Merge branch 'linus' into pci-for-jesse
[linux-2.6] / arch / x86 / kernel / genx2apic_uv_x.c
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  * SGI UV APIC functions (note: not an Intel compatible APIC)
7  *
8  * Copyright (C) 2007-2008 Silicon Graphics, Inc. All rights reserved.
9  */
10
11 #include <linux/kernel.h>
12 #include <linux/threads.h>
13 #include <linux/cpumask.h>
14 #include <linux/string.h>
15 #include <linux/kernel.h>
16 #include <linux/ctype.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/bootmem.h>
20 #include <linux/module.h>
21 #include <asm/smp.h>
22 #include <asm/ipi.h>
23 #include <asm/genapic.h>
24 #include <asm/pgtable.h>
25 #include <asm/uv/uv_mmrs.h>
26 #include <asm/uv/uv_hub.h>
27
28 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
29 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
30
31 struct uv_blade_info *uv_blade_info;
32 EXPORT_SYMBOL_GPL(uv_blade_info);
33
34 short *uv_node_to_blade;
35 EXPORT_SYMBOL_GPL(uv_node_to_blade);
36
37 short *uv_cpu_to_blade;
38 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
39
40 short uv_possible_blades;
41 EXPORT_SYMBOL_GPL(uv_possible_blades);
42
43 /* Start with all IRQs pointing to boot CPU.  IRQ balancing will shift them. */
44
45 static cpumask_t uv_target_cpus(void)
46 {
47         return cpumask_of_cpu(0);
48 }
49
50 static cpumask_t uv_vector_allocation_domain(int cpu)
51 {
52         cpumask_t domain = CPU_MASK_NONE;
53         cpu_set(cpu, domain);
54         return domain;
55 }
56
57 int uv_wakeup_secondary(int phys_apicid, unsigned int start_rip)
58 {
59         unsigned long val;
60         int pnode;
61
62         pnode = uv_apicid_to_pnode(phys_apicid);
63         val = (1UL << UVH_IPI_INT_SEND_SHFT) |
64             (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
65             (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
66             APIC_DM_INIT;
67         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
68         mdelay(10);
69
70         val = (1UL << UVH_IPI_INT_SEND_SHFT) |
71             (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
72             (((long)start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
73             APIC_DM_STARTUP;
74         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
75         return 0;
76 }
77
78 static void uv_send_IPI_one(int cpu, int vector)
79 {
80         unsigned long val, apicid, lapicid;
81         int pnode;
82
83         apicid = per_cpu(x86_cpu_to_apicid, cpu); /* ZZZ - cache node-local ? */
84         lapicid = apicid & 0x3f;                /* ZZZ macro needed */
85         pnode = uv_apicid_to_pnode(apicid);
86         val =
87             (1UL << UVH_IPI_INT_SEND_SHFT) | (lapicid <<
88                                               UVH_IPI_INT_APIC_ID_SHFT) |
89             (vector << UVH_IPI_INT_VECTOR_SHFT);
90         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
91 }
92
93 static void uv_send_IPI_mask(cpumask_t mask, int vector)
94 {
95         unsigned int cpu;
96
97         for (cpu = 0; cpu < NR_CPUS; ++cpu)
98                 if (cpu_isset(cpu, mask))
99                         uv_send_IPI_one(cpu, vector);
100 }
101
102 static void uv_send_IPI_allbutself(int vector)
103 {
104         cpumask_t mask = cpu_online_map;
105
106         cpu_clear(smp_processor_id(), mask);
107
108         if (!cpus_empty(mask))
109                 uv_send_IPI_mask(mask, vector);
110 }
111
112 static void uv_send_IPI_all(int vector)
113 {
114         uv_send_IPI_mask(cpu_online_map, vector);
115 }
116
117 static int uv_apic_id_registered(void)
118 {
119         return 1;
120 }
121
122 static unsigned int uv_cpu_mask_to_apicid(cpumask_t cpumask)
123 {
124         int cpu;
125
126         /*
127          * We're using fixed IRQ delivery, can only return one phys APIC ID.
128          * May as well be the first.
