1 /*
2  * Machine specific setup for xen
3  *
4  * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
5  */
6 
7 #include <linux/module.h>
8 #include <linux/sched.h>
9 #include <linux/mm.h>
10 #include <linux/pm.h>
11 #include <linux/memblock.h>
12 #include <linux/cpuidle.h>
13 #include <linux/cpufreq.h>
14 
15 #include <asm/elf.h>
16 #include <asm/vdso.h>
17 #include <asm/e820.h>
18 #include <asm/setup.h>
19 #include <asm/acpi.h>
20 #include <asm/numa.h>
21 #include <asm/xen/hypervisor.h>
22 #include <asm/xen/hypercall.h>
23 
24 #include <xen/xen.h>
25 #include <xen/page.h>
26 #include <xen/interface/callback.h>
27 #include <xen/interface/memory.h>
28 #include <xen/interface/physdev.h>
29 #include <xen/features.h>
30 
31 #include "xen-ops.h"
32 #include "vdso.h"
33 
34 /* These are code, but not functions.  Defined in entry.S */
35 extern const char xen_hypervisor_callback[];
36 extern const char xen_failsafe_callback[];
37 extern void xen_sysenter_target(void);
38 extern void xen_syscall_target(void);
39 extern void xen_syscall32_target(void);
40 
41 /* Amount of extra memory space we add to the e820 ranges */
42 struct xen_memory_region xen_extra_mem[XEN_EXTRA_MEM_MAX_REGIONS] __initdata;
43 
44 /* Number of pages released from the initial allocation. */
45 unsigned long xen_released_pages;
46 
47 /*
48  * The maximum amount of extra memory compared to the base size.  The
49  * main scaling factor is the size of struct page.  At extreme ratios
50  * of base:extra, all the base memory can be filled with page
51  * structures for the extra memory, leaving no space for anything
52  * else.
53  *
54  * 10x seems like a reasonable balance between scaling flexibility and
55  * leaving a practically usable system.
56  */
57 #define EXTRA_MEM_RATIO		(10)
58 
xen_add_extra_mem(u64 start,u64 size)59 static void __init xen_add_extra_mem(u64 start, u64 size)
60 {
61 	unsigned long pfn;
62 	int i;
63 
64 	for (i = 0; i < XEN_EXTRA_MEM_MAX_REGIONS; i++) {
65 		/* Add new region. */
66 		if (xen_extra_mem[i].size == 0) {
67 			xen_extra_mem[i].start = start;
68 			xen_extra_mem[i].size  = size;
69 			break;
70 		}
71 		/* Append to existing region. */
72 		if (xen_extra_mem[i].start + xen_extra_mem[i].size == start) {
73 			xen_extra_mem[i].size += size;
74 			break;
75 		}
76 	}
77 	if (i == XEN_EXTRA_MEM_MAX_REGIONS)
78 		printk(KERN_WARNING "Warning: not enough extra memory regions\n");
79 
80 	memblock_reserve(start, size);
81 
82 	xen_max_p2m_pfn = PFN_DOWN(start + size);
83 	for (pfn = PFN_DOWN(start); pfn < xen_max_p2m_pfn; pfn++) {
84 		unsigned long mfn = pfn_to_mfn(pfn);
85 
86 		if (WARN(mfn == pfn, "Trying to over-write 1-1 mapping (pfn: %lx)\n", pfn))
87 			continue;
88 		WARN(mfn != INVALID_P2M_ENTRY, "Trying to remove %lx which has %lx mfn!\n",
89 			pfn, mfn);
90 
91 		__set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
92 	}
93 }
94 
xen_release_chunk(unsigned long start,unsigned long end)95 static unsigned long __init xen_release_chunk(unsigned long start,
96 					      unsigned long end)
97 {
98 	struct xen_memory_reservation reservation = {
99 		.address_bits = 0,
100 		.extent_order = 0,
101 		.domid        = DOMID_SELF
102 	};
103 	unsigned long len = 0;
104 	unsigned long pfn;
105 	int ret;
106 
107 	for(pfn = start; pfn < end; pfn++) {
108 		unsigned long mfn = pfn_to_mfn(pfn);
109 
110 		/* Make sure pfn exists to start with */
111 		if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
112 			continue;
113 
114 		set_xen_guest_handle(reservation.extent_start, &mfn);
115 		reservation.nr_extents = 1;
116 
117 		ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
