET_MASK ((1<<IIO_ITTE_WIDGET_BITS)-1) 16#define IIO_ITTE_WIDGET_MASK ((1<<IIO_ITTE_WIDGET_BITS)-1)
17#define IIO_ITTE_WIDGET_SHIFT 8 17#define IIO_ITTE_WIDGET_SHIFT 8
18 18
19#define IIO_ITTE_WIDGET(itte) \
20 (((itte) >> IIO_ITTE_WIDGET_SHIFT) & IIO_ITTE_WIDGET_MASK)
21
19/* 22/*
20 * Use the top big window as a surrogate for the first small window 23 * Use the top big window as a surrogate for the first small window
21 */ 24 */
@@ -34,7 +37,8 @@ struct sn_flush_device_list {
34 unsigned long sfdl_force_int_addr; 37 unsigned long sfdl_force_int_addr;
35 unsigned long sfdl_flush_value; 38 unsigned long sfdl_flush_value;
36 volatile unsigned long *sfdl_flush_addr; 39 volatile unsigned long *sfdl_flush_addr;
37 uint64_t sfdl_persistent_busnum; 40 uint32_t sfdl_persistent_busnum;
41 uint32_t sfdl_persistent_segment;
38 struct pcibus_info *sfdl_pcibus_info; 42 struct pcibus_info *sfdl_pcibus_info;
39 spinlock_t sfdl_flush_lock; 43 spinlock_t sfdl_flush_lock;
40}; 44};
@@ -58,7 +62,8 @@ struct hubdev_info {
58 62
59 void *hdi_nodepda; 63 void *hdi_nodepda;
60 void *hdi_node_vertex; 64 void *hdi_node_vertex;
61 void *hdi_xtalk_vertex; 65 uint32_t max_segment_number;
66 uint32_t max_pcibus_number;
62}; 67};
63 68
64extern void hubdev_init_node(nodepda_t *, cnodeid_t); 69extern void hubdev_init_node(nodepda_t *, cnodeid_t);
diff --git a/arch/ia64/sn/kernel/bte.c b/arch/ia64/sn/kernel/bte.c
index 647deae9bfcd..45854c637e9c 100644
--- a/arch/ia64/sn/kernel/bte.c
+++ b/arch/ia64/sn/kernel/bte.c
@@ -29,16 +29,30 @@
29 29
30/* two interfaces on two btes */ 30/* two interfaces on two btes */
31#define MAX_INTERFACES_TO_TRY 4 31#define MAX_INTERFACES_TO_TRY 4
32#define MAX_NODES_TO_TRY 2
32 33
33static struct bteinfo_s *bte_if_on_node(nasid_t nasid, int interface) 34static struct bteinfo_s *bte_if_on_node(nasid_t nasid, int interface)
34{ 35{
35 nodepda_t *tmp_nodepda; 36 nodepda_t *tmp_nodepda;
36 37
38 if (nasid_to_cnodeid(nasid) == -1)
39 return (struct bteinfo_s *)NULL;;
40
37 tmp_nodepda = NODEPDA(nasid_to_cnodeid(nasid)); 41 tmp_nodepda = NODEPDA(nasid_to_cnodeid(nasid));
38 return &tmp_nodepda->bte_if[interface]; 42 return &tmp_nodepda->bte_if[interface];
39 43
40} 44}
41 45
46static inline void bte_start_transfer(struct bteinfo_s *bte, u64 len, u64 mode)
47{
48 if (is_shub2()) {
49 BTE_CTRL_STORE(bte, (IBLS_BUSY | ((len) | (mode) << 24)));
50 } else {
51 BTE_LNSTAT_STORE(bte, len);
52 BTE_CTRL_STORE(bte, mode);
53 }
54}
55
42/************************************************************************ 56/************************************************************************
43 * Block Transfer Engine copy related functions. 57 * Block Transfer Engine copy related functions.
44 * 58 *
@@ -67,13 +81,15 @@ bte_result_t bte_copy(u64 src, u64 dest, u64 len, u64 mode, void *notification)
67{ 81{
68 u64 transfer_size; 82 u64 transfer_size;
69 u64 transfer_stat; 83 u64 transfer_stat;
84 u64 notif_phys_addr;
70 struct bteinfo_s *bte; 85 struct bteinfo_s *bte;
71 bte_result_t bte_status; 86 bte_result_t bte_status;
72 unsigned long irq_flags; 87 unsigned long irq_flags;
73 unsigned long itc_end = 0; 88 unsigned long itc_end = 0;
74 struct bteinfo_s *btes_to_try[MAX_INTERFACES_TO_TRY]; 89 int nasid_to_try[MAX_NODES_TO_TRY];
75 int bte_if_index; 90 int my_nasid = get_nasid();
76 int bte_pri, bte_sec; 91 int bte_if_index, nasid_index;
92 int bte_first, btes_per_node = BTES_PER_NODE;
77 93
78 BTE_PRINTK(("bte_copy(0x%lx, 0x%lx, 0x%lx, 0x%lx, 0x%p)\n", 94 BTE_PRINTK(("bte_copy(0x%lx, 0x%lx, 0x%lx, 0x%lx, 0x%p)\n",
79 src, dest, len, mode, notification)); 95 src, dest, len, mode, notification));
@@ -86,36 +102,26 @@ bte_result_t bte_copy(u64 src, u64 dest, u64 len, u64 mode, void *notification)
86 (src & L1_CACHE_MASK) || (dest & L1_CACHE_MASK)); 102 (src & L1_CACHE_MASK) || (dest & L1_CACHE_MASK));
87 BUG_ON(!(len < ((BTE_LEN_MASK + 1) << L1_CACHE_SHIFT))); 103 BUG_ON(!(len < ((BTE_LEN_MASK + 1) << L1_CACHE_SHIFT)));
88 104
89 /* CPU 0 (per node) tries bte0 first, CPU 1 try bte1 first */ 105 /*
90 if (cpuid_to_subnode(smp_processor_id()) == 0) { 106 * Start with interface corresponding to cpu number
91 bte_pri = 0; 107 */
92 bte_sec = 1; 108 bte_first = raw_smp_processor_id() % btes_per_node;
93 } else {
94 bte_pri = 1;
95 bte_sec = 0;
96 }
97 109
98 if (mode & BTE_USE_DEST) { 110 if (mode & BTE_USE_DEST) {
99 /* try remote then local */ 111 /* try remote then local */
100 btes_to_try[0] = bte_if_on_node(NASID_GET(dest), bte_pri); 112 nasid_to_try[0] = NASID_GET(dest);
101 btes_to_try[1] = bte_if_on_node(NASID_GET(dest), bte_sec);
102 if (mode & BTE_USE_ANY) { 113 if (mode & BTE_USE_ANY) {
103 btes_to_try[2] = bte_if_on_node(get_nasid(), bte_pri); 114 nasid_to_try[1] = my_nasid;
104 btes_to_try[3] = bte_if_on_node(get_nasid(), bte_sec);
105 } else { 115 } else {
106 btes_to_try[2] = NULL; 116 nasid_to_try[1] = (int)NULL;
107 btes_to_try[3] = NULL;
108 } 117 }
109 } else { 118 } else {
110 /* try local then remote */ 119 /* try local then remote */
111 btes_to_try[0] = bte_if_on_node(get_nasid(), bte_pri); 120 nasid_to_try[0] = my_nasid;
112 btes_to_try[1] = bte_if_on_node(get_nasid(), bte_sec);
113 if (mode & BTE_USE_ANY) { 121 if (mode & BTE_USE_ANY) {
114 btes_to_try[2] = bte_if_on_node(NASID_GET(dest), bte_pri); 122 nasid_to_try[1] = NASID_GET(dest);
115 btes_to_try[3] = bte_if_on_node(NASID_GET(dest), bte_sec);
116 } else { 123 } else {
117 btes_to_try[2] = NULL; 124 nasid_to_try[1] = (int)NULL;
118 btes_to_try[3] = NULL;
119 } 125 }
120 } 126 }
121 127
@@ -123,11 +129,12 @@ retry_bteop:
123 do { 129 do {
124 local_irq_save(irq_flags); 130 local_irq_save(irq_flags);
125 131
126 bte_if_index = 0; 132 bte_if_index = bte_first;
133 nasid_index = 0;
127 134
128 /* Attempt to lock one of the BTE interfaces. */ 135 /* Attempt to lock one of the BTE interfaces. */
129 while (bte_if_index < MAX_INTERFACES_TO_TRY) { 136 while (nasid_index < MAX_NODES_TO_TRY) {
130 bte = btes_to_try[bte_if_index++]; 137 bte = bte_if_on_node(nasid_to_try[nasid_index],bte_if_index);
131 138
132 if (bte == NULL) { 139 if (bte == NULL) {
133 continue; 140 continue;
@@ -143,6 +150,15 @@ retry_bteop:
143 break; 150 break;
144 } 151 }
145 } 152 }
153
154 bte_if_index = (bte_if_index + 1) % btes_per_node; /* Next interface */
155 if (bte_if_index == bte_first) {
156 /*
157 * We've tried all interfaces on this node
158 */
159 nasid_index++;
160 }
161
146 bte = NULL; 162 bte = NULL;
147 } 163 }
148 164
@@ -169,7 +185,13 @@ retry_bteop:
169 185
170 /* Initialize the notification to a known value. */ 186 /* Initialize the notification to a known value. */
171 *bte->most_rcnt_na = BTE_WORD_BUSY; 187 *bte->most_rcnt_na = BTE_WORD_BUSY;
188 notif_phys_addr = TO_PHYS(ia64_tpa((unsigned long)bte->most_rcnt_na));
172 189
190 if (is_shub2()) {
191 src = SH2_TIO_PHYS_TO_DMA(src);
192 dest = SH2_TIO_PHYS_TO_DMA(dest);
193 notif_phys_addr = SH2_TIO_PHYS_TO_DMA(notif_phys_addr);
194 }
173 /* Set the source and destination registers */ 195 /* Set the source and destination registers */
174 BTE_PRINTKV(("IBSA = 0x%lx)\n", (TO_PHYS(src)))); 196 BTE_PRINTKV(("IBSA = 0x%lx)\n", (TO_PHYS(src))));
175 BTE_SRC_STORE(bte, TO_PHYS(src)); 197 BTE_SRC_STORE(bte, TO_PHYS(src));
@@ -177,14 +199,12 @@ retry_bteop:
177 BTE_DEST_STORE(bte, TO_PHYS(dest)); 199 BTE_DEST_STORE(bte, TO_PHYS(dest));
178 200
179 /* Set the notification register */ 201 /* Set the notification register */
180 BTE_PRINTKV(("IBNA = 0x%lx)\n", 202 BTE_PRINTKV(("IBNA = 0x%lx)\n", notif_phys_addr));
181 TO_PHYS(ia64_tpa((unsigned long)bte->most_rcnt_na)))); 203 BTE_NOTIF_STORE(bte, notif_phys_addr);
182 BTE_NOTIF_STORE(bte,
183 TO_PHYS(ia64_tpa((unsigned long)bte->most_rcnt_na)));
184 204
185 /* Initiate the transfer */ 205 /* Initiate the transfer */
186 BTE_PRINTK(("IBCT = 0x%lx)\n", BTE_VALID_MODE(mode))); 206 BTE_PRINTK(("IBCT = 0x%lx)\n", BTE_VALID_MODE(mode)));
187 BTE_START_TRANSFER(bte, transfer_size, BTE_VALID_MODE(mode)); 207 bte_start_transfer(bte, transfer_size, BTE_VALID_MODE(mode));
188 208
189 itc_end = ia64_get_itc() + (40000000 * local_cpu_data->cyc_per_usec); 209 itc_end = ia64_get_itc() + (40000000 * local_cpu_data->cyc_per_usec);
190 210
@@ -195,6 +215,7 @@ retry_bteop:
195 } 215 }
196 216
197 while ((transfer_stat = *bte->most_rcnt_na) == BTE_WORD_BUSY) { 217 while ((transfer_stat = *bte->most_rcnt_na) == BTE_WORD_BUSY) {
218 cpu_relax();
198 if (ia64_get_itc() > itc_end) { 219 if (ia64_get_itc() > itc_end) {
199 BTE_PRINTK(("BTE timeout nasid 0x%x bte%d IBLS = 0x%lx na 0x%lx\n", 220 BTE_PRINTK(("BTE timeout nasid 0x%x bte%d IBLS = 0x%lx na 0x%lx\n",
200 NASID_GET(bte->bte_base_addr), bte->bte_num, 221 NASID_GET(bte->bte_base_addr), bte->bte_num,
diff --git a/arch/ia64/sn/kernel/huberror.c b/arch/ia64/sn/kernel/huberror.c
index 5c39b43ba3c0..5c5eb01c50f0 100644
--- a/arch/ia64/sn/kernel/huberror.c
+++ b/arch/ia64/sn/kernel/huberror.c
@@ -76,7 +76,7 @@ void hubiio_crb_free(struct hubdev_info *hubdev_info, int crbnum)
76 */ 76 */
77 REMOTE_HUB_S(hubdev_info->hdi_nasid, IIO_ICDR, (IIO_ICDR_PND | crbnum)); 77 REMOTE_HUB_S(hubdev_info->hdi_nasid, IIO_ICDR, (IIO_ICDR_PND | crbnum));
78 while (REMOTE_HUB_L(hubdev_info->hdi_nasid, IIO_ICDR) & IIO_ICDR_PND) 78 while (REMOTE_HUB_L(hubdev_info->hdi_nasid, IIO_ICDR) & IIO_ICDR_PND)
79 udelay(1); 79 cpu_relax();
80 80
81} 81}
82 82
diff --git a/arch/ia64/sn/kernel/io_init.c b/arch/ia64/sn/kernel/io_init.c
index 414cdf2e3c96..4564ed0b5ff3 100644
--- a/arch/ia64/sn/kernel/io_init.c
+++ b/arch/ia64/sn/kernel/io_init.c
@@ -18,6 +18,7 @@
18#include <asm/sn/simulator.h> 18#include <asm/sn/simulator.h>
19#include <asm/sn/sn_sal.h> 19#include <asm/sn/sn_sal.h>
20#include <asm/sn/tioca_provider.h> 20#include <asm/sn/tioca_provider.h>
21#include <asm/sn/tioce_provider.h>
21#include "xtalk/hubdev.h" 22#include "xtalk/hubdev.h"
22#include "xtalk/xwidgetdev.h" 23#include "xtalk/xwidgetdev.h"
23 24
@@ -44,6 +45,9 @@ int sn_ioif_inited = 0; /* SN I/O infrastructure initialized? */
44 45
45struct sn_pcibus_provider *sn_pci_provider[PCIIO_ASIC_MAX_TYPES]; /* indexed by asic type */ 46struct sn_pcibus_provider *sn_pci_provider[PCIIO_ASIC_MAX_TYPES]; /* indexed by asic type */
46 47
48static int max_segment_number = 0; /* Default highest segment number */
49static int max_pcibus_number = 255; /* Default highest pci bus number */
50
47/* 51/*
48 * Hooks and struct for unsupported pci providers 52 * Hooks and struct for unsupported pci providers
49 */ 53 */
@@ -157,13 +161,28 @@ static void sn_fixup_ionodes(void)
157 uint64_t nasid; 161 uint64_t nasid;
158 int i, widget; 162 int i, widget;
159 163
164 /*
165 * Get SGI Specific HUB chipset information.
