blob: 880862c7d9ddba51e1b6964bc80dcf49d6a8b6ff [file] [log] [blame]
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001/*
2 * Core of Xen paravirt_ops implementation.
3 *
4 * This file contains the xen_paravirt_ops structure itself, and the
5 * implementations for:
6 * - privileged instructions
7 * - interrupt flags
8 * - segment operations
9 * - booting and setup
10 *
11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
12 */
13
Sheng Yang38e20b02010-05-14 12:40:51 +010014#include <linux/cpu.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070015#include <linux/kernel.h>
16#include <linux/init.h>
17#include <linux/smp.h>
18#include <linux/preempt.h>
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -070019#include <linux/hardirq.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070020#include <linux/percpu.h>
21#include <linux/delay.h>
22#include <linux/start_kernel.h>
23#include <linux/sched.h>
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -070024#include <linux/kprobes.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070025#include <linux/bootmem.h>
26#include <linux/module.h>
Jeremy Fitzhardingef4f97b32007-07-17 18:37:05 -070027#include <linux/mm.h>
28#include <linux/page-flags.h>
29#include <linux/highmem.h>
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +010030#include <linux/console.h>
Chris Wright5d990b62009-12-04 12:15:21 -080031#include <linux/pci.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090032#include <linux/gfp.h>
Jeremy Fitzhardinge236260b2010-10-06 15:52:29 -070033#include <linux/memblock.h>
David Vrabel96f28bc2013-04-03 17:31:50 +010034#include <linux/edd.h>
Josh Poimboeuf983bb6d2016-02-28 22:22:36 -060035#include <linux/frame.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070036
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +020037#ifdef CONFIG_KEXEC_CORE
38#include <linux/kexec.h>
39#endif
40
Jeremy Fitzhardinge1ccbf532009-10-06 15:11:14 -070041#include <xen/xen.h>
Stefano Stabellini0ec53ec2012-09-14 13:37:32 +000042#include <xen/events.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070043#include <xen/interface/xen.h>
Jeremy Fitzhardingeecbf29c2008-12-16 12:37:07 -080044#include <xen/interface/version.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070045#include <xen/interface/physdev.h>
46#include <xen/interface/vcpu.h>
Sheng Yangbee6ab52010-05-14 12:39:33 +010047#include <xen/interface/memory.h>
Jan Beulichf221b042015-01-13 07:40:05 +000048#include <xen/interface/nmi.h>
Liu, Jinsongcef12ee2012-06-07 19:56:51 +080049#include <xen/interface/xen-mca.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070050#include <xen/features.h>
51#include <xen/page.h>
Sheng Yang38e20b02010-05-14 12:40:51 +010052#include <xen/hvm.h>
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -070053#include <xen/hvc-console.h>
Konrad Rzeszutek Wilk211063d2011-12-08 17:32:23 +080054#include <xen/acpi.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070055
56#include <asm/paravirt.h>
Ingo Molnar7b6aa332009-02-17 13:58:15 +010057#include <asm/apic.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070058#include <asm/page.h>
Alex Nixonb5401a92010-03-18 16:31:34 -040059#include <asm/xen/pci.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070060#include <asm/xen/hypercall.h>
61#include <asm/xen/hypervisor.h>
62#include <asm/fixmap.h>
63#include <asm/processor.h>
Jeremy Fitzhardinge707ebbc2009-03-27 11:29:02 -070064#include <asm/proto.h>
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -070065#include <asm/msr-index.h>
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -070066#include <asm/traps.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070067#include <asm/setup.h>
68#include <asm/desc.h>
Ian Campbell817a8242010-02-26 17:16:00 +000069#include <asm/pgalloc.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070070#include <asm/pgtable.h>
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -070071#include <asm/tlbflush.h>
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -070072#include <asm/reboot.h>
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -070073#include <asm/stackprotector.h>
Sheng Yangbee6ab52010-05-14 12:39:33 +010074#include <asm/hypervisor.h>
Jan Beulichf221b042015-01-13 07:40:05 +000075#include <asm/mach_traps.h>
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -050076#include <asm/mwait.h>
David Vrabel76a8df72012-05-04 14:29:46 +010077#include <asm/pci_x86.h>
Konrad Rzeszutek Wilkc79c4982013-02-26 12:51:27 -050078#include <asm/pat.h>
Stefano Stabellinia314e3e2015-10-22 16:20:46 +000079#include <asm/cpu.h>
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -050080
81#ifdef CONFIG_ACPI
82#include <linux/acpi.h>
83#include <asm/acpi.h>
84#include <acpi/pdc_intel.h>
85#include <acpi/processor.h>
86#include <xen/interface/platform.h>
87#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070088
89#include "xen-ops.h"
Jeremy Fitzhardinge3b827c12007-07-17 18:37:04 -070090#include "mmu.h"
Ben Guthrof447d562012-04-21 00:11:04 +080091#include "smp.h"
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070092#include "multicalls.h"
Boris Ostrovsky65d0cf02015-08-10 16:34:34 -040093#include "pmu.h"
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070094
95EXPORT_SYMBOL_GPL(hypercall_page);
96
Konrad Rzeszutek Wilka5209962013-05-05 08:51:47 -040097/*
98 * Pointer to the xen_vcpu_info structure or
99 * &HYPERVISOR_shared_info->vcpu_info[cpu]. See xen_hvm_init_shared_info
100 * and xen_vcpu_setup for details. By default it points to share_info->vcpu_info
101 * but if the hypervisor supports VCPUOP_register_vcpu_info then it can point
102 * to xen_vcpu_info. The pointer is used in __xen_evtchn_do_upcall to
103 * acknowledge pending events.
104 * Also more subtly it is used by the patched version of irq enable/disable
105 * e.g. xen_irq_enable_direct and xen_iret in PV mode.
106 *
107 * The desire to be able to do those mask/unmask operations as a single
108 * instruction by using the per-cpu offset held in %gs is the real reason
109 * vcpu info is in a per-cpu pointer and the original reason for this
110 * hypercall.
111 *
112 */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700113DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
Konrad Rzeszutek Wilka5209962013-05-05 08:51:47 -0400114
115/*
116 * Per CPU pages used if hypervisor supports VCPUOP_register_vcpu_info
117 * hypercall. This can be used both in PV and PVHVM mode. The structure
118 * overrides the default per_cpu(xen_vcpu, cpu) value.
119 */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700120DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
Jeremy Fitzhardinge9f799912007-10-16 11:51:30 -0700121
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -0700122enum xen_domain_type xen_domain_type = XEN_NATIVE;
123EXPORT_SYMBOL_GPL(xen_domain_type);
124
Ian Campbell7e775062010-09-30 12:37:26 +0100125unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
126EXPORT_SYMBOL(machine_to_phys_mapping);
Jan Beulichccbcdf72011-08-16 15:07:41 +0100127unsigned long machine_to_phys_nr;
128EXPORT_SYMBOL(machine_to_phys_nr);
Ian Campbell7e775062010-09-30 12:37:26 +0100129
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700130struct start_info *xen_start_info;
131EXPORT_SYMBOL_GPL(xen_start_info);
132
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +0100133struct shared_info xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700134
Jeremy Fitzhardinge38341432009-02-02 13:55:31 -0800135void *xen_initial_gdt;
136
Sheng Yangbee6ab52010-05-14 12:39:33 +0100137RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
Sheng Yang38e20b02010-05-14 12:40:51 +0100138__read_mostly int xen_have_vector_callback;
139EXPORT_SYMBOL_GPL(xen_have_vector_callback);
Sheng Yangbee6ab52010-05-14 12:39:33 +0100140
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700141/*
142 * Point at some empty memory to start with. We map the real shared_info
143 * page as soon as fixmap is up and running.
144 */
Olaf Hering4648da72012-07-17 11:57:41 +0200145struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700146
147/*
148 * Flag to determine whether vcpu info placement is available on all
149 * VCPUs. We assume it is to start with, and then set it to zero on
150 * the first failure. This is because it can succeed on some VCPUs
151 * and not others, since it can involve hypervisor memory allocation,
152 * or because the guest failed to guarantee all the appropriate
153 * constraints on all VCPUs (ie buffer can't cross a page boundary).
