2 * Machine specific setup for xen
4 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
7 #include <linux/module.h>
8 #include <linux/sched.h>
11 #include <linux/memblock.h>
16 #include <asm/setup.h>
18 #include <asm/xen/hypervisor.h>
19 #include <asm/xen/hypercall.h>
23 #include <xen/interface/callback.h>
24 #include <xen/interface/memory.h>
25 #include <xen/interface/physdev.h>
26 #include <xen/interface/memory.h>
27 #include <xen/features.h>
32 /* These are code, but not functions. Defined in entry.S */
33 extern const char xen_hypervisor_callback[];
34 extern const char xen_failsafe_callback[];
35 extern void xen_sysenter_target(void);
36 extern void xen_syscall_target(void);
37 extern void xen_syscall32_target(void);
39 /* Amount of extra memory space we add to the e820 ranges */
40 phys_addr_t xen_extra_mem_start, xen_extra_mem_size;
43 * The maximum amount of extra memory compared to the base size. The
44 * main scaling factor is the size of struct page. At extreme ratios
45 * of base:extra, all the base memory can be filled with page
46 * structures for the extra memory, leaving no space for anything
49 * 10x seems like a reasonable balance between scaling flexibility and
50 * leaving a practically usable system.
52 #define EXTRA_MEM_RATIO (10)
54 static __init void xen_add_extra_mem(unsigned long pages)
56 u64 size = (u64)pages * PAGE_SIZE;
57 u64 extra_start = xen_extra_mem_start + xen_extra_mem_size;
62 e820_add_region(extra_start, size, E820_RAM);
63 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
65 memblock_x86_reserve_range(extra_start, extra_start + size, "XEN EXTRA");
67 xen_extra_mem_size += size;
69 xen_max_p2m_pfn = PFN_DOWN(extra_start + size);
72 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
75 struct xen_memory_reservation reservation = {
80 unsigned long start, end;
81 unsigned long len = 0;
85 start = PFN_UP(start_addr);
86 end = PFN_DOWN(end_addr);
91 printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
93 for(pfn = start; pfn < end; pfn++) {
94 unsigned long mfn = pfn_to_mfn(pfn);
96 /* Make sure pfn exists to start with */
97 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
100 set_xen_guest_handle(reservation.extent_start, &mfn);
101 reservation.nr_extents = 1;
103 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
105 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
108 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
112 printk(KERN_CONT "%ld pages freed\n", len);
117 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
118 const struct e820map *e820)
120 phys_addr_t max_addr = PFN_PHYS(max_pfn);
121 phys_addr_t last_end = 0;
122 unsigned long released = 0;
125 for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
126 phys_addr_t end = e820->map[i].addr;
127 end = min(max_addr, end);
129 released += xen_release_chunk(last_end, end);
130 last_end = e820->map[i].addr + e820->map[i].size;
133 if (last_end < max_addr)
134 released += xen_release_chunk(last_end, max_addr);
136 printk(KERN_INFO "released %ld pages of unused memory\n", released);
141 * machine_specific_memory_setup - Hook for machine specific memory setup.
143 char * __init xen_memory_setup(void)
145 static struct e820entry map[E820MAX] __initdata;
147 unsigned long max_pfn = xen_start_info->nr_pages;
148 unsigned long long mem_end;
150 struct xen_memory_map memmap;
151 unsigned long extra_pages = 0;
152 unsigned long extra_limit;
156 max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
157 mem_end = PFN_PHYS(max_pfn);
159 memmap.nr_entries = E820MAX;
160 set_xen_guest_handle(memmap.buffer, map);
162 op = xen_initial_domain() ?