166 * Inform Prom that this kernel can support domain bus numbering.
167 */
160 for (i = 0; i < numionodes; i++) { 168 for (i = 0; i < numionodes; i++) {
161 hubdev = (struct hubdev_info *)(NODEPDA(i)->pdinfo); 169 hubdev = (struct hubdev_info *)(NODEPDA(i)->pdinfo);
162 nasid = cnodeid_to_nasid(i); 170 nasid = cnodeid_to_nasid(i);
171 hubdev->max_segment_number = 0xffffffff;
172 hubdev->max_pcibus_number = 0xff;
163 status = sal_get_hubdev_info(nasid, (uint64_t) __pa(hubdev)); 173 status = sal_get_hubdev_info(nasid, (uint64_t) __pa(hubdev));
164 if (status) 174 if (status)
165 continue; 175 continue;
166 176
177 /* Save the largest Domain and pcibus numbers found. */
178 if (hubdev->max_segment_number) {
179 /*
180 * Dealing with a Prom that supports segments.
181 */
182 max_segment_number = hubdev->max_segment_number;
183 max_pcibus_number = hubdev->max_pcibus_number;
184 }
185
167 /* Attach the error interrupt handlers */ 186 /* Attach the error interrupt handlers */
168 if (nasid & 1) 187 if (nasid & 1)
169 ice_error_init(hubdev); 188 ice_error_init(hubdev);
@@ -230,7 +249,7 @@ void sn_pci_unfixup_slot(struct pci_dev *dev)
230void sn_pci_fixup_slot(struct pci_dev *dev) 249void sn_pci_fixup_slot(struct pci_dev *dev)
231{ 250{
232 int idx; 251 int idx;
233 int segment = 0; 252 int segment = pci_domain_nr(dev->bus);
234 int status = 0; 253 int status = 0;
235 struct pcibus_bussoft *bs; 254 struct pcibus_bussoft *bs;
236 struct pci_bus *host_pci_bus; 255 struct pci_bus *host_pci_bus;
@@ -283,9 +302,9 @@ void sn_pci_fixup_slot(struct pci_dev *dev)
283 * PCI host_pci_dev struct and set up host bus linkages 302 * PCI host_pci_dev struct and set up host bus linkages
284 */ 303 */
285 304
286 bus_no = SN_PCIDEV_INFO(dev)->pdi_slot_host_handle >> 32; 305 bus_no = (SN_PCIDEV_INFO(dev)->pdi_slot_host_handle >> 32) & 0xff;
287 devfn = SN_PCIDEV_INFO(dev)->pdi_slot_host_handle & 0xffffffff; 306 devfn = SN_PCIDEV_INFO(dev)->pdi_slot_host_handle & 0xffffffff;
288 host_pci_bus = pci_find_bus(pci_domain_nr(dev->bus), bus_no); 307 host_pci_bus = pci_find_bus(segment, bus_no);
289 host_pci_dev = pci_get_slot(host_pci_bus, devfn); 308 host_pci_dev = pci_get_slot(host_pci_bus, devfn);
290 309
291 SN_PCIDEV_INFO(dev)->host_pci_dev = host_pci_dev; 310 SN_PCIDEV_INFO(dev)->host_pci_dev = host_pci_dev;
@@ -333,6 +352,7 @@ void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus)
333 prom_bussoft_ptr = __va(prom_bussoft_ptr); 352 prom_bussoft_ptr = __va(prom_bussoft_ptr);
334 353
335 controller = kcalloc(1,sizeof(struct pci_controller), GFP_KERNEL); 354 controller = kcalloc(1,sizeof(struct pci_controller), GFP_KERNEL);
355 controller->segment = segment;
336 if (!controller) 356 if (!controller)
337 BUG(); 357 BUG();
338 358
@@ -390,7 +410,7 @@ void sn_pci_controller_fixup(int segment, int busnum, struct pci_bus *bus)
390 if (controller->node >= num_online_nodes()) { 410 if (controller->node >= num_online_nodes()) {
391 struct pcibus_bussoft *b = SN_PCIBUS_BUSSOFT(bus); 411 struct pcibus_bussoft *b = SN_PCIBUS_BUSSOFT(bus);
392 412
393 printk(KERN_WARNING "Device ASIC=%u XID=%u PBUSNUM=%lu" 413 printk(KERN_WARNING "Device ASIC=%u XID=%u PBUSNUM=%u"
394 "L_IO=%lx L_MEM=%lx BASE=%lx\n", 414 "L_IO=%lx L_MEM=%lx BASE=%lx\n",
395 b->bs_asic_type, b->bs_xid, b->bs_persist_busnum, 415 b->bs_asic_type, b->bs_xid, b->bs_persist_busnum,
396 b->bs_legacy_io, b->bs_legacy_mem, b->bs_base); 416 b->bs_legacy_io, b->bs_legacy_mem, b->bs_base);
@@ -445,6 +465,7 @@ sn_sysdata_free_start:
445static int __init sn_pci_init(void) 465static int __init sn_pci_init(void)
446{ 466{
447 int i = 0; 467 int i = 0;
468 int j = 0;
448 struct pci_dev *pci_dev = NULL; 469 struct pci_dev *pci_dev = NULL;
449 extern void sn_init_cpei_timer(void); 470 extern void sn_init_cpei_timer(void);
450#ifdef CONFIG_PROC_FS 471#ifdef CONFIG_PROC_FS
@@ -464,6 +485,7 @@ static int __init sn_pci_init(void)
464 485
465 pcibr_init_provider(); 486 pcibr_init_provider();
466 tioca_init_provider(); 487 tioca_init_provider();
488 tioce_init_provider();
467 489
468 /* 490 /*
469 * This is needed to avoid bounce limit checks in the blk layer 491 * This is needed to avoid bounce limit checks in the blk layer
@@ -479,8 +501,9 @@ static int __init sn_pci_init(void)
479#endif 501#endif
480 502
481 /* busses are not known yet ... */ 503 /* busses are not known yet ... */
482 for (i = 0; i < PCI_BUSES_TO_SCAN; i++) 504 for (i = 0; i <= max_segment_number; i++)
483 sn_pci_controller_fixup(0, i, NULL); 505 for (j = 0; j <= max_pcibus_number; j++)
506 sn_pci_controller_fixup(i, j, NULL);
484 507
485 /* 508 /*
486 * Generic Linux PCI Layer has created the pci_bus and pci_dev 509 * Generic Linux PCI Layer has created the pci_bus and pci_dev
diff --git a/arch/ia64/sn/kernel/irq.c b/arch/ia64/sn/kernel/irq.c
index 392bf8a072b6..01d18b7b5bb3 100644
--- a/arch/ia64/sn/kernel/irq.c
+++ b/arch/ia64/sn/kernel/irq.c
@@ -5,7 +5,7 @@
5 * License. See the file "COPYING" in the main directory of this archive 5 * License. See the file "COPYING" in the main directory of this archive
6 * for more details. 6 * for more details.
7 * 7 *
8 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved. 8 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
9 */ 9 */
10 10
11#include <linux/irq.h> 11#include <linux/irq.h>
@@ -76,16 +76,14 @@ static void sn_enable_irq(unsigned int irq)
76 76
77static void sn_ack_irq(unsigned int irq) 77static void sn_ack_irq(unsigned int irq)
78{ 78{
79 uint64_t event_occurred, mask = 0; 79 u64 event_occurred, mask = 0;
80 int nasid;
81 80
82 irq = irq & 0xff; 81 irq = irq & 0xff;
83 nasid = get_nasid();
84 event_occurred = 82 event_occurred =
85 HUB_L((uint64_t *) GLOBAL_MMR_ADDR(nasid, SH_EVENT_OCCURRED)); 83 HUB_L((u64*)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED));
86 mask = event_occurred & SH_ALL_INT_MASK; 84 mask = event_occurred & SH_ALL_INT_MASK;
87 HUB_S((uint64_t *) GLOBAL_MMR_ADDR(nasid, SH_EVENT_OCCURRED_ALIAS), 85 HUB_S((u64*)LOCAL_MMR_ADDR(SH_EVENT_OCCURRED_ALIAS),
88 mask); 86 mask);
89 __set_bit(irq, (volatile void *)pda->sn_in_service_ivecs); 87 __set_bit(irq, (volatile void *)pda->sn_in_service_ivecs);
90 88
91 move_irq(irq); 89 move_irq(irq);
@@ -93,15 +91,12 @@ static void sn_ack_irq(unsigned int irq)
93 91
94static void sn_end_irq(unsigned int irq) 92static void sn_end_irq(unsigned int irq)
95{ 93{
96 int nasid;
97 int ivec; 94 int ivec;
98 uint64_t event_occurred; 95 u64 event_occurred;
99 96
100 ivec = irq & 0xff; 97 ivec = irq & 0xff;
101 if (ivec == SGI_UART_VECTOR) { 98 if (ivec == SGI_UART_VECTOR) {
102 nasid = get_nasid(); 99 event_occurred = HUB_L((u64*)LOCAL_MMR_ADDR (SH_EVENT_OCCURRED));
103 event_occurred = HUB_L((uint64_t *) GLOBAL_MMR_ADDR
104 (nasid, SH_EVENT_OCCURRED));
105 /* If the UART bit is set here, we may have received an 100 /* If the UART bit is set here, we may have received an
106 * interrupt from the UART that the driver missed. To 101 * interrupt from the UART that the driver missed. To
107 * make sure, we IPI ourselves to force us to look again. 102 * make sure, we IPI ourselves to force us to look again.
@@ -132,6 +127,7 @@ static void sn_set_affinity_irq(unsigned int irq, cpumask_t mask)
132 int local_widget, status; 127 int local_widget, status;
133 nasid_t local_nasid; 128 nasid_t local_nasid;
134 struct sn_irq_info *new_irq_info; 129 struct sn_irq_info *new_irq_info;
130 struct sn_pcibus_provider *pci_provider;
135 131
136 new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC); 132 new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC);
137 if (new_irq_info == NULL) 133 if (new_irq_info == NULL)
@@ -171,8 +167,9 @@ static void sn_set_affinity_irq(unsigned int irq, cpumask_t mask)
171 new_irq_info->irq_cpuid = cpuid; 167 new_irq_info->irq_cpuid = cpuid;
172 register_intr_pda(new_irq_info); 168 register_intr_pda(new_irq_info);
173 169
174 if (IS_PCI_BRIDGE_ASIC(new_irq_info->irq_bridge_type)) 170 pci_provider = sn_pci_provider[new_irq_info->irq_bridge_type];
175 pcibr_change_devices_irq(new_irq_info); 171 if (pci_provider && pci_provider->target_interrupt)
172 (pci_provider->target_interrupt)(new_irq_info);
176 173
177 spin_lock(&sn_irq_info_lock); 174 spin_lock(&sn_irq_info_lock);
178 list_replace_rcu(&sn_irq_info->list, &new_irq_info->list); 175 list_replace_rcu(&sn_irq_info->list, &new_irq_info->list);
@@ -317,6 +314,16 @@ void sn_irq_unfixup(struct pci_dev *pci_dev)
317 pci_dev_put(pci_dev); 314 pci_dev_put(pci_dev);
318} 315}
319 316
317static inline void
318sn_call_force_intr_provider(struct sn_irq_info *sn_irq_info)
319{
320 struct sn_pcibus_provider *pci_provider;
321
322 pci_provider = sn_pci_provider[sn_irq_info->irq_bridge_type];
323 if (pci_provider && pci_provider->force_interrupt)
324 (*pci_provider->force_interrupt)(sn_irq_info);
325}
326
320static void force_interrupt(int irq) 327static void force_interrupt(int irq)
321{ 328{
322 struct sn_irq_info *sn_irq_info; 329 struct sn_irq_info *sn_irq_info;
@@ -325,11 +332,9 @@ static void force_interrupt(int irq)
325 return; 332 return;
326 333
327 rcu_read_lock(); 334 rcu_read_lock();
328 list_for_each_entry_rcu(sn_irq_info, sn_irq_lh[irq], list) { 335 list_for_each_entry_rcu(sn_irq_info, sn_irq_lh[irq], list)
329 if (IS_PCI_BRIDGE_ASIC(sn_irq_info->irq_bridge_type) && 336 sn_call_force_intr_provider(sn_irq_info);
330 (sn_irq_info->irq_bridge != NULL)) 337
331 pcibr_force_interrupt(sn_irq_info);
332 }
333 rcu_read_unlock(); 338 rcu_read_unlock();
334} 339}
335 340
@@ -351,6 +356,14 @@ static void sn_check_intr(int irq, struct sn_irq_info *sn_irq_info)
351 struct pcidev_info *pcidev_info; 356 struct pcidev_info *pcidev_info;
352 struct pcibus_info *pcibus_info; 357 struct pcibus_info *pcibus_info;
353 358
359 /*
360 * Bridge types attached to TIO (anything but PIC) do not need this WAR
361 * since they do not target Shub II interrupt registers. If that
362 * ever changes, this check needs to accomodate.
363 */
364 if (sn_irq_info->irq_bridge_type != PCIIO_ASIC_TYPE_PIC)
365 return;
366
354 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo; 367 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
355 if (!pcidev_info) 368 if (!pcidev_info)
356 return; 369 return;
@@ -377,16 +390,12 @@ static void sn_check_intr(int irq, struct sn_irq_info *sn_irq_info)
377 break; 390 break;
378 } 391 }
379 if (!test_bit(irr_bit, &irr_reg)) { 392 if (!test_bit(irr_bit, &irr_reg)) {
380 if (!test_bit(irq, pda->sn_soft_irr)) { 393 if (!test_bit(irq, pda->sn_in_service_ivecs)) {
381 if (!test_bit(irq, pda->sn_in_service_ivecs)) { 394 regval &= 0xff;
382 regval &= 0xff; 395 if (sn_irq_info->irq_int_bit & regval &
383 if (sn_irq_info->irq_int_bit & regval & 396 sn_irq_info->irq_last_intr) {
384 sn_irq_info->irq_last_intr) { 397 regval &= ~(sn_irq_info->irq_int_bit & regval);
385 regval &= 398 sn_call_force_intr_provider(sn_irq_info);
386 ~(sn_irq_info->
387 irq_int_bit & regval);
388 pcibr_force_interrupt(sn_irq_info);
389 }
390 } 399 }
391 } 400 }
392 } 401 }
@@ -404,13 +413,7 @@ void sn_lb_int_war_check(void)
404 rcu_read_lock(); 413 rcu_read_lock();
405 for (i = pda->sn_first_irq; i <= pda->sn_last_irq; i++) { 414 for (i = pda->sn_first_irq; i <= pda->sn_last_irq; i++) {
406 list_for_each_entry_rcu(sn_irq_info, sn_irq_lh[i], list) { 415 list_for_each_entry_rcu(sn_irq_info, sn_irq_lh[i], list) {
407 /* 416 sn_check_intr(i, sn_irq_info);
408 * Only call for PCI bridges that are fully
409 * initialized.
