xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 68db0cf1)
1 /*
2  * Architecture-specific setup.
3  *
4  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  *	Stephane Eranian <eranian@hpl.hp.com>
7  * Copyright (C) 2000, 2004 Intel Corp
8  * 	Rohit Seth <rohit.seth@intel.com>
9  * 	Suresh Siddha <suresh.b.siddha@intel.com>
10  * 	Gordon Jin <gordon.jin@intel.com>
11  * Copyright (C) 1999 VA Linux Systems
12  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13  *
14  * 12/26/04 S.Siddha, G.Jin, R.Seth
15  *			Add multi-threading and multi-core detection
16  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
17  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
18  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
19  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
20  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
21  * 01/07/99 S.Eranian	added the support for command line argument
22  * 06/24/99 W.Drummond	added boot_cpu_data.
23  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
24  */
25 #include <linux/module.h>
26 #include <linux/init.h>
27 
28 #include <linux/acpi.h>
29 #include <linux/bootmem.h>
30 #include <linux/console.h>
31 #include <linux/delay.h>
32 #include <linux/kernel.h>
33 #include <linux/reboot.h>
34 #include <linux/sched.h>
35 #include <linux/sched/clock.h>
36 #include <linux/sched/task_stack.h>
37 #include <linux/seq_file.h>
38 #include <linux/string.h>
39 #include <linux/threads.h>
40 #include <linux/screen_info.h>
41 #include <linux/dmi.h>
42 #include <linux/serial.h>
43 #include <linux/serial_core.h>
44 #include <linux/efi.h>
45 #include <linux/initrd.h>
46 #include <linux/pm.h>
47 #include <linux/cpufreq.h>
48 #include <linux/kexec.h>
49 #include <linux/crash_dump.h>
50 
51 #include <asm/machvec.h>
52 #include <asm/mca.h>
53 #include <asm/meminit.h>
54 #include <asm/page.h>
55 #include <asm/patch.h>
56 #include <asm/pgtable.h>
57 #include <asm/processor.h>
58 #include <asm/sal.h>
59 #include <asm/sections.h>
60 #include <asm/setup.h>
61 #include <asm/smp.h>
62 #include <asm/tlbflush.h>
63 #include <asm/unistd.h>
64 #include <asm/hpsim.h>
65 
66 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
67 # error "struct cpuinfo_ia64 too big!"
68 #endif
69 
70 #ifdef CONFIG_SMP
71 unsigned long __per_cpu_offset[NR_CPUS];
72 EXPORT_SYMBOL(__per_cpu_offset);
73 #endif
74 
75 DEFINE_PER_CPU(struct cpuinfo_ia64, ia64_cpu_info);
76 EXPORT_SYMBOL(ia64_cpu_info);
77 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
78 #ifdef CONFIG_SMP
79 EXPORT_SYMBOL(local_per_cpu_offset);
80 #endif
81 unsigned long ia64_cycles_per_usec;
82 struct ia64_boot_param *ia64_boot_param;
83 struct screen_info screen_info;
84 unsigned long vga_console_iobase;
85 unsigned long vga_console_membase;
86 
87 static struct resource data_resource = {
88 	.name	= "Kernel data",
89 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
90 };
91 
92 static struct resource code_resource = {
93 	.name	= "Kernel code",
94 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
95 };
96 
97 static struct resource bss_resource = {
98 	.name	= "Kernel bss",
99 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
100 };
101 
102 unsigned long ia64_max_cacheline_size;
103 
104 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
105 EXPORT_SYMBOL(ia64_iobase);
106 struct io_space io_space[MAX_IO_SPACES];
107 EXPORT_SYMBOL(io_space);
108 unsigned int num_io_spaces;
109 
110 /*
111  * "flush_icache_range()" needs to know what processor dependent stride size to use
112  * when it makes i-cache(s) coherent with d-caches.
113  */
114 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
115 unsigned long ia64_i_cache_stride_shift = ~0;
116 /*
117  * "clflush_cache_range()" needs to know what processor dependent stride size to
118  * use when it flushes cache lines including both d-cache and i-cache.
