1 /*
2  * IBM Summit-Specific Code
3  *
4  * Written By: Matthew Dobson, IBM Corporation
5  *
6  * Copyright (c) 2003 IBM Corp.
7  *
8  * All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or (at
13  * your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
18  * NON INFRINGEMENT.  See the GNU General Public License for more
19  * details.
20  *
21  * You should have received a copy of the GNU General Public License
22  * along with this program; if not, write to the Free Software
23  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24  *
25  * Send feedback to <colpatch@us.ibm.com>
26  *
27  */
28 
29 #include <linux/mm.h>
30 #include <linux/init.h>
31 #include <asm/io.h>
32 #include <asm/bios_ebda.h>
33 
34 /*
35  * APIC driver for the IBM "Summit" chipset.
36  */
37 #include <linux/threads.h>
38 #include <linux/cpumask.h>
39 #include <asm/mpspec.h>
40 #include <asm/apic.h>
41 #include <asm/smp.h>
42 #include <asm/fixmap.h>
43 #include <asm/apicdef.h>
44 #include <asm/ipi.h>
45 #include <linux/kernel.h>
46 #include <linux/string.h>
47 #include <linux/gfp.h>
48 #include <linux/smp.h>
49 
summit_get_apic_id(unsigned long x)50 static unsigned summit_get_apic_id(unsigned long x)
51 {
52 	return (x >> 24) & 0xFF;
53 }
54 
summit_send_IPI_mask(const struct cpumask * mask,int vector)55 static inline void summit_send_IPI_mask(const struct cpumask *mask, int vector)
56 {
57 	default_send_IPI_mask_sequence_logical(mask, vector);
58 }
59 
summit_send_IPI_allbutself(int vector)60 static void summit_send_IPI_allbutself(int vector)
61 {
62 	default_send_IPI_mask_allbutself_logical(cpu_online_mask, vector);
63 }
64 
summit_send_IPI_all(int vector)65 static void summit_send_IPI_all(int vector)
66 {
67 	summit_send_IPI_mask(cpu_online_mask, vector);
68 }
69 
70 #include <asm/tsc.h>
71 
72 extern int use_cyclone;
73 
74 #ifdef CONFIG_X86_SUMMIT_NUMA
75 static void setup_summit(void);
76 #else
setup_summit(void)77 static inline void setup_summit(void) {}
78 #endif
79 
summit_mps_oem_check(struct mpc_table * mpc,char * oem,char * productid)80 static int summit_mps_oem_check(struct mpc_table *mpc, char *oem,
81 		char *productid)
82 {
83 	if (!strncmp(oem, "IBM ENSW", 8) &&
84 			(!strncmp(productid, "VIGIL SMP", 9)
85 			 || !strncmp(productid, "EXA", 3)
86 			 || !strncmp(productid, "RUTHLESS SMP", 12))){
87 		mark_tsc_unstable("Summit based system");
88 		use_cyclone = 1; /*enable cyclone-timer*/
89 		setup_summit();
90 		return 1;
91 	}
92 	return 0;
93 }
94 
95 /* Hook from generic ACPI tables.c */
summit_acpi_madt_oem_check(char * oem_id,char * oem_table_id)96 static int summit_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
97 {
98 	if (!strncmp(oem_id, "IBM", 3) &&
99 	    (!strncmp(oem_table_id, "SERVIGIL", 8)
100 	     || !strncmp(oem_table_id, "EXA", 3))){
101 		mark_tsc_unstable("Summit based system");
102 		use_cyclone = 1; /*enable cyclone-timer*/
103 		setup_summit();
104 		return 1;
105 	}
106 	return 0;
107 }
108 
109 struct rio_table_hdr {
110 	unsigned char version;      /* Version number of this data structure           */
111 	                            /* Version 3 adds chassis_num & WP_index           */
112 	unsigned char num_scal_dev; /* # of Scalability devices (Twisters for Vigil)   */
113 	unsigned char num_rio_dev;  /* # of RIO I/O devices (Cyclones and Winnipegs)   */
114 } __attribute__((packed));
115 
116 struct scal_detail {
117 	unsigned char node_id;      /* Scalability Node ID                             */
118 	unsigned long CBAR;         /* Address of 1MB register space                   */