410 */
411 if (IS_PCI_BRIDGE_ASIC(sn_irq_info->irq_bridge_type) &&
412 (sn_irq_info->irq_bridge != NULL))
413 sn_check_intr(i, sn_irq_info);
414 } 417 }
415 } 418 }
416 rcu_read_unlock(); 419 rcu_read_unlock();
diff --git a/arch/ia64/sn/kernel/setup.c b/arch/ia64/sn/kernel/setup.c
index 7c7fe441d623..a594aca959e6 100644
--- a/arch/ia64/sn/kernel/setup.c
+++ b/arch/ia64/sn/kernel/setup.c
@@ -80,8 +80,6 @@ EXPORT_PER_CPU_SYMBOL(__sn_cnodeid_to_nasid);
80DEFINE_PER_CPU(struct nodepda_s *, __sn_nodepda); 80DEFINE_PER_CPU(struct nodepda_s *, __sn_nodepda);
81EXPORT_PER_CPU_SYMBOL(__sn_nodepda); 81EXPORT_PER_CPU_SYMBOL(__sn_nodepda);
82 82
83partid_t sn_partid = -1;
84EXPORT_SYMBOL(sn_partid);
85char sn_system_serial_number_string[128]; 83char sn_system_serial_number_string[128];
86EXPORT_SYMBOL(sn_system_serial_number_string); 84EXPORT_SYMBOL(sn_system_serial_number_string);
87u64 sn_partition_serial_number; 85u64 sn_partition_serial_number;
@@ -403,6 +401,7 @@ static void __init sn_init_pdas(char **cmdline_p)
403 memset(nodepdaindr[cnode], 0, sizeof(nodepda_t)); 401 memset(nodepdaindr[cnode], 0, sizeof(nodepda_t));
404 memset(nodepdaindr[cnode]->phys_cpuid, -1, 402 memset(nodepdaindr[cnode]->phys_cpuid, -1,
405 sizeof(nodepdaindr[cnode]->phys_cpuid)); 403 sizeof(nodepdaindr[cnode]->phys_cpuid));
404 spin_lock_init(&nodepdaindr[cnode]->ptc_lock);
406 } 405 }
407 406
408 /* 407 /*
@@ -532,8 +531,8 @@ void __init sn_cpu_init(void)
532 */ 531 */
533 { 532 {
534 u64 pio1[] = {SH1_PIO_WRITE_STATUS_0, 0, SH1_PIO_WRITE_STATUS_1, 0}; 533 u64 pio1[] = {SH1_PIO_WRITE_STATUS_0, 0, SH1_PIO_WRITE_STATUS_1, 0};
535 u64 pio2[] = {SH2_PIO_WRITE_STATUS_0, SH2_PIO_WRITE_STATUS_1, 534 u64 pio2[] = {SH2_PIO_WRITE_STATUS_0, SH2_PIO_WRITE_STATUS_2,
536 SH2_PIO_WRITE_STATUS_2, SH2_PIO_WRITE_STATUS_3}; 535 SH2_PIO_WRITE_STATUS_1, SH2_PIO_WRITE_STATUS_3};
537 u64 *pio; 536 u64 *pio;
538 pio = is_shub1() ? pio1 : pio2; 537 pio = is_shub1() ? pio1 : pio2;
539 pda->pio_write_status_addr = (volatile unsigned long *) LOCAL_MMR_ADDR(pio[slice]); 538 pda->pio_write_status_addr = (volatile unsigned long *) LOCAL_MMR_ADDR(pio[slice]);
diff --git a/arch/ia64/sn/kernel/sn2/ptc_deadlock.S b/arch/ia64/sn/kernel/sn2/ptc_deadlock.S
index 96cb71d15682..3fa95065a446 100644
--- a/arch/ia64/sn/kernel/sn2/ptc_deadlock.S
+++ b/arch/ia64/sn/kernel/sn2/ptc_deadlock.S
@@ -3,7 +3,7 @@
3 * License. See the file "COPYING" in the main directory of this archive 3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details. 4 * for more details.
5 * 5 *
6 * Copyright (C) 2000-2004 Silicon Graphics, Inc. All rights reserved. 6 * Copyright (C) 2000-2005 Silicon Graphics, Inc. All rights reserved.
7 */ 7 */
8 8
9#include <asm/types.h> 9#include <asm/types.h>
@@ -11,7 +11,7 @@
11 11
12#define DEADLOCKBIT SH_PIO_WRITE_STATUS_WRITE_DEADLOCK_SHFT 12#define DEADLOCKBIT SH_PIO_WRITE_STATUS_WRITE_DEADLOCK_SHFT
13#define WRITECOUNTMASK SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK 13#define WRITECOUNTMASK SH_PIO_WRITE_STATUS_PENDING_WRITE_COUNT_MASK
14#define ALIAS_OFFSET (SH1_PIO_WRITE_STATUS_0_ALIAS-SH1_PIO_WRITE_STATUS_0) 14#define ALIAS_OFFSET 8
15 15
16 16
17 .global sn2_ptc_deadlock_recovery_core 17 .global sn2_ptc_deadlock_recovery_core
@@ -36,13 +36,15 @@ sn2_ptc_deadlock_recovery_core:
36 extr.u piowcphy=piowc,0,61;; // Convert piowc to uncached physical address 36 extr.u piowcphy=piowc,0,61;; // Convert piowc to uncached physical address
37 dep piowcphy=-1,piowcphy,63,1 37 dep piowcphy=-1,piowcphy,63,1
38 movl mask=WRITECOUNTMASK 38 movl mask=WRITECOUNTMASK
39 mov r8=r0
39 40
401: 411:
41 add scr2=ALIAS_OFFSET,piowc // Address of WRITE_STATUS alias register 42 add scr2=ALIAS_OFFSET,piowc // Address of WRITE_STATUS alias register
42 mov scr1=7;; // Clear DEADLOCK, WRITE_ERROR, MULTI_WRITE_ERROR 43 ;;
43 st8.rel [scr2]=scr1;; 44 ld8.acq scr1=[scr2];;
44 45
455: ld8.acq scr1=[piowc];; // Wait for PIOs to complete. 465: ld8.acq scr1=[piowc];; // Wait for PIOs to complete.
47 hint @pause
46 and scr2=scr1,mask;; // mask of writecount bits 48 and scr2=scr1,mask;; // mask of writecount bits
47 cmp.ne p6,p0=zeroval,scr2 49 cmp.ne p6,p0=zeroval,scr2
48(p6) br.cond.sptk 5b 50(p6) br.cond.sptk 5b
@@ -57,6 +59,7 @@ sn2_ptc_deadlock_recovery_core:
57 st8.rel [ptc0]=data0 // Write PTC0 & wait for completion. 59 st8.rel [ptc0]=data0 // Write PTC0 & wait for completion.
58 60
595: ld8.acq scr1=[piowcphy];; // Wait for PIOs to complete. 615: ld8.acq scr1=[piowcphy];; // Wait for PIOs to complete.
62 hint @pause
60 and scr2=scr1,mask;; // mask of writecount bits 63 and scr2=scr1,mask;; // mask of writecount bits
61 cmp.ne p6,p0=zeroval,scr2 64 cmp.ne p6,p0=zeroval,scr2
62(p6) br.cond.sptk 5b;; 65(p6) br.cond.sptk 5b;;
@@ -67,6 +70,7 @@ sn2_ptc_deadlock_recovery_core:
67(p7) st8.rel [ptc1]=data1;; // Now write PTC1. 70(p7) st8.rel [ptc1]=data1;; // Now write PTC1.
68 71
695: ld8.acq scr1=[piowcphy];; // Wait for PIOs to complete. 725: ld8.acq scr1=[piowcphy];; // Wait for PIOs to complete.
73 hint @pause
70 and scr2=scr1,mask;; // mask of writecount bits 74 and scr2=scr1,mask;; // mask of writecount bits
71 cmp.ne p6,p0=zeroval,scr2 75 cmp.ne p6,p0=zeroval,scr2
72(p6) br.cond.sptk 5b 76(p6) br.cond.sptk 5b
@@ -77,6 +81,7 @@ sn2_ptc_deadlock_recovery_core:
77 srlz.i;; 81 srlz.i;;
78 ////////////// END PHYSICAL MODE //////////////////// 82 ////////////// END PHYSICAL MODE ////////////////////
79 83
84(p8) add r8=1,r8
80(p8) br.cond.spnt 1b;; // Repeat if DEADLOCK occurred. 85(p8) br.cond.spnt 1b;; // Repeat if DEADLOCK occurred.
81 86
82 br.ret.sptk rp 87 br.ret.sptk rp
diff --git a/arch/ia64/sn/kernel/sn2/sn2_smp.c b/arch/ia64/sn/kernel/sn2/sn2_smp.c
index 7af05a7ac743..0a4ee50c302f 100644
--- a/arch/ia64/sn/kernel/sn2/sn2_smp.c
+++ b/arch/ia64/sn/kernel/sn2/sn2_smp.c
@@ -5,7 +5,7 @@
5 * License. See the file "COPYING" in the main directory of this archive 5 * License. See the file "COPYING" in the main directory of this archive
6 * for more details. 6 * for more details.
7 * 7 *
8 * Copyright (C) 2000-2004 Silicon Graphics, Inc. All rights reserved. 8 * Copyright (C) 2000-2005 Silicon Graphics, Inc. All rights reserved.
9 */ 9 */
10 10
11#include <linux/init.h> 11#include <linux/init.h>
@@ -20,6 +20,8 @@
20#include <linux/module.h> 20#include <linux/module.h>
21#include <linux/bitops.h> 21#include <linux/bitops.h>
22#include <linux/nodemask.h> 22#include <linux/nodemask.h>
23#include <linux/proc_fs.h>
24#include <linux/seq_file.h>
23 25
24#include <asm/processor.h> 26#include <asm/processor.h>
25#include <asm/irq.h> 27#include <asm/irq.h>
@@ -39,12 +41,120 @@
39#include <asm/sn/nodepda.h> 41#include <asm/sn/nodepda.h>
40#include <asm/sn/rw_mmr.h> 42#include <asm/sn/rw_mmr.h>
41 43
42void sn2_ptc_deadlock_recovery(volatile unsigned long *, unsigned long data0, 44DEFINE_PER_CPU(struct ptc_stats, ptcstats);
43 volatile unsigned long *, unsigned long data1); 45DECLARE_PER_CPU(struct ptc_stats, ptcstats);
44 46
45static __cacheline_aligned DEFINE_SPINLOCK(sn2_global_ptc_lock); 47static __cacheline_aligned DEFINE_SPINLOCK(sn2_global_ptc_lock);
46 48
47static unsigned long sn2_ptc_deadlock_count; 49void sn2_ptc_deadlock_recovery(short *, short, int, volatile unsigned long *, unsigned long data0,
50 volatile unsigned long *, unsigned long data1);
51
52#ifdef DEBUG_PTC
53/*
54 * ptctest:
55 *
56 * xyz - 3 digit hex number:
57 * x - Force PTC purges to use shub:
58 * 0 - no force
59 * 1 - force
60 * y - interupt enable
61 * 0 - disable interrupts
62 * 1 - leave interuupts enabled
63 * z - type of lock:
64 * 0 - global lock
65 * 1 - node local lock
66 * 2 - no lock
67 *
68 * Note: on shub1, only ptctest == 0 is supported. Don't try other values!