154 *
155 * Note that any particular CPU may be using a placed vcpu structure,
156 * but we can only optimise if the all are.
157 *
158 * 0: not available, 1: available
159 */
Jeremy Fitzhardingee4d04072009-02-02 13:55:54 -0800160static int have_vcpu_info_placement = 1;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700161
David Vrabel1c32cdc2012-07-09 11:39:08 +0100162struct tls_descs {
163 struct desc_struct desc[3];
164};
165
166/*
167 * Updating the 3 TLS descriptors in the GDT on every task switch is
168 * surprisingly expensive so we avoid updating them if they haven't
169 * changed. Since Xen writes different descriptors than the one
170 * passed in the update_descriptor hypercall we keep shadow copies to
171 * compare against.
172 */
173static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
174
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700175static void clamp_max_cpus(void)
176{
177#ifdef CONFIG_SMP
178 if (setup_max_cpus > MAX_VIRT_CPUS)
179 setup_max_cpus = MAX_VIRT_CPUS;
180#endif
181}
182
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100183static void xen_vcpu_setup(int cpu)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700184{
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700185 struct vcpu_register_vcpu_info info;
186 int err;
187 struct vcpu_info *vcpup;
188
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +0100189 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700190
Konrad Rzeszutek Wilk7f1fc262013-05-05 09:30:09 -0400191 /*
192 * This path is called twice on PVHVM - first during bootup via
193 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
194 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
195 * As we can only do the VCPUOP_register_vcpu_info once lets
196 * not over-write its result.
197 *
198 * For PV it is called during restore (xen_vcpu_restore) and bootup
199 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
200 * use this function.
201 */
202 if (xen_hvm_domain()) {
203 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
204 return;
205 }
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700206 if (cpu < MAX_VIRT_CPUS)
207 per_cpu(xen_vcpu,cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
208
209 if (!have_vcpu_info_placement) {
210 if (cpu >= MAX_VIRT_CPUS)
211 clamp_max_cpus();
212 return;
213 }
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700214
215 vcpup = &per_cpu(xen_vcpu_info, cpu);
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800216 info.mfn = arbitrary_virt_to_mfn(vcpup);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700217 info.offset = offset_in_page(vcpup);
218
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700219 /* Check to see if the hypervisor will put the vcpu_info
220 structure where we want it, which allows direct access via
Konrad Rzeszutek Wilka5209962013-05-05 08:51:47 -0400221 a percpu-variable.
222 N.B. This hypercall can _only_ be called once per CPU. Subsequent
223 calls will error out with -EINVAL. This is due to the fact that
224 hypervisor has no unregister variant and this hypercall does not
225 allow to over-write info.mfn and info.offset.
226 */
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700227 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, cpu, &info);
228
229 if (err) {
230 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
231 have_vcpu_info_placement = 0;
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700232 clamp_max_cpus();
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700233 } else {
234 /* This cpu is using the registered vcpu info, even if
235 later ones fail to. */
236 per_cpu(xen_vcpu, cpu) = vcpup;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700237 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700238}
239
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100240/*
241 * On restore, set the vcpu placement up again.
242 * If it fails, then we're in a bad state, since
243 * we can't back out from using it...
244 */
245void xen_vcpu_restore(void)
246{
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800247 int cpu;
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100248
Wei Liu9d328a92012-12-13 15:33:05 +0000249 for_each_possible_cpu(cpu) {
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800250 bool other_cpu = (cpu != smp_processor_id());
Wei Liu9d328a92012-12-13 15:33:05 +0000251 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL);
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100252
Wei Liu9d328a92012-12-13 15:33:05 +0000253 if (other_cpu && is_up &&
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800254 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL))
255 BUG();
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100256
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800257 xen_setup_runstate_info(cpu);
Ian Campbellbe012922009-11-21 08:35:55 +0800258
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800259 if (have_vcpu_info_placement)
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100260 xen_vcpu_setup(cpu);
261
Wei Liu9d328a92012-12-13 15:33:05 +0000262 if (other_cpu && is_up &&
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800263 HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL))
264 BUG();
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100265 }
266}
267
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700268static void __init xen_banner(void)
269{
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700270 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
271 struct xen_extraversion extra;
272 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
273
Mukesh Rathord285d682013-12-13 12:45:31 -0500274 pr_info("Booting paravirtualized kernel %son %s\n",
275 xen_feature(XENFEAT_auto_translated_physmap) ?
276 "with PVH extensions " : "", pv_info.name);
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700277 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
278 version >> 16, version & 0xffff, extra.extraversion,
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -0700279 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700280}
Konrad Rzeszutek Wilk394b40f2012-11-27 11:39:40 -0500281/* Check if running on Xen version (major, minor) or later */
282bool
283xen_running_on_version_or_later(unsigned int major, unsigned int minor)
284{
285 unsigned int version;
286
287 if (!xen_domain())
288 return false;
289
290 version = HYPERVISOR_xen_version(XENVER_version, NULL);
291 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
292 ((version >> 16) > major))
293 return true;
294 return false;
295}
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700296
Andre Przywara5e626252012-05-29 13:07:31 +0200297#define CPUID_THERM_POWER_LEAF 6
298#define APERFMPERF_PRESENT 0
299
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800300static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
301static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
302
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500303static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
304static __read_mostly unsigned int cpuid_leaf5_ecx_val;
305static __read_mostly unsigned int cpuid_leaf5_edx_val;
306
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100307static void xen_cpuid(unsigned int *ax, unsigned int *bx,
308 unsigned int *cx, unsigned int *dx)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700309{
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700310 unsigned maskebx = ~0;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800311 unsigned maskecx = ~0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700312 unsigned maskedx = ~0;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500313 unsigned setecx = 0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700314 /*
315 * Mask out inconvenient features, to try and disable as many
316 * unsupported kernel subsystems as possible.
317 */
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700318 switch (*ax) {
319 case 1:
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800320 maskecx = cpuid_leaf1_ecx_mask;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500321 setecx = cpuid_leaf1_ecx_set_mask;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800322 maskedx = cpuid_leaf1_edx_mask;
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700323 break;
324
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500325 case CPUID_MWAIT_LEAF:
326 /* Synthesize the values.. */
327 *ax = 0;
328 *bx = 0;
329 *cx = cpuid_leaf5_ecx_val;
330 *dx = cpuid_leaf5_edx_val;
331 return;
332
Andre Przywara5e626252012-05-29 13:07:31 +0200333 case CPUID_THERM_POWER_LEAF:
334 /* Disabling APERFMPERF for kernel usage */
335 maskecx = ~(1 << APERFMPERF_PRESENT);
336 break;
337
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700338 case 0xb:
339 /* Suppress extended topology stuff */
340 maskebx = 0;
341 break;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800342 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700343
344 asm(XEN_EMULATE_PREFIX "cpuid"
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100345 : "=a" (*ax),
346 "=b" (*bx),
347 "=c" (*cx),
348 "=d" (*dx)
349 : "0" (*ax), "2" (*cx));
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800350
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700351 *bx &= maskebx;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800352 *cx &= maskecx;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500353 *cx |= setecx;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100354 *dx &= maskedx;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700355}
Josh Poimboeuf983bb6d2016-02-28 22:22:36 -0600356STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700357
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500358static bool __init xen_check_mwait(void)
359{
Liu, Jinsonge3aa4e62012-11-08 05:44:28 +0000360#ifdef CONFIG_ACPI
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500361 struct xen_platform_op op = {
362 .cmd = XENPF_set_processor_pminfo,
363 .u.set_pminfo.id = -1,
364 .u.set_pminfo.type = XEN_PM_PDC,
365 };
366 uint32_t buf[3];
367 unsigned int ax, bx, cx, dx;
368 unsigned int mwait_mask;
369
370 /* We need to determine whether it is OK to expose the MWAIT
371 * capability to the kernel to harvest deeper than C3 states from ACPI
372 * _CST using the processor_harvest_xen.c module. For this to work, we
373 * need to gather the MWAIT_LEAF values (which the cstate.c code
374 * checks against). The hypervisor won't expose the MWAIT flag because
375 * it would break backwards compatibility; so we will find out directly
376 * from the hardware and hypercall.
