xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 6de298ff)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Architecture-specific setup.
4  *
5  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
6  *	David Mosberger-Tang <davidm@hpl.hp.com>
7  *	Stephane Eranian <eranian@hpl.hp.com>
8  * Copyright (C) 2000, 2004 Intel Corp
9  * 	Rohit Seth <rohit.seth@intel.com>
10  * 	Suresh Siddha <suresh.b.siddha@intel.com>
11  * 	Gordon Jin <gordon.jin@intel.com>
12  * Copyright (C) 1999 VA Linux Systems
13  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
14  *
15  * 12/26/04 S.Siddha, G.Jin, R.Seth
16  *			Add multi-threading and multi-core detection
17  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
18  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
19  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
20  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
21  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
22  * 01/07/99 S.Eranian	added the support for command line argument
23  * 06/24/99 W.Drummond	added boot_cpu_data.
24  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
25  */
26 #include <linux/module.h>
27 #include <linux/init.h>
28 #include <linux/pgtable.h>
29 
30 #include <linux/acpi.h>
31 #include <linux/console.h>
32 #include <linux/delay.h>
33 #include <linux/cpu.h>
34 #include <linux/kdev_t.h>
35 #include <linux/kernel.h>
36 #include <linux/memblock.h>
37 #include <linux/reboot.h>
38 #include <linux/sched/mm.h>
39 #include <linux/sched/clock.h>
40 #include <linux/sched/task_stack.h>
41 #include <linux/seq_file.h>
42 #include <linux/string.h>
43 #include <linux/threads.h>
44 #include <linux/screen_info.h>
45 #include <linux/dmi.h>
46 #include <linux/root_dev.h>
47 #include <linux/serial.h>
48 #include <linux/serial_core.h>
49 #include <linux/efi.h>
50 #include <linux/initrd.h>
51 #include <linux/pm.h>
52 #include <linux/cpufreq.h>
53 #include <linux/kexec.h>
54 #include <linux/crash_dump.h>
55 
56 #include <asm/mca.h>
57 #include <asm/meminit.h>
58 #include <asm/page.h>
59 #include <asm/patch.h>
60 #include <asm/processor.h>
61 #include <asm/sal.h>
62 #include <asm/sections.h>
63 #include <asm/setup.h>
64 #include <asm/smp.h>
65 #include <asm/tlbflush.h>
66 #include <asm/unistd.h>
67 #include <asm/uv/uv.h>
68 #include <asm/xtp.h>
69 
70 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
71 # error "struct cpuinfo_ia64 too big!"
72 #endif
73 
74 char ia64_platform_name[64];
75 
76 #ifdef CONFIG_SMP
77 unsigned long __per_cpu_offset[NR_CPUS];
78 EXPORT_SYMBOL(__per_cpu_offset);
79 #endif
80 
81 DEFINE_PER_CPU(struct cpuinfo_ia64, ia64_cpu_info);
82 EXPORT_SYMBOL(ia64_cpu_info);
83 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
84 #ifdef CONFIG_SMP
85 EXPORT_SYMBOL(local_per_cpu_offset);
86 #endif
87 unsigned long ia64_cycles_per_usec;
88 struct ia64_boot_param *ia64_boot_param;
89 struct screen_info screen_info;
90 unsigned long vga_console_iobase;
91 unsigned long vga_console_membase;
92 
93 static struct resource data_resource = {
94 	.name	= "Kernel data",
95 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
96 };
97 
98 static struct resource code_resource = {
99 	.name	= "Kernel code",
100 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
101 };
102 
103 static struct resource bss_resource = {
104 	.name	= "Kernel bss",
105 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
106 };
107 
108 unsigned long ia64_max_cacheline_size;
109 
110 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
111 EXPORT_SYMBOL(ia64_iobase);
112 struct io_space io_space[MAX_IO_SPACES];
113 EXPORT_SYMBOL(io_space);
114 unsigned int num_io_spaces;
115 
116 /*
117  * "flush_icache_range()" needs to know what processor dependent stride size to use
118  * when it makes i-cache(s) coherent with d-caches.
119  */
120 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
121 unsigned long ia64_i_cache_stride_shift = ~0;
122 /*
123  * "clflush_cache_range()" needs to know what processor dependent stride size to
124  * use when it flushes cache lines including both d-cache and i-cache.