69 */
70
71static unsigned int sn2_ptctest = 0;
72
73static int __init ptc_test(char *str)
74{
75 get_option(&str, &sn2_ptctest);
76 return 1;
77}
78__setup("ptctest=", ptc_test);
79
80static inline int ptc_lock(unsigned long *flagp)
81{
82 unsigned long opt = sn2_ptctest & 255;
83
84 switch (opt) {
85 case 0x00:
86 spin_lock_irqsave(&sn2_global_ptc_lock, *flagp);
87 break;
88 case 0x01:
89 spin_lock_irqsave(&sn_nodepda->ptc_lock, *flagp);
90 break;
91 case 0x02:
92 local_irq_save(*flagp);
93 break;
94 case 0x10:
95 spin_lock(&sn2_global_ptc_lock);
96 break;
97 case 0x11:
98 spin_lock(&sn_nodepda->ptc_lock);
99 break;
100 case 0x12:
101 break;
102 default:
103 BUG();
104 }
105 return opt;
106}
107
108static inline void ptc_unlock(unsigned long flags, int opt)
109{
110 switch (opt) {
111 case 0x00:
112 spin_unlock_irqrestore(&sn2_global_ptc_lock, flags);
113 break;
114 case 0x01:
115 spin_unlock_irqrestore(&sn_nodepda->ptc_lock, flags);
116 break;
117 case 0x02:
118 local_irq_restore(flags);
119 break;
120 case 0x10:
121 spin_unlock(&sn2_global_ptc_lock);
122 break;
123 case 0x11:
124 spin_unlock(&sn_nodepda->ptc_lock);
125 break;
126 case 0x12:
127 break;
128 default:
129 BUG();
130 }
131}
132#else
133
134#define sn2_ptctest 0
135
136static inline int ptc_lock(unsigned long *flagp)
137{
138 spin_lock_irqsave(&sn2_global_ptc_lock, *flagp);
139 return 0;
140}
141
142static inline void ptc_unlock(unsigned long flags, int opt)
143{
144 spin_unlock_irqrestore(&sn2_global_ptc_lock, flags);
145}
146#endif
147
148struct ptc_stats {
149 unsigned long ptc_l;
150 unsigned long change_rid;
151 unsigned long shub_ptc_flushes;
152 unsigned long nodes_flushed;
153 unsigned long deadlocks;
154 unsigned long lock_itc_clocks;
155 unsigned long shub_itc_clocks;
156 unsigned long shub_itc_clocks_max;
157};
48 158
49static inline unsigned long wait_piowc(void) 159static inline unsigned long wait_piowc(void)
50{ 160{
@@ -89,9 +199,9 @@ void
89sn2_global_tlb_purge(unsigned long start, unsigned long end, 199sn2_global_tlb_purge(unsigned long start, unsigned long end,
90 unsigned long nbits) 200 unsigned long nbits)
91{ 201{
92 int i, shub1, cnode, mynasid, cpu, lcpu = 0, nasid, flushed = 0; 202 int i, opt, shub1, cnode, mynasid, cpu, lcpu = 0, nasid, flushed = 0;
93 volatile unsigned long *ptc0, *ptc1; 203 volatile unsigned long *ptc0, *ptc1;
94 unsigned long flags = 0, data0 = 0, data1 = 0; 204 unsigned long itc, itc2, flags, data0 = 0, data1 = 0;
95 struct mm_struct *mm = current->active_mm; 205 struct mm_struct *mm = current->active_mm;
96 short nasids[MAX_NUMNODES], nix; 206 short nasids[MAX_NUMNODES], nix;
97 nodemask_t nodes_flushed; 207 nodemask_t nodes_flushed;
@@ -114,16 +224,19 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end,
114 start += (1UL << nbits); 224 start += (1UL << nbits);
115 } while (start < end); 225 } while (start < end);
116 ia64_srlz_i(); 226 ia64_srlz_i();
227 __get_cpu_var(ptcstats).ptc_l++;
117 preempt_enable(); 228 preempt_enable();
118 return; 229 return;
119 } 230 }
120 231
121 if (atomic_read(&mm->mm_users) == 1) { 232 if (atomic_read(&mm->mm_users) == 1) {
122 flush_tlb_mm(mm); 233 flush_tlb_mm(mm);
234 __get_cpu_var(ptcstats).change_rid++;
123 preempt_enable(); 235 preempt_enable();
124 return; 236 return;
125 } 237 }
126 238
239 itc = ia64_get_itc();
127 nix = 0; 240 nix = 0;
128 for_each_node_mask(cnode, nodes_flushed) 241 for_each_node_mask(cnode, nodes_flushed)
129 nasids[nix++] = cnodeid_to_nasid(cnode); 242 nasids[nix++] = cnodeid_to_nasid(cnode);
@@ -148,7 +261,12 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end,
148 261
149 mynasid = get_nasid(); 262 mynasid = get_nasid();
150 263
151 spin_lock_irqsave(&sn2_global_ptc_lock, flags); 264 itc = ia64_get_itc();
265 opt = ptc_lock(&flags);
266 itc2 = ia64_get_itc();
267 __get_cpu_var(ptcstats).lock_itc_clocks += itc2 - itc;
268 __get_cpu_var(ptcstats).shub_ptc_flushes++;
269 __get_cpu_var(ptcstats).nodes_flushed += nix;
152 270
153 do { 271 do {
154 if (shub1) 272 if (shub1)
@@ -157,7 +275,7 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end,
157 data0 = (data0 & ~SH2_PTC_ADDR_MASK) | (start & SH2_PTC_ADDR_MASK); 275 data0 = (data0 & ~SH2_PTC_ADDR_MASK) | (start & SH2_PTC_ADDR_MASK);
158 for (i = 0; i < nix; i++) { 276 for (i = 0; i < nix; i++) {
159 nasid = nasids[i]; 277 nasid = nasids[i];
160 if (unlikely(nasid == mynasid)) { 278 if ((!(sn2_ptctest & 3)) && unlikely(nasid == mynasid)) {
161 ia64_ptcga(start, nbits << 2); 279 ia64_ptcga(start, nbits << 2);
162 ia64_srlz_i(); 280 ia64_srlz_i();
163 } else { 281 } else {
@@ -169,18 +287,22 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end,
169 flushed = 1; 287 flushed = 1;
170 } 288 }
171 } 289 }
172
173 if (flushed 290 if (flushed
174 && (wait_piowc() & 291 && (wait_piowc() &
175 SH_PIO_WRITE_STATUS_WRITE_DEADLOCK_MASK)) { 292 (SH_PIO_WRITE_STATUS_WRITE_DEADLOCK_MASK))) {
176 sn2_ptc_deadlock_recovery(ptc0, data0, ptc1, data1); 293 sn2_ptc_deadlock_recovery(nasids, nix, mynasid, ptc0, data0, ptc1, data1);
177 } 294 }
178 295
179 start += (1UL << nbits); 296 start += (1UL << nbits);
180 297
181 } while (start < end); 298 } while (start < end);
182 299
183 spin_unlock_irqrestore(&sn2_global_ptc_lock, flags); 300 itc2 = ia64_get_itc() - itc2;
301 __get_cpu_var(ptcstats).shub_itc_clocks += itc2;
302 if (itc2 > __get_cpu_var(ptcstats).shub_itc_clocks_max)
303 __get_cpu_var(ptcstats).shub_itc_clocks_max = itc2;
304
305 ptc_unlock(flags, opt);
184 306
185 preempt_enable(); 307 preempt_enable();
186} 308}
@@ -192,31 +314,29 @@ sn2_global_tlb_purge(unsigned long start, unsigned long end,
192 * TLB flush transaction. The recovery sequence is somewhat tricky & is 314 * TLB flush transaction. The recovery sequence is somewhat tricky & is
193 * coded in assembly language. 315 * coded in assembly language.
194 */ 316 */
195void sn2_ptc_deadlock_recovery(volatile unsigned long *ptc0, unsigned long data0, 317void sn2_ptc_deadlock_recovery(short *nasids, short nix, int mynasid, volatile unsigned long *ptc0, unsigned long data0,
196 volatile unsigned long *ptc1, unsigned long data1) 318 volatile unsigned long *ptc1, unsigned long data1)
197{ 319{
198 extern void sn2_ptc_deadlock_recovery_core(volatile unsigned long *, unsigned long, 320 extern void sn2_ptc_deadlock_recovery_core(volatile unsigned long *, unsigned long,
199 volatile unsigned long *, unsigned long, volatile unsigned long *, unsigned long); 321 volatile unsigned long *, unsigned long, volatile unsigned long *, unsigned long);
200 int cnode, mycnode, nasid; 322 short nasid, i;
201 volatile unsigned long *piows; 323 unsigned long *piows, zeroval;
202 volatile unsigned long zeroval;
203 324
204 sn2_ptc_deadlock_count++; 325 __get_cpu_var(ptcstats).deadlocks++;
205 326
206 piows = pda->pio_write_status_addr; 327 piows = (unsigned long *) pda->pio_write_status_addr;
207 zeroval = pda->pio_write_status_val; 328 zeroval = pda->pio_write_status_val;
208 329
209 mycnode = numa_node_id(); 330 for (i=0; i < nix; i++) {
210 331 nasid = nasids[i];
211 for_each_online_node(cnode) { 332 if (!(sn2_ptctest & 3) && nasid == mynasid)
212 if (is_headless_node(cnode) || cnode == mycnode)
213 continue; 333 continue;
214 nasid = cnodeid_to_nasid(cnode);
215 ptc0 = CHANGE_NASID(nasid, ptc0); 334 ptc0 = CHANGE_NASID(nasid, ptc0);
216 if (ptc1) 335 if (ptc1)
217 ptc1 = CHANGE_NASID(nasid, ptc1); 336 ptc1 = CHANGE_NASID(nasid, ptc1);
218 sn2_ptc_deadlock_recovery_core(ptc0, data0, ptc1, data1, piows, zeroval); 337 sn2_ptc_deadlock_recovery_core(ptc0, data0, ptc1, data1, piows, zeroval);
219 } 338 }
339
220} 340}
221 341
222/** 342/**
@@ -293,3 +413,93 @@ void sn2_send_IPI(int cpuid, int vector, int delivery_mode, int redirect)
293 413
294 sn_send_IPI_phys(nasid, physid, vector, delivery_mode); 414 sn_send_IPI_phys(nasid, physid, vector, delivery_mode);
295} 415}
416
417#ifdef CONFIG_PROC_FS
418
419#define PTC_BASENAME "sgi_sn/ptc_statistics"
420
421static void *sn2_ptc_seq_start(struct seq_file *file, loff_t * offset)
422{
423 if (*offset < NR_CPUS)
424 return offset;
425 return NULL;
426}
427
428static void *sn2_ptc_seq_next(struct seq_file *file, void *data, loff_t * offset)
429{
430 (*offset)++;
431 if (*offset < NR_CPUS)
432 return offset;
433 return NULL;
434}
435
436static void sn2_ptc_seq_stop(struct seq_file *file, void *data)
437{
438}
439
440static int sn2_ptc_seq_show(struct seq_file *file, void *data)
441{
442 struct ptc_stats *stat;
443 int cpu;
444
445 cpu = *(loff_t *) data;
446
447 if (!cpu) {
448 seq_printf(file, "# ptc_l change_rid shub_ptc_flushes shub_nodes_flushed deadlocks lock_nsec shub_nsec shub_nsec_max\n");
449 seq_printf(file, "# ptctest %d\n", sn2_ptctest);
450 }
451
452 if (cpu < NR_CPUS && cpu_online(cpu)) {
453 stat = &per_cpu(ptcstats, cpu);
454 seq_printf(file, "cpu %d %ld %ld %ld %ld %ld %ld %ld %ld\n", cpu, stat->ptc_l,
455 stat->change_rid, stat->shub_ptc_flushes, stat->nodes_flushed,
456 stat->deadlocks,
457 1000 * stat->lock_itc_clocks / per_cpu(cpu_info, cpu).cyc_per_usec,
458 1000 * stat->shub_itc_clocks / per_cpu(cpu_info, cpu).cyc_per_usec,
459 1000 * stat->shub_itc_clocks_max / per_cpu(cpu_info, cpu).cyc_per_usec);
460 }
461
462 return 0;
463}
464
465static struct seq_operations sn2_ptc_seq_ops = {
466 .start = sn2_ptc_seq_start,
467 .next = sn2_ptc_seq_next,
468 .stop = sn2_ptc_seq_stop,
469 .show = sn2_ptc_seq_show
470};
471
472int sn2_ptc_proc_open(struct inode *inode, struct file *file)
473{
474 return seq_open(file, &sn2_ptc_seq_ops);
475}
476
477static struct file_operations proc_sn2_ptc_operations = {
478 .open = sn2_ptc_proc_open,
479 .read = seq_read,
480 .llseek = seq_lseek,
481 .release = seq_release,
482};
483
484static struct proc_dir_entry *proc_sn2_ptc;
485
486static int __init sn2_ptc_init(void)
487{
488 if (!(proc_sn2_ptc = create_proc_entry(PTC_BASENAME, 0444, NULL))) {
489 printk(KERN_ERR "unable to create %s proc entry", PTC_BASENAME);
490 return -EINVAL;
491 }
492 proc_sn2_ptc->proc_fops = &proc_sn2_ptc_operations;
493 spin_lock_init(&sn2_global_ptc_lock);
494 return 0;
495}
496
497static void __exit sn2_ptc_exit(void)
498{
499 remove_proc_entry(PTC_BASENAME, NULL);
500}
501
502module_init(sn2_ptc_init);
503module_exit(sn2_ptc_exit);
504#endif /* CONFIG_PROC_FS */
505
diff --git a/arch/ia64/sn/kernel/sn2/sn_hwperf.c b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
index 833e700fdac9..0513aacac8c1 100644
--- a/arch/ia64/sn/kernel/sn2/sn_hwperf.c
+++ b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
@@ -36,7 +36,6 @@
36#include <asm/topology.h> 36#include <asm/topology.h>
37#include <asm/smp.h> 37#include <asm/smp.h>
38#include <asm/semaphore.h> 38#include <asm/semaphore.h>
39#include <asm/segment.h>
40#include <asm/uaccess.h> 39#include <asm/uaccess.h>
41#include <asm/sal.h> 40#include <asm/sal.h>
42#include <asm/sn/io.h> 41#include <asm/sn/io.h>
@@ -59,7 +58,7 @@ static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
59 struct sn_hwperf_object_info *objbuf = NULL; 58 struct sn_hwperf_object_info *objbuf = NULL;
60 59
61 if ((e = sn_hwperf_init()) < 0) { 60 if ((e = sn_hwperf_init()) < 0) {
62 printk("sn_hwperf_init failed: err %d\n", e); 61 printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
63 goto out; 62 goto out;
64 } 63 }
65 64
@@ -111,7 +110,7 @@ static int sn_hwperf_geoid_to_cnode(char *location)
111 if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab)) 110 if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
112 return -1; 111 return -1;
113 112
114 for (cnode = 0; cnode < numionodes; cnode++) { 113 for_each_node(cnode) {
115 geoid = cnodeid_get_geoid(cnode); 114 geoid = cnodeid_get_geoid(cnode);
116 module_id = geo_module(geoid); 115 module_id = geo_module(geoid);
117 this_rack = MODULE_GET_RACK(module_id); 116 this_rack = MODULE_GET_RACK(module_id);
@@ -124,11 +123,13 @@ static int sn_hwperf_geoid_to_cnode(char *location)
124 } 123 }
125 } 124 }
126 125
127 return cnode < numionodes ? cnode : -1; 126 return node_possible(cnode) ? cnode : -1;
128} 127}
129 128
130static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj) 129static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
131{ 130{
131 if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
132 BUG();
132 if (!obj->sn_hwp_this_part) 133 if (!obj->sn_hwp_this_part)
133 return -1; 134 return -1;
134 return sn_hwperf_geoid_to_cnode(obj->location); 135 return sn_hwperf_geoid_to_cnode(obj->location);
@@ -174,31 +175,199 @@ static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj,
174 return slabname; 175 return slabname;
175} 176}
176 177
177static void print_pci_topology(struct seq_file *s, 178static void print_pci_topology(struct seq_file *s)
178 struct sn_hwperf_object_info *obj, int *ordinal, 179{
179 u64 rack, u64 bay, u64 slot, u64 slab) 180 char *p;
181 size_t sz;
182 int e;
183
184 for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) {
185 if (!(p = (char *)kmalloc(sz, GFP_KERNEL)))
186 break;
187 e = ia64_sn_ioif_get_pci_topology(__pa(p), sz);
188 if (e == SALRET_OK)
189 seq_puts(s, p);
190 kfree(p);
191 if (e == SALRET_OK || e == SALRET_NOT_IMPLEMENTED)
192 break;
193 }
194}
195
196static inline int sn_hwperf_has_cpus(cnodeid_t node)
197{
198 return node_online(node) && nr_cpus_node(node);
199}
200
201static inline int sn_hwperf_has_mem(cnodeid_t node)
202{
203 return node_online(node) && NODE_DATA(node)->node_present_pages;
204}
205
206static struct sn_hwperf_object_info *
207sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
208 int nobj, int id)
180{ 209{
181 char *p1; 210 int i;
182 char *p2; 211 struct sn_hwperf_object_info *p = objbuf;
183 char *pg; 212
184 213 for (i=0; i < nobj; i++, p++) {
185 if (!(pg = (char *)get_zeroed_page(GFP_KERNEL))) 214 if (p->id == id)