377 */
378 if (!xen_initial_domain())
379 return false;
380
Liu, Jinsonge3aa4e62012-11-08 05:44:28 +0000381 /*
382 * When running under platform earlier than Xen4.2, do not expose
383 * mwait, to avoid the risk of loading native acpi pad driver
384 */
385 if (!xen_running_on_version_or_later(4, 2))
386 return false;
387
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500388 ax = 1;
389 cx = 0;
390
391 native_cpuid(&ax, &bx, &cx, &dx);
392
393 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
394 (1 << (X86_FEATURE_MWAIT % 32));
395
396 if ((cx & mwait_mask) != mwait_mask)
397 return false;
398
399 /* We need to emulate the MWAIT_LEAF and for that we need both
400 * ecx and edx. The hypercall provides only partial information.
401 */
402
403 ax = CPUID_MWAIT_LEAF;
404 bx = 0;
405 cx = 0;
406 dx = 0;
407
408 native_cpuid(&ax, &bx, &cx, &dx);
409
410 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
411 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
412 */
413 buf[0] = ACPI_PDC_REVISION_ID;
414 buf[1] = 1;
415 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
416
417 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
418
Stefano Stabellinicfafae92015-11-23 10:36:12 +0000419 if ((HYPERVISOR_platform_op(&op) == 0) &&
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500420 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
421 cpuid_leaf5_ecx_val = cx;
422 cpuid_leaf5_edx_val = dx;
423 }
424 return true;
425#else
426 return false;
427#endif
428}
Daniel Kiperad3062a2011-05-04 20:15:18 +0200429static void __init xen_init_cpuid_mask(void)
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800430{
431 unsigned int ax, bx, cx, dx;
Shan Haitao947ccf92010-11-09 11:43:36 -0800432 unsigned int xsave_mask;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800433
434 cpuid_leaf1_edx_mask =
Liu, Jinsongcef12ee2012-06-07 19:56:51 +0800435 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800436 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
437
438 if (!xen_initial_domain())
439 cpuid_leaf1_edx_mask &=
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400440 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
Zhenzhong Duan4ea9b9a2013-03-29 09:14:00 +0800441
442 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
443
Shan Haitao947ccf92010-11-09 11:43:36 -0800444 ax = 1;
Konrad Rzeszutek Wilk5e287832011-09-29 13:06:42 -0400445 cx = 0;
Mukesh Rathord285d682013-12-13 12:45:31 -0500446 cpuid(1, &ax, &bx, &cx, &dx);
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800447
Shan Haitao947ccf92010-11-09 11:43:36 -0800448 xsave_mask =
449 (1 << (X86_FEATURE_XSAVE % 32)) |
450 (1 << (X86_FEATURE_OSXSAVE % 32));
451
452 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
453 if ((cx & xsave_mask) != xsave_mask)
454 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500455 if (xen_check_mwait())
456 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800457}
458
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700459static void xen_set_debugreg(int reg, unsigned long val)
460{
461 HYPERVISOR_set_debugreg(reg, val);
462}
463
464static unsigned long xen_get_debugreg(int reg)
465{
466 return HYPERVISOR_get_debugreg(reg);
467}
468
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800469static void xen_end_context_switch(struct task_struct *next)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700470{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700471 xen_mc_flush();
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800472 paravirt_end_context_switch(next);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700473}
474
475static unsigned long xen_store_tr(void)
476{
477 return 0;
478}
479
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700480/*
Jeremy Fitzhardingecef43bf2008-07-28 13:33:44 -0700481 * Set the page permissions for a particular virtual address. If the
482 * address is a vmalloc mapping (or other non-linear mapping), then
483 * find the linear mapping of the page and also set its protections to
484 * match.
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700485 */
486static void set_aliased_prot(void *v, pgprot_t prot)
487{
488 int level;
489 pte_t *ptep;
490 pte_t pte;
491 unsigned long pfn;
492 struct page *page;
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700493 unsigned char dummy;
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700494
495 ptep = lookup_address((unsigned long)v, &level);
496 BUG_ON(ptep == NULL);
497
498 pfn = pte_pfn(*ptep);
499 page = pfn_to_page(pfn);
500
501 pte = pfn_pte(pfn, prot);
502
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700503 /*
504 * Careful: update_va_mapping() will fail if the virtual address
505 * we're poking isn't populated in the page tables. We don't
506 * need to worry about the direct map (that's always in the page
507 * tables), but we need to be careful about vmap space. In
508 * particular, the top level page table can lazily propagate
509 * entries between processes, so if we've switched mms since we
510 * vmapped the target in the first place, we might not have the
511 * top-level page table entry populated.
512 *
513 * We disable preemption because we want the same mm active when
514 * we probe the target and when we issue the hypercall. We'll
515 * have the same nominal mm, but if we're a kernel thread, lazy
516 * mm dropping could change our pgd.
517 *
518 * Out of an abundance of caution, this uses __get_user() to fault
519 * in the target address just in case there's some obscure case
520 * in which the target address isn't readable.
521 */
522
523 preempt_disable();
524
525 pagefault_disable(); /* Avoid warnings due to being atomic. */
526 __get_user(dummy, (unsigned char __user __force *)v);
527 pagefault_enable();
528
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700529 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
530 BUG();
531
532 if (!PageHighMem(page)) {
533 void *av = __va(PFN_PHYS(pfn));
534
535 if (av != v)
536 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
537 BUG();
538 } else
539 kmap_flush_unused();
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700540
541 preempt_enable();
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700542}
543
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700544static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
545{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700546 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700547 int i;
548
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700549 /*
550 * We need to mark the all aliases of the LDT pages RO. We
551 * don't need to call vm_flush_aliases(), though, since that's
552 * only responsible for flushing aliases out the TLBs, not the
553 * page tables, and Xen will flush the TLB for us if needed.
554 *
555 * To avoid confusing future readers: none of this is necessary
556 * to load the LDT. The hypervisor only checks this when the
557 * LDT is faulted in due to subsequent descriptor access.
558 */
559
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700560 for(i = 0; i < entries; i += entries_per_page)
561 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700562}
563
564static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
565{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700566 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700567 int i;
568
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700569 for(i = 0; i < entries; i += entries_per_page)
570 set_aliased_prot(ldt + i, PAGE_KERNEL);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700571}
572
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700573static void xen_set_ldt(const void *addr, unsigned entries)
574{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700575 struct mmuext_op *op;
576 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
577
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800578 trace_xen_cpu_set_ldt(addr, entries);
579
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700580 op = mcs.args;
581 op->cmd = MMUEXT_SET_LDT;
Jan Beulich4dbf7af2008-01-30 13:33:14 +0100582 op->arg1.linear_addr = (unsigned long)addr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700583 op->arg2.nr_ents = entries;
584
585 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
586
587 xen_mc_issue(PARAVIRT_LAZY_CPU);
588}
589
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100590static void xen_load_gdt(const struct desc_ptr *dtr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700591{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700592 unsigned long va = dtr->address;
593 unsigned int size = dtr->size + 1;
594 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
Jeremy Fitzhardinge3ce5fa72009-03-04 15:26:00 -0800595 unsigned long frames[pages];
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700596 int f;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700597
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700598 /*
599 * A GDT can be up to 64k in size, which corresponds to 8192
600 * 8-byte entries, or 16 4k pages..
601 */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700602
603 BUG_ON(size > 65536);
604 BUG_ON(va & ~PAGE_MASK);
605
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700606 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800607 int level;
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700608 pte_t *ptep;
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800609 unsigned long pfn, mfn;
610 void *virt;
611
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700612 /*
613 * The GDT is per-cpu and is in the percpu data area.
614 * That can be virtually mapped, so we need to do a
615 * page-walk to get the underlying MFN for the
616 * hypercall. The page can also be in the kernel's
617 * linear range, so we need to RO that mapping too.