119  */
120 /* Safest way to go: 32 bytes by 32 bytes */
121 #define	CACHE_STRIDE_SHIFT	5
122 unsigned long ia64_cache_stride_shift = ~0;
123 
124 /*
125  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
126  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
127  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
128  * address of the second buffer must be aligned to (merge_mask+1) in order to be
129  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
130  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
131  * page-size of 2^64.
132  */
133 unsigned long ia64_max_iommu_merge_mask = ~0UL;
134 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
135 
136 /*
137  * We use a special marker for the end of memory and it uses the extra (+1) slot
138  */
139 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
140 int num_rsvd_regions __initdata;
141 
142 
143 /*
144  * Filter incoming memory segments based on the primitive map created from the boot
145  * parameters. Segments contained in the map are removed from the memory ranges. A
146  * caller-specified function is called with the memory ranges that remain after filtering.
147  * This routine does not assume the incoming segments are sorted.
148  */
149 int __init
150 filter_rsvd_memory (u64 start, u64 end, void *arg)
151 {
152 	u64 range_start, range_end, prev_start;
153 	void (*func)(unsigned long, unsigned long, int);
154 	int i;
155 
156 #if IGNORE_PFN0
157 	if (start == PAGE_OFFSET) {
158 		printk(KERN_WARNING "warning: skipping physical page 0\n");
159 		start += PAGE_SIZE;
160 		if (start >= end) return 0;
161 	}
162 #endif
163 	/*
164 	 * lowest possible address(walker uses virtual)
165 	 */
166 	prev_start = PAGE_OFFSET;
167 	func = arg;
168 
169 	for (i = 0; i < num_rsvd_regions; ++i) {
170 		range_start = max(start, prev_start);
171 		range_end   = min(end, rsvd_region[i].start);
172 
173 		if (range_start < range_end)
174 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
175 
176 		/* nothing more available in this segment */
177 		if (range_end == end) return 0;
178 
179 		prev_start = rsvd_region[i].end;
180 	}
181 	/* end of memory marker allows full processing inside loop body */
182 	return 0;
183 }
184 
185 /*
186  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
187  * are not filtered out.
188  */
189 int __init
190 filter_memory(u64 start, u64 end, void *arg)
191 {
192 	void (*func)(unsigned long, unsigned long, int);
193 
194 #if IGNORE_PFN0
195 	if (start == PAGE_OFFSET) {
196 		printk(KERN_WARNING "warning: skipping physical page 0\n");
197 		start += PAGE_SIZE;
198 		if (start >= end)
199 			return 0;
200 	}
201 #endif
202 	func = arg;
203 	if (start < end)
204 		call_pernode_memory(__pa(start), end - start, func);
205 	return 0;
206 }
207 
208 static void __init
209 sort_regions (struct rsvd_region *rsvd_region, int max)
210 {
211 	int j;
212 
213 	/* simple bubble sorting */
214 	while (max--) {
215 		for (j = 0; j < max; ++j) {
216 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
217 				struct rsvd_region tmp;
218 				tmp = rsvd_region[j];
219 				rsvd_region[j] = rsvd_region[j + 1];
220 				rsvd_region[j + 1] = tmp;
221 			}
222 		}
223 	}
224 }
225 
226 /* merge overlaps */
227 static int __init
228 merge_regions (struct rsvd_region *rsvd_region, int max)
229 {
230 	int i;
231 	for (i = 1; i < max; ++i) {
232 		if (rsvd_region[i].start >= rsvd_region[i-1].end)
233 			continue;
234 		if (rsvd_region[i].end > rsvd_region[i-1].end)
235 			rsvd_region[i-1].end = rsvd_region[i].end;
236 		--max;
237 		memmove(&rsvd_region[i], &rsvd_region[i+1],
238 			(max - i) * sizeof(struct rsvd_region));
239 	}
240 	return max;
241 }
242 
243 /*
244  * Request address space for all standard resources
245  */
246 static int __init register_memory(void)
247 {
248 	code_resource.start = ia64_tpa(_text);
249 	code_resource.end   = ia64_tpa(_etext) - 1;
250 	data_resource.start = ia64_tpa(_etext);
251 	data_resource.end   = ia64_tpa(_edata) - 1;
252 	bss_resource.start  = ia64_tpa(__bss_start);
253 	bss_resource.end    = ia64_tpa(_end) - 1;
254 	efi_initialize_iomem_resources(&code_resource, &data_resource,
255 			&bss_resource);
256 
257 	return 0;
258 }
259 
260 __initcall(register_memory);
261 
262 
263 #ifdef CONFIG_KEXEC
264 
265 /*
266  * This function checks if the reserved crashkernel is allowed on the specific
267  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
268  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
269  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
270  * in kdump case. See the comment in sba_init() in sba_iommu.c.