186 return; /* ignore */ 215 return p;
187 if (ia64_sn_ioif_get_pci_topology(rack, bay, slot, slab, 216 }
188 __pa(pg), PAGE_SIZE) == SN_HWPERF_OP_OK) { 217
189 for (p1=pg; *p1 && p1 < pg + PAGE_SIZE;) { 218 return NULL;
190 if (!(p2 = strchr(p1, '\n'))) 219
220}
221
222static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf,
223 int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
224{
225 int e;
226 struct sn_hwperf_object_info *nodeobj = NULL;
227 struct sn_hwperf_object_info *op;
228 struct sn_hwperf_object_info *dest;
229 struct sn_hwperf_object_info *router;
230 struct sn_hwperf_port_info ptdata[16];
231 int sz, i, j;
232 cnodeid_t c;
233 int found_mem = 0;
234 int found_cpu = 0;
235
236 if (!node_possible(node))
237 return -EINVAL;
238
239 if (sn_hwperf_has_cpus(node)) {
240 if (near_cpu_node)
241 *near_cpu_node = node;
242 found_cpu++;
243 }
244
245 if (sn_hwperf_has_mem(node)) {
246 if (near_mem_node)
247 *near_mem_node = node;
248 found_mem++;
249 }
250
251 if (found_cpu && found_mem)
252 return 0; /* trivially successful */
253
254 /* find the argument node object */
255 for (i=0, op=objbuf; i < nobj; i++, op++) {
256 if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op))
257 continue;
258 if (node == sn_hwperf_obj_to_cnode(op)) {
259 nodeobj = op;
260 break;
261 }
262 }
263 if (!nodeobj) {
264 e = -ENOENT;
265 goto err;
266 }
267
268 /* get it's interconnect topology */
269 sz = op->ports * sizeof(struct sn_hwperf_port_info);
270 if (sz > sizeof(ptdata))
271 BUG();
272 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
273 SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
274 (u64)&ptdata, 0, 0, NULL);
275 if (e != SN_HWPERF_OP_OK) {
276 e = -EINVAL;
277 goto err;
278 }
279
280 /* find nearest node with cpus and nearest memory */
281 for (router=NULL, j=0; j < op->ports; j++) {
282 dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
283 if (!dest || SN_HWPERF_FOREIGN(dest) ||
284 !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
285 continue;
286 }
287 c = sn_hwperf_obj_to_cnode(dest);
288 if (!found_cpu && sn_hwperf_has_cpus(c)) {
289 if (near_cpu_node)
290 *near_cpu_node = c;
291 found_cpu++;
292 }
293 if (!found_mem && sn_hwperf_has_mem(c)) {
294 if (near_mem_node)
295 *near_mem_node = c;
296 found_mem++;
297 }
298 if (SN_HWPERF_IS_ROUTER(dest))
299 router = dest;
300 }
301
302 if (router && (!found_cpu || !found_mem)) {
303 /* search for a node connected to the same router */
304 sz = router->ports * sizeof(struct sn_hwperf_port_info);
305 if (sz > sizeof(ptdata))
306 BUG();
307 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
308 SN_HWPERF_ENUM_PORTS, router->id, sz,
309 (u64)&ptdata, 0, 0, NULL);
310 if (e != SN_HWPERF_OP_OK) {
311 e = -EINVAL;
312 goto err;
313 }
314 for (j=0; j < router->ports; j++) {
315 dest = sn_hwperf_findobj_id(objbuf, nobj,
316 ptdata[j].conn_id);
317 if (!dest || dest->id == node ||
318 SN_HWPERF_FOREIGN(dest) ||
319 !SN_HWPERF_IS_NODE(dest) ||
320 SN_HWPERF_IS_IONODE(dest)) {
321 continue;
322 }
323 c = sn_hwperf_obj_to_cnode(dest);
324 if (!found_cpu && sn_hwperf_has_cpus(c)) {
325 if (near_cpu_node)
326 *near_cpu_node = c;
327 found_cpu++;
328 }
329 if (!found_mem && sn_hwperf_has_mem(c)) {
330 if (near_mem_node)
331 *near_mem_node = c;
332 found_mem++;
333 }
334 if (found_cpu && found_mem)
335 break;
336 }
337 }
338
339 if (!found_cpu || !found_mem) {
340 /* resort to _any_ node with CPUs and memory */
341 for (i=0, op=objbuf; i < nobj; i++, op++) {
342 if (SN_HWPERF_FOREIGN(op) ||
343 SN_HWPERF_IS_IONODE(op) ||
344 !SN_HWPERF_IS_NODE(op)) {
345 continue;
346 }
347 c = sn_hwperf_obj_to_cnode(op);
348 if (!found_cpu && sn_hwperf_has_cpus(c)) {
349 if (near_cpu_node)
350 *near_cpu_node = c;
351 found_cpu++;
352 }
353 if (!found_mem && sn_hwperf_has_mem(c)) {
354 if (near_mem_node)
355 *near_mem_node = c;
356 found_mem++;
357 }
358 if (found_cpu && found_mem)
191 break; 359 break;
192 *p2 = '\0';
193 seq_printf(s, "pcibus %d %s-%s\n",
194 *ordinal, obj->location, p1);
195 (*ordinal)++;
196 p1 = p2 + 1;
197 } 360 }
198 } 361 }
199 free_page((unsigned long)pg); 362
363 if (!found_cpu || !found_mem)
364 e = -ENODATA;
365
366err:
367 return e;
200} 368}
201 369
370
202static int sn_topology_show(struct seq_file *s, void *d) 371static int sn_topology_show(struct seq_file *s, void *d)
203{ 372{
204 int sz; 373 int sz;
@@ -215,7 +384,6 @@ static int sn_topology_show(struct seq_file *s, void *d)
215 struct sn_hwperf_object_info *p; 384 struct sn_hwperf_object_info *p;
216 struct sn_hwperf_object_info *obj = d; /* this object */ 385 struct sn_hwperf_object_info *obj = d; /* this object */
217 struct sn_hwperf_object_info *objs = s->private; /* all objects */ 386 struct sn_hwperf_object_info *objs = s->private; /* all objects */
218 int rack, bay, slot, slab;
219 u8 shubtype; 387 u8 shubtype;
220 u8 system_size; 388 u8 system_size;
221 u8 sharing_size; 389 u8 sharing_size;
@@ -225,7 +393,6 @@ static int sn_topology_show(struct seq_file *s, void *d)
225 u8 region_size; 393 u8 region_size;
226 u16 nasid_mask; 394 u16 nasid_mask;
227 int nasid_msb; 395 int nasid_msb;
228 int pci_bus_ordinal = 0;
229 396
230 if (obj == objs) { 397 if (obj == objs) {
231 seq_printf(s, "# sn_topology version 2\n"); 398 seq_printf(s, "# sn_topology version 2\n");
@@ -253,6 +420,8 @@ static int sn_topology_show(struct seq_file *s, void *d)
253 shubtype ? "shub2" : "shub1", 420 shubtype ? "shub2" : "shub1",
254 (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift, 421 (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
255 system_size, sharing_size, coher, region_size); 422 system_size, sharing_size, coher, region_size);
423
424 print_pci_topology(s);
256 } 425 }
257 426
258 if (SN_HWPERF_FOREIGN(obj)) { 427 if (SN_HWPERF_FOREIGN(obj)) {
@@ -272,11 +441,24 @@ static int sn_topology_show(struct seq_file *s, void *d)
272 if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj)) 441 if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
273 seq_putc(s, '\n'); 442 seq_putc(s, '\n');
274 else { 443 else {
444 cnodeid_t near_mem = -1;
445 cnodeid_t near_cpu = -1;
446
275 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal)); 447 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
276 for (i=0; i < numionodes; i++) { 448
277 seq_printf(s, i ? ":%d" : ", dist %d", 449 if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt,
278 node_distance(ordinal, i)); 450 ordinal, &near_mem, &near_cpu) == 0) {
451 seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d",
452 near_mem, near_cpu);
453 }
454
455 if (!SN_HWPERF_IS_IONODE(obj)) {
456 for_each_online_node(i) {
457 seq_printf(s, i ? ":%d" : ", dist %d",
458 node_distance(ordinal, i));
459 }
279 } 460 }
461
280 seq_putc(s, '\n'); 462 seq_putc(s, '\n');
281 463
282 /* 464 /*
@@ -300,17 +482,6 @@ static int sn_topology_show(struct seq_file *s, void *d)
300 seq_putc(s, '\n'); 482 seq_putc(s, '\n');
301 } 483 }
302 } 484 }
303
304 /*
305 * PCI busses attached to this node, if any
306 */
307 if (sn_hwperf_location_to_bpos(obj->location,
308 &rack, &bay, &slot, &slab)) {
309 /* export pci bus info */
310 print_pci_topology(s, obj, &pci_bus_ordinal,
311 rack, bay, slot, slab);
312
313 }
314 } 485 }
315 486
316 if (obj->ports) { 487 if (obj->ports) {
@@ -572,6 +743,8 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
572 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) { 743 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
573 memset(p, 0, a.sz); 744 memset(p, 0, a.sz);
574 for (i = 0; i < nobj; i++) { 745 for (i = 0; i < nobj; i++) {
746 if (!SN_HWPERF_IS_NODE(objs + i))
747 continue;
575 node = sn_hwperf_obj_to_cnode(objs + i); 748 node = sn_hwperf_obj_to_cnode(objs + i);
576 for_each_online_cpu(j) { 749 for_each_online_cpu(j) {
577 if (node != cpu_to_node(j)) 750 if (node != cpu_to_node(j))
@@ -598,7 +771,7 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
598 771
599 case SN_HWPERF_GET_NODE_NASID: 772 case SN_HWPERF_GET_NODE_NASID:
600 if (a.sz != sizeof(u64) || 773 if (a.sz != sizeof(u64) ||
601 (node = a.arg) < 0 || node >= numionodes) { 774 (node = a.arg) < 0 || !node_possible(node)) {
602 r = -EINVAL; 775 r = -EINVAL;
603 goto error; 776 goto error;
604 } 777 }
@@ -627,6 +800,14 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
627 vfree(objs); 800 vfree(objs);
628 goto error; 801 goto error;
629 } 802 }
803
804 if (!SN_HWPERF_IS_NODE(objs + i) &&
805 !SN_HWPERF_IS_IONODE(objs + i)) {
806 r = -ENOENT;
807 vfree(objs);
808 goto error;
809 }
810
630 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i); 811 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
631 vfree(objs); 812 vfree(objs);
632 } 813 }
@@ -692,6 +873,7 @@ static int sn_hwperf_init(void)
692 873
693 /* single threaded, once-only initialization */ 874 /* single threaded, once-only initialization */
694 down(&sn_hwperf_init_mutex); 875 down(&sn_hwperf_init_mutex);
876
695 if (sn_hwperf_salheap) { 877 if (sn_hwperf_salheap) {
696 up(&sn_hwperf_init_mutex); 878 up(&sn_hwperf_init_mutex);
697 return e; 879 return e;
@@ -742,19 +924,6 @@ out:
742 sn_hwperf_salheap = NULL; 924 sn_hwperf_salheap = NULL;
743 sn_hwperf_obj_cnt = 0; 925 sn_hwperf_obj_cnt = 0;
744 } 926 }
745
746 if (!e) {
747 /*
748 * Register a dynamic misc device for ioctl. Platforms
749 * supporting hotplug will create /dev/sn_hwperf, else
750 * user can to look up the minor number in /proc/misc.
751 */
752 if ((e = misc_register(&sn_hwperf_dev)) != 0) {
753 printk(KERN_ERR "sn_hwperf_init: misc register "
754 "for \"sn_hwperf\" failed, err %d\n", e);
755 }
756 }
757
758 up(&sn_hwperf_init_mutex); 927 up(&sn_hwperf_init_mutex);
759 return e; 928 return e;
760} 929}
@@ -782,3 +951,41 @@ int sn_topology_release(struct inode *inode, struct file *file)
782 vfree(seq->private); 951 vfree(seq->private);
783 return seq_release(inode, file); 952 return seq_release(inode, file);
784} 953}
954
955int sn_hwperf_get_nearest_node(cnodeid_t node,
956 cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
957{
958 int e;
959 int nobj;
960 struct sn_hwperf_object_info *objbuf;
961
962 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
963 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
964 node, near_mem_node, near_cpu_node);
965 vfree(objbuf);
966 }
967
968 return e;
969}
970
971static int __devinit sn_hwperf_misc_register_init(void)
972{
973 int e;
974
975 sn_hwperf_init();
976
977 /*
978 * Register a dynamic misc device for hwperf ioctls. Platforms
979 * supporting hotplug will create /dev/sn_hwperf, else user
980 * can to look up the minor number in /proc/misc.
981 */
982 if ((e = misc_register(&sn_hwperf_dev)) != 0) {
983 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
984 "register misc device for \"%s\"\n", sn_hwperf_dev.name);
985 }
986
987 return e;
988}
989
990device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
991EXPORT_SYMBOL(sn_hwperf_get_nearest_node);
diff --git a/arch/ia64/sn/kernel/sn2/sn_proc_fs.c b/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
index 266a3a84c01d..a06719d752a0 100644
--- a/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
+++ b/arch/ia64/sn/kernel/sn2/sn_proc_fs.c
@@ -3,7 +3,7 @@
3 * License. See the file "COPYING" in the main directory of this archive 3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details. 4 * for more details.
5 * 5 *
6 * Copyright (C) 2000-2004 Silicon Graphics, Inc. All rights reserved. 6 * Copyright (C) 2000-2005 Silicon Graphics, Inc. All rights reserved.
7 */ 7 */
8#include <linux/config.h> 8#include <linux/config.h>
9#include <asm/uaccess.h> 9#include <asm/uaccess.h>
@@ -15,7 +15,7 @@
15 15
16static int partition_id_show(struct seq_file *s, void *p) 16static int partition_id_show(struct seq_file *s, void *p)
17{ 17{
18 seq_printf(s, "%d\n", sn_local_partid()); 18 seq_printf(s, "%d\n", sn_partition_id);
19 return 0; 19 return 0;
20} 20}
21 21
diff --git a/arch/ia64/sn/kernel/sn2/timer_interrupt.c b/arch/ia64/sn/kernel/sn2/timer_interrupt.c
index cde7375390b0..adf5db2e2afe 100644
--- a/arch/ia64/sn/kernel/sn2/timer_interrupt.c
+++ b/arch/ia64/sn/kernel/sn2/timer_interrupt.c
@@ -1,7 +1,7 @@
1/* 1/*
2 * 2 *
3 * 3 *
4 * Copyright (c) 2003 Silicon Graphics, Inc. All Rights Reserved. 4 * Copyright (c) 2005 Silicon Graphics, Inc. All Rights Reserved.
5 * 5 *
6 * This program is free software; you can redistribute it and/or modify it 6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2 of the GNU General Public License 7 * under the terms of version 2 of the GNU General Public License
@@ -50,14 +50,16 @@ void sn_timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
50 LED_CPU_HEARTBEAT, LED_CPU_HEARTBEAT); 50 LED_CPU_HEARTBEAT, LED_CPU_HEARTBEAT);
51 } 51 }
52 52
53 if (enable_shub_wars_1_1()) { 53 if (is_shub1()) {
54 /* Bugfix code for SHUB 1.1 */ 54 if (enable_shub_wars_1_1()) {
55 if (pda->pio_shub_war_cam_addr) 55 /* Bugfix code for SHUB 1.1 */
56 *pda->pio_shub_war_cam_addr = 0x8000000000000010UL; 56 if (pda->pio_shub_war_cam_addr)
57 *pda->pio_shub_war_cam_addr = 0x8000000000000010UL;
58 }
59 if (pda->sn_lb_int_war_ticks == 0)
60 sn_lb_int_war_check();
61 pda->sn_lb_int_war_ticks++;
62 if (pda->sn_lb_int_war_ticks >= SN_LB_INT_WAR_INTERVAL)
63 pda->sn_lb_int_war_ticks = 0;
57 } 64 }
58 if (pda->sn_lb_int_war_ticks == 0)
59 sn_lb_int_war_check();
60 pda->sn_lb_int_war_ticks++;
61 if (pda->sn_lb_int_war_ticks >= SN_LB_INT_WAR_INTERVAL)
62 pda->sn_lb_int_war_ticks = 0;
63} 65}
diff --git a/arch/ia64/sn/pci/Makefile b/arch/ia64/sn/pci/Makefile
index 2f915bce25f9..321576b1b425 100644
--- a/arch/ia64/sn/pci/Makefile
+++ b/arch/ia64/sn/pci/Makefile
@@ -7,4 +7,4 @@
7# 7#
8# Makefile for the sn pci general routines. 8# Makefile for the sn pci general routines.