119 	unsigned char port0node;    /* Node ID port connected to: 0xFF=None            */
120 	unsigned char port0port;    /* Port num port connected to: 0,1,2, or 0xFF=None */
121 	unsigned char port1node;    /* Node ID port connected to: 0xFF = None          */
122 	unsigned char port1port;    /* Port num port connected to: 0,1,2, or 0xFF=None */
123 	unsigned char port2node;    /* Node ID port connected to: 0xFF = None          */
124 	unsigned char port2port;    /* Port num port connected to: 0,1,2, or 0xFF=None */
125 	unsigned char chassis_num;  /* 1 based Chassis number (1 = boot node)          */
126 } __attribute__((packed));
127 
128 struct rio_detail {
129 	unsigned char node_id;      /* RIO Node ID                                     */
130 	unsigned long BBAR;         /* Address of 1MB register space                   */
131 	unsigned char type;         /* Type of device                                  */
132 	unsigned char owner_id;     /* For WPEG: Node ID of Cyclone that owns this WPEG*/
133 	                            /* For CYC:  Node ID of Twister that owns this CYC */
134 	unsigned char port0node;    /* Node ID port connected to: 0xFF=None            */
135 	unsigned char port0port;    /* Port num port connected to: 0,1,2, or 0xFF=None */
136 	unsigned char port1node;    /* Node ID port connected to: 0xFF=None            */
137 	unsigned char port1port;    /* Port num port connected to: 0,1,2, or 0xFF=None */
138 	unsigned char first_slot;   /* For WPEG: Lowest slot number below this WPEG    */
139 	                            /* For CYC:  0                                     */
140 	unsigned char status;       /* For WPEG: Bit 0 = 1 : the XAPIC is used         */
141 	                            /*                 = 0 : the XAPIC is not used, ie:*/
142 	                            /*                     ints fwded to another XAPIC */
143 	                            /*           Bits1:7 Reserved                      */
144 	                            /* For CYC:  Bits0:7 Reserved                      */
145 	unsigned char WP_index;     /* For WPEG: WPEG instance index - lower ones have */
146 	                            /*           lower slot numbers/PCI bus numbers    */
147 	                            /* For CYC:  No meaning                            */
148 	unsigned char chassis_num;  /* 1 based Chassis number                          */
149 	                            /* For LookOut WPEGs this field indicates the      */
150 	                            /* Expansion Chassis #, enumerated from Boot       */
151 	                            /* Node WPEG external port, then Boot Node CYC     */
152 	                            /* external port, then Next Vigil chassis WPEG     */
153 	                            /* external port, etc.                             */
154 	                            /* Shared Lookouts have only 1 chassis number (the */
155 	                            /* first one assigned)                             */
156 } __attribute__((packed));
157 
158 
159 typedef enum {
160 	CompatTwister = 0,  /* Compatibility Twister               */
161 	AltTwister    = 1,  /* Alternate Twister of internal 8-way */
162 	CompatCyclone = 2,  /* Compatibility Cyclone               */
163 	AltCyclone    = 3,  /* Alternate Cyclone of internal 8-way */
164 	CompatWPEG    = 4,  /* Compatibility WPEG                  */
165 	AltWPEG       = 5,  /* Second Planar WPEG                  */
166 	LookOutAWPEG  = 6,  /* LookOut WPEG                        */
167 	LookOutBWPEG  = 7,  /* LookOut WPEG                        */
168 } node_type;