125  */
126 /* Safest way to go: 32 bytes by 32 bytes */
127 #define	CACHE_STRIDE_SHIFT	5
128 unsigned long ia64_cache_stride_shift = ~0;
129 
130 /*
131  * We use a special marker for the end of memory and it uses the extra (+1) slot
132  */
133 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
134 static int num_rsvd_regions __initdata;
135 
136 
137 /*
138  * Filter incoming memory segments based on the primitive map created from the boot
139  * parameters. Segments contained in the map are removed from the memory ranges. A
140  * caller-specified function is called with the memory ranges that remain after filtering.
141  * This routine does not assume the incoming segments are sorted.
142  */
143 int __init
144 filter_rsvd_memory (u64 start, u64 end, void *arg)
145 {
146 	u64 range_start, range_end, prev_start;
147 	void (*func)(unsigned long, unsigned long, int);
148 	int i;
149 
150 #if IGNORE_PFN0
151 	if (start == PAGE_OFFSET) {
152 		printk(KERN_WARNING "warning: skipping physical page 0\n");
153 		start += PAGE_SIZE;
154 		if (start >= end) return 0;
155 	}
156 #endif
157 	/*
158 	 * lowest possible address(walker uses virtual)
159 	 */
160 	prev_start = PAGE_OFFSET;
161 	func = arg;
162 
163 	for (i = 0; i < num_rsvd_regions; ++i) {
164 		range_start = max(start, prev_start);
165 		range_end   = min(end, rsvd_region[i].start);
166 
167 		if (range_start < range_end)
168 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
169 
170 		/* nothing more available in this segment */
171 		if (range_end == end) return 0;
172 
173 		prev_start = rsvd_region[i].end;
174 	}
175 	/* end of memory marker allows full processing inside loop body */
176 	return 0;
177 }
178 
179 /*
180  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
181  * are not filtered out.
182  */
183 int __init
184 filter_memory(u64 start, u64 end, void *arg)
185 {
186 	void (*func)(unsigned long, unsigned long, int);
187 
188 #if IGNORE_PFN0
189 	if (start == PAGE_OFFSET) {
190 		printk(KERN_WARNING "warning: skipping physical page 0\n");
191 		start += PAGE_SIZE;
192 		if (start >= end)
193 			return 0;
194 	}
195 #endif
196 	func = arg;
197 	if (start < end)
198 		call_pernode_memory(__pa(start), end - start, func);
199 	return 0;
200 }
201 
202 static void __init
203 sort_regions (struct rsvd_region *rsvd_region, int max)
204 {
205 	int j;
206 
207 	/* simple bubble sorting */
208 	while (max--) {
209 		for (j = 0; j < max; ++j) {
210 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
211 				swap(rsvd_region[j], rsvd_region[j + 1]);
212 			}
213 		}
214 	}
215 }
216 
217 /* merge overlaps */
218 static int __init
219 merge_regions (struct rsvd_region *rsvd_region, int max)
220 {
221 	int i;
222 	for (i = 1; i < max; ++i) {
223 		if (rsvd_region[i].start >= rsvd_region[i-1].end)
224 			continue;
225 		if (rsvd_region[i].end > rsvd_region[i-1].end)
226 			rsvd_region[i-1].end = rsvd_region[i].end;
227 		--max;
228 		memmove(&rsvd_region[i], &rsvd_region[i+1],
229 			(max - i) * sizeof(struct rsvd_region));
230 	}
231 	return max;
232 }
233 
234 /*
235  * Request address space for all standard resources
236  */
237 static int __init register_memory(void)
238 {
239 	code_resource.start = ia64_tpa(_text);
240 	code_resource.end   = ia64_tpa(_etext) - 1;
241 	data_resource.start = ia64_tpa(_etext);
242 	data_resource.end   = ia64_tpa(_edata) - 1;
243 	bss_resource.start  = ia64_tpa(__bss_start);
244 	bss_resource.end    = ia64_tpa(_end) - 1;
245 	efi_initialize_iomem_resources(&code_resource, &data_resource,
246 			&bss_resource);
247 
248 	return 0;
249 }
250 
251 __initcall(register_memory);
252 
253 
254 #ifdef CONFIG_KEXEC
255 
256 /*
257  * This function checks if the reserved crashkernel is allowed on the specific
258  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
259  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
260  * kernel/dma/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
261  * in kdump case. See the comment in sba_init() in sba_iommu.c.