9 9
10obj-y := pci_dma.o tioca_provider.o pcibr/ 10obj-y := pci_dma.o tioca_provider.o tioce_provider.o pcibr/
diff --git a/arch/ia64/sn/pci/pcibr/pcibr_dma.c b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
index b058dc2a0b9d..34093476e965 100644
--- a/arch/ia64/sn/pci/pcibr/pcibr_dma.c
+++ b/arch/ia64/sn/pci/pcibr/pcibr_dma.c
@@ -3,7 +3,7 @@
3 * License. See the file "COPYING" in the main directory of this archive 3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details. 4 * for more details.
5 * 5 *
6 * Copyright (C) 2001-2004 Silicon Graphics, Inc. All rights reserved. 6 * Copyright (C) 2001-2005 Silicon Graphics, Inc. All rights reserved.
7 */ 7 */
8 8
9#include <linux/types.h> 9#include <linux/types.h>
@@ -215,8 +215,8 @@ void sn_dma_flush(uint64_t addr)
215 int is_tio; 215 int is_tio;
216 int wid_num; 216 int wid_num;
217 int i, j; 217 int i, j;
218 int bwin;
219 uint64_t flags; 218 uint64_t flags;
219 uint64_t itte;
220 struct hubdev_info *hubinfo; 220 struct hubdev_info *hubinfo;
221 volatile struct sn_flush_device_list *p; 221 volatile struct sn_flush_device_list *p;
222 struct sn_flush_nasid_entry *flush_nasid_list; 222 struct sn_flush_nasid_entry *flush_nasid_list;
@@ -233,31 +233,36 @@ void sn_dma_flush(uint64_t addr)
233 if (!hubinfo) { 233 if (!hubinfo) {
234 BUG(); 234 BUG();
235 } 235 }
236 is_tio = (nasid & 1);
237 if (is_tio) {
238 wid_num = TIO_SWIN_WIDGETNUM(addr);
239 bwin = TIO_BWIN_WINDOWNUM(addr);
240 } else {
241 wid_num = SWIN_WIDGETNUM(addr);
242 bwin = BWIN_WINDOWNUM(addr);
243 }
244 236
245 flush_nasid_list = &hubinfo->hdi_flush_nasid_list; 237 flush_nasid_list = &hubinfo->hdi_flush_nasid_list;
246 if (flush_nasid_list->widget_p == NULL) 238 if (flush_nasid_list->widget_p == NULL)
247 return; 239 return;
248 if (bwin > 0) {
249 uint64_t itte = flush_nasid_list->iio_itte[bwin];
250 240
251 if (is_tio) { 241 is_tio = (nasid & 1);
252 wid_num = (itte >> TIO_ITTE_WIDGET_SHIFT) & 242 if (is_tio) {
253 TIO_ITTE_WIDGET_MASK; 243 int itte_index;
254 } else { 244
255 wid_num = (itte >> IIO_ITTE_WIDGET_SHIFT) & 245 if (TIO_HWIN(addr))
256 IIO_ITTE_WIDGET_MASK; 246 itte_index = 0;
257 } 247 else if (TIO_BWIN_WINDOWNUM(addr))
248 itte_index = TIO_BWIN_WINDOWNUM(addr);
249 else
250 itte_index = -1;
251
252 if (itte_index >= 0) {
253 itte = flush_nasid_list->iio_itte[itte_index];
254 if (! TIO_ITTE_VALID(itte))
255 return;
256 wid_num = TIO_ITTE_WIDGET(itte);
257 } else
258 wid_num = TIO_SWIN_WIDGETNUM(addr);
259 } else {
260 if (BWIN_WINDOWNUM(addr)) {
261 itte = flush_nasid_list->iio_itte[BWIN_WINDOWNUM(addr)];
262 wid_num = IIO_ITTE_WIDGET(itte);
263 } else
264 wid_num = SWIN_WIDGETNUM(addr);
258 } 265 }
259 if (flush_nasid_list->widget_p == NULL)
260 return;
261 if (flush_nasid_list->widget_p[wid_num] == NULL) 266 if (flush_nasid_list->widget_p[wid_num] == NULL)
262 return; 267 return;
263 p = &flush_nasid_list->widget_p[wid_num][0]; 268 p = &flush_nasid_list->widget_p[wid_num][0];
@@ -283,10 +288,16 @@ void sn_dma_flush(uint64_t addr)
283 /* 288 /*
284 * For TIOCP use the Device(x) Write Request Buffer Flush Bridge 289 * For TIOCP use the Device(x) Write Request Buffer Flush Bridge
285 * register since it ensures the data has entered the coherence 290 * register since it ensures the data has entered the coherence
286 * domain, unlike PIC 291 * domain, unlike PIC.
287 */ 292 */
288 if (is_tio) { 293 if (is_tio) {
289 uint32_t tio_id = REMOTE_HUB_L(nasid, TIO_NODE_ID); 294 /*
295 * Note: devices behind TIOCE should never be matched in the
296 * above code, and so the following code is PIC/CP centric.
297 * If CE ever needs the sn_dma_flush mechanism, we will have
298 * to account for that here and in tioce_bus_fixup().
299 */
300 uint32_t tio_id = HUB_L(TIO_IOSPACE_ADDR(nasid, TIO_NODE_ID));
290 uint32_t revnum = XWIDGET_PART_REV_NUM(tio_id); 301 uint32_t revnum = XWIDGET_PART_REV_NUM(tio_id);
291 302
292 /* TIOCP BRINGUP WAR (PV907516): Don't write buffer flush reg */ 303 /* TIOCP BRINGUP WAR (PV907516): Don't write buffer flush reg */
@@ -306,7 +317,8 @@ void sn_dma_flush(uint64_t addr)
306 *(volatile uint32_t *)(p->sfdl_force_int_addr) = 1; 317 *(volatile uint32_t *)(p->sfdl_force_int_addr) = 1;
307 318
308 /* wait for the interrupt to come back. */ 319 /* wait for the interrupt to come back. */
309 while (*(p->sfdl_flush_addr) != 0x10f) ; 320 while (*(p->sfdl_flush_addr) != 0x10f)
321 cpu_relax();
310 322
311 /* okay, everything is synched up. */ 323 /* okay, everything is synched up. */
312 spin_unlock_irqrestore((spinlock_t *)&p->sfdl_flush_lock, flags); 324 spin_unlock_irqrestore((spinlock_t *)&p->sfdl_flush_lock, flags);
diff --git a/arch/ia64/sn/pci/pcibr/pcibr_provider.c b/arch/ia64/sn/pci/pcibr/pcibr_provider.c
index b95e928636a1..7b03b8084ffc 100644
--- a/arch/ia64/sn/pci/pcibr/pcibr_provider.c
+++ b/arch/ia64/sn/pci/pcibr/pcibr_provider.c
@@ -15,6 +15,7 @@
15#include <asm/sn/pcibus_provider_defs.h> 15#include <asm/sn/pcibus_provider_defs.h>
16#include <asm/sn/pcidev.h> 16#include <asm/sn/pcidev.h>
17#include <asm/sn/sn_sal.h> 17#include <asm/sn/sn_sal.h>
18#include <asm/sn/sn2/sn_hwperf.h>
18#include "xtalk/xwidgetdev.h" 19#include "xtalk/xwidgetdev.h"
19#include "xtalk/hubdev.h" 20#include "xtalk/hubdev.h"
20 21
@@ -60,7 +61,7 @@ static int sal_pcibr_error_interrupt(struct pcibus_info *soft)
60 ret_stuff.status = 0; 61 ret_stuff.status = 0;
61 ret_stuff.v0 = 0; 62 ret_stuff.v0 = 0;
62 63
63 segment = 0; 64 segment = soft->pbi_buscommon.bs_persist_segment;
64 busnum = soft->pbi_buscommon.bs_persist_busnum; 65 busnum = soft->pbi_buscommon.bs_persist_busnum;
65 SAL_CALL_NOLOCK(ret_stuff, 66 SAL_CALL_NOLOCK(ret_stuff,
66 (u64) SN_SAL_IOIF_ERROR_INTERRUPT, 67 (u64) SN_SAL_IOIF_ERROR_INTERRUPT,
@@ -88,6 +89,7 @@ void *
88pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *controller) 89pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *controller)
89{ 90{
90 int nasid, cnode, j; 91 int nasid, cnode, j;
92 cnodeid_t near_cnode;
91 struct hubdev_info *hubdev_info; 93 struct hubdev_info *hubdev_info;
92 struct pcibus_info *soft; 94 struct pcibus_info *soft;
93 struct sn_flush_device_list *sn_flush_device_list; 95 struct sn_flush_device_list *sn_flush_device_list;
@@ -115,7 +117,7 @@ pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
115 /* 117 /*
116 * register the bridge's error interrupt handler 118 * register the bridge's error interrupt handler
117 */ 119 */
118 if (request_irq(SGI_PCIBR_ERROR, (void *)pcibr_error_intr_handler, 120 if (request_irq(SGI_PCIASIC_ERROR, (void *)pcibr_error_intr_handler,
119 SA_SHIRQ, "PCIBR error", (void *)(soft))) { 121 SA_SHIRQ, "PCIBR error", (void *)(soft))) {
120 printk(KERN_WARNING 122 printk(KERN_WARNING
121 "pcibr cannot allocate interrupt for error handler\n"); 123 "pcibr cannot allocate interrupt for error handler\n");
@@ -142,9 +144,12 @@ pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
142 j++, sn_flush_device_list++) { 144 j++, sn_flush_device_list++) {
143 if (sn_flush_device_list->sfdl_slot == -1) 145 if (sn_flush_device_list->sfdl_slot == -1)
144 continue; 146 continue;
145 if (sn_flush_device_list-> 147 if ((sn_flush_device_list->
146 sfdl_persistent_busnum == 148 sfdl_persistent_segment ==
147 soft->pbi_buscommon.bs_persist_busnum) 149 soft->pbi_buscommon.bs_persist_segment) &&
150 (sn_flush_device_list->
151 sfdl_persistent_busnum ==
152 soft->pbi_buscommon.bs_persist_busnum))
148 sn_flush_device_list->sfdl_pcibus_info = 153 sn_flush_device_list->sfdl_pcibus_info =
149 soft; 154 soft;
150 } 155 }
@@ -158,12 +163,18 @@ pcibr_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
158 memset(soft->pbi_int_ate_resource.ate, 0, 163 memset(soft->pbi_int_ate_resource.ate, 0,
159 (soft->pbi_int_ate_size * sizeof(uint64_t))); 164 (soft->pbi_int_ate_size * sizeof(uint64_t)));
160 165
161 if (prom_bussoft->bs_asic_type == PCIIO_ASIC_TYPE_TIOCP) 166 if (prom_bussoft->bs_asic_type == PCIIO_ASIC_TYPE_TIOCP) {
162 /* 167 /* TIO PCI Bridge: find nearest node with CPUs */
163 * TIO PCI Bridge with no closest node information. 168 int e = sn_hwperf_get_nearest_node(cnode, NULL, &near_cnode);
164 * FIXME: Find another way to determine the closest node 169
165 */ 170 if (e < 0) {
166 controller->node = -1; 171 near_cnode = (cnodeid_t)-1; /* use any node */
172 printk(KERN_WARNING "pcibr_bus_fixup: failed to find "
173 "near node with CPUs to TIO node %d, err=%d\n",
174 cnode, e);
175 }
176 controller->node = near_cnode;
177 }
167 else 178 else
168 controller->node = cnode; 179 controller->node = cnode;
169 return soft; 180 return soft;
@@ -175,6 +186,9 @@ void pcibr_force_interrupt(struct sn_irq_info *sn_irq_info)
175 struct pcibus_info *pcibus_info; 186 struct pcibus_info *pcibus_info;
176 int bit = sn_irq_info->irq_int_bit; 187 int bit = sn_irq_info->irq_int_bit;
177 188
189 if (! sn_irq_info->irq_bridge)
190 return;
191
178 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo; 192 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
179 if (pcidev_info) { 193 if (pcidev_info) {
180 pcibus_info = 194 pcibus_info =
@@ -184,7 +198,7 @@ void pcibr_force_interrupt(struct sn_irq_info *sn_irq_info)
184 } 198 }
185} 199}
186 200
187void pcibr_change_devices_irq(struct sn_irq_info *sn_irq_info) 201void pcibr_target_interrupt(struct sn_irq_info *sn_irq_info)
188{ 202{
189 struct pcidev_info *pcidev_info; 203 struct pcidev_info *pcidev_info;
190 struct pcibus_info *pcibus_info; 204 struct pcibus_info *pcibus_info;
@@ -219,6 +233,8 @@ struct sn_pcibus_provider pcibr_provider = {
219 .dma_map_consistent = pcibr_dma_map_consistent, 233 .dma_map_consistent = pcibr_dma_map_consistent,
220 .dma_unmap = pcibr_dma_unmap, 234 .dma_unmap = pcibr_dma_unmap,
221 .bus_fixup = pcibr_bus_fixup, 235 .bus_fixup = pcibr_bus_fixup,
236 .force_interrupt = pcibr_force_interrupt,
237 .target_interrupt = pcibr_target_interrupt
222}; 238};
223 239
224int 240int
diff --git a/arch/ia64/sn/pci/tioca_provider.c b/arch/ia64/sn/pci/tioca_provider.c
index 5d76a7581465..ea09c12f0258 100644
--- a/arch/ia64/sn/pci/tioca_provider.c
+++ b/arch/ia64/sn/pci/tioca_provider.c
@@ -559,7 +559,7 @@ tioca_error_intr_handler(int irq, void *arg, struct pt_regs *pt)
559 ret_stuff.status = 0; 559 ret_stuff.status = 0;
560 ret_stuff.v0 = 0; 560 ret_stuff.v0 = 0;
561 561
562 segment = 0; 562 segment = soft->ca_common.bs_persist_segment;
563 busnum = soft->ca_common.bs_persist_busnum; 563 busnum = soft->ca_common.bs_persist_busnum;
564 564
565 SAL_CALL_NOLOCK(ret_stuff, 565 SAL_CALL_NOLOCK(ret_stuff,
@@ -622,7 +622,8 @@ tioca_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *cont
622 nasid_to_cnodeid(tioca_common->ca_closest_nasid); 622 nasid_to_cnodeid(tioca_common->ca_closest_nasid);
623 tioca_common->ca_kernel_private = (uint64_t) tioca_kern; 623 tioca_common->ca_kernel_private = (uint64_t) tioca_kern;
624 624
625 bus = pci_find_bus(0, tioca_common->ca_common.bs_persist_busnum); 625 bus = pci_find_bus(tioca_common->ca_common.bs_persist_segment,
626 tioca_common->ca_common.bs_persist_busnum);
626 BUG_ON(!bus); 627 BUG_ON(!bus);
627 tioca_kern->ca_devices = &bus->devices; 628 tioca_kern->ca_devices = &bus->devices;
628 629
@@ -656,6 +657,8 @@ static struct sn_pcibus_provider tioca_pci_interfaces = {
656 .dma_map_consistent = tioca_dma_map, 657 .dma_map_consistent = tioca_dma_map,
657 .dma_unmap = tioca_dma_unmap, 658 .dma_unmap = tioca_dma_unmap,
658 .bus_fixup = tioca_bus_fixup, 659 .bus_fixup = tioca_bus_fixup,
660 .force_interrupt = NULL,
661 .target_interrupt = NULL
659}; 662};
660 663
661/** 664/**
diff --git a/arch/ia64/sn/pci/tioce_provider.c b/arch/ia64/sn/pci/tioce_provider.c
new file mode 100644
index 000000000000..8e75db2b825d
--- /dev/null
+++ b/arch/ia64/sn/pci/tioce_provider.c
@@ -0,0 +1,771 @@
1/*
2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
4 * for more details.