169 
is_WPEG(struct rio_detail * rio)170 static inline int is_WPEG(struct rio_detail *rio){
171 	return (rio->type == CompatWPEG || rio->type == AltWPEG ||
172 		rio->type == LookOutAWPEG || rio->type == LookOutBWPEG);
173 }
174 
175 #define SUMMIT_APIC_DFR_VALUE	(APIC_DFR_CLUSTER)
176 
summit_target_cpus(void)177 static const struct cpumask *summit_target_cpus(void)
178 {
179 	/* CPU_MASK_ALL (0xff) has undefined behaviour with
180 	 * dest_LowestPrio mode logical clustered apic interrupt routing
181 	 * Just start on cpu 0.  IRQ balancing will spread load
182 	 */
183 	return cpumask_of(0);
184 }
185 
summit_check_apicid_used(physid_mask_t * map,int apicid)186 static unsigned long summit_check_apicid_used(physid_mask_t *map, int apicid)
187 {
188 	return 0;
189 }
190 
191 /* we don't use the phys_cpu_present_map to indicate apicid presence */
summit_check_apicid_present(int bit)192 static unsigned long summit_check_apicid_present(int bit)
193 {
194 	return 1;
195 }
196 
summit_early_logical_apicid(int cpu)197 static int summit_early_logical_apicid(int cpu)
198 {
199 	int count = 0;
200 	u8 my_id = early_per_cpu(x86_cpu_to_apicid, cpu);
201 	u8 my_cluster = APIC_CLUSTER(my_id);
202 #ifdef CONFIG_SMP
203 	u8 lid;
204 	int i;
205 
206 	/* Create logical APIC IDs by counting CPUs already in cluster. */
207 	for (count = 0, i = nr_cpu_ids; --i >= 0; ) {
208 		lid = early_per_cpu(x86_cpu_to_logical_apicid, i);
209 		if (lid != BAD_APICID && APIC_CLUSTER(lid) == my_cluster)
210 			++count;
211 	}
212 #endif
213 	/* We only have a 4 wide bitmap in cluster mode.  If a deranged
214 	 * BIOS puts 5 CPUs in one APIC cluster, we're hosed. */
215 	BUG_ON(count >= XAPIC_DEST_CPUS_SHIFT);
216 	return my_cluster | (1UL << count);
217 }
218 
summit_init_apic_ldr(void)219 static void summit_init_apic_ldr(void)
220 {
221 	int cpu = smp_processor_id();
222 	unsigned long id = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
223 	unsigned long val;
224 
225 	apic_write(APIC_DFR, SUMMIT_APIC_DFR_VALUE);
226 	val = apic_read(APIC_LDR) & ~APIC_LDR_MASK;
227 	val |= SET_APIC_LOGICAL_ID(id);
228 	apic_write(APIC_LDR, val);
229 }
230 
summit_apic_id_registered(void)231 static int summit_apic_id_registered(void)
232 {
233 	return 1;
234 }
235 
summit_setup_apic_routing(void)236 static void summit_setup_apic_routing(void)
237 {
238 	printk("Enabling APIC mode:  Summit.  Using %d I/O APICs\n",
239 						nr_ioapics);
240 }
241 
summit_cpu_present_to_apicid(int mps_cpu)242 static int summit_cpu_present_to_apicid(int mps_cpu)
243 {
244 	if (mps_cpu < nr_cpu_ids)
245 		return (int)per_cpu(x86_bios_cpu_apicid, mps_cpu);
246 	else
247 		return BAD_APICID;
248 }
249 
summit_ioapic_phys_id_map(physid_mask_t * phys_id_map,physid_mask_t * retmap)250 static void summit_ioapic_phys_id_map(physid_mask_t *phys_id_map, physid_mask_t *retmap)
251 {
252 	/* For clustered we don't have a good way to do this yet - hack */
253 	physids_promote(0x0FL, retmap);
254 }
255 
summit_apicid_to_cpu_present(int apicid,physid_mask_t * retmap)256 static void summit_apicid_to_cpu_present(int apicid, physid_mask_t *retmap)
257 {
258 	physid_set_mask_of_physid(0, retmap);
259 }
260 
summit_check_phys_apicid_present(int physical_apicid)261 static int summit_check_phys_apicid_present(int physical_apicid)
262 {
263 	return 1;
264 }
265 
summit_cpu_mask_to_apicid(const struct cpumask * cpumask)266 static unsigned int summit_cpu_mask_to_apicid(const struct cpumask *cpumask)
267 {
268 	unsigned int round = 0;
269 	int cpu, apicid = 0;
270 
271 	/*
272 	 * The cpus in the mask must all be on the apic cluster.