129          */
130         cpu = first_cpu(cpumask);
131         if ((unsigned)cpu < NR_CPUS)
132                 return per_cpu(x86_cpu_to_apicid, cpu);
133         else
134                 return BAD_APICID;
135 }
136
137 static unsigned int phys_pkg_id(int index_msb)
138 {
139         return GET_APIC_ID(read_apic_id()) >> index_msb;
140 }
141
142 #ifdef ZZZ              /* Needs x2apic patch */
143 static void uv_send_IPI_self(int vector)
144 {
145         apic_write(APIC_SELF_IPI, vector);
146 }
147 #endif
148
149 struct genapic apic_x2apic_uv_x = {
150         .name = "UV large system",
151         .int_delivery_mode = dest_Fixed,
152         .int_dest_mode = (APIC_DEST_PHYSICAL != 0),
153         .target_cpus = uv_target_cpus,
154         .vector_allocation_domain = uv_vector_allocation_domain,/* Fixme ZZZ */
155         .apic_id_registered = uv_apic_id_registered,
156         .send_IPI_all = uv_send_IPI_all,
157         .send_IPI_allbutself = uv_send_IPI_allbutself,
158         .send_IPI_mask = uv_send_IPI_mask,
159         /* ZZZ.send_IPI_self = uv_send_IPI_self, */
160         .cpu_mask_to_apicid = uv_cpu_mask_to_apicid,
161         .phys_pkg_id = phys_pkg_id,     /* Fixme ZZZ */
162 };
163
164 static __cpuinit void set_x2apic_extra_bits(int pnode)
165 {
166         __get_cpu_var(x2apic_extra_bits) = (pnode << 6);
167 }
168
169 /*
170  * Called on boot cpu.
171  */
172 static __init int boot_pnode_to_blade(int pnode)
173 {
174         int blade;
175
176         for (blade = 0; blade < uv_num_possible_blades(); blade++)
177                 if (pnode == uv_blade_info[blade].pnode)
178                         return blade;
179         BUG();
180 }
181
182 struct redir_addr {
183         unsigned long redirect;
184         unsigned long alias;
185 };
186
187 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
188
189 static __initdata struct redir_addr redir_addrs[] = {
190         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_SI_ALIAS0_OVERLAY_CONFIG},
191         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_SI_ALIAS1_OVERLAY_CONFIG},
192         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_SI_ALIAS2_OVERLAY_CONFIG},
193 };
194
195 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
196 {
197         union uvh_si_alias0_overlay_config_u alias;
198         union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
199         int i;
200
201         for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
202                 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
203                 if (alias.s.base == 0) {
204                         *size = (1UL << alias.s.m_alias);
205                         redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
206                         *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
207                         return;
208                 }
209         }
210         BUG();
211 }
212
213 static __init void map_low_mmrs(void)
214 {
215         init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
216         init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
217 }
218
219 enum map_type {map_wb, map_uc};
220
221 static void map_high(char *id, unsigned long base, int shift, enum map_type map_type)
222 {
223         unsigned long bytes, paddr;
224
225         paddr = base << shift;
226         bytes = (1UL << shift);
227         printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
228                                                 paddr + bytes);
229         if (map_type == map_uc)
230                 init_extra_mapping_uc(paddr, bytes);
231         else
232                 init_extra_mapping_wb(paddr, bytes);
233
234 }
235 static __init void map_gru_high(int max_pnode)
236 {
237         union uvh_rh_gam_gru_overlay_config_mmr_u gru;
238         int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
239
240         gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
241         if (gru.s.enable)
242                 map_high("GRU", gru.s.base, shift, map_wb);
243 }
244
245 static __init void map_config_high(int max_pnode)
246 {
247         union uvh_rh_gam_cfg_overlay_config_mmr_u cfg;
248         int shift = UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR_BASE_SHFT;
249
250         cfg.v = uv_read_local_mmr(UVH_RH_GAM_CFG_OVERLAY_CONFIG_MMR);
251         if (cfg.s.enable)
252                 map_high("CONFIG", cfg.s.base, shift, map_uc);
253 }
254
255 static __init void map_mmr_high(int max_pnode)
256 {
257         union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
258         int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
259
260         mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
261         if (mmr.s.enable)
262                 map_high("MMR", mmr.s.base, shift, map_uc);
263 }
264
265 static __init void map_mmioh_high(int max_pnode)
266 {
267         union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