262  *
263  * So, the only machvec that really supports loading the kdump kernel
264  * over 4 GB is "uv".
265  */
266 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
267 {
268 	if (is_uv_system())
269 		return 1;
270 	else
271 		return pbase < (1UL << 32);
272 }
273 
274 static void __init setup_crashkernel(unsigned long total, int *n)
275 {
276 	unsigned long long base = 0, size = 0;
277 	int ret;
278 
279 	ret = parse_crashkernel(boot_command_line, total,
280 			&size, &base);
281 	if (ret == 0 && size > 0) {
282 		if (!base) {
283 			sort_regions(rsvd_region, *n);
284 			*n = merge_regions(rsvd_region, *n);
285 			base = kdump_find_rsvd_region(size,
286 					rsvd_region, *n);
287 		}
288 
289 		if (!check_crashkernel_memory(base, size)) {
290 			pr_warn("crashkernel: There would be kdump memory "
291 				"at %ld GB but this is unusable because it "
292 				"must\nbe below 4 GB. Change the memory "
293 				"configuration of the machine.\n",
294 				(unsigned long)(base >> 30));
295 			return;
296 		}
297 
298 		if (base != ~0UL) {
299 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
300 					"for crashkernel (System RAM: %ldMB)\n",
301 					(unsigned long)(size >> 20),
302 					(unsigned long)(base >> 20),
303 					(unsigned long)(total >> 20));
304 			rsvd_region[*n].start =
305 				(unsigned long)__va(base);
306 			rsvd_region[*n].end =
307 				(unsigned long)__va(base + size);
308 			(*n)++;
309 			crashk_res.start = base;
310 			crashk_res.end = base + size - 1;
311 		}
312 	}
313 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
314 	efi_memmap_res.end = efi_memmap_res.start +
315 		ia64_boot_param->efi_memmap_size;
316 	boot_param_res.start = __pa(ia64_boot_param);
317 	boot_param_res.end = boot_param_res.start +
318 		sizeof(*ia64_boot_param);
319 }
320 #else
321 static inline void __init setup_crashkernel(unsigned long total, int *n)
322 {}
323 #endif
324 
325 #ifdef CONFIG_CRASH_DUMP
326 static int __init reserve_elfcorehdr(u64 *start, u64 *end)
327 {
328 	u64 length;
329 
330 	/* We get the address using the kernel command line,
331 	 * but the size is extracted from the EFI tables.
332 	 * Both address and size are required for reservation
333 	 * to work properly.
334 	 */
335 
336 	if (!is_vmcore_usable())
337 		return -EINVAL;
338 
339 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
340 		vmcore_unusable();
341 		return -EINVAL;
342 	}
343 
344 	*start = (unsigned long)__va(elfcorehdr_addr);
345 	*end = *start + length;
346 	return 0;
347 }
348 #endif /* CONFIG_CRASH_DUMP */
349 
350 /**
351  * reserve_memory - setup reserved memory areas
352  *
353  * Setup the reserved memory areas set aside for the boot parameters,
354  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
355  * see arch/ia64/include/asm/meminit.h if you need to define more.
356  */
357 void __init
358 reserve_memory (void)
359 {
360 	int n = 0;
361 	unsigned long total_memory;
362 
363 	/*
364 	 * none of the entries in this table overlap
365 	 */
366 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
367 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
368 	n++;
369 
370 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
371 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
372 	n++;
373 
374 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
375 	rsvd_region[n].end   = (rsvd_region[n].start
376 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
377 	n++;
378 
379 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
380 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
381 	n++;
382 
383 #ifdef CONFIG_BLK_DEV_INITRD
384 	if (ia64_boot_param->initrd_start) {
385 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
386 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
387 		n++;
388 	}
389 #endif
390 
391 #ifdef CONFIG_CRASH_DUMP
392 	if (reserve_elfcorehdr(&rsvd_region[n].start,
393 			       &rsvd_region[n].end) == 0)
394 		n++;
395 #endif
396 
397 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
398 	n++;
399 
400 	setup_crashkernel(total_memory, &n);
401 
402 	/* end of memory marker */
403 	rsvd_region[n].start = ~0UL;
404 	rsvd_region[n].end   = ~0UL;
405 	n++;
406 
407 	num_rsvd_regions = n;
408 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
409 
410 	sort_regions(rsvd_region, num_rsvd_regions);
411 	num_rsvd_regions = merge_regions(rsvd_region, num_rsvd_regions);
412 
413 	/* reserve all regions except the end of memory marker with memblock */
414 	for (n = 0; n < num_rsvd_regions - 1; n++) {
415 		struct rsvd_region *region = &rsvd_region[n];
416 		phys_addr_t addr = __pa(region->start);
417 		phys_addr_t size = region->end - region->start;
418 
419 		memblock_reserve(addr, size);
420 	}
421 }
422 
423 /**
424  * find_initrd - get initrd parameters from the boot parameter structure
425  *
426  * Grab the initrd start and end from the boot parameter struct given us by
427  * the boot loader.