271  *
272  * So, the only machvec that really supports loading the kdump kernel
273  * over 4 GB is "sn2".
274  */
275 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
276 {
277 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
278 		return 1;
279 	else
280 		return pbase < (1UL << 32);
281 }
282 
283 static void __init setup_crashkernel(unsigned long total, int *n)
284 {
285 	unsigned long long base = 0, size = 0;
286 	int ret;
287 
288 	ret = parse_crashkernel(boot_command_line, total,
289 			&size, &base);
290 	if (ret == 0 && size > 0) {
291 		if (!base) {
292 			sort_regions(rsvd_region, *n);
293 			*n = merge_regions(rsvd_region, *n);
294 			base = kdump_find_rsvd_region(size,
295 					rsvd_region, *n);
296 		}
297 
298 		if (!check_crashkernel_memory(base, size)) {
299 			pr_warning("crashkernel: There would be kdump memory "
300 				"at %ld GB but this is unusable because it "
301 				"must\nbe below 4 GB. Change the memory "
302 				"configuration of the machine.\n",
303 				(unsigned long)(base >> 30));
304 			return;
305 		}
306 
307 		if (base != ~0UL) {
308 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
309 					"for crashkernel (System RAM: %ldMB)\n",
310 					(unsigned long)(size >> 20),
311 					(unsigned long)(base >> 20),
312 					(unsigned long)(total >> 20));
313 			rsvd_region[*n].start =
314 				(unsigned long)__va(base);
315 			rsvd_region[*n].end =
316 				(unsigned long)__va(base + size);
317 			(*n)++;
318 			crashk_res.start = base;
319 			crashk_res.end = base + size - 1;
320 		}
321 	}
322 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
323 	efi_memmap_res.end = efi_memmap_res.start +
324 		ia64_boot_param->efi_memmap_size;
325 	boot_param_res.start = __pa(ia64_boot_param);
326 	boot_param_res.end = boot_param_res.start +
327 		sizeof(*ia64_boot_param);
328 }
329 #else
330 static inline void __init setup_crashkernel(unsigned long total, int *n)
331 {}
332 #endif
333 
334 /**
335  * reserve_memory - setup reserved memory areas
336  *
337  * Setup the reserved memory areas set aside for the boot parameters,
338  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
339  * see arch/ia64/include/asm/meminit.h if you need to define more.
340  */
341 void __init
342 reserve_memory (void)
343 {
344 	int n = 0;
345 	unsigned long total_memory;
346 
347 	/*
348 	 * none of the entries in this table overlap
349 	 */
350 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
351 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
352 	n++;
353 
354 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
355 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
356 	n++;
357 
358 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
359 	rsvd_region[n].end   = (rsvd_region[n].start
360 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
361 	n++;
362 
363 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
364 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
365 	n++;
366 
367 #ifdef CONFIG_BLK_DEV_INITRD
368 	if (ia64_boot_param->initrd_start) {
369 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
370 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
371 		n++;
372 	}
373 #endif
374 
375 #ifdef CONFIG_CRASH_DUMP
376 	if (reserve_elfcorehdr(&rsvd_region[n].start,
377 			       &rsvd_region[n].end) == 0)
378 		n++;
379 #endif
380 
381 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
382 	n++;
383 
384 	setup_crashkernel(total_memory, &n);
385 
386 	/* end of memory marker */
387 	rsvd_region[n].start = ~0UL;
388 	rsvd_region[n].end   = ~0UL;
389 	n++;
390 
391 	num_rsvd_regions = n;
392 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
393 
394 	sort_regions(rsvd_region, num_rsvd_regions);
395 	num_rsvd_regions = merge_regions(rsvd_region, num_rsvd_regions);
396 }
397 
398 
399 /**
400  * find_initrd - get initrd parameters from the boot parameter structure
401  *
402  * Grab the initrd start and end from the boot parameter struct given us by
403  * the boot loader.