163 XENMEM_machine_memory_map :
165 rc = HYPERVISOR_memory_op(op, &memmap);
167 memmap.nr_entries = 1;
169 map[0].size = mem_end;
170 /* 8MB slack (to balance backend allocations). */
171 map[0].size += 8ULL << 20;
172 map[0].type = E820_RAM;
178 xen_extra_mem_start = mem_end;
179 for (i = 0; i < memmap.nr_entries; i++) {
180 unsigned long long end = map[i].addr + map[i].size;
182 if (map[i].type == E820_RAM) {
183 if (map[i].addr < mem_end && end > mem_end) {
184 /* Truncate region to max_mem. */
185 u64 delta = end - mem_end;
187 map[i].size -= delta;
188 extra_pages += PFN_DOWN(delta);
194 if (end > xen_extra_mem_start)
195 xen_extra_mem_start = end;
197 /* If region is non-RAM or below mem_end, add what remains */
198 if ((map[i].type != E820_RAM || map[i].addr < mem_end) &&
200 e820_add_region(map[i].addr, map[i].size, map[i].type);
204 * Even though this is normal, usable memory under Xen, reserve
205 * ISA memory anyway because too many things think they can poke
208 * In a dom0 kernel, this region is identity mapped with the
209 * hardware ISA area, so it really is out of bounds.
211 e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
218 * See comment above "struct start_info" in <xen/interface/xen.h>
220 memblock_x86_reserve_range(__pa(xen_start_info->mfn_list),
221 __pa(xen_start_info->pt_base),
224 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
226 extra_pages += xen_return_unused_memory(xen_start_info->nr_pages, &e820);
229 * Clamp the amount of extra memory to a EXTRA_MEM_RATIO
230 * factor the base size. On non-highmem systems, the base
231 * size is the full initial memory allocation; on highmem it
232 * is limited to the max size of lowmem, so that it doesn't
233 * get completely filled.
235 * In principle there could be a problem in lowmem systems if
236 * the initial memory is also very large with respect to
237 * lowmem, but we won't try to deal with that here.
239 extra_limit = min(EXTRA_MEM_RATIO * min(max_pfn, PFN_DOWN(MAXMEM)),
240 max_pfn + extra_pages);
242 if (extra_limit >= max_pfn)
243 extra_pages = extra_limit - max_pfn;
247 if (!xen_initial_domain())
248 xen_add_extra_mem(extra_pages);
253 static void xen_idle(void)
260 current_thread_info()->status &= ~TS_POLLING;
261 smp_mb__after_clear_bit();
263 current_thread_info()->status |= TS_POLLING;
268 * Set the bit indicating "nosegneg" library variants should be used.
269 * We only need to bother in pure 32-bit mode; compat 32-bit processes
270 * can have un-truncated segments, so wrapping around is allowed.
272 static void __init fiddle_vdso(void)
276 mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
277 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
278 mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
279 *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
283 static __cpuinit int register_callback(unsigned type, const void *func)
285 struct callback_register callback = {
287 .address = XEN_CALLBACK(__KERNEL_CS, func),
288 .flags = CALLBACKF_mask_events,
291 return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
294 void __cpuinit xen_enable_sysenter(void)
297 unsigned sysenter_feature;
300 sysenter_feature = X86_FEATURE_SEP;
302 sysenter_feature = X86_FEATURE_SYSENTER32;
305 if (!boot_cpu_has(sysenter_feature))
308 ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
310 setup_clear_cpu_cap(sysenter_feature);
313 void __cpuinit xen_enable_syscall(void)
318 ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
320 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
321 /* Pretty fatal; 64-bit userspace has no other
322 mechanism for syscalls. */
325 if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
326 ret = register_callback(CALLBACKTYPE_syscall32,
327 xen_syscall32_target);
329 setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
331 #endif /* CONFIG_X86_64 */
334 void __init xen_arch_setup(void)
336 struct physdev_set_iopl set_iopl;
339 xen_panic_handler_init();
341 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
342 HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
344 if (!xen_feature(XENFEAT_auto_translated_physmap))
345 HYPERVISOR_vm_assist(VMASST_CMD_enable,
346 VMASST_TYPE_pae_extended_cr3);
348 if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
349 register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
352 xen_enable_sysenter();
353 xen_enable_syscall();
356 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
358 printk(KERN_INFO "physdev_op failed %d\n", rc);
361 if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
362 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
367 memcpy(boot_command_line, xen_start_info->cmd_line,
368 MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
369 COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);