5 *
6 * Copyright (C) 2003-2005 Silicon Graphics, Inc. All Rights Reserved.
7 */
8
9#include <linux/types.h>
10#include <linux/interrupt.h>
11#include <linux/pci.h>
12#include <asm/sn/sn_sal.h>
13#include <asm/sn/addrs.h>
14#include <asm/sn/pcidev.h>
15#include <asm/sn/pcibus_provider_defs.h>
16#include <asm/sn/tioce_provider.h>
17
18/**
19 * Bus address ranges for the 5 flavors of TIOCE DMA
20 */
21
22#define TIOCE_D64_MIN 0x8000000000000000UL
23#define TIOCE_D64_MAX 0xffffffffffffffffUL
24#define TIOCE_D64_ADDR(a) ((a) >= TIOCE_D64_MIN)
25
26#define TIOCE_D32_MIN 0x0000000080000000UL
27#define TIOCE_D32_MAX 0x00000000ffffffffUL
28#define TIOCE_D32_ADDR(a) ((a) >= TIOCE_D32_MIN && (a) <= TIOCE_D32_MAX)
29
30#define TIOCE_M32_MIN 0x0000000000000000UL
31#define TIOCE_M32_MAX 0x000000007fffffffUL
32#define TIOCE_M32_ADDR(a) ((a) >= TIOCE_M32_MIN && (a) <= TIOCE_M32_MAX)
33
34#define TIOCE_M40_MIN 0x0000004000000000UL
35#define TIOCE_M40_MAX 0x0000007fffffffffUL
36#define TIOCE_M40_ADDR(a) ((a) >= TIOCE_M40_MIN && (a) <= TIOCE_M40_MAX)
37
38#define TIOCE_M40S_MIN 0x0000008000000000UL
39#define TIOCE_M40S_MAX 0x000000ffffffffffUL
40#define TIOCE_M40S_ADDR(a) ((a) >= TIOCE_M40S_MIN && (a) <= TIOCE_M40S_MAX)
41
42/*
43 * ATE manipulation macros.
44 */
45
46#define ATE_PAGESHIFT(ps) (__ffs(ps))
47#define ATE_PAGEMASK(ps) ((ps)-1)
48
49#define ATE_PAGE(x, ps) ((x) >> ATE_PAGESHIFT(ps))
50#define ATE_NPAGES(start, len, pagesize) \
51 (ATE_PAGE((start)+(len)-1, pagesize) - ATE_PAGE(start, pagesize) + 1)
52
53#define ATE_VALID(ate) ((ate) & (1UL << 63))
54#define ATE_MAKE(addr, ps) (((addr) & ~ATE_PAGEMASK(ps)) | (1UL << 63))
55
56/*
57 * Flavors of ate-based mapping supported by tioce_alloc_map()
58 */
59
60#define TIOCE_ATE_M32 1
61#define TIOCE_ATE_M40 2
62#define TIOCE_ATE_M40S 3
63
64#define KB(x) ((x) << 10)
65#define MB(x) ((x) << 20)
66#define GB(x) ((x) << 30)
67
68/**
69 * tioce_dma_d64 - create a DMA mapping using 64-bit direct mode
70 * @ct_addr: system coretalk address
71 *
72 * Map @ct_addr into 64-bit CE bus space. No device context is necessary
73 * and no CE mapping are consumed.
74 *
75 * Bits 53:0 come from the coretalk address. The remaining bits are set as
76 * follows:
77 *
78 * 63 - must be 1 to indicate d64 mode to CE hardware
79 * 62 - barrier bit ... controlled with tioce_dma_barrier()
80 * 61 - 0 since this is not an MSI transaction
81 * 60:54 - reserved, MBZ
82 */
83static uint64_t
84tioce_dma_d64(unsigned long ct_addr)
85{
86 uint64_t bus_addr;
87
88 bus_addr = ct_addr | (1UL << 63);
89
90 return bus_addr;
91}
92
93/**
94 * pcidev_to_tioce - return misc ce related pointers given a pci_dev
95 * @pci_dev: pci device context
96 * @base: ptr to store struct tioce_mmr * for the CE holding this device
97 * @kernel: ptr to store struct tioce_kernel * for the CE holding this device
98 * @port: ptr to store the CE port number that this device is on
99 *
100 * Return pointers to various CE-related structures for the CE upstream of
101 * @pci_dev.
102 */
103static inline void
104pcidev_to_tioce(struct pci_dev *pdev, struct tioce **base,
105 struct tioce_kernel **kernel, int *port)
106{
107 struct pcidev_info *pcidev_info;
108 struct tioce_common *ce_common;
109 struct tioce_kernel *ce_kernel;
110
111 pcidev_info = SN_PCIDEV_INFO(pdev);
112 ce_common = (struct tioce_common *)pcidev_info->pdi_pcibus_info;
113 ce_kernel = (struct tioce_kernel *)ce_common->ce_kernel_private;
114
115 if (base)
116 *base = (struct tioce *)ce_common->ce_pcibus.bs_base;
117 if (kernel)
118 *kernel = ce_kernel;
119
120 /*
121 * we use port as a zero-based value internally, even though the
122 * documentation is 1-based.
123 */
124 if (port)
125 *port =
126 (pdev->bus->number < ce_kernel->ce_port1_secondary) ? 0 : 1;
127}
128
129/**
130 * tioce_alloc_map - Given a coretalk address, map it to pcie bus address
131 * space using one of the various ATE-based address modes.
132 * @ce_kern: tioce context
133 * @type: map mode to use
134 * @port: 0-based port that the requesting device is downstream of
135 * @ct_addr: the coretalk address to map
136 * @len: number of bytes to map
137 *
138 * Given the addressing type, set up various paramaters that define the
139 * ATE pool to use. Search for a contiguous block of entries to cover the
140 * length, and if enough resources exist, fill in the ATE's and construct a
141 * tioce_dmamap struct to track the mapping.
142 */
143static uint64_t
144tioce_alloc_map(struct tioce_kernel *ce_kern, int type, int port,
145 uint64_t ct_addr, int len)
146{
147 int i;
148 int j;
149 int first;
150 int last;
151 int entries;
152 int nates;
153 int pagesize;
154 uint64_t *ate_shadow;
155 uint64_t *ate_reg;
156 uint64_t addr;
157 struct tioce *ce_mmr;
158 uint64_t bus_base;
159 struct tioce_dmamap *map;
160
161 ce_mmr = (struct tioce *)ce_kern->ce_common->ce_pcibus.bs_base;
162
163 switch (type) {
164 case TIOCE_ATE_M32:
165 /*
166 * The first 64 entries of the ate3240 pool are dedicated to
167 * super-page (TIOCE_ATE_M40S) mode.
168 */
169 first = 64;
170 entries = TIOCE_NUM_M3240_ATES - 64;
171 ate_shadow = ce_kern->ce_ate3240_shadow;
172 ate_reg = ce_mmr->ce_ure_ate3240;
173 pagesize = ce_kern->ce_ate3240_pagesize;
174 bus_base = TIOCE_M32_MIN;
175 break;
176 case TIOCE_ATE_M40:
177 first = 0;
178 entries = TIOCE_NUM_M40_ATES;
179 ate_shadow = ce_kern->ce_ate40_shadow;
180 ate_reg = ce_mmr->ce_ure_ate40;
181 pagesize = MB(64);
182 bus_base = TIOCE_M40_MIN;
183 break;
184 case TIOCE_ATE_M40S:
185 /*
186 * ate3240 entries 0-31 are dedicated to port1 super-page
187 * mappings. ate3240 entries 32-63 are dedicated to port2.
188 */
189 first = port * 32;
190 entries = 32;
191 ate_shadow = ce_kern->ce_ate3240_shadow;
192 ate_reg = ce_mmr->ce_ure_ate3240;
193 pagesize = GB(16);
194 bus_base = TIOCE_M40S_MIN;
195 break;
196 default:
197 return 0;
198 }
199
200 nates = ATE_NPAGES(ct_addr, len, pagesize);
201 if (nates > entries)
202 return 0;
203
204 last = first + entries - nates;
205 for (i = first; i <= last; i++) {
206 if (ATE_VALID(ate_shadow[i]))
207 continue;
208
209 for (j = i; j < i + nates; j++)
210 if (ATE_VALID(ate_shadow[j]))
211 break;
212
213 if (j >= i + nates)
214 break;
215 }
216
217 if (i > last)
218 return 0;
219
220 map = kcalloc(1, sizeof(struct tioce_dmamap), GFP_ATOMIC);
221 if (!map)
222 return 0;
223
224 addr = ct_addr;
225 for (j = 0; j < nates; j++) {
226 uint64_t ate;
227
228 ate = ATE_MAKE(addr, pagesize);
229 ate_shadow[i + j] = ate;
230 ate_reg[i + j] = ate;
231 addr += pagesize;
232 }
233
234 map->refcnt = 1;
235 map->nbytes = nates * pagesize;
236 map->ct_start = ct_addr & ~ATE_PAGEMASK(pagesize);
237 map->pci_start = bus_base + (i * pagesize);
238 map->ate_hw = &ate_reg[i];
239 map->ate_shadow = &ate_shadow[i];
240 map->ate_count = nates;
241
242 list_add(&map->ce_dmamap_list, &ce_kern->ce_dmamap_list);
243
244 return (map->pci_start + (ct_addr - map->ct_start));
245}
246
247/**
248 * tioce_dma_d32 - create a DMA mapping using 32-bit direct mode
249 * @pdev: linux pci_dev representing the function
250 * @paddr: system physical address
251 *
252 * Map @paddr into 32-bit bus space of the CE associated with @pcidev_info.
253 */
254static uint64_t
255tioce_dma_d32(struct pci_dev *pdev, uint64_t ct_addr)
256{
257 int dma_ok;
258 int port;
259 struct tioce *ce_mmr;
260 struct tioce_kernel *ce_kern;
261 uint64_t ct_upper;
262 uint64_t ct_lower;
263 dma_addr_t bus_addr;
264
265 ct_upper = ct_addr & ~0x3fffffffUL;
266 ct_lower = ct_addr & 0x3fffffffUL;
267
268 pcidev_to_tioce(pdev, &ce_mmr, &ce_kern, &port);
269
270 if (ce_kern->ce_port[port].dirmap_refcnt == 0) {
271 volatile uint64_t tmp;
272
273 ce_kern->ce_port[port].dirmap_shadow = ct_upper;
274 ce_mmr->ce_ure_dir_map[port] = ct_upper;
275 tmp = ce_mmr->ce_ure_dir_map[port];
276 dma_ok = 1;
277 } else
278 dma_ok = (ce_kern->ce_port[port].dirmap_shadow == ct_upper);
279
280 if (dma_ok) {
281 ce_kern->ce_port[port].dirmap_refcnt++;
282 bus_addr = TIOCE_D32_MIN + ct_lower;
283 } else
284 bus_addr = 0;
285
286 return bus_addr;
287}
288
289/**
290 * tioce_dma_barrier - swizzle a TIOCE bus address to include or exclude
291 * the barrier bit.
292 * @bus_addr: bus address to swizzle
293 *
294 * Given a TIOCE bus address, set the appropriate bit to indicate barrier
295 * attributes.
296 */
297static uint64_t
298tioce_dma_barrier(uint64_t bus_addr, int on)
299{
300 uint64_t barrier_bit;
301
302 /* barrier not supported in M40/M40S mode */
303 if (TIOCE_M40_ADDR(bus_addr) || TIOCE_M40S_ADDR(bus_addr))
304 return bus_addr;
305
306 if (TIOCE_D64_ADDR(bus_addr))
307 barrier_bit = (1UL << 62);
308 else /* must be m32 or d32 */
309 barrier_bit = (1UL << 30);
310
311 return (on) ? (bus_addr | barrier_bit) : (bus_addr & ~barrier_bit);
312}
313
314/**
315 * tioce_dma_unmap - release CE mapping resources
316 * @pdev: linux pci_dev representing the function
317 * @bus_addr: bus address returned by an earlier tioce_dma_map
318 * @dir: mapping direction (unused)
319 *
320 * Locate mapping resources associated with @bus_addr and release them.
321 * For mappings created using the direct modes there are no resources
322 * to release.
323 */
324void
325tioce_dma_unmap(struct pci_dev *pdev, dma_addr_t bus_addr, int dir)
326{
327 int i;
328 int port;
329 struct tioce_kernel *ce_kern;
330 struct tioce *ce_mmr;
331 unsigned long flags;
332
333 bus_addr = tioce_dma_barrier(bus_addr, 0);
334 pcidev_to_tioce(pdev, &ce_mmr, &ce_kern, &port);
335
336 /* nothing to do for D64 */
337
338 if (TIOCE_D64_ADDR(bus_addr))
339 return;
340
341 spin_lock_irqsave(&ce_kern->ce_lock, flags);
342
343 if (TIOCE_D32_ADDR(bus_addr)) {
344 if (--ce_kern->ce_port[port].dirmap_refcnt == 0) {
345 ce_kern->ce_port[port].dirmap_shadow = 0;
346 ce_mmr->ce_ure_dir_map[port] = 0;
347 }
348 } else {
349 struct tioce_dmamap *map;
350
351 list_for_each_entry(map, &ce_kern->ce_dmamap_list,
352 ce_dmamap_list) {
353 uint64_t last;
354
355 last = map->pci_start + map->nbytes - 1;
356 if (bus_addr >= map->pci_start && bus_addr <= last)
357 break;
358 }
359
360 if (&map->ce_dmamap_list == &ce_kern->ce_dmamap_list) {
361 printk(KERN_WARNING
362 "%s: %s - no map found for bus_addr 0x%lx\n",
363 __FUNCTION__, pci_name(pdev), bus_addr);
364 } else if (--map->refcnt == 0) {
365 for (i = 0; i < map->ate_count; i++) {
366 map->ate_shadow[i] = 0;
367 map->ate_hw[i] = 0;
368 }
369
370 list_del(&map->ce_dmamap_list);
371 kfree(map);
372 }
373 }
374
375 spin_unlock_irqrestore(&ce_kern->ce_lock, flags);
376}
377
378/**
379 * tioce_do_dma_map - map pages for PCI DMA
380 * @pdev: linux pci_dev representing the function
381 * @paddr: host physical address to map
382 * @byte_count: bytes to map
383 *
384 * This is the main wrapper for mapping host physical pages to CE PCI space.