268         int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
269
270         mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
271         if (mmioh.s.enable)
272                 map_high("MMIOH", mmioh.s.base, shift, map_uc);
273 }
274
275 static __init void uv_system_init(void)
276 {
277         union uvh_si_addr_map_config_u m_n_config;
278         union uvh_node_id_u node_id;
279         unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
280         int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
281         int max_pnode = 0;
282         unsigned long mmr_base, present;
283
284         map_low_mmrs();
285
286         m_n_config.v = uv_read_local_mmr(UVH_SI_ADDR_MAP_CONFIG);
287         m_val = m_n_config.s.m_skt;
288         n_val = m_n_config.s.n_skt;
289         mmr_base =
290             uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
291             ~UV_MMR_ENABLE;
292         printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
293
294         for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
295                 uv_possible_blades +=
296                   hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
297         printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
298
299         bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
300         uv_blade_info = alloc_bootmem_pages(bytes);
301
302         get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
303
304         bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
305         uv_node_to_blade = alloc_bootmem_pages(bytes);
306         memset(uv_node_to_blade, 255, bytes);
307
308         bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
309         uv_cpu_to_blade = alloc_bootmem_pages(bytes);
310         memset(uv_cpu_to_blade, 255, bytes);
311
312         blade = 0;
313         for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
314                 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
315                 for (j = 0; j < 64; j++) {
316                         if (!test_bit(j, &present))
317                                 continue;
318                         uv_blade_info[blade].pnode = (i * 64 + j);
319                         uv_blade_info[blade].nr_possible_cpus = 0;
320                         uv_blade_info[blade].nr_online_cpus = 0;
321                         blade++;
322                 }
323         }
324
325         node_id.v = uv_read_local_mmr(UVH_NODE_ID);
326         gnode_upper = (((unsigned long)node_id.s.node_id) &
327                        ~((1 << n_val) - 1)) << m_val;
328
329         for_each_present_cpu(cpu) {
330                 nid = cpu_to_node(cpu);
331                 pnode = uv_apicid_to_pnode(per_cpu(x86_cpu_to_apicid, cpu));
332                 blade = boot_pnode_to_blade(pnode);
333                 lcpu = uv_blade_info[blade].nr_possible_cpus;
334                 uv_blade_info[blade].nr_possible_cpus++;
335
336                 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
337                 uv_cpu_hub_info(cpu)->lowmem_remap_top =
338                                         lowmem_redir_base + lowmem_redir_size;
339                 uv_cpu_hub_info(cpu)->m_val = m_val;
340                 uv_cpu_hub_info(cpu)->n_val = m_val;
341                 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
342                 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
343                 uv_cpu_hub_info(cpu)->pnode = pnode;
344                 uv_cpu_hub_info(cpu)->pnode_mask = (1 << n_val) - 1;
345                 uv_cpu_hub_info(cpu)->gpa_mask = (1 << (m_val + n_val)) - 1;
346                 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
347                 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
348                 uv_cpu_hub_info(cpu)->coherency_domain_number = 0;/* ZZZ */
349                 uv_node_to_blade[nid] = blade;
350                 uv_cpu_to_blade[cpu] = blade;
351                 max_pnode = max(pnode, max_pnode);
352
353                 printk(KERN_DEBUG "UV: cpu %d, apicid 0x%x, pnode %d, nid %d, "
354                         "lcpu %d, blade %d\n",
355                         cpu, per_cpu(x86_cpu_to_apicid, cpu), pnode, nid,
356                         lcpu, blade);
357         }
358
359         map_gru_high(max_pnode);
360         map_mmr_high(max_pnode);
361         map_config_high(max_pnode);
362         map_mmioh_high(max_pnode);
363 }
364
365 /*
366  * Called on each cpu to initialize the per_cpu UV data area.
367  *      ZZZ hotplug not supported yet
368  */
369 void __cpuinit uv_cpu_init(void)
370 {
371         if (!uv_node_to_blade)
372                 uv_system_init();
373
374         uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
375
376         if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
377                 set_x2apic_extra_bits(uv_hub_info->pnode);
378 }