428  */
429 void __init
430 find_initrd (void)
431 {
432 #ifdef CONFIG_BLK_DEV_INITRD
433 	if (ia64_boot_param->initrd_start) {
434 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
435 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
436 
437 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%llu bytes)\n",
438 		       initrd_start, ia64_boot_param->initrd_size);
439 	}
440 #endif
441 }
442 
443 static void __init
444 io_port_init (void)
445 {
446 	unsigned long phys_iobase;
447 
448 	/*
449 	 * Set `iobase' based on the EFI memory map or, failing that, the
450 	 * value firmware left in ar.k0.
451 	 *
452 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
453 	 * the port's virtual address, so ia32_load_state() loads it with a
454 	 * user virtual address.  But in ia64 mode, glibc uses the
455 	 * *physical* address in ar.k0 to mmap the appropriate area from
456 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
457 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
458 	 *
459 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
460 	 * any of the I/O port spaces and are done via MMIO using the
461 	 * virtual mmio_base from the appropriate io_space[].
462 	 */
463 	phys_iobase = efi_get_iobase();
464 	if (!phys_iobase) {
465 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
466 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
467 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
468 	}
469 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
470 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
471 
472 	/* setup legacy IO port space */
473 	io_space[0].mmio_base = ia64_iobase;
474 	io_space[0].sparse = 1;
475 	num_io_spaces = 1;
476 }
477 
478 /**
479  * early_console_setup - setup debugging console
480  *
481  * Consoles started here require little enough setup that we can start using
482  * them very early in the boot process, either right after the machine
483  * vector initialization, or even before if the drivers can detect their hw.
484  *
485  * Returns non-zero if a console couldn't be setup.
486  */
487 static inline int __init
488 early_console_setup (char *cmdline)
489 {
490 #ifdef CONFIG_EFI_PCDP
491 	if (!efi_setup_pcdp_console(cmdline))
492 		return 0;
493 #endif
494 	return -1;
495 }
496 
497 static void __init
498 screen_info_setup(void)
499 {
500 	unsigned int orig_x, orig_y, num_cols, num_rows, font_height;
501 
502 	memset(&screen_info, 0, sizeof(screen_info));
503 
504 	if (!ia64_boot_param->console_info.num_rows ||
505 	    !ia64_boot_param->console_info.num_cols) {
506 		printk(KERN_WARNING "invalid screen-info, guessing 80x25\n");
507 		orig_x = 0;
508 		orig_y = 0;
509 		num_cols = 80;
510 		num_rows = 25;
511 		font_height = 16;
512 	} else {
513 		orig_x = ia64_boot_param->console_info.orig_x;
514 		orig_y = ia64_boot_param->console_info.orig_y;
515 		num_cols = ia64_boot_param->console_info.num_cols;
516 		num_rows = ia64_boot_param->console_info.num_rows;
517 		font_height = 400 / num_rows;
518 	}
519 
520 	screen_info.orig_x = orig_x;
521 	screen_info.orig_y = orig_y;
522 	screen_info.orig_video_cols  = num_cols;
523 	screen_info.orig_video_lines = num_rows;
524 	screen_info.orig_video_points = font_height;
525 	screen_info.orig_video_mode = 3;	/* XXX fake */
526 	screen_info.orig_video_isVGA = 1;	/* XXX fake */
527 	screen_info.orig_video_ega_bx = 3;	/* XXX fake */
528 }
529 
530 static inline void
531 mark_bsp_online (void)
532 {
533 #ifdef CONFIG_SMP
534 	/* If we register an early console, allow CPU 0 to printk */
535 	set_cpu_online(smp_processor_id(), true);
536 #endif
537 }
538 
539 static __initdata int nomca;
540 static __init int setup_nomca(char *s)
541 {
542 	nomca = 1;
543 	return 0;
544 }
545 early_param("nomca", setup_nomca);
546 
547 void __init
548 setup_arch (char **cmdline_p)
549 {
550 	unw_init();
551 
552 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
553 
554 	*cmdline_p = __va(ia64_boot_param->command_line);
555 	strscpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
556 
557 	efi_init();
558 	io_port_init();
559 
560 	uv_probe_system_type();
561 	parse_early_param();
562 
563 	if (early_console_setup(*cmdline_p) == 0)
564 		mark_bsp_online();
565 
566 	/* Initialize the ACPI boot-time table parser */
567 	acpi_table_init();
568 	early_acpi_boot_init();
569 #ifdef CONFIG_ACPI_NUMA
570 	acpi_numa_init();
571 	acpi_numa_fixup();
572 #ifdef CONFIG_ACPI_HOTPLUG_CPU
573 	prefill_possible_map();
574 #endif
575 	per_cpu_scan_finalize((cpumask_empty(&early_cpu_possible_map) ?