118 					   &reservation);
119 		WARN(ret != 1, "Failed to release pfn %lx err=%d\n", pfn, ret);
120 		if (ret == 1) {
121 			__set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
122 			len++;
123 		}
124 	}
125 	printk(KERN_INFO "Freeing  %lx-%lx pfn range: %lu pages freed\n",
126 	       start, end, len);
127 
128 	return len;
129 }
130 
xen_set_identity_and_release(const struct e820entry * list,size_t map_size,unsigned long nr_pages)131 static unsigned long __init xen_set_identity_and_release(
132 	const struct e820entry *list, size_t map_size, unsigned long nr_pages)
133 {
134 	phys_addr_t start = 0;
135 	unsigned long released = 0;
136 	unsigned long identity = 0;
137 	const struct e820entry *entry;
138 	int i;
139 
140 	/*
141 	 * Combine non-RAM regions and gaps until a RAM region (or the
142 	 * end of the map) is reached, then set the 1:1 map and
143 	 * release the pages (if available) in those non-RAM regions.
144 	 *
145 	 * The combined non-RAM regions are rounded to a whole number
146 	 * of pages so any partial pages are accessible via the 1:1
147 	 * mapping.  This is needed for some BIOSes that put (for
148 	 * example) the DMI tables in a reserved region that begins on
149 	 * a non-page boundary.
150 	 */
151 	for (i = 0, entry = list; i < map_size; i++, entry++) {
152 		phys_addr_t end = entry->addr + entry->size;
153 
154 		if (entry->type == E820_RAM || i == map_size - 1) {
155 			unsigned long start_pfn = PFN_DOWN(start);
156 			unsigned long end_pfn = PFN_UP(end);
157 
158 			if (entry->type == E820_RAM)
159 				end_pfn = PFN_UP(entry->addr);
160 
161 			if (start_pfn < end_pfn) {
162 				if (start_pfn < nr_pages)
163 					released += xen_release_chunk(
164 						start_pfn, min(end_pfn, nr_pages));
165 
166 				identity += set_phys_range_identity(
167 					start_pfn, end_pfn);
168 			}
169 			start = end;
170 		}
171 	}
172 
173 	printk(KERN_INFO "Released %lu pages of unused memory\n", released);
174 	printk(KERN_INFO "Set %ld page(s) to 1-1 mapping\n", identity);
175 
176 	return released;
177 }
178 
xen_get_max_pages(void)179 static unsigned long __init xen_get_max_pages(void)
180 {
181 	unsigned long max_pages = MAX_DOMAIN_PAGES;
182 	domid_t domid = DOMID_SELF;
183 	int ret;
184 
185 	/*
186 	 * For the initial domain we use the maximum reservation as
187 	 * the maximum page.
188 	 *
189 	 * For guest domains the current maximum reservation reflects
190 	 * the current maximum rather than the static maximum. In this
191 	 * case the e820 map provided to us will cover the static
192 	 * maximum region.
193 	 */
194 	if (xen_initial_domain()) {
195 		ret = HYPERVISOR_memory_op(XENMEM_maximum_reservation, &domid);
196 		if (ret > 0)
197 			max_pages = ret;
198 	}
199 
200 	return min(max_pages, MAX_DOMAIN_PAGES);
201 }
202 
xen_align_and_add_e820_region(u64 start,u64 size,int type)203 static void xen_align_and_add_e820_region(u64 start, u64 size, int type)
204 {
205 	u64 end = start + size;
206 
207 	/* Align RAM regions to page boundaries. */
208 	if (type == E820_RAM) {
209 		start = PAGE_ALIGN(start);
210 		end &= ~((u64)PAGE_SIZE - 1);
211 	}
212 
213 	e820_add_region(start, end - start, type);
214 }
215 
xen_ignore_unusable(struct e820entry * list,size_t map_size)216 void xen_ignore_unusable(struct e820entry *list, size_t map_size)
217 {
218 	struct e820entry *entry;
219 	unsigned int i;
220 
221 	for (i = 0, entry = list; i < map_size; i++, entry++) {
222 		if (entry->type == E820_UNUSABLE)
223 			entry->type = E820_RAM;
224 	}
225 }
226 
227 /**
228  * machine_specific_memory_setup - Hook for machine specific memory setup.