273 	 */
274 	for_each_cpu(cpu, cpumask) {
275 		int new_apicid = early_per_cpu(x86_cpu_to_logical_apicid, cpu);
276 
277 		if (round && APIC_CLUSTER(apicid) != APIC_CLUSTER(new_apicid)) {
278 			printk("%s: Not a valid mask!\n", __func__);
279 			return BAD_APICID;
280 		}
281 		apicid |= new_apicid;
282 		round++;
283 	}
284 	return apicid;
285 }
286 
summit_cpu_mask_to_apicid_and(const struct cpumask * inmask,const struct cpumask * andmask)287 static unsigned int summit_cpu_mask_to_apicid_and(const struct cpumask *inmask,
288 			      const struct cpumask *andmask)
289 {
290 	int apicid = early_per_cpu(x86_cpu_to_logical_apicid, 0);
291 	cpumask_var_t cpumask;
292 
293 	if (!alloc_cpumask_var(&cpumask, GFP_ATOMIC))
294 		return apicid;
295 
296 	cpumask_and(cpumask, inmask, andmask);
297 	cpumask_and(cpumask, cpumask, cpu_online_mask);
298 	apicid = summit_cpu_mask_to_apicid(cpumask);
299 
300 	free_cpumask_var(cpumask);
301 
302 	return apicid;
303 }
304 
305 /*
306  * cpuid returns the value latched in the HW at reset, not the APIC ID
307  * register's value.  For any box whose BIOS changes APIC IDs, like
308  * clustered APIC systems, we must use hard_smp_processor_id.
309  *
310  * See Intel's IA-32 SW Dev's Manual Vol2 under CPUID.
311  */
summit_phys_pkg_id(int cpuid_apic,int index_msb)312 static int summit_phys_pkg_id(int cpuid_apic, int index_msb)
313 {
314 	return hard_smp_processor_id() >> index_msb;
315 }
316 
probe_summit(void)317 static int probe_summit(void)
318 {
319 	/* probed later in mptable/ACPI hooks */
320 	return 0;
321 }
322 
summit_vector_allocation_domain(int cpu,struct cpumask * retmask)323 static void summit_vector_allocation_domain(int cpu, struct cpumask *retmask)
324 {
325 	/* Careful. Some cpus do not strictly honor the set of cpus
326 	 * specified in the interrupt destination when using lowest
327 	 * priority interrupt delivery mode.
328 	 *
329 	 * In particular there was a hyperthreading cpu observed to
330 	 * deliver interrupts to the wrong hyperthread when only one
331 	 * hyperthread was specified in the interrupt desitination.
332 	 */
333 	cpumask_clear(retmask);
334 	cpumask_bits(retmask)[0] = APIC_ALL_CPUS;
335 }
336 
337 #ifdef CONFIG_X86_SUMMIT_NUMA
338 static struct rio_table_hdr *rio_table_hdr;
339 static struct scal_detail   *scal_devs[MAX_NUMNODES];
340 static struct rio_detail    *rio_devs[MAX_NUMNODES*4];
341 
342 #ifndef CONFIG_X86_NUMAQ
343 static int mp_bus_id_to_node[MAX_MP_BUSSES];
344 #endif
345 
setup_pci_node_map_for_wpeg(int wpeg_num,int last_bus)346 static int setup_pci_node_map_for_wpeg(int wpeg_num, int last_bus)
347 {
348 	int twister = 0, node = 0;
349 	int i, bus, num_buses;
350 
351 	for (i = 0; i < rio_table_hdr->num_rio_dev; i++) {
352 		if (rio_devs[i]->node_id == rio_devs[wpeg_num]->owner_id) {
353 			twister = rio_devs[i]->owner_id;
354 			break;
355 		}
356 	}
357 	if (i == rio_table_hdr->num_rio_dev) {
358 		printk(KERN_ERR "%s: Couldn't find owner Cyclone for Winnipeg!\n", __func__);
359 		return last_bus;
360 	}
361 
362 	for (i = 0; i < rio_table_hdr->num_scal_dev; i++) {
363 		if (scal_devs[i]->node_id == twister) {
364 			node = scal_devs[i]->node_id;
365 			break;
366 		}
367 	}
368 	if (i == rio_table_hdr->num_scal_dev) {
369 		printk(KERN_ERR "%s: Couldn't find owner Twister for Cyclone!\n", __func__);
370 		return last_bus;
371 	}
372 
373 	switch (rio_devs[wpeg_num]->type) {
374 	case CompatWPEG:
375 		/*
376 		 * The Compatibility Winnipeg controls the 2 legacy buses,
377 		 * the 66MHz PCI bus [2 slots] and the 2 "extra" buses in case
378 		 * a PCI-PCI bridge card is used in either slot: total 5 buses.