404  */
405 void __init
406 find_initrd (void)
407 {
408 #ifdef CONFIG_BLK_DEV_INITRD
409 	if (ia64_boot_param->initrd_start) {
410 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
411 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
412 
413 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%llu bytes)\n",
414 		       initrd_start, ia64_boot_param->initrd_size);
415 	}
416 #endif
417 }
418 
419 static void __init
420 io_port_init (void)
421 {
422 	unsigned long phys_iobase;
423 
424 	/*
425 	 * Set `iobase' based on the EFI memory map or, failing that, the
426 	 * value firmware left in ar.k0.
427 	 *
428 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
429 	 * the port's virtual address, so ia32_load_state() loads it with a
430 	 * user virtual address.  But in ia64 mode, glibc uses the
431 	 * *physical* address in ar.k0 to mmap the appropriate area from
432 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
433 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
434 	 *
435 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
436 	 * any of the I/O port spaces and are done via MMIO using the
437 	 * virtual mmio_base from the appropriate io_space[].
438 	 */
439 	phys_iobase = efi_get_iobase();
440 	if (!phys_iobase) {
441 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
442 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
443 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
444 	}
445 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
446 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
447 
448 	/* setup legacy IO port space */
449 	io_space[0].mmio_base = ia64_iobase;
450 	io_space[0].sparse = 1;
451 	num_io_spaces = 1;
452 }
453 
454 /**
455  * early_console_setup - setup debugging console
456  *
457  * Consoles started here require little enough setup that we can start using
458  * them very early in the boot process, either right after the machine
459  * vector initialization, or even before if the drivers can detect their hw.
460  *
461  * Returns non-zero if a console couldn't be setup.
462  */
463 static inline int __init
464 early_console_setup (char *cmdline)
465 {
466 	int earlycons = 0;
467 
468 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
469 	{
470 		extern int sn_serial_console_early_setup(void);
471 		if (!sn_serial_console_early_setup())
472 			earlycons++;
473 	}
474 #endif
475 #ifdef CONFIG_EFI_PCDP
476 	if (!efi_setup_pcdp_console(cmdline))
477 		earlycons++;
478 #endif
479 	if (!simcons_register())
480 		earlycons++;
481 
482 	return (earlycons) ? 0 : -1;
483 }
484 
485 static inline void
486 mark_bsp_online (void)
487 {
488 #ifdef CONFIG_SMP
489 	/* If we register an early console, allow CPU 0 to printk */
490 	set_cpu_online(smp_processor_id(), true);
491 #endif
492 }
493 
494 static __initdata int nomca;
495 static __init int setup_nomca(char *s)
496 {
497 	nomca = 1;
498 	return 0;
499 }
500 early_param("nomca", setup_nomca);
501 
502 #ifdef CONFIG_CRASH_DUMP
503 int __init reserve_elfcorehdr(u64 *start, u64 *end)
504 {
505 	u64 length;
506 
507 	/* We get the address using the kernel command line,
508 	 * but the size is extracted from the EFI tables.
509 	 * Both address and size are required for reservation
510 	 * to work properly.
511 	 */
512 
513 	if (!is_vmcore_usable())
514 		return -EINVAL;
515 
516 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
517 		vmcore_unusable();
518 		return -EINVAL;
519 	}
520 
521 	*start = (unsigned long)__va(elfcorehdr_addr);
522 	*end = *start + length;
523 	return 0;
524 }
525 
526 #endif /* CONFIG_PROC_VMCORE */
527 
528 void __init
529 setup_arch (char **cmdline_p)
530 {
531 	unw_init();
532 
533 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
534 
535 	*cmdline_p = __va(ia64_boot_param->command_line);
536 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
537 
538 	efi_init();
539 	io_port_init();
540 
541 #ifdef CONFIG_IA64_GENERIC
542 	/* machvec needs to be parsed from the command line
543 	 * before parse_early_param() is called to ensure
544 	 * that ia64_mv is initialised before any command line
545 	 * settings may cause console setup to occur
546 	 */
547 	machvec_init_from_cmdline(*cmdline_p);
548 #endif
549 
550 	parse_early_param();
551 
552 	if (early_console_setup(*cmdline_p) == 0)
553 		mark_bsp_online();
554 
555 #ifdef CONFIG_ACPI
556 	/* Initialize the ACPI boot-time table parser */
557 	acpi_table_init();
558 	early_acpi_boot_init();
559 # ifdef CONFIG_ACPI_NUMA
560 	acpi_numa_init();
561 	acpi_numa_fixup();
562 #  ifdef CONFIG_ACPI_HOTPLUG_CPU
563 	prefill_possible_map();
564 #  endif
565 	per_cpu_scan_finalize((cpumask_weight(&early_cpu_possible_map) == 0 ?