229  **/
xen_memory_setup(void)230 char * __init xen_memory_setup(void)
231 {
232 	static struct e820entry map[E820MAX] __initdata;
233 
234 	unsigned long max_pfn = xen_start_info->nr_pages;
235 	unsigned long long mem_end;
236 	int rc;
237 	struct xen_memory_map memmap;
238 	unsigned long max_pages;
239 	unsigned long extra_pages = 0;
240 	int i;
241 	int op;
242 
243 	max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
244 	mem_end = PFN_PHYS(max_pfn);
245 
246 	memmap.nr_entries = E820MAX;
247 	set_xen_guest_handle(memmap.buffer, map);
248 
249 	op = xen_initial_domain() ?
250 		XENMEM_machine_memory_map :
251 		XENMEM_memory_map;
252 	rc = HYPERVISOR_memory_op(op, &memmap);
253 	if (rc == -ENOSYS) {
254 		BUG_ON(xen_initial_domain());
255 		memmap.nr_entries = 1;
256 		map[0].addr = 0ULL;
257 		map[0].size = mem_end;
258 		/* 8MB slack (to balance backend allocations). */
259 		map[0].size += 8ULL << 20;
260 		map[0].type = E820_RAM;
261 		rc = 0;
262 	}
263 	BUG_ON(rc);
264 
265 	/*
266 	 * Xen won't allow a 1:1 mapping to be created to UNUSABLE
267 	 * regions, so if we're using the machine memory map leave the
268 	 * region as RAM as it is in the pseudo-physical map.
269 	 *
270 	 * UNUSABLE regions in domUs are not handled and will need
271 	 * a patch in the future.
272 	 */
273 	if (xen_initial_domain())
274 		xen_ignore_unusable(map, memmap.nr_entries);
275 
276 	/* Make sure the Xen-supplied memory map is well-ordered. */
277 	sanitize_e820_map(map, memmap.nr_entries, &memmap.nr_entries);
278 
279 	max_pages = xen_get_max_pages();
280 	if (max_pages > max_pfn)
281 		extra_pages += max_pages - max_pfn;
282 
283 	/*
284 	 * Set P2M for all non-RAM pages and E820 gaps to be identity
285 	 * type PFNs.  Any RAM pages that would be made inaccesible by
286 	 * this are first released.
287 	 */
288 	xen_released_pages = xen_set_identity_and_release(
289 		map, memmap.nr_entries, max_pfn);
290 	extra_pages += xen_released_pages;
291 
292 	/*
293 	 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
294 	 * factor the base size.  On non-highmem systems, the base
295 	 * size is the full initial memory allocation; on highmem it
296 	 * is limited to the max size of lowmem, so that it doesn't
297 	 * get completely filled.
298 	 *
299 	 * In principle there could be a problem in lowmem systems if
300 	 * the initial memory is also very large with respect to
301 	 * lowmem, but we won't try to deal with that here.
302 	 */
303 	extra_pages = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
304 			  extra_pages);
305 
306 	i = 0;
307 	while (i < memmap.nr_entries) {
308 		u64 addr = map[i].addr;
309 		u64 size = map[i].size;
310 		u32 type = map[i].type;
311 
312 		if (type == E820_RAM) {
313 			if (addr < mem_end) {
314 				size = min(size, mem_end - addr);
315 			} else if (extra_pages) {
316 				size = min(size, (u64)extra_pages * PAGE_SIZE);
317 				extra_pages -= size / PAGE_SIZE;
318 				xen_add_extra_mem(addr, size);
319 			} else
320 				type = E820_UNUSABLE;
321 		}
322 
323 		xen_align_and_add_e820_region(addr, size, type);
324 
325 		map[i].addr += size;
326 		map[i].size -= size;
327 		if (map[i].size == 0)
328 			i++;
329 	}
330 
331 	/*
332 	 * In domU, the ISA region is normal, usable memory, but we
333 	 * reserve ISA memory anyway because too many things poke
334 	 * about in there.