385 * The mapping mode used is based on the device's dma_mask.
386 */
387static uint64_t
388tioce_do_dma_map(struct pci_dev *pdev, uint64_t paddr, size_t byte_count,
389 int barrier)
390{
391 unsigned long flags;
392 uint64_t ct_addr;
393 uint64_t mapaddr = 0;
394 struct tioce_kernel *ce_kern;
395 struct tioce_dmamap *map;
396 int port;
397 uint64_t dma_mask;
398
399 dma_mask = (barrier) ? pdev->dev.coherent_dma_mask : pdev->dma_mask;
400
401 /* cards must be able to address at least 31 bits */
402 if (dma_mask < 0x7fffffffUL)
403 return 0;
404
405 ct_addr = PHYS_TO_TIODMA(paddr);
406
407 /*
408 * If the device can generate 64 bit addresses, create a D64 map.
409 * Since this should never fail, bypass the rest of the checks.
410 */
411 if (dma_mask == ~0UL) {
412 mapaddr = tioce_dma_d64(ct_addr);
413 goto dma_map_done;
414 }
415
416 pcidev_to_tioce(pdev, NULL, &ce_kern, &port);
417
418 spin_lock_irqsave(&ce_kern->ce_lock, flags);
419
420 /*
421 * D64 didn't work ... See if we have an existing map that covers
422 * this address range. Must account for devices dma_mask here since
423 * an existing map might have been done in a mode using more pci
424 * address bits than this device can support.
425 */
426 list_for_each_entry(map, &ce_kern->ce_dmamap_list, ce_dmamap_list) {
427 uint64_t last;
428
429 last = map->ct_start + map->nbytes - 1;
430 if (ct_addr >= map->ct_start &&
431 ct_addr + byte_count - 1 <= last &&
432 map->pci_start <= dma_mask) {
433 map->refcnt++;
434 mapaddr = map->pci_start + (ct_addr - map->ct_start);
435 break;
436 }
437 }
438
439 /*
440 * If we don't have a map yet, and the card can generate 40
441 * bit addresses, try the M40/M40S modes. Note these modes do not
442 * support a barrier bit, so if we need a consistent map these
443 * won't work.
444 */
445 if (!mapaddr && !barrier && dma_mask >= 0xffffffffffUL) {
446 /*
447 * We have two options for 40-bit mappings: 16GB "super" ATE's
448 * and 64MB "regular" ATE's. We'll try both if needed for a
449 * given mapping but which one we try first depends on the
450 * size. For requests >64MB, prefer to use a super page with
451 * regular as the fallback. Otherwise, try in the reverse order.
452 */
453
454 if (byte_count > MB(64)) {
455 mapaddr = tioce_alloc_map(ce_kern, TIOCE_ATE_M40S,
456 port, ct_addr, byte_count);
457 if (!mapaddr)
458 mapaddr =
459 tioce_alloc_map(ce_kern, TIOCE_ATE_M40, -1,
460 ct_addr, byte_count);
461 } else {
462 mapaddr = tioce_alloc_map(ce_kern, TIOCE_ATE_M40, -1,
463 ct_addr, byte_count);
464 if (!mapaddr)
465 mapaddr =
466 tioce_alloc_map(ce_kern, TIOCE_ATE_M40S,
467 port, ct_addr, byte_count);
468 }
469 }
470
471 /*
472 * 32-bit direct is the next mode to try
473 */
474 if (!mapaddr && dma_mask >= 0xffffffffUL)
475 mapaddr = tioce_dma_d32(pdev, ct_addr);
476
477 /*
478 * Last resort, try 32-bit ATE-based map.
479 */
480 if (!mapaddr)
481 mapaddr =
482 tioce_alloc_map(ce_kern, TIOCE_ATE_M32, -1, ct_addr,
483 byte_count);
484
485 spin_unlock_irqrestore(&ce_kern->ce_lock, flags);
486
487dma_map_done:
488 if (mapaddr & barrier)
489 mapaddr = tioce_dma_barrier(mapaddr, 1);
490
491 return mapaddr;
492}
493
494/**
495 * tioce_dma - standard pci dma map interface
496 * @pdev: pci device requesting the map
497 * @paddr: system physical address to map into pci space
498 * @byte_count: # bytes to map
499 *
500 * Simply call tioce_do_dma_map() to create a map with the barrier bit clear
501 * in the address.
502 */
503static uint64_t
504tioce_dma(struct pci_dev *pdev, uint64_t paddr, size_t byte_count)
505{
506 return tioce_do_dma_map(pdev, paddr, byte_count, 0);
507}
508
509/**
510 * tioce_dma_consistent - consistent pci dma map interface
511 * @pdev: pci device requesting the map
512 * @paddr: system physical address to map into pci space
513 * @byte_count: # bytes to map
514 *
515 * Simply call tioce_do_dma_map() to create a map with the barrier bit set
516 * in the address.
517 */ static uint64_t
518tioce_dma_consistent(struct pci_dev *pdev, uint64_t paddr, size_t byte_count)
519{
520 return tioce_do_dma_map(pdev, paddr, byte_count, 1);
521}
522
523/**
524 * tioce_error_intr_handler - SGI TIO CE error interrupt handler
525 * @irq: unused
526 * @arg: pointer to tioce_common struct for the given CE
527 * @pt: unused
528 *
529 * Handle a CE error interrupt. Simply a wrapper around a SAL call which
530 * defers processing to the SGI prom.
531 */ static irqreturn_t
532tioce_error_intr_handler(int irq, void *arg, struct pt_regs *pt)
533{
534 struct tioce_common *soft = arg;
535 struct ia64_sal_retval ret_stuff;
536 ret_stuff.status = 0;
537 ret_stuff.v0 = 0;
538
539 SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_ERROR_INTERRUPT,
540 soft->ce_pcibus.bs_persist_segment,
541 soft->ce_pcibus.bs_persist_busnum, 0, 0, 0, 0, 0);
542
543 return IRQ_HANDLED;
544}
545
546/**
547 * tioce_kern_init - init kernel structures related to a given TIOCE
548 * @tioce_common: ptr to a cached tioce_common struct that originated in prom
549 */ static struct tioce_kernel *
550tioce_kern_init(struct tioce_common *tioce_common)
551{
552 int i;
553 uint32_t tmp;
554 struct tioce *tioce_mmr;
555 struct tioce_kernel *tioce_kern;
556
557 tioce_kern = kcalloc(1, sizeof(struct tioce_kernel), GFP_KERNEL);
558 if (!tioce_kern) {
559 return NULL;
560 }
561
562 tioce_kern->ce_common = tioce_common;
563 spin_lock_init(&tioce_kern->ce_lock);
564 INIT_LIST_HEAD(&tioce_kern->ce_dmamap_list);
565 tioce_common->ce_kernel_private = (uint64_t) tioce_kern;
566
567 /*
568 * Determine the secondary bus number of the port2 logical PPB.
569 * This is used to decide whether a given pci device resides on
570 * port1 or port2. Note: We don't have enough plumbing set up
571 * here to use pci_read_config_xxx() so use the raw_pci_ops vector.
572 */
573
574 raw_pci_ops->read(tioce_common->ce_pcibus.bs_persist_segment,
575 tioce_common->ce_pcibus.bs_persist_busnum,
576 PCI_DEVFN(2, 0), PCI_SECONDARY_BUS, 1, &tmp);
577 tioce_kern->ce_port1_secondary = (uint8_t) tmp;
578
579 /*
580 * Set PMU pagesize to the largest size available, and zero out
581 * the ate's.
582 */
583
584 tioce_mmr = (struct tioce *)tioce_common->ce_pcibus.bs_base;
585 tioce_mmr->ce_ure_page_map &= ~CE_URE_PAGESIZE_MASK;
586 tioce_mmr->ce_ure_page_map |= CE_URE_256K_PAGESIZE;
587 tioce_kern->ce_ate3240_pagesize = KB(256);
588
589 for (i = 0; i < TIOCE_NUM_M40_ATES; i++) {
590 tioce_kern->ce_ate40_shadow[i] = 0;
591 tioce_mmr->ce_ure_ate40[i] = 0;
592 }
593
594 for (i = 0; i < TIOCE_NUM_M3240_ATES; i++) {
595 tioce_kern->ce_ate3240_shadow[i] = 0;
596 tioce_mmr->ce_ure_ate3240[i] = 0;
597 }
598
599 return tioce_kern;
600}
601
602/**
603 * tioce_force_interrupt - implement altix force_interrupt() backend for CE
604 * @sn_irq_info: sn asic irq that we need an interrupt generated for
605 *
606 * Given an sn_irq_info struct, set the proper bit in ce_adm_force_int to
607 * force a secondary interrupt to be generated. This is to work around an
608 * asic issue where there is a small window of opportunity for a legacy device
609 * interrupt to be lost.
610 */
611static void
612tioce_force_interrupt(struct sn_irq_info *sn_irq_info)
613{
614 struct pcidev_info *pcidev_info;
615 struct tioce_common *ce_common;
616 struct tioce *ce_mmr;
617 uint64_t force_int_val;
618
619 if (!sn_irq_info->irq_bridge)
620 return;
621
622 if (sn_irq_info->irq_bridge_type != PCIIO_ASIC_TYPE_TIOCE)
623 return;
624
625 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
626 if (!pcidev_info)
627 return;
628
629 ce_common = (struct tioce_common *)pcidev_info->pdi_pcibus_info;
630 ce_mmr = (struct tioce *)ce_common->ce_pcibus.bs_base;
631
632 /*
633 * irq_int_bit is originally set up by prom, and holds the interrupt
634 * bit shift (not mask) as defined by the bit definitions in the
635 * ce_adm_int mmr. These shifts are not the same for the
636 * ce_adm_force_int register, so do an explicit mapping here to make
637 * things clearer.
638 */
639
640 switch (sn_irq_info->irq_int_bit) {
641 case CE_ADM_INT_PCIE_PORT1_DEV_A_SHFT:
642 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT1_DEV_A_SHFT;
643 break;
644 case CE_ADM_INT_PCIE_PORT1_DEV_B_SHFT:
645 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT1_DEV_B_SHFT;
646 break;
647 case CE_ADM_INT_PCIE_PORT1_DEV_C_SHFT:
648 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT1_DEV_C_SHFT;
649 break;
650 case CE_ADM_INT_PCIE_PORT1_DEV_D_SHFT:
651 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT1_DEV_D_SHFT;
652 break;
653 case CE_ADM_INT_PCIE_PORT2_DEV_A_SHFT:
654 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT2_DEV_A_SHFT;
655 break;
656 case CE_ADM_INT_PCIE_PORT2_DEV_B_SHFT:
657 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT2_DEV_B_SHFT;
658 break;
659 case CE_ADM_INT_PCIE_PORT2_DEV_C_SHFT:
660 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT2_DEV_C_SHFT;
661 break;
662 case CE_ADM_INT_PCIE_PORT2_DEV_D_SHFT:
663 force_int_val = 1UL << CE_ADM_FORCE_INT_PCIE_PORT2_DEV_D_SHFT;
664 break;
665 default:
666 return;
667 }
668 ce_mmr->ce_adm_force_int = force_int_val;
669}
670
671/**
672 * tioce_target_interrupt - implement set_irq_affinity for tioce resident
673 * functions. Note: only applies to line interrupts, not MSI's.
674 *
675 * @sn_irq_info: SN IRQ context
676 *
677 * Given an sn_irq_info, set the associated CE device's interrupt destination
678 * register. Since the interrupt destination registers are on a per-ce-slot
679 * basis, this will retarget line interrupts for all functions downstream of
680 * the slot.
681 */
682static void
683tioce_target_interrupt(struct sn_irq_info *sn_irq_info)
684{
685 struct pcidev_info *pcidev_info;
686 struct tioce_common *ce_common;
687 struct tioce *ce_mmr;
688 int bit;
689
690 pcidev_info = (struct pcidev_info *)sn_irq_info->irq_pciioinfo;
691 if (!pcidev_info)
692 return;
693
694 ce_common = (struct tioce_common *)pcidev_info->pdi_pcibus_info;
695 ce_mmr = (struct tioce *)ce_common->ce_pcibus.bs_base;
696
697 bit = sn_irq_info->irq_int_bit;
698
699 ce_mmr->ce_adm_int_mask |= (1UL << bit);
700 ce_mmr->ce_adm_int_dest[bit] =
701 ((uint64_t)sn_irq_info->irq_irq << INTR_VECTOR_SHFT) |
702 sn_irq_info->irq_xtalkaddr;
703 ce_mmr->ce_adm_int_mask &= ~(1UL << bit);
704
705 tioce_force_interrupt(sn_irq_info);
706}
707
708/**
709 * tioce_bus_fixup - perform final PCI fixup for a TIO CE bus
710 * @prom_bussoft: Common prom/kernel struct representing the bus
711 *
712 * Replicates the tioce_common pointed to by @prom_bussoft in kernel
713 * space. Allocates and initializes a kernel-only area for a given CE,
714 * and sets up an irq for handling CE error interrupts.
715 *
716 * On successful setup, returns the kernel version of tioce_common back to
717 * the caller.
718 */
719static void *
720tioce_bus_fixup(struct pcibus_bussoft *prom_bussoft, struct pci_controller *controller)
721{
722 struct tioce_common *tioce_common;
723
724 /*
725 * Allocate kernel bus soft and copy from prom.
726 */
727
728 tioce_common = kcalloc(1, sizeof(struct tioce_common), GFP_KERNEL);
729 if (!tioce_common)
730 return NULL;
731
732 memcpy(tioce_common, prom_bussoft, sizeof(struct tioce_common));
733 tioce_common->ce_pcibus.bs_base |= __IA64_UNCACHED_OFFSET;
734
735 if (tioce_kern_init(tioce_common) == NULL) {
736 kfree(tioce_common);
737 return NULL;
738 }
739
740 if (request_irq(SGI_PCIASIC_ERROR,
741 tioce_error_intr_handler,
742 SA_SHIRQ, "TIOCE error", (void *)tioce_common))
743 printk(KERN_WARNING
744 "%s: Unable to get irq %d. "
745 "Error interrupts won't be routed for "
746 "TIOCE bus %04x:%02x\n",
747 __FUNCTION__, SGI_PCIASIC_ERROR,
748 tioce_common->ce_pcibus.bs_persist_segment,
749 tioce_common->ce_pcibus.bs_persist_busnum);
750
751 return tioce_common;
752}
753
754static struct sn_pcibus_provider tioce_pci_interfaces = {
755 .dma_map = tioce_dma,
756 .dma_map_consistent = tioce_dma_consistent,
757 .dma_unmap = tioce_dma_unmap,
758 .bus_fixup = tioce_bus_fixup,
759 .force_interrupt = tioce_force_interrupt,
760 .target_interrupt = tioce_target_interrupt
761};
762
763/**
764 * tioce_init_provider - init SN PCI provider ops for TIO CE
765 */
766int
767tioce_init_provider(void)
768{
769 sn_pci_provider[PCIIO_ASIC_TYPE_TIOCE] = &tioce_pci_interfaces;
770 return 0;
771}
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-27 00:44:17 -0500