379 		 */
380 		num_buses = 5;
381 		break;
382 	case AltWPEG:
383 		/*
384 		 * The Alternate Winnipeg controls the 2 133MHz buses [1 slot
385 		 * each], their 2 "extra" buses, the 100MHz bus [2 slots] and
386 		 * the "extra" buses for each of those slots: total 7 buses.
387 		 */
388 		num_buses = 7;
389 		break;
390 	case LookOutAWPEG:
391 	case LookOutBWPEG:
392 		/*
393 		 * A Lookout Winnipeg controls 3 100MHz buses [2 slots each]
394 		 * & the "extra" buses for each of those slots: total 9 buses.
395 		 */
396 		num_buses = 9;
397 		break;
398 	default:
399 		printk(KERN_INFO "%s: Unsupported Winnipeg type!\n", __func__);
400 		return last_bus;
401 	}
402 
403 	for (bus = last_bus; bus < last_bus + num_buses; bus++)
404 		mp_bus_id_to_node[bus] = node;
405 	return bus;
406 }
407 
build_detail_arrays(void)408 static int build_detail_arrays(void)
409 {
410 	unsigned long ptr;
411 	int i, scal_detail_size, rio_detail_size;
412 
413 	if (rio_table_hdr->num_scal_dev > MAX_NUMNODES) {
414 		printk(KERN_WARNING "%s: MAX_NUMNODES too low!  Defined as %d, but system has %d nodes.\n", __func__, MAX_NUMNODES, rio_table_hdr->num_scal_dev);
415 		return 0;
416 	}
417 
418 	switch (rio_table_hdr->version) {
419 	default:
420 		printk(KERN_WARNING "%s: Invalid Rio Grande Table Version: %d\n", __func__, rio_table_hdr->version);
421 		return 0;
422 	case 2:
423 		scal_detail_size = 11;
424 		rio_detail_size = 13;
425 		break;
426 	case 3:
427 		scal_detail_size = 12;
428 		rio_detail_size = 15;
429 		break;
430 	}
431 
432 	ptr = (unsigned long)rio_table_hdr + 3;
433 	for (i = 0; i < rio_table_hdr->num_scal_dev; i++, ptr += scal_detail_size)
434 		scal_devs[i] = (struct scal_detail *)ptr;
435 
436 	for (i = 0; i < rio_table_hdr->num_rio_dev; i++, ptr += rio_detail_size)
437 		rio_devs[i] = (struct rio_detail *)ptr;
438 
439 	return 1;
440 }
441 
setup_summit(void)442 void setup_summit(void)
443 {
444 	unsigned long		ptr;
445 	unsigned short		offset;
446 	int			i, next_wpeg, next_bus = 0;
447 
448 	/* The pointer to the EBDA is stored in the word @ phys 0x40E(40:0E) */
449 	ptr = get_bios_ebda();
450 	ptr = (unsigned long)phys_to_virt(ptr);
451 
452 	rio_table_hdr = NULL;
453 	offset = 0x180;
454 	while (offset) {
455 		/* The block id is stored in the 2nd word */
456 		if (*((unsigned short *)(ptr + offset + 2)) == 0x4752) {
457 			/* set the pointer past the offset & block id */
458 			rio_table_hdr = (struct rio_table_hdr *)(ptr + offset + 4);
459 			break;
460 		}
461 		/* The next offset is stored in the 1st word.  0 means no more */
462 		offset = *((unsigned short *)(ptr + offset));
463 	}
464 	if (!rio_table_hdr) {
465 		printk(KERN_ERR "%s: Unable to locate Rio Grande Table in EBDA - bailing!\n", __func__);
466 		return;
467 	}
468 
469 	if (!build_detail_arrays())
470 		return;
471 
472 	/* The first Winnipeg we're looking for has an index of 0 */
473 	next_wpeg = 0;
474 	do {
475 		for (i = 0; i < rio_table_hdr->num_rio_dev; i++) {
476 			if (is_WPEG(rio_devs[i]) && rio_devs[i]->WP_index == next_wpeg) {
477 				/* It's the Winnipeg we're looking for! */
478 				next_bus = setup_pci_node_map_for_wpeg(i, next_bus);
479 				next_wpeg++;
480 				break;
481 			}
482 		}
483 		/*
484 		 * If we go through all Rio devices and don't find one with
485 		 * the next index, it means we've found all the Winnipegs,
486 		 * and thus all the PCI buses.