566 		32 : cpumask_weight(&early_cpu_possible_map)),
567 		additional_cpus > 0 ? additional_cpus : 0);
568 # endif
569 #endif /* CONFIG_APCI_BOOT */
570 
571 #ifdef CONFIG_SMP
572 	smp_build_cpu_map();
573 #endif
574 	find_memory();
575 
576 	/* process SAL system table: */
577 	ia64_sal_init(__va(efi.sal_systab));
578 
579 #ifdef CONFIG_ITANIUM
580 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
581 #else
582 	{
583 		unsigned long num_phys_stacked;
584 
585 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
586 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
587 	}
588 #endif
589 
590 #ifdef CONFIG_SMP
591 	cpu_physical_id(0) = hard_smp_processor_id();
592 #endif
593 
594 	cpu_init();	/* initialize the bootstrap CPU */
595 	mmu_context_init();	/* initialize context_id bitmap */
596 
597 #ifdef CONFIG_VT
598 	if (!conswitchp) {
599 # if defined(CONFIG_DUMMY_CONSOLE)
600 		conswitchp = &dummy_con;
601 # endif
602 # if defined(CONFIG_VGA_CONSOLE)
603 		/*
604 		 * Non-legacy systems may route legacy VGA MMIO range to system
605 		 * memory.  vga_con probes the MMIO hole, so memory looks like
606 		 * a VGA device to it.  The EFI memory map can tell us if it's
607 		 * memory so we can avoid this problem.
608 		 */
609 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
610 			conswitchp = &vga_con;
611 # endif
612 	}
613 #endif
614 
615 	/* enable IA-64 Machine Check Abort Handling unless disabled */
616 	if (!nomca)
617 		ia64_mca_init();
618 
619 	platform_setup(cmdline_p);
620 #ifndef CONFIG_IA64_HP_SIM
621 	check_sal_cache_flush();
622 #endif
623 	paging_init();
624 
625 	clear_sched_clock_stable();
626 }
627 
628 /*
629  * Display cpu info for all CPUs.
630  */
631 static int
632 show_cpuinfo (struct seq_file *m, void *v)
633 {
634 #ifdef CONFIG_SMP
635 #	define lpj	c->loops_per_jiffy
636 #	define cpunum	c->cpu
637 #else
638 #	define lpj	loops_per_jiffy
639 #	define cpunum	0
640 #endif
641 	static struct {
642 		unsigned long mask;
643 		const char *feature_name;
644 	} feature_bits[] = {
645 		{ 1UL << 0, "branchlong" },
646 		{ 1UL << 1, "spontaneous deferral"},
647 		{ 1UL << 2, "16-byte atomic ops" }
648 	};
649 	char features[128], *cp, *sep;
650 	struct cpuinfo_ia64 *c = v;
651 	unsigned long mask;
652 	unsigned long proc_freq;
653 	int i, size;
654 
655 	mask = c->features;
656 
657 	/* build the feature string: */
658 	memcpy(features, "standard", 9);
659 	cp = features;
660 	size = sizeof(features);
661 	sep = "";
662 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
663 		if (mask & feature_bits[i].mask) {
664 			cp += snprintf(cp, size, "%s%s", sep,
665 				       feature_bits[i].feature_name),
666 			sep = ", ";
667 			mask &= ~feature_bits[i].mask;
668 			size = sizeof(features) - (cp - features);
669 		}
670 	}
671 	if (mask && size > 1) {
672 		/* print unknown features as a hex value */
673 		snprintf(cp, size, "%s0x%lx", sep, mask);
674 	}
675 
676 	proc_freq = cpufreq_quick_get(cpunum);
677 	if (!proc_freq)
678 		proc_freq = c->proc_freq / 1000;
679 
680 	seq_printf(m,
681 		   "processor  : %d\n"
682 		   "vendor     : %s\n"
683 		   "arch       : IA-64\n"
684 		   "family     : %u\n"
685 		   "model      : %u\n"
686 		   "model name : %s\n"
687 		   "revision   : %u\n"
688 		   "archrev    : %u\n"
689 		   "features   : %s\n"
690 		   "cpu number : %lu\n"
691 		   "cpu regs   : %u\n"