335 	 */
336 	e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
337 			E820_RESERVED);
338 
339 	/*
340 	 * Reserve Xen bits:
341 	 *  - mfn_list
342 	 *  - xen_start_info
343 	 * See comment above "struct start_info" in <xen/interface/xen.h>
344 	 */
345 	memblock_reserve(__pa(xen_start_info->mfn_list),
346 			 xen_start_info->pt_base - xen_start_info->mfn_list);
347 
348 	sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
349 
350 	return "Xen";
351 }
352 
353 /*
354  * Set the bit indicating "nosegneg" library variants should be used.
355  * We only need to bother in pure 32-bit mode; compat 32-bit processes
356  * can have un-truncated segments, so wrapping around is allowed.
357  */
fiddle_vdso(void)358 static void __init fiddle_vdso(void)
359 {
360 #ifdef CONFIG_X86_32
361 	u32 *mask;
362 	mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
363 	*mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
364 	mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
365 	*mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
366 #endif
367 }
368 
register_callback(unsigned type,const void * func)369 static int __cpuinit register_callback(unsigned type, const void *func)
370 {
371 	struct callback_register callback = {
372 		.type = type,
373 		.address = XEN_CALLBACK(__KERNEL_CS, func),
374 		.flags = CALLBACKF_mask_events,
375 	};
376 
377 	return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
378 }
379 
xen_enable_sysenter(void)380 void __cpuinit xen_enable_sysenter(void)
381 {
382 	int ret;
383 	unsigned sysenter_feature;
384 
385 #ifdef CONFIG_X86_32
386 	sysenter_feature = X86_FEATURE_SEP;
387 #else
388 	sysenter_feature = X86_FEATURE_SYSENTER32;
389 #endif
390 
391 	if (!boot_cpu_has(sysenter_feature))
392 		return;
393 
394 	ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
395 	if(ret != 0)
396 		setup_clear_cpu_cap(sysenter_feature);
397 }
398 
xen_enable_syscall(void)399 void __cpuinit xen_enable_syscall(void)
400 {
401 #ifdef CONFIG_X86_64
402 	int ret;
403 
404 	ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
405 	if (ret != 0) {
406 		printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
407 		/* Pretty fatal; 64-bit userspace has no other
408 		   mechanism for syscalls. */
409 	}
410 
411 	if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
412 		ret = register_callback(CALLBACKTYPE_syscall32,
413 					xen_syscall32_target);
414 		if (ret != 0)
415 			setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
416 	}
417 #endif /* CONFIG_X86_64 */
418 }
419 
xen_arch_setup(void)420 void __init xen_arch_setup(void)
421 {
422 	xen_panic_handler_init();
423 
424 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
425 	HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
426 
427 	if (!xen_feature(XENFEAT_auto_translated_physmap))
428 		HYPERVISOR_vm_assist(VMASST_CMD_enable,
429 				     VMASST_TYPE_pae_extended_cr3);
430 
431 	if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
432 	    register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
433 		BUG();
434 
435 	xen_enable_sysenter();
436 	xen_enable_syscall();
437 
438 #ifdef CONFIG_ACPI
439 	if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
440 		printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
441 		disable_acpi();
442 	}
443 #endif
444 
445 	memcpy(boot_command_line, xen_start_info->cmd_line,
446 	       MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
447 	       COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
448 
449 	/* Set up idle, making sure it calls safe_halt() pvop */
450 #ifdef CONFIG_X86_32
451 	boot_cpu_data.hlt_works_ok = 1;
452 #endif
453 	disable_cpuidle();
454 	disable_cpufreq();
455 	WARN_ON(set_pm_idle_to_default());
456 	fiddle_vdso();
457 #ifdef CONFIG_NUMA
458 	numa_off = 1;
459 #endif
460 }
461