487 		 */
488 		if (i == rio_table_hdr->num_rio_dev)
489 			next_wpeg = 0;
490 	} while (next_wpeg != 0);
491 }
492 #endif
493 
494 struct apic apic_summit = {
495 
496 	.name				= "summit",
497 	.probe				= probe_summit,
498 	.acpi_madt_oem_check		= summit_acpi_madt_oem_check,
499 	.apic_id_registered		= summit_apic_id_registered,
500 
501 	.irq_delivery_mode		= dest_LowestPrio,
502 	/* logical delivery broadcast to all CPUs: */
503 	.irq_dest_mode			= 1,
504 
505 	.target_cpus			= summit_target_cpus,
506 	.disable_esr			= 1,
507 	.dest_logical			= APIC_DEST_LOGICAL,
508 	.check_apicid_used		= summit_check_apicid_used,
509 	.check_apicid_present		= summit_check_apicid_present,
510 
511 	.vector_allocation_domain	= summit_vector_allocation_domain,
512 	.init_apic_ldr			= summit_init_apic_ldr,
513 
514 	.ioapic_phys_id_map		= summit_ioapic_phys_id_map,
515 	.setup_apic_routing		= summit_setup_apic_routing,
516 	.multi_timer_check		= NULL,
517 	.cpu_present_to_apicid		= summit_cpu_present_to_apicid,
518 	.apicid_to_cpu_present		= summit_apicid_to_cpu_present,
519 	.setup_portio_remap		= NULL,
520 	.check_phys_apicid_present	= summit_check_phys_apicid_present,
521 	.enable_apic_mode		= NULL,
522 	.phys_pkg_id			= summit_phys_pkg_id,
523 	.mps_oem_check			= summit_mps_oem_check,
524 
525 	.get_apic_id			= summit_get_apic_id,
526 	.set_apic_id			= NULL,
527 	.apic_id_mask			= 0xFF << 24,
528 
529 	.cpu_mask_to_apicid		= summit_cpu_mask_to_apicid,
530 	.cpu_mask_to_apicid_and		= summit_cpu_mask_to_apicid_and,
531 
532 	.send_IPI_mask			= summit_send_IPI_mask,
533 	.send_IPI_mask_allbutself	= NULL,
534 	.send_IPI_allbutself		= summit_send_IPI_allbutself,
535 	.send_IPI_all			= summit_send_IPI_all,
536 	.send_IPI_self			= default_send_IPI_self,
537 
538 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
539 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
540 
541 	.wait_for_init_deassert		= default_wait_for_init_deassert,
542 
543 	.smp_callin_clear_local_apic	= NULL,
544 	.inquire_remote_apic		= default_inquire_remote_apic,
545 
546 	.read				= native_apic_mem_read,
547 	.write				= native_apic_mem_write,
548 	.icr_read			= native_apic_icr_read,
549 	.icr_write			= native_apic_icr_write,
550 	.wait_icr_idle			= native_apic_wait_icr_idle,
551 	.safe_wait_icr_idle		= native_safe_apic_wait_icr_idle,
552 
553 	.x86_32_early_logical_apicid	= summit_early_logical_apicid,
554 	.x86_32_numa_cpu_node		= default_x86_32_numa_cpu_node,
555 };
556