576 		32 : cpumask_weight(&early_cpu_possible_map)),
577 		additional_cpus > 0 ? additional_cpus : 0);
578 #endif /* CONFIG_ACPI_NUMA */
579 
580 #ifdef CONFIG_SMP
581 	smp_build_cpu_map();
582 #endif
583 	find_memory();
584 
585 	/* process SAL system table: */
586 	ia64_sal_init(__va(sal_systab_phys));
587 
588 #ifdef CONFIG_ITANIUM
589 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
590 #else
591 	{
592 		unsigned long num_phys_stacked;
593 
594 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
595 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
596 	}
597 #endif
598 
599 #ifdef CONFIG_SMP
600 	cpu_physical_id(0) = hard_smp_processor_id();
601 #endif
602 
603 	cpu_init();	/* initialize the bootstrap CPU */
604 	mmu_context_init();	/* initialize context_id bitmap */
605 
606 #ifdef CONFIG_VT
607 	if (!conswitchp) {
608 # if defined(CONFIG_VGA_CONSOLE)
609 		/*
610 		 * Non-legacy systems may route legacy VGA MMIO range to system
611 		 * memory.  vga_con probes the MMIO hole, so memory looks like
612 		 * a VGA device to it.  The EFI memory map can tell us if it's
613 		 * memory so we can avoid this problem.
614 		 */
615 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
616 			conswitchp = &vga_con;
617 # endif
618 	}
619 #endif
620 
621 	/* enable IA-64 Machine Check Abort Handling unless disabled */
622 	if (!nomca)
623 		ia64_mca_init();
624 
625 	/*
626 	 * Default to /dev/sda2.  This assumes that the EFI partition
627 	 * is physical disk 1 partition 1 and the Linux root disk is
628 	 * physical disk 1 partition 2.
629 	 */
630 	ROOT_DEV = MKDEV(SCSI_DISK0_MAJOR, 2);
631 
632 	if (is_uv_system())
633 		uv_setup(cmdline_p);
634 #ifdef CONFIG_SMP
635 	else
636 		init_smp_config();
637 #endif
638 
639 	screen_info_setup();
640 	paging_init();
641 
642 	clear_sched_clock_stable();
643 }
644 
645 /*
646  * Display cpu info for all CPUs.
647  */
648 static int
649 show_cpuinfo (struct seq_file *m, void *v)
650 {
651 #ifdef CONFIG_SMP
652 #	define lpj	c->loops_per_jiffy
653 #	define cpunum	c->cpu
654 #else
655 #	define lpj	loops_per_jiffy
656 #	define cpunum	0
657 #endif
658 	static struct {
659 		unsigned long mask;
660 		const char *feature_name;
661 	} feature_bits[] = {
662 		{ 1UL << 0, "branchlong" },
663 		{ 1UL << 1, "spontaneous deferral"},
664 		{ 1UL << 2, "16-byte atomic ops" }
665 	};
666 	char features[128], *cp, *sep;
667 	struct cpuinfo_ia64 *c = v;
668 	unsigned long mask;
669 	unsigned long proc_freq;
670 	int i, size;
671 
672 	mask = c->features;
673 
674 	/* build the feature string: */
675 	memcpy(features, "standard", 9);
676 	cp = features;
677 	size = sizeof(features);
678 	sep = "";
679 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
680 		if (mask & feature_bits[i].mask) {
681 			cp += snprintf(cp, size, "%s%s", sep,
682 				       feature_bits[i].feature_name),
683 			sep = ", ";
684 			mask &= ~feature_bits[i].mask;
685 			size = sizeof(features) - (cp - features);
686 		}
687 	}
688 	if (mask && size > 1) {
689 		/* print unknown features as a hex value */
690 		snprintf(cp, size, "%s0x%lx", sep, mask);
691 	}
692 
693 	proc_freq = cpufreq_quick_get(cpunum);
694 	if (!proc_freq)
695 		proc_freq = c->proc_freq / 1000;
696 
697 	seq_printf(m,
698 		   "processor  : %d\n"
699 		   "vendor     : %s\n"
700 		   "arch       : IA-64\n"
701 		   "family     : %u\n"
702 		   "model      : %u\n"
703 		   "model name : %s\n"
704 		   "revision   : %u\n"
705 		   "archrev    : %u\n"
706 		   "features   : %s\n"
707 		   "cpu number : %lu\n"
708 		   "cpu regs   : %u\n"
709 		   "cpu MHz    : %lu.%03lu\n"
710 		   "itc MHz    : %lu.%06lu\n"
711 		   "BogoMIPS   : %lu.%02lu\n",
712 		   cpunum, c->vendor, c->family, c->model,
713 		   c->model_name, c->revision, c->archrev,
714 		   features, c->ppn, c->number,
715 		   proc_freq / 1000, proc_freq % 1000,
716 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
717 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
718 #ifdef CONFIG_SMP
719 	seq_printf(m, "siblings   : %u\n",
720 		   cpumask_weight(&cpu_core_map[cpunum]));
721 	if (c->socket_id != -1)
722 		seq_printf(m, "physical id: %u\n", c->socket_id);
723 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
724 		seq_printf(m,
725 			   "core id    : %u\n"
726 			   "thread id  : %u\n",
727 			   c->core_id, c->thread_id);
728 #endif
729 	seq_printf(m,"\n");
730 
731 	return 0;
732 }
733 
734 static void *
735 c_start (struct seq_file *m, loff_t *pos)
736 {
737 #ifdef CONFIG_SMP
738 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
739 		++*pos;
740 #endif
741 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
742 }
743 
744 static void *
745 c_next (struct seq_file *m, void *v, loff_t *pos)
746 {
747 	++*pos;
748 	return c_start(m, pos);
749 }
750 
751 static void
752 c_stop (struct seq_file *m, void *v)
753 {
754 }
755 
756 const struct seq_operations cpuinfo_op = {
757 	.start =	c_start,
758 	.next =		c_next,
759 	.stop =		c_stop,
760 	.show =		show_cpuinfo
761 };
762 
763 #define MAX_BRANDS	8
764 static char brandname[MAX_BRANDS][128];
765 
766 static char *
767 get_model_name(__u8 family, __u8 model)
768 {
769 	static int overflow;
770 	char brand[128];
771 	int i;
772 
773 	memcpy(brand, "Unknown", 8);
774 	if (ia64_pal_get_brand_info(brand)) {
775 		if (family == 0x7)
776 			memcpy(brand, "Merced", 7);
777 		else if (family == 0x1f) switch (model) {
778 			case 0: memcpy(brand, "McKinley", 9); break;
779 			case 1: memcpy(brand, "Madison", 8); break;
780 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
781 		}
782 	}
783 	for (i = 0; i < MAX_BRANDS; i++)
784 		if (strcmp(brandname[i], brand) == 0)
785 			return brandname[i];
786 	for (i = 0; i < MAX_BRANDS; i++)
787 		if (brandname[i][0] == '\0')
788 			return strcpy(brandname[i], brand);
789 	if (overflow++ == 0)
790 		printk(KERN_ERR
791 		       "%s: Table overflow. Some processor model information will be missing\n",
792 		       __func__);
793 	return "Unknown";
794 }
795 
796 static void
797 identify_cpu (struct cpuinfo_ia64 *c)
798 {
799 	union {
800 		unsigned long bits[5];
801 		struct {
802 			/* id 0 & 1: */
803 			char vendor[16];
804 
805 			/* id 2 */
806 			u64 ppn;		/* processor serial number */
807 
808 			/* id 3: */
809 			unsigned number		:  8;
810 			unsigned revision	:  8;
811 			unsigned model		:  8;
812 			unsigned family		:  8;
813 			unsigned archrev	:  8;
814 			unsigned reserved	: 24;
815 
816 			/* id 4: */
817 			u64 features;
818 		} field;
819 	} cpuid;
820 	pal_vm_info_1_u_t vm1;
821 	pal_vm_info_2_u_t vm2;
822 	pal_status_t status;
823 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
824 	int i;
825 	for (i = 0; i < 5; ++i)
826 		cpuid.bits[i] = ia64_get_cpuid(i);
827 
828 	memcpy(c->vendor, cpuid.field.vendor, 16);
829 #ifdef CONFIG_SMP
830 	c->cpu = smp_processor_id();
831 
832 	/* below default values will be overwritten  by identify_siblings()
833 	 * for Multi-Threading/Multi-Core capable CPUs
834 	 */
835 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
836 	c->socket_id = -1;
837 
838 	identify_siblings(c);
839 
840 	if (c->threads_per_core > smp_num_siblings)
841 		smp_num_siblings = c->threads_per_core;
842 #endif
843 	c->ppn = cpuid.field.ppn;
844 	c->number = cpuid.field.number;
845 	c->revision = cpuid.field.revision;
846 	c->model = cpuid.field.model;
847 	c->family = cpuid.field.family;
848 	c->archrev = cpuid.field.archrev;
849 	c->features = cpuid.field.features;
850 	c->model_name = get_model_name(c->family, c->model);
851 
852 	status = ia64_pal_vm_summary(&vm1, &vm2);
853 	if (status == PAL_STATUS_SUCCESS) {
854 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
855 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
856 	}
857 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
858 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
859 }
860 
861 /*
862  * Do the following calculations:
863  *
864  * 1. the max. cache line size.
865  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
866  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
867  */
868 static void
869 get_cache_info(void)
870 {
871 	unsigned long line_size, max = 1;
872 	unsigned long l, levels, unique_caches;
873 	pal_cache_config_info_t cci;
874 	long status;
875 
876         status = ia64_pal_cache_summary(&levels, &unique_caches);
877         if (status != 0) {
878                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
879                        __func__, status);
880                 max = SMP_CACHE_BYTES;
881 		/* Safest setup for "flush_icache_range()" */
882 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
883 		/* Safest setup for "clflush_cache_range()" */
884 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
885 		goto out;
886         }
887 
888 	for (l = 0; l < levels; ++l) {
889 		/* cache_type (data_or_unified)=2 */
890 		status = ia64_pal_cache_config_info(l, 2, &cci);
891 		if (status != 0) {
892 			printk(KERN_ERR "%s: ia64_pal_cache_config_info"
893 				"(l=%lu, 2) failed (status=%ld)\n",
894 				__func__, l, status);
895 			max = SMP_CACHE_BYTES;
896 			/* The safest setup for "flush_icache_range()" */
897 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
898 			/* The safest setup for "clflush_cache_range()" */
899 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
900 			cci.pcci_unified = 1;
901 		} else {
902 			if (cci.pcci_stride < ia64_cache_stride_shift)
903 				ia64_cache_stride_shift = cci.pcci_stride;
904 
905 			line_size = 1 << cci.pcci_line_size;
906 			if (line_size > max)
907 				max = line_size;
908 		}
909 
910 		if (!cci.pcci_unified) {
911 			/* cache_type (instruction)=1*/
912 			status = ia64_pal_cache_config_info(l, 1, &cci);
913 			if (status != 0) {
914 				printk(KERN_ERR "%s: ia64_pal_cache_config_info"
915 					"(l=%lu, 1) failed (status=%ld)\n",
916 					__func__, l, status);
917 				/* The safest setup for flush_icache_range() */
918 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
919 			}
920 		}
921 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
922 			ia64_i_cache_stride_shift = cci.pcci_stride;
923 	}
924   out:
925 	if (max > ia64_max_cacheline_size)
926 		ia64_max_cacheline_size = max;
927 }
928 
929 /*
930  * cpu_init() initializes state that is per-CPU.  This function acts
931  * as a 'CPU state barrier', nothing should get across.