618 */
619 ptep = lookup_address(va, &level);
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800620 BUG_ON(ptep == NULL);
621
622 pfn = pte_pfn(*ptep);
623 mfn = pfn_to_mfn(pfn);
624 virt = __va(PFN_PHYS(pfn));
625
626 frames[f] = mfn;
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800627
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700628 make_lowmem_page_readonly((void *)va);
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800629 make_lowmem_page_readonly(virt);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700630 }
631
Jeremy Fitzhardinge3ce5fa72009-03-04 15:26:00 -0800632 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
633 BUG();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700634}
635
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700636/*
637 * load_gdt for early boot, when the gdt is only mapped once
638 */
Daniel Kiperad3062a2011-05-04 20:15:18 +0200639static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700640{
641 unsigned long va = dtr->address;
642 unsigned int size = dtr->size + 1;
643 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
644 unsigned long frames[pages];
645 int f;
646
647 /*
648 * A GDT can be up to 64k in size, which corresponds to 8192
649 * 8-byte entries, or 16 4k pages..
650 */
651
652 BUG_ON(size > 65536);
653 BUG_ON(va & ~PAGE_MASK);
654
655 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
656 pte_t pte;
657 unsigned long pfn, mfn;
658
659 pfn = virt_to_pfn(va);
660 mfn = pfn_to_mfn(pfn);
661
662 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
663
664 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
665 BUG();
666
667 frames[f] = mfn;
668 }
669
670 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
671 BUG();
672}
673
David Vrabel59290362012-07-09 11:39:07 +0100674static inline bool desc_equal(const struct desc_struct *d1,
675 const struct desc_struct *d2)
676{
677 return d1->a == d2->a && d1->b == d2->b;
678}
679
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700680static void load_TLS_descriptor(struct thread_struct *t,
681 unsigned int cpu, unsigned int i)
682{
David Vrabel1c32cdc2012-07-09 11:39:08 +0100683 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
684 struct desc_struct *gdt;
685 xmaddr_t maddr;
686 struct multicall_space mc;
687
688 if (desc_equal(shadow, &t->tls_array[i]))
689 return;
690
691 *shadow = t->tls_array[i];
692
693 gdt = get_cpu_gdt_table(cpu);
694 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
695 mc = __xen_mc_entry(0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700696
697 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
698}
699
700static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
701{
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700702 /*
Tejun Heoccbeed32009-02-09 22:17:40 +0900703 * XXX sleazy hack: If we're being called in a lazy-cpu zone
704 * and lazy gs handling is enabled, it means we're in a
705 * context switch, and %gs has just been saved. This means we
706 * can zero it out to prevent faults on exit from the
707 * hypervisor if the next process has no %gs. Either way, it
708 * has been saved, and the new value will get loaded properly.
709 * This will go away as soon as Xen has been modified to not
710 * save/restore %gs for normal hypercalls.
Eduardo Habkost8a954082008-07-08 15:07:10 -0700711 *
712 * On x86_64, this hack is not used for %gs, because gs points
713 * to KERNEL_GS_BASE (and uses it for PDA references), so we
714 * must not zero %gs on x86_64
715 *
716 * For x86_64, we need to zero %fs, otherwise we may get an
717 * exception between the new %fs descriptor being loaded and
718 * %fs being effectively cleared at __switch_to().
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700719 */
Eduardo Habkost8a954082008-07-08 15:07:10 -0700720 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
721#ifdef CONFIG_X86_32
Tejun Heoccbeed32009-02-09 22:17:40 +0900722 lazy_load_gs(0);
Eduardo Habkost8a954082008-07-08 15:07:10 -0700723#else
724 loadsegment(fs, 0);
725#endif
726 }
727
728 xen_mc_batch();
729
730 load_TLS_descriptor(t, cpu, 0);
731 load_TLS_descriptor(t, cpu, 1);
732 load_TLS_descriptor(t, cpu, 2);
733
734 xen_mc_issue(PARAVIRT_LAZY_CPU);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700735}
736
Eduardo Habkosta8fc1082008-07-08 15:07:05 -0700737#ifdef CONFIG_X86_64
738static void xen_load_gs_index(unsigned int idx)
739{
740 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
741 BUG();
742}
743#endif
744
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700745static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100746 const void *ptr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700747{
Jeremy Fitzhardingecef43bf2008-07-28 13:33:44 -0700748 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100749 u64 entry = *(u64 *)ptr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700750
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800751 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
752
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700753 preempt_disable();
754
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700755 xen_mc_flush();
756 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
757 BUG();
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700758
759 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700760}
761
Eduardo Habkoste176d362008-07-08 15:06:59 -0700762static int cvt_gate_to_trap(int vector, const gate_desc *val,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700763 struct trap_info *info)
764{
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700765 unsigned long addr;
766
Jeremy Fitzhardinge6d02c422009-03-29 22:57:15 -0700767 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700768 return 0;
769
770 info->vector = vector;
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700771
772 addr = gate_offset(*val);
773#ifdef CONFIG_X86_64
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700774 /*
775 * Look for known traps using IST, and substitute them
776 * appropriately. The debugger ones are the only ones we care
Liu, Jinsong05e36002012-06-07 20:10:37 +0800777 * about. Xen will handle faults like double_fault,
778 * so we should never see them. Warn if
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700779 * there's an unexpected IST-using fault handler.
780 */
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700781 if (addr == (unsigned long)debug)
782 addr = (unsigned long)xen_debug;
783 else if (addr == (unsigned long)int3)
784 addr = (unsigned long)xen_int3;
785 else if (addr == (unsigned long)stack_segment)
786 addr = (unsigned long)xen_stack_segment;
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400787 else if (addr == (unsigned long)double_fault) {
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700788 /* Don't need to handle these */
789 return 0;
790#ifdef CONFIG_X86_MCE
791 } else if (addr == (unsigned long)machine_check) {
Liu, Jinsong05e36002012-06-07 20:10:37 +0800792 /*
793 * when xen hypervisor inject vMCE to guest,
794 * use native mce handler to handle it
795 */
796 ;
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700797#endif
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400798 } else if (addr == (unsigned long)nmi)
799 /*
800 * Use the native version as well.
801 */
802 ;
803 else {
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700804 /* Some other trap using IST? */
805 if (WARN_ON(val->ist != 0))
806 return 0;
807 }
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700808#endif /* CONFIG_X86_64 */
809 info->address = addr;
810
Eduardo Habkoste176d362008-07-08 15:06:59 -0700811 info->cs = gate_segment(*val);
812 info->flags = val->dpl;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700813 /* interrupt gates clear IF */
Jeremy Fitzhardinge6d02c422009-03-29 22:57:15 -0700814 if (val->type == GATE_INTERRUPT)
815 info->flags |= 1 << 2;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700816
817 return 1;
818}
819
820/* Locations of each CPU's IDT */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100821static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700822
823/* Set an IDT entry. If the entry is part of the current IDT, then
824 also update Xen. */
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100825static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700826{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700827 unsigned long p = (unsigned long)&dt[entrynum];
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700828 unsigned long start, end;
829
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800830 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
831
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700832 preempt_disable();
833
Christoph Lameter780f36d2010-12-06 11:16:29 -0600834 start = __this_cpu_read(idt_desc.address);
835 end = start + __this_cpu_read(idt_desc.size) + 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700836
837 xen_mc_flush();
838
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100839 native_write_idt_entry(dt, entrynum, g);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700840
841 if (p >= start && (p + 8) <= end) {
842 struct trap_info info[2];
843
844 info[1].address = 0;
845
Eduardo Habkoste176d362008-07-08 15:06:59 -0700846 if (cvt_gate_to_trap(entrynum, g, &info[0]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700847 if (HYPERVISOR_set_trap_table(info))
848 BUG();
849 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700850
851 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700852}
853
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100854static void xen_convert_trap_info(const struct desc_ptr *desc,
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700855 struct trap_info *traps)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700856{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700857 unsigned in, out, count;
858
Eduardo Habkoste176d362008-07-08 15:06:59 -0700859 count = (desc->size+1) / sizeof(gate_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700860 BUG_ON(count > 256);
861
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700862 for (in = out = 0; in < count; in++) {
Eduardo Habkoste176d362008-07-08 15:06:59 -0700863 gate_desc *entry = (gate_desc*)(desc->address) + in;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700864
Eduardo Habkoste176d362008-07-08 15:06:59 -0700865 if (cvt_gate_to_trap(in, entry, &traps[out]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700866 out++;
867 }
868 traps[out].address = 0;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700869}
870
871void xen_copy_trap_info(struct trap_info *traps)
872{
Christoph Lameter89cbc762014-08-17 12:30:40 -0500873 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700874
875 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700876}
877
878/* Load a new IDT into Xen. In principle this can be per-CPU, so we
879 hold a spinlock to protect the static traps[] array (static because
880 it avoids allocation, and saves stack space). */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100881static void xen_load_idt(const struct desc_ptr *desc)
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700882{
883 static DEFINE_SPINLOCK(lock);
884 static struct trap_info traps[257];
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700885
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800886 trace_xen_cpu_load_idt(desc);
887
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700888 spin_lock(&lock);
889
Christoph Lameter89cbc762014-08-17 12:30:40 -0500890 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700891
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700892 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700893
894 xen_mc_flush();
895 if (HYPERVISOR_set_trap_table(traps))
896 BUG();
897
898 spin_unlock(&lock);
899}
900
901/* Write a GDT descriptor entry. Ignore LDT descriptors, since
902 they're handled differently. */
903static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100904 const void *desc, int type)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700905{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800906 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
907
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700908 preempt_disable();
909
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100910 switch (type) {
911 case DESC_LDT:
912 case DESC_TSS:
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700913 /* ignore */
914 break;
915
916 default: {
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800917 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700918
919 xen_mc_flush();
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100920 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700921 BUG();
922 }
923
924 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700925
926 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700927}
928
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700929/*
930 * Version of write_gdt_entry for use at early boot-time needed to
931 * update an entry as simply as possible.
932 */
Daniel Kiperad3062a2011-05-04 20:15:18 +0200933static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700934 const void *desc, int type)
935{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800936 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
937
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700938 switch (type) {
939 case DESC_LDT:
940 case DESC_TSS:
941 /* ignore */
942 break;
943
944 default: {
945 xmaddr_t maddr = virt_to_machine(&dt[entry]);
946
947 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
948 dt[entry] = *(struct desc_struct *)desc;
949 }
950
951 }
952}
953
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100954static void xen_load_sp0(struct tss_struct *tss,
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700955 struct thread_struct *thread)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700956{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800957 struct multicall_space mcs;
958
959 mcs = xen_mc_entry(0);
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100960 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700961 xen_mc_issue(PARAVIRT_LAZY_CPU);
Andy Lutomirski8ef46a62015-03-05 19:19:02 -0800962 tss->x86_tss.sp0 = thread->sp0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700963}
964
Andy Lutomirskib7a584592016-03-16 14:14:21 -0700965void xen_set_iopl_mask(unsigned mask)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700966{
967 struct physdev_set_iopl set_iopl;
968
969 /* Force the change at ring 0. */
970 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
971 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
972}
973
974static void xen_io_delay(void)
975{
976}
977
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +0100978static void xen_clts(void)
979{
980 struct multicall_space mcs;
981
982 mcs = xen_mc_entry(0);
983
984 MULTI_fpu_taskswitch(mcs.mc, 0);
985
986 xen_mc_issue(PARAVIRT_LAZY_CPU);
987}
988
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -0700989static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
990
991static unsigned long xen_read_cr0(void)
992{
Alex Shi2113f462012-01-13 23:53:35 +0800993 unsigned long cr0 = this_cpu_read(xen_cr0_value);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -0700994
995 if (unlikely(cr0 == 0)) {
996 cr0 = native_read_cr0();
Alex Shi2113f462012-01-13 23:53:35 +0800997 this_cpu_write(xen_cr0_value, cr0);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -0700998 }
999
1000 return cr0;
1001}
1002
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001003static void xen_write_cr0(unsigned long cr0)
1004{
1005 struct multicall_space mcs;
1006
Alex Shi2113f462012-01-13 23:53:35 +08001007 this_cpu_write(xen_cr0_value, cr0);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001008
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001009 /* Only pay attention to cr0.TS; everything else is
1010 ignored. */
1011 mcs = xen_mc_entry(0);
1012
1013 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1014
1015 xen_mc_issue(PARAVIRT_LAZY_CPU);
1016}
1017
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001018static void xen_write_cr4(unsigned long cr4)
1019{
Boris Ostrovsky3375d822015-08-10 16:34:38 -04001020 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
Jeremy Fitzhardinge2956a352008-05-26 23:31:04 +01001021
1022 native_write_cr4(cr4);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001023}
Konrad Rzeszutek Wilk1a7bbda2012-10-10 13:25:48 -04001024#ifdef CONFIG_X86_64
1025static inline unsigned long xen_read_cr8(void)
1026{
1027 return 0;
1028}
1029static inline void xen_write_cr8(unsigned long val)
1030{
1031 BUG_ON(val);
1032}
1033#endif
Boris Ostrovsky31795b472015-02-11 14:39:18 -05001034
1035static u64 xen_read_msr_safe(unsigned int msr, int *err)
1036{
1037 u64 val;
1038
Boris Ostrovsky6b08cd62015-08-10 16:34:36 -04001039 if (pmu_msr_read(msr, &val, err))
1040 return val;
1041
Boris Ostrovsky31795b472015-02-11 14:39:18 -05001042 val = native_read_msr_safe(msr, err);
1043 switch (msr) {
1044 case MSR_IA32_APICBASE:
1045#ifdef CONFIG_X86_X2APIC
1046 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1047#endif
1048 val &= ~X2APIC_ENABLE;
1049 break;
1050 }
1051 return val;
1052}
1053
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001054static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1055{
1056 int ret;
1057
1058 ret = 0;
1059
Tejf63c2f22008-12-16 11:56:06 -08001060 switch (msr) {
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001061#ifdef CONFIG_X86_64
1062 unsigned which;
1063 u64 base;
1064
1065 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1066 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1067 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1068
1069 set:
1070 base = ((u64)high << 32) | low;
1071 if (HYPERVISOR_set_segment_base(which, base) != 0)
H. Peter Anvin0cc02132009-08-31 14:23:29 -07001072 ret = -EIO;
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001073 break;
1074#endif
Jeremy Fitzhardinged89961e2008-07-24 13:48:58 -07001075
1076 case MSR_STAR:
1077 case MSR_CSTAR:
1078 case MSR_LSTAR:
1079 case MSR_SYSCALL_MASK:
1080 case MSR_IA32_SYSENTER_CS:
1081 case MSR_IA32_SYSENTER_ESP:
1082 case MSR_IA32_SYSENTER_EIP:
1083 /* Fast syscall setup is all done in hypercalls, so
1084 these are all ignored. Stub them out here to stop
1085 Xen console noise. */
Boris Ostrovsky2ecf91b2015-09-21 09:09:28 -04001086 break;
Jeremy Fitzhardinge41f2e472010-03-30 11:47:40 -07001087
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001088 default:
Boris Ostrovsky6b08cd62015-08-10 16:34:36 -04001089 if (!pmu_msr_write(msr, low, high, &ret))
1090 ret = native_write_msr_safe(msr, low, high);
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001091 }
1092
1093 return ret;
1094}
1095
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +01001096void xen_setup_shared_info(void)
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001097{
1098 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001099 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1100 xen_start_info->shared_info);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001101
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001102 HYPERVISOR_shared_info =
1103 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001104 } else
1105 HYPERVISOR_shared_info =
1106 (struct shared_info *)__va(xen_start_info->shared_info);
1107
1108#ifndef CONFIG_SMP
1109 /* In UP this is as good a place as any to set up shared info */
1110 xen_setup_vcpu_info_placement();
1111#endif
Jeremy Fitzhardinged5edbc12008-05-26 23:31:22 +01001112
1113 xen_setup_mfn_list_list();
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001114}
1115
Rusty Russell5f054e32012-03-29 15:38:31 +10301116/* This is called once we have the cpu_possible_mask */
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +01001117void xen_setup_vcpu_info_placement(void)
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001118{
1119 int cpu;
1120
1121 for_each_possible_cpu(cpu)
1122 xen_vcpu_setup(cpu);
1123
1124 /* xen_vcpu_setup managed to place the vcpu_info within the
Mukesh Rathor27713742013-12-11 15:36:51 -05001125 * percpu area for all cpus, so make use of it. Note that for
1126 * PVH we want to use native IRQ mechanism. */
1127 if (have_vcpu_info_placement && !xen_pvh_domain()) {
Jeremy Fitzhardingeecb93d12009-01-28 14:35:05 -08001128 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1129 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1130 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1131 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001132 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001133 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001134}
1135
Andi Kleenab144f52007-08-10 22:31:03 +02001136static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1137 unsigned long addr, unsigned len)
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001138{
1139 char *start, *end, *reloc;
1140 unsigned ret;
1141
1142 start = end = reloc = NULL;
1143
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001144#define SITE(op, x) \
1145 case PARAVIRT_PATCH(op.x): \
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001146 if (have_vcpu_info_placement) { \
1147 start = (char *)xen_##x##_direct; \
1148 end = xen_##x##_direct_end; \
1149 reloc = xen_##x##_direct_reloc; \
1150 } \
1151 goto patch_site
1152
1153 switch (type) {
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001154 SITE(pv_irq_ops, irq_enable);
1155 SITE(pv_irq_ops, irq_disable);
1156 SITE(pv_irq_ops, save_fl);
1157 SITE(pv_irq_ops, restore_fl);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001158#undef SITE
1159
1160 patch_site:
1161 if (start == NULL || (end-start) > len)
1162 goto default_patch;
1163
Andi Kleenab144f52007-08-10 22:31:03 +02001164 ret = paravirt_patch_insns(insnbuf, len, start, end);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001165
1166 /* Note: because reloc is assigned from something that
1167 appears to be an array, gcc assumes it's non-null,
1168 but doesn't know its relationship with start and
1169 end. */
1170 if (reloc > start && reloc < end) {
1171 int reloc_off = reloc - start;
Andi Kleenab144f52007-08-10 22:31:03 +02001172 long *relocp = (long *)(insnbuf + reloc_off);
1173 long delta = start - (char *)addr;
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001174
1175 *relocp += delta;
1176 }
1177 break;
1178
1179 default_patch:
1180 default:
Andi Kleenab144f52007-08-10 22:31:03 +02001181 ret = paravirt_patch_default(type, clobbers, insnbuf,
1182 addr, len);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001183 break;
1184 }
1185
1186 return ret;
1187}
1188
Daniel Kiperad3062a2011-05-04 20:15:18 +02001189static const struct pv_info xen_info __initconst = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001190 .paravirt_enabled = 1,
1191 .shared_kernel_pmd = 0,
1192
Andy Lutomirski318f5a22011-08-03 09:31:53 -04001193#ifdef CONFIG_X86_64
1194 .extra_user_64bit_cs = FLAT_USER_CS64,
1195#endif
David Vrabeld8c98a12015-12-11 09:07:53 -05001196 .features = 0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001197 .name = "Xen",
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001198};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001199
Daniel Kiperad3062a2011-05-04 20:15:18 +02001200static const struct pv_init_ops xen_init_ops __initconst = {
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001201 .patch = xen_patch,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001202};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001203
Daniel Kiperad3062a2011-05-04 20:15:18 +02001204static const struct pv_cpu_ops xen_cpu_ops __initconst = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001205 .cpuid = xen_cpuid,
1206
1207 .set_debugreg = xen_set_debugreg,
1208 .get_debugreg = xen_get_debugreg,
1209
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001210 .clts = xen_clts,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001211
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001212 .read_cr0 = xen_read_cr0,
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001213 .write_cr0 = xen_write_cr0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001214
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001215 .read_cr4 = native_read_cr4,
1216 .read_cr4_safe = native_read_cr4_safe,
1217 .write_cr4 = xen_write_cr4,
1218
Konrad Rzeszutek Wilk1a7bbda2012-10-10 13:25:48 -04001219#ifdef CONFIG_X86_64
1220 .read_cr8 = xen_read_cr8,
1221 .write_cr8 = xen_write_cr8,
1222#endif
1223
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001224 .wbinvd = native_wbinvd,
1225
Boris Ostrovsky31795b472015-02-11 14:39:18 -05001226 .read_msr = xen_read_msr_safe,
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001227 .write_msr = xen_write_msr_safe,
Konrad Rzeszutek Wilk1ab46fd2012-05-30 18:23:56 -04001228
Boris Ostrovsky65d0cf02015-08-10 16:34:34 -04001229 .read_pmc = xen_read_pmc,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001230
Jeremy Fitzhardinge81e103f2008-04-17 17:40:51 +02001231 .iret = xen_iret,
Jeremy Fitzhardinge6fcac6d2008-07-08 15:07:14 -07001232#ifdef CONFIG_X86_64
Jeremy Fitzhardinge6fcac6d2008-07-08 15:07:14 -07001233 .usergs_sysret64 = xen_sysret64,
1234#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001235
1236 .load_tr_desc = paravirt_nop,
1237 .set_ldt = xen_set_ldt,
1238 .load_gdt = xen_load_gdt,
1239 .load_idt = xen_load_idt,
1240 .load_tls = xen_load_tls,
Eduardo Habkosta8fc1082008-07-08 15:07:05 -07001241#ifdef CONFIG_X86_64
1242 .load_gs_index = xen_load_gs_index,
1243#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001244
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -07001245 .alloc_ldt = xen_alloc_ldt,
1246 .free_ldt = xen_free_ldt,
1247
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001248 .store_idt = native_store_idt,
1249 .store_tr = xen_store_tr,
1250
1251 .write_ldt_entry = xen_write_ldt_entry,
1252 .write_gdt_entry = xen_write_gdt_entry,
1253 .write_idt_entry = xen_write_idt_entry,
H. Peter Anvinfaca6222008-01-30 13:31:02 +01001254 .load_sp0 = xen_load_sp0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001255
1256 .set_iopl_mask = xen_set_iopl_mask,
1257 .io_delay = xen_io_delay,
1258
Jeremy Fitzhardinge952d1d72008-07-08 15:07:01 -07001259 /* Xen takes care of %gs when switching to usermode for us */
1260 .swapgs = paravirt_nop,
1261
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08001262 .start_context_switch = paravirt_start_context_switch,
1263 .end_context_switch = xen_end_context_switch,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001264};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001265
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001266static void xen_reboot(int reason)
1267{
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001268 struct sched_shutdown r = { .reason = reason };
Boris Ostrovsky65d0cf02015-08-10 16:34:34 -04001269 int cpu;
1270
1271 for_each_online_cpu(cpu)
1272 xen_pmu_finish(cpu);
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001273
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001274 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001275 BUG();
1276}
1277
1278static void xen_restart(char *msg)
1279{
1280 xen_reboot(SHUTDOWN_reboot);
1281}
1282
1283static void xen_emergency_restart(void)
1284{
1285 xen_reboot(SHUTDOWN_reboot);
1286}
1287
1288static void xen_machine_halt(void)
1289{
1290 xen_reboot(SHUTDOWN_poweroff);
1291}
1292
Tom Goetzb2abe502011-05-16 15:06:26 -04001293static void xen_machine_power_off(void)
1294{
1295 if (pm_power_off)
1296 pm_power_off();
1297 xen_reboot(SHUTDOWN_poweroff);
1298}
1299
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001300static void xen_crash_shutdown(struct pt_regs *regs)
1301{
1302 xen_reboot(SHUTDOWN_crash);
1303}
1304
Donald Dutilef09f6d12010-07-15 14:56:49 -04001305static int
1306xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1307{
Ian Campbell086748e2010-08-03 14:55:14 -07001308 xen_reboot(SHUTDOWN_crash);
Donald Dutilef09f6d12010-07-15 14:56:49 -04001309 return NOTIFY_DONE;
1310}
1311
1312static struct notifier_block xen_panic_block = {
1313 .notifier_call= xen_panic_event,
Radim Krčmářbc5eb202014-05-13 18:56:25 +02001314 .priority = INT_MIN
Donald Dutilef09f6d12010-07-15 14:56:49 -04001315};
1316
1317int xen_panic_handler_init(void)
1318{
1319 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1320 return 0;
1321}
1322
Daniel Kiperad3062a2011-05-04 20:15:18 +02001323static const struct machine_ops xen_machine_ops __initconst = {
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001324 .restart = xen_restart,
1325 .halt = xen_machine_halt,
Tom Goetzb2abe502011-05-16 15:06:26 -04001326 .power_off = xen_machine_power_off,
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001327 .shutdown = xen_machine_halt,
1328 .crash_shutdown = xen_crash_shutdown,
1329 .emergency_restart = xen_emergency_restart,
1330};
1331
Jan Beulichf221b042015-01-13 07:40:05 +00001332static unsigned char xen_get_nmi_reason(void)
1333{
1334 unsigned char reason = 0;
1335
1336 /* Construct a value which looks like it came from port 0x61. */
1337 if (test_bit(_XEN_NMIREASON_io_error,
1338 &HYPERVISOR_shared_info->arch.nmi_reason))
1339 reason |= NMI_REASON_IOCHK;
1340 if (test_bit(_XEN_NMIREASON_pci_serr,
1341 &HYPERVISOR_shared_info->arch.nmi_reason))
1342 reason |= NMI_REASON_SERR;
1343
1344 return reason;
1345}
1346
David Vrabel96f28bc2013-04-03 17:31:50 +01001347static void __init xen_boot_params_init_edd(void)
1348{
1349#if IS_ENABLED(CONFIG_EDD)
1350 struct xen_platform_op op;
1351 struct edd_info *edd_info;
1352 u32 *mbr_signature;
1353 unsigned nr;
1354 int ret;
1355
1356 edd_info = boot_params.eddbuf;
1357 mbr_signature = boot_params.edd_mbr_sig_buffer;
1358
1359 op.cmd = XENPF_firmware_info;
1360
1361 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1362 for (nr = 0; nr < EDDMAXNR; nr++) {
1363 struct edd_info *info = edd_info + nr;
1364
1365 op.u.firmware_info.index = nr;
1366 info->params.length = sizeof(info->params);
1367 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1368 &info->params);
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001369 ret = HYPERVISOR_platform_op(&op);
David Vrabel96f28bc2013-04-03 17:31:50 +01001370 if (ret)
1371 break;
1372
1373#define C(x) info->x = op.u.firmware_info.u.disk_info.x
1374 C(device);
1375 C(version);
1376 C(interface_support);
1377 C(legacy_max_cylinder);
1378 C(legacy_max_head);
1379 C(legacy_sectors_per_track);
1380#undef C
1381 }
1382 boot_params.eddbuf_entries = nr;
1383
1384 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1385 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1386 op.u.firmware_info.index = nr;
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001387 ret = HYPERVISOR_platform_op(&op);
David Vrabel96f28bc2013-04-03 17:31:50 +01001388 if (ret)
1389 break;
1390 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1391 }
1392 boot_params.edd_mbr_sig_buf_entries = nr;
1393#endif
1394}
1395
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001396/*
1397 * Set up the GDT and segment registers for -fstack-protector. Until
1398 * we do this, we have to be careful not to call any stack-protected
1399 * function, which is most of the kernel.
Mukesh Rathor5840c842013-12-13 11:48:08 -05001400 *
1401 * Note, that it is __ref because the only caller of this after init
1402 * is PVH which is not going to use xen_load_gdt_boot or other
1403 * __init functions.
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001404 */
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001405static void __ref xen_setup_gdt(int cpu)
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001406{
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001407 if (xen_feature(XENFEAT_auto_translated_physmap)) {
1408#ifdef CONFIG_X86_64
1409 unsigned long dummy;
1410
Mukesh Rathor5840c842013-12-13 11:48:08 -05001411 load_percpu_segment(cpu); /* We need to access per-cpu area */
1412 switch_to_new_gdt(cpu); /* GDT and GS set */
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001413
1414 /* We are switching of the Xen provided GDT to our HVM mode
1415 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1416 * and we are jumping to reload it.
1417 */
1418 asm volatile ("pushq %0\n"
1419 "leaq 1f(%%rip),%0\n"
1420 "pushq %0\n"
1421 "lretq\n"
1422 "1:\n"
1423 : "=&r" (dummy) : "0" (__KERNEL_CS));
1424
1425 /*
1426 * While not needed, we also set the %es, %ds, and %fs
1427 * to zero. We don't care about %ss as it is NULL.
1428 * Strictly speaking this is not needed as Xen zeros those
1429 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1430 *
1431 * Linux zeros them in cpu_init() and in secondary_startup_64
1432 * (for BSP).
1433 */
1434 loadsegment(es, 0);
1435 loadsegment(ds, 0);
1436 loadsegment(fs, 0);
1437#else
1438 /* PVH: TODO Implement. */
1439 BUG();
1440#endif
1441 return; /* PVH does not need any PV GDT ops. */
1442 }
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001443 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1444 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1445
1446 setup_stack_canary_segment(0);
1447 switch_to_new_gdt(0);
1448
1449 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1450 pv_cpu_ops.load_gdt = xen_load_gdt;
1451}
1452
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001453#ifdef CONFIG_XEN_PVH
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001454/*
1455 * A PV guest starts with default flags that are not set for PVH, set them
1456 * here asap.
1457 */
1458static void xen_pvh_set_cr_flags(int cpu)
1459{
1460
1461 /* Some of these are setup in 'secondary_startup_64'. The others:
1462 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1463 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1464 write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
Mukesh Rathorafca5012014-01-29 16:15:18 -08001465
1466 if (!cpu)
1467 return;
1468 /*
1469 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
Ingo Molnar21c4cd12015-04-26 14:27:17 +02001470 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
Mukesh Rathorafca5012014-01-29 16:15:18 -08001471 */
1472 if (cpu_has_pse)
Andy Lutomirski375074c2014-10-24 15:58:07 -07001473 cr4_set_bits_and_update_boot(X86_CR4_PSE);
Mukesh Rathorafca5012014-01-29 16:15:18 -08001474
1475 if (cpu_has_pge)
Andy Lutomirski375074c2014-10-24 15:58:07 -07001476 cr4_set_bits_and_update_boot(X86_CR4_PGE);
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001477}
1478
1479/*
1480 * Note, that it is ref - because the only caller of this after init
1481 * is PVH which is not going to use xen_load_gdt_boot or other
1482 * __init functions.
1483 */
1484void __ref xen_pvh_secondary_vcpu_init(int cpu)
1485{
1486 xen_setup_gdt(cpu);
1487 xen_pvh_set_cr_flags(cpu);
1488}
1489
Mukesh Rathord285d682013-12-13 12:45:31 -05001490static void __init xen_pvh_early_guest_init(void)
1491{
1492 if (!xen_feature(XENFEAT_auto_translated_physmap))
1493 return;
1494
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001495 if (!xen_feature(XENFEAT_hvm_callback_vector))
1496 return;
1497
1498 xen_have_vector_callback = 1;
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001499
1500 xen_pvh_early_cpu_init(0, false);
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001501 xen_pvh_set_cr_flags(0);
Mukesh Rathord285d682013-12-13 12:45:31 -05001502
1503#ifdef CONFIG_X86_32
1504 BUG(); /* PVH: Implement proper support. */
1505#endif
1506}
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001507#endif /* CONFIG_XEN_PVH */
Mukesh Rathord285d682013-12-13 12:45:31 -05001508
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001509/* First C function to be called on Xen boot */
Andi Kleen2605fc22014-05-02 00:44:37 +02001510asmlinkage __visible void __init xen_start_kernel(void)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001511{
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001512 struct physdev_set_iopl set_iopl;
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001513 unsigned long initrd_start = 0;
Borislav Petkov9cd25aa2015-06-04 18:55:10 +02001514 u64 pat;
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001515 int rc;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001516
1517 if (!xen_start_info)
1518 return;
1519
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07001520 xen_domain_type = XEN_PV_DOMAIN;
1521
Mukesh Rathord285d682013-12-13 12:45:31 -05001522 xen_setup_features();
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001523#ifdef CONFIG_XEN_PVH
Mukesh Rathord285d682013-12-13 12:45:31 -05001524 xen_pvh_early_guest_init();
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001525#endif
Ian Campbell7e775062010-09-30 12:37:26 +01001526 xen_setup_machphys_mapping();
1527
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001528 /* Install Xen paravirt ops */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001529 pv_info = xen_info;
David Vrabeld8c98a12015-12-11 09:07:53 -05001530 if (xen_initial_domain())
1531 pv_info.features |= PV_SUPPORTED_RTC;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001532 pv_init_ops = xen_init_ops;
Jan Beulichf221b042015-01-13 07:40:05 +00001533 if (!xen_pvh_domain()) {
Mukesh Rathord285d682013-12-13 12:45:31 -05001534 pv_cpu_ops = xen_cpu_ops;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001535
Jan Beulichf221b042015-01-13 07:40:05 +00001536 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1537 }
1538
David Vrabelabacaad2014-06-02 17:58:01 +01001539 if (xen_feature(XENFEAT_auto_translated_physmap))
1540 x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1541 else
1542 x86_init.resources.memory_setup = xen_memory_setup;
Thomas Gleixner42bbdb42009-08-20 13:04:10 +02001543 x86_init.oem.arch_setup = xen_arch_setup;
Thomas Gleixner6f30c1a2009-08-20 13:19:57 +02001544 x86_init.oem.banner = xen_banner;
Thomas Gleixner845b3942009-08-19 15:37:03 +02001545
Stefano Stabellini409771d2010-05-14 12:48:19 +01001546 xen_init_time_ops();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001547
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001548 /*
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001549 * Set up some pagetable state before starting to set any ptes.
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001550 */
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001551
Jeremy Fitzhardinge973df352009-10-27 16:54:19 -07001552 xen_init_mmu_ops();
1553
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001554 /* Prevent unwanted bits from being set in PTEs. */
1555 __supported_pte_mask &= ~_PAGE_GLOBAL;
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001556
Ian Campbell817a8242010-02-26 17:16:00 +00001557 /*
1558 * Prevent page tables from being allocated in highmem, even
1559 * if CONFIG_HIGHPTE is enabled.
1560 */
1561 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1562
Jeremy Fitzhardingeb75fe4e2009-09-21 13:34:06 -07001563 /* Work out if we support NX */
H. Peter Anvin4763ed42009-11-13 15:28:16 -08001564 x86_configure_nx();
Jeremy Fitzhardingeb75fe4e2009-09-21 13:34:06 -07001565
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001566 /* Get mfn list */
Konrad Rzeszutek Wilk696fd7c2013-12-15 12:37:46 -05001567 xen_build_dynamic_phys_to_machine();
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001568
1569 /*
1570 * Set up kernel GDT and segment registers, mainly so that
1571 * -fstack-protector code can be executed.
1572 */
Mukesh Rathor5840c842013-12-13 11:48:08 -05001573 xen_setup_gdt(0);
Jeremy Fitzhardinge0d1edf42008-07-28 11:53:57 -07001574
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001575 xen_init_irq_ops();
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -08001576 xen_init_cpuid_mask();
1577
Yinghai Lu94a8c3c2008-07-13 22:19:35 -07001578#ifdef CONFIG_X86_LOCAL_APIC
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001579 /*
Yinghai Lu94a8c3c2008-07-13 22:19:35 -07001580 * set up the basic apic ops.
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001581 */
Konrad Rzeszutek Wilkfeb44f12015-03-02 12:06:23 -05001582 xen_init_apic();
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001583#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001584
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -07001585 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1586 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1587 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1588 }
1589
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001590 machine_ops = xen_machine_ops;
1591
Jeremy Fitzhardinge38341432009-02-02 13:55:31 -08001592 /*
1593 * The only reliable way to retain the initial address of the
1594 * percpu gdt_page is to remember it here, so we can go and
1595 * mark it RW later, when the initial percpu area is freed.
1596 */
1597 xen_initial_gdt = &per_cpu(gdt_page, 0);
Jeremy Fitzhardinge795f99b2009-01-30 17:47:54 +09001598
Jeremy Fitzhardingea9e70622008-07-08 15:06:41 -07001599 xen_smp_init();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001600
Ian Campbellc1f5db12010-12-03 09:54:03 +00001601#ifdef CONFIG_ACPI_NUMA
1602 /*
1603 * The pages we from Xen are not related to machine pages, so
1604 * any NUMA information the kernel tries to get from ACPI will
1605 * be meaningless. Prevent it from trying.
1606 */
1607 acpi_numa = -1;
1608#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001609 /* Don't do the full vcpu_info placement stuff until we have a
1610 possible map and a non-dummy shared_info. */
1611 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1612
Jeremy Fitzhardinge55d80852009-02-25 09:42:25 -08001613 local_irq_disable();
Tejun Heo2ce802f2011-01-20 12:06:35 +01001614 early_boot_irqs_disabled = true;
Jeremy Fitzhardinge55d80852009-02-25 09:42:25 -08001615
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001616 xen_raw_console_write("mapping kernel into physical memory\n");
Juergen Gross6c2681c2015-07-17 06:51:34 +02001617 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1618 xen_start_info->nr_pages);
1619 xen_reserve_special_pages();
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001620
Juergen Gross47591df2014-11-03 14:02:04 +01001621 /*
1622 * Modify the cache mode translation tables to match Xen's PAT
1623 * configuration.
1624 */
Borislav Petkov9cd25aa2015-06-04 18:55:10 +02001625 rdmsrl(MSR_IA32_CR_PAT, pat);
1626 pat_init_cache_modes(pat);
Juergen Gross47591df2014-11-03 14:02:04 +01001627
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001628 /* keep using Xen gdt for now; no urgent need to change it */
1629
Ian Campbelle68266b2010-01-13 10:16:08 +00001630#ifdef CONFIG_X86_32
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001631 pv_info.kernel_rpl = 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001632 if (xen_feature(XENFEAT_supervisor_mode_kernel))
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001633 pv_info.kernel_rpl = 0;
Ian Campbelle68266b2010-01-13 10:16:08 +00001634#else
1635 pv_info.kernel_rpl = 0;
1636#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001637 /* set the limit of our address space */
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001638 xen_reserve_top();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001639
Mukesh Rathord285d682013-12-13 12:45:31 -05001640 /* PVH: runs at default kernel iopl of 0 */
1641 if (!xen_pvh_domain()) {
1642 /*
1643 * We used to do this in xen_arch_setup, but that is too late
1644 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1645 * early_amd_init which pokes 0xcf8 port.
1646 */
1647 set_iopl.iopl = 1;
1648 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1649 if (rc != 0)
1650 xen_raw_printk("physdev_op failed %d\n", rc);
1651 }
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001652
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001653#ifdef CONFIG_X86_32
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001654 /* set up basic CPUID stuff */
1655 cpu_detect(&new_cpu_data);
H. Peter Anvin60e019e2013-04-29 16:04:20 +02001656 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
Jeremy Fitzhardinged560bc62009-08-25 12:53:02 -07001657 new_cpu_data.wp_works_ok = 1;
Huaitong Han16aaa532016-01-25 20:41:47 +01001658 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001659#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001660
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001661 if (xen_start_info->mod_start) {
1662 if (xen_start_info->flags & SIF_MOD_START_PFN)
1663 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1664 else
1665 initrd_start = __pa(xen_start_info->mod_start);
1666 }
1667
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001668 /* Poke various useful things into boot_params */
H. Peter Anvin30c82642007-10-15 17:13:22 -07001669 boot_params.hdr.type_of_loader = (9 << 4) | 0;
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001670 boot_params.hdr.ramdisk_image = initrd_start;
H. Peter Anvin30c82642007-10-15 17:13:22 -07001671 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
Jeremy Fitzhardingeb7c3c5c2008-07-08 15:07:07 -07001672 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001673
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07001674 if (!xen_initial_domain()) {
Jeremy Fitzhardinge83abc702008-05-26 23:31:12 +01001675 add_preferred_console("xenboot", 0, NULL);
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001676 add_preferred_console("tty", 0, NULL);
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +01001677 add_preferred_console("hvc", 0, NULL);
Alex Nixonb5401a92010-03-18 16:31:34 -04001678 if (pci_xen)
1679 x86_init.pci.arch_init = pci_xen_init;
Chris Wright5d990b62009-12-04 12:15:21 -08001680 } else {
Jeremy Fitzhardