xen: make sure pages are really part of domain before freeing
[linux-3.10.git] / arch / x86 / xen / setup.c
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
12 #include <asm/elf.h>
13 #include <asm/vdso.h>
14 #include <asm/e820.h>
15 #include <asm/setup.h>
16 #include <asm/acpi.h>
17 #include <asm/xen/hypervisor.h>
18 #include <asm/xen/hypercall.h>
19
20 #include <xen/page.h>
21 #include <xen/interface/callback.h>
22 #include <xen/interface/physdev.h>
23 #include <xen/interface/memory.h>
24 #include <xen/features.h>
25
26 #include "xen-ops.h"
27 #include "vdso.h"
28
29 /* These are code, but not functions.  Defined in entry.S */
30 extern const char xen_hypervisor_callback[];
31 extern const char xen_failsafe_callback[];
32 extern void xen_sysenter_target(void);
33 extern void xen_syscall_target(void);
34 extern void xen_syscall32_target(void);
35
36 static unsigned long __init xen_release_chunk(phys_addr_t start_addr,
37                                               phys_addr_t end_addr)
38 {
39         struct xen_memory_reservation reservation = {
40                 .address_bits = 0,
41                 .extent_order = 0,
42                 .domid        = DOMID_SELF
43         };
44         unsigned long start, end;
45         unsigned long len = 0;
46         unsigned long pfn;
47         int ret;
48
49         start = PFN_UP(start_addr);
50         end = PFN_DOWN(end_addr);
51
52         if (end <= start)
53                 return 0;
54
55         printk(KERN_INFO "xen_release_chunk: looking at area pfn %lx-%lx: ",
56                start, end);
57         for(pfn = start; pfn < end; pfn++) {
58                 unsigned long mfn = pfn_to_mfn(pfn);
59
60                 /* Make sure pfn exists to start with */
61                 if (mfn == INVALID_P2M_ENTRY || mfn_to_pfn(mfn) != pfn)
62                         continue;
63
64                 set_xen_guest_handle(reservation.extent_start, &mfn);
65                 reservation.nr_extents = 1;
66
67                 ret = HYPERVISOR_memory_op(XENMEM_decrease_reservation,
68                                            &reservation);
69                 WARN(ret != 1, "Failed to release memory %lx-%lx err=%d\n",
70                      start, end, ret);
71                 if (ret == 1) {
72                         set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
73                         len++;
74                 }
75         }
76         printk(KERN_CONT "%ld pages freed\n", len);
77
78         return len;
79 }
80
81 static unsigned long __init xen_return_unused_memory(unsigned long max_pfn,
82                                                      const struct e820map *e820)
83 {
84         phys_addr_t max_addr = PFN_PHYS(max_pfn);
85         phys_addr_t last_end = 0;
86         unsigned long released = 0;
87         int i;
88
89         for (i = 0; i < e820->nr_map && last_end < max_addr; i++) {
90                 phys_addr_t end = e820->map[i].addr;
91                 end = min(max_addr, end);
92
93                 released += xen_release_chunk(last_end, end);
94                 last_end = e820->map[i].addr + e820->map[i].size;
95         }
96
97         if (last_end < max_addr)
98                 released += xen_release_chunk(last_end, max_addr);
99
100         printk(KERN_INFO "released %ld pages of unused memory\n", released);
101         return released;
102 }
103
104 /**
105  * machine_specific_memory_setup - Hook for machine specific memory setup.
106  **/
107
108 char * __init xen_memory_setup(void)
109 {
110         unsigned long max_pfn = xen_start_info->nr_pages;
111
112         max_pfn = min(MAX_DOMAIN_PAGES, max_pfn);
113
114         e820.nr_map = 0;
115
116         e820_add_region(0, PFN_PHYS((u64)max_pfn), E820_RAM);
117
118         /*
119          * Even though this is normal, usable memory under Xen, reserve
120          * ISA memory anyway because too many things think they can poke
121          * about in there.
122          */
123         e820_add_region(ISA_START_ADDRESS, ISA_END_ADDRESS - ISA_START_ADDRESS,
124                         E820_RESERVED);
125
126         /*
127          * Reserve Xen bits:
128          *  - mfn_list
129          *  - xen_start_info
130          * See comment above "struct start_info" in <xen/interface/xen.h>
131          */
132         reserve_early(__pa(xen_start_info->mfn_list),
133                       __pa(xen_start_info->pt_base),
134                         "XEN START INFO");
135
136         sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
137
138         xen_return_unused_memory(xen_start_info->nr_pages, &e820);
139
140         return "Xen";
141 }
142
143 static void xen_idle(void)
144 {
145         local_irq_disable();
146
147         if (need_resched())
148                 local_irq_enable();
149         else {
150                 current_thread_info()->status &= ~TS_POLLING;
151                 smp_mb__after_clear_bit();
152                 safe_halt();
153                 current_thread_info()->status |= TS_POLLING;
154         }
155 }
156
157 /*
158  * Set the bit indicating "nosegneg" library variants should be used.
159  * We only need to bother in pure 32-bit mode; compat 32-bit processes
160  * can have un-truncated segments, so wrapping around is allowed.
161  */
162 static void __init fiddle_vdso(void)
163 {
164 #ifdef CONFIG_X86_32
165         u32 *mask;
166         mask = VDSO32_SYMBOL(&vdso32_int80_start, NOTE_MASK);
167         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
168         mask = VDSO32_SYMBOL(&vdso32_sysenter_start, NOTE_MASK);
169         *mask |= 1 << VDSO_NOTE_NONEGSEG_BIT;
170 #endif
171 }
172
173 static __cpuinit int register_callback(unsigned type, const void *func)
174 {
175         struct callback_register callback = {
176                 .type = type,
177                 .address = XEN_CALLBACK(__KERNEL_CS, func),
178                 .flags = CALLBACKF_mask_events,
179         };
180
181         return HYPERVISOR_callback_op(CALLBACKOP_register, &callback);
182 }
183
184 void __cpuinit xen_enable_sysenter(void)
185 {
186         int ret;
187         unsigned sysenter_feature;
188
189 #ifdef CONFIG_X86_32
190         sysenter_feature = X86_FEATURE_SEP;
191 #else
192         sysenter_feature = X86_FEATURE_SYSENTER32;
193 #endif
194
195         if (!boot_cpu_has(sysenter_feature))
196                 return;
197
198         ret = register_callback(CALLBACKTYPE_sysenter, xen_sysenter_target);
199         if(ret != 0)
200                 setup_clear_cpu_cap(sysenter_feature);
201 }
202
203 void __cpuinit xen_enable_syscall(void)
204 {
205 #ifdef CONFIG_X86_64
206         int ret;
207
208         ret = register_callback(CALLBACKTYPE_syscall, xen_syscall_target);
209         if (ret != 0) {
210                 printk(KERN_ERR "Failed to set syscall callback: %d\n", ret);
211                 /* Pretty fatal; 64-bit userspace has no other
212                    mechanism for syscalls. */
213         }
214
215         if (boot_cpu_has(X86_FEATURE_SYSCALL32)) {
216                 ret = register_callback(CALLBACKTYPE_syscall32,
217                                         xen_syscall32_target);
218                 if (ret != 0)
219                         setup_clear_cpu_cap(X86_FEATURE_SYSCALL32);
220         }
221 #endif /* CONFIG_X86_64 */
222 }
223
224 void __init xen_arch_setup(void)
225 {
226         struct physdev_set_iopl set_iopl;
227         int rc;
228
229         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_4gb_segments);
230         HYPERVISOR_vm_assist(VMASST_CMD_enable, VMASST_TYPE_writable_pagetables);
231
232         if (!xen_feature(XENFEAT_auto_translated_physmap))
233                 HYPERVISOR_vm_assist(VMASST_CMD_enable,
234                                      VMASST_TYPE_pae_extended_cr3);
235
236         if (register_callback(CALLBACKTYPE_event, xen_hypervisor_callback) ||
237             register_callback(CALLBACKTYPE_failsafe, xen_failsafe_callback))
238                 BUG();
239
240         xen_enable_sysenter();
241         xen_enable_syscall();
242
243         set_iopl.iopl = 1;
244         rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
245         if (rc != 0)
246                 printk(KERN_INFO "physdev_op failed %d\n", rc);
247
248 #ifdef CONFIG_ACPI
249         if (!(xen_start_info->flags & SIF_INITDOMAIN)) {
250                 printk(KERN_INFO "ACPI in unprivileged domain disabled\n");
251                 disable_acpi();
252         }
253 #endif
254
255         memcpy(boot_command_line, xen_start_info->cmd_line,
256                MAX_GUEST_CMDLINE > COMMAND_LINE_SIZE ?
257                COMMAND_LINE_SIZE : MAX_GUEST_CMDLINE);
258
259         pm_idle = xen_idle;
260
261         paravirt_disable_iospace();
262
263         fiddle_vdso();
264 }