932  */
933 void
934 cpu_init (void)
935 {
936 	extern void ia64_mmu_init(void *);
937 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
938 	unsigned long num_phys_stacked;
939 	pal_vm_info_2_u_t vmi;
940 	unsigned int max_ctx;
941 	struct cpuinfo_ia64 *cpu_info;
942 	void *cpu_data;
943 
944 	cpu_data = per_cpu_init();
945 #ifdef CONFIG_SMP
946 	/*
947 	 * insert boot cpu into sibling and core mapes
948 	 * (must be done after per_cpu area is setup)
949 	 */
950 	if (smp_processor_id() == 0) {
951 		cpumask_set_cpu(0, &per_cpu(cpu_sibling_map, 0));
952 		cpumask_set_cpu(0, &cpu_core_map[0]);
953 	} else {
954 		/*
955 		 * Set ar.k3 so that assembly code in MCA handler can compute
956 		 * physical addresses of per cpu variables with a simple:
957 		 *   phys = ar.k3 + &per_cpu_var
958 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
959 		 * the TLB when needed. head.S already did this for cpu0.
960 		 */
961 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
962 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
963 	}
964 #endif
965 
966 	get_cache_info();
967 
968 	/*
969 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
970 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
971 	 * depends on the data returned by identify_cpu().  We break the dependency by
972 	 * accessing cpu_data() through the canonical per-CPU address.
973 	 */
974 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(ia64_cpu_info) - __per_cpu_start);
975 	identify_cpu(cpu_info);
976 
977 #ifdef CONFIG_MCKINLEY
978 	{
979 #		define FEATURE_SET 16
980 		struct ia64_pal_retval iprv;
981 
982 		if (cpu_info->family == 0x1f) {
983 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
984 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
985 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
986 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
987 		}
988 	}
989 #endif
990 
991 	/* Clear the stack memory reserved for pt_regs: */
992 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
993 
994 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
995 
996 	/*
997 	 * Initialize the page-table base register to a global
998 	 * directory with all zeroes.  This ensure that we can handle
999 	 * TLB-misses to user address-space even before we created the
1000 	 * first user address-space.  This may happen, e.g., due to
1001 	 * aggressive use of lfetch.fault.
1002 	 */
1003 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
1004 
1005 	/*
1006 	 * Initialize default control register to defer speculative faults except
1007 	 * for those arising from TLB misses, which are not deferred.  The
1008 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
1009 	 * the kernel must have recovery code for all speculative accesses).  Turn on
1010 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
1011 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
1012 	 * be fine).
1013 	 */
1014 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
1015 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
1016 	mmgrab(&init_mm);
1017 	current->active_mm = &init_mm;
1018 	BUG_ON(current->mm);
1019 
1020 	ia64_mmu_init(ia64_imva(cpu_data));
1021 	ia64_mca_cpu_init(ia64_imva(cpu_data));
1022 
1023 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
1024 	ia64_set_itc(0);
1025 
1026 	/* disable all local interrupt sources: */
1027 	ia64_set_itv(1 << 16);
1028 	ia64_set_lrr0(1 << 16);
1029 	ia64_set_lrr1(1 << 16);
1030 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1031 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1032 
1033 	/* clear TPR & XTP to enable all interrupt classes: */
1034 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1035 
1036 	/* Clear any pending interrupts left by SAL/EFI */
1037 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1038 		ia64_eoi();
1039 
1040 #ifdef CONFIG_SMP
1041 	normal_xtp();
1042 #endif
1043 
1044 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1045 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1046 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1047 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1048 	} else {
1049 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1050 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1051 	}
1052 	while (max_ctx < ia64_ctx.max_ctx) {
1053 		unsigned int old = ia64_ctx.max_ctx;
1054 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1055 			break;
1056 	}
1057 
1058 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1059 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1060 		       "stacked regs\n");
1061 		num_phys_stacked = 96;
1062 	}
1063 	/* size of physical stacked register partition plus 8 bytes: */
1064 	if (num_phys_stacked > max_num_phys_stacked) {
1065 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1066 		max_num_phys_stacked = num_phys_stacked;
1067 	}
1068 }
1069 
1070 void __init arch_cpu_finalize_init(void)
1071 {
1072 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1073 			       (unsigned long) __end___mckinley_e9_bundles);
1074 }
1075 
1076 static int __init run_dmi_scan(void)
1077 {
1078 	dmi_setup();
1079 	return 0;
1080 }
1081 core_initcall(run_dmi_scan);
1082