692 		   "cpu MHz    : %lu.%03lu\n"
693 		   "itc MHz    : %lu.%06lu\n"
694 		   "BogoMIPS   : %lu.%02lu\n",
695 		   cpunum, c->vendor, c->family, c->model,
696 		   c->model_name, c->revision, c->archrev,
697 		   features, c->ppn, c->number,
698 		   proc_freq / 1000, proc_freq % 1000,
699 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
700 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
701 #ifdef CONFIG_SMP
702 	seq_printf(m, "siblings   : %u\n",
703 		   cpumask_weight(&cpu_core_map[cpunum]));
704 	if (c->socket_id != -1)
705 		seq_printf(m, "physical id: %u\n", c->socket_id);
706 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
707 		seq_printf(m,
708 			   "core id    : %u\n"
709 			   "thread id  : %u\n",
710 			   c->core_id, c->thread_id);
711 #endif
712 	seq_printf(m,"\n");
713 
714 	return 0;
715 }
716 
717 static void *
718 c_start (struct seq_file *m, loff_t *pos)
719 {
720 #ifdef CONFIG_SMP
721 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
722 		++*pos;
723 #endif
724 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
725 }
726 
727 static void *
728 c_next (struct seq_file *m, void *v, loff_t *pos)
729 {
730 	++*pos;
731 	return c_start(m, pos);
732 }
733 
734 static void
735 c_stop (struct seq_file *m, void *v)
736 {
737 }
738 
739 const struct seq_operations cpuinfo_op = {
740 	.start =	c_start,
741 	.next =		c_next,
742 	.stop =		c_stop,
743 	.show =		show_cpuinfo
744 };
745 
746 #define MAX_BRANDS	8
747 static char brandname[MAX_BRANDS][128];
748 
749 static char *
750 get_model_name(__u8 family, __u8 model)
751 {
752 	static int overflow;
753 	char brand[128];
754 	int i;
755 
756 	memcpy(brand, "Unknown", 8);
757 	if (ia64_pal_get_brand_info(brand)) {
758 		if (family == 0x7)
759 			memcpy(brand, "Merced", 7);
760 		else if (family == 0x1f) switch (model) {
761 			case 0: memcpy(brand, "McKinley", 9); break;
762 			case 1: memcpy(brand, "Madison", 8); break;
763 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
764 		}
765 	}
766 	for (i = 0; i < MAX_BRANDS; i++)
767 		if (strcmp(brandname[i], brand) == 0)
768 			return brandname[i];
769 	for (i = 0; i < MAX_BRANDS; i++)
770 		if (brandname[i][0] == '\0')
771 			return strcpy(brandname[i], brand);
772 	if (overflow++ == 0)
773 		printk(KERN_ERR
774 		       "%s: Table overflow. Some processor model information will be missing\n",
775 		       __func__);
776 	return "Unknown";
777 }
778 
779 static void
780 identify_cpu (struct cpuinfo_ia64 *c)
781 {
782 	union {
783 		unsigned long bits[5];
784 		struct {
785 			/* id 0 & 1: */
786 			char vendor[16];
787 
788 			/* id 2 */
789 			u64 ppn;		/* processor serial number */
790 
791 			/* id 3: */
792 			unsigned number		:  8;
793 			unsigned revision	:  8;
794 			unsigned model		:  8;
795 			unsigned family		:  8;
796 			unsigned archrev	:  8;
797 			unsigned reserved	: 24;
798 
799 			/* id 4: */
800 			u64 features;
801 		} field;
802 	} cpuid;
803 	pal_vm_info_1_u_t vm1;
804 	pal_vm_info_2_u_t vm2;
805 	pal_status_t status;
806 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
807 	int i;
808 	for (i = 0; i < 5; ++i)
809 		cpuid.bits[i] = ia64_get_cpuid(i);
810 
811 	memcpy(c->vendor, cpuid.field.vendor, 16);
812 #ifdef CONFIG_SMP
813 	c->cpu = smp_processor_id();
814 
815 	/* below default values will be overwritten  by identify_siblings()
816 	 * for Multi-Threading/Multi-Core capable CPUs
817 	 */
818 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
819 	c->socket_id = -1;
820 
821 	identify_siblings(c);
822 
823 	if (c->threads_per_core > smp_num_siblings)
824 		smp_num_siblings = c->threads_per_core;
825 #endif
826 	c->ppn = cpuid.field.ppn;
827 	c->number = cpuid.field.number;
828 	c->revision = cpuid.field.revision;
829 	c->model = cpuid.field.model;
830 	c->family = cpuid.field.family;
831 	c->archrev = cpuid.field.archrev;
832 	c->features = cpuid.field.features;
833 	c->model_name = get_model_name(c->family, c->model);
834 
835 	status = ia64_pal_vm_summary(&vm1, &vm2);
836 	if (status == PAL_STATUS_SUCCESS) {
837 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
838 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
839 	}
840 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
841 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
842 }
843 
844 /*
845  * Do the following calculations:
846  *
847  * 1. the max. cache line size.
848  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
849  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
850  */
851 static void
852 get_cache_info(void)
853 {
854 	unsigned long line_size, max = 1;
855 	unsigned long l, levels, unique_caches;
856 	pal_cache_config_info_t cci;
857 	long status;
858 
859         status = ia64_pal_cache_summary(&levels, &unique_caches);
860         if (status != 0) {
861                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
862                        __func__, status);
863                 max = SMP_CACHE_BYTES;
864 		/* Safest setup for "flush_icache_range()" */
865 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
866 		/* Safest setup for "clflush_cache_range()" */
867 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
868 		goto out;
869         }
870 
871 	for (l = 0; l < levels; ++l) {
872 		/* cache_type (data_or_unified)=2 */
873 		status = ia64_pal_cache_config_info(l, 2, &cci);
874 		if (status != 0) {
875 			printk(KERN_ERR "%s: ia64_pal_cache_config_info"
876 				"(l=%lu, 2) failed (status=%ld)\n",
877 				__func__, l, status);
878 			max = SMP_CACHE_BYTES;
879 			/* The safest setup for "flush_icache_range()" */
880 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
881 			/* The safest setup for "clflush_cache_range()" */
882 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
883 			cci.pcci_unified = 1;
884 		} else {
885 			if (cci.pcci_stride < ia64_cache_stride_shift)
886 				ia64_cache_stride_shift = cci.pcci_stride;
887 
888 			line_size = 1 << cci.pcci_line_size;
889 			if (line_size > max)
890 				max = line_size;
891 		}
892 
893 		if (!cci.pcci_unified) {
894 			/* cache_type (instruction)=1*/
895 			status = ia64_pal_cache_config_info(l, 1, &cci);
896 			if (status != 0) {
897 				printk(KERN_ERR "%s: ia64_pal_cache_config_info"
898 					"(l=%lu, 1) failed (status=%ld)\n",
899 					__func__, l, status);
900 				/* The safest setup for flush_icache_range() */
901 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
902 			}
903 		}
904 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
905 			ia64_i_cache_stride_shift = cci.pcci_stride;
906 	}
907   out:
908 	if (max > ia64_max_cacheline_size)
909 		ia64_max_cacheline_size = max;
910 }
911 
912 /*
913  * cpu_init() initializes state that is per-CPU.  This function acts
914  * as a 'CPU state barrier', nothing should get across.
915  */
916 void
917 cpu_init (void)
918 {
919 	extern void ia64_mmu_init(void *);
920 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
921 	unsigned long num_phys_stacked;
922 	pal_vm_info_2_u_t vmi;
923 	unsigned int max_ctx;
924 	struct cpuinfo_ia64 *cpu_info;
925 	void *cpu_data;
926 
927 	cpu_data = per_cpu_init();
928 #ifdef CONFIG_SMP
929 	/*
930 	 * insert boot cpu into sibling and core mapes
931 	 * (must be done after per_cpu area is setup)
932 	 */
933 	if (smp_processor_id() == 0) {
934 		cpumask_set_cpu(0, &per_cpu(cpu_sibling_map, 0));
935 		cpumask_set_cpu(0, &cpu_core_map[0]);
936 	} else {
937 		/*
938 		 * Set ar.k3 so that assembly code in MCA handler can compute
939 		 * physical addresses of per cpu variables with a simple:
940 		 *   phys = ar.k3 + &per_cpu_var
941 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
942 		 * the TLB when needed. head.S already did this for cpu0.
943 		 */
944 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
945 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
946 	}
947 #endif
948 
949 	get_cache_info();
950 
951 	/*
952 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
953 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
954 	 * depends on the data returned by identify_cpu().  We break the dependency by
955 	 * accessing cpu_data() through the canonical per-CPU address.
956 	 */
957 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(ia64_cpu_info) - __per_cpu_start);
958 	identify_cpu(cpu_info);
959 
960 #ifdef CONFIG_MCKINLEY
961 	{
962 #		define FEATURE_SET 16
963 		struct ia64_pal_retval iprv;
964 
965 		if (cpu_info->family == 0x1f) {
966 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
967 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
968 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
969 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
970 		}
971 	}
972 #endif
973 
974 	/* Clear the stack memory reserved for pt_regs: */
975 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
976 
977 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
978 
979 	/*
980 	 * Initialize the page-table base register to a global
981 	 * directory with all zeroes.  This ensure that we can handle
982 	 * TLB-misses to user address-space even before we created the
983 	 * first user address-space.  This may happen, e.g., due to
984 	 * aggressive use of lfetch.fault.
985 	 */
986 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
987 
988 	/*
989 	 * Initialize default control register to defer speculative faults except
990 	 * for those arising from TLB misses, which are not deferred.  The
991 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
992 	 * the kernel must have recovery code for all speculative accesses).  Turn on
993 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
994 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
995 	 * be fine).
996 	 */
997 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
998 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
999 	mmgrab(&init_mm);
1000 	current->active_mm = &init_mm;
1001 	BUG_ON(current->mm);
1002 
1003 	ia64_mmu_init(ia64_imva(cpu_data));
1004 	ia64_mca_cpu_init(ia64_imva(cpu_data));
1005 
1006 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
1007 	ia64_set_itc(0);
1008 
1009 	/* disable all local interrupt sources: */
1010 	ia64_set_itv(1 << 16);
1011 	ia64_set_lrr0(1 << 16);
1012 	ia64_set_lrr1(1 << 16);
1013 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1014 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1015 
1016 	/* clear TPR & XTP to enable all interrupt classes: */
1017 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1018 
1019 	/* Clear any pending interrupts left by SAL/EFI */
1020 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1021 		ia64_eoi();
1022 
1023 #ifdef CONFIG_SMP
1024 	normal_xtp();
1025 #endif
1026 
1027 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1028 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1029 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1030 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1031 	} else {
1032 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1033 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1034 	}
1035 	while (max_ctx < ia64_ctx.max_ctx) {
1036 		unsigned int old = ia64_ctx.max_ctx;
1037 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1038 			break;
1039 	}
1040 
1041 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1042 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1043 		       "stacked regs\n");
1044 		num_phys_stacked = 96;
1045 	}
1046 	/* size of physical stacked register partition plus 8 bytes: */
1047 	if (num_phys_stacked > max_num_phys_stacked) {
1048 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1049 		max_num_phys_stacked = num_phys_stacked;
1050 	}
1051 	platform_cpu_init();
1052 }
1053 
1054 void __init
1055 check_bugs (void)
1056 {
1057 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1058 			       (unsigned long) __end___mckinley_e9_bundles);
1059 }
1060 
1061 static int __init run_dmi_scan(void)
1062 {
1063 	dmi_scan_machine();
1064 	dmi_memdev_walk();
1065 	dmi_set_dump_stack_arch_desc();
1066 	return 0;
1067 }
1068 core_initcall(run_dmi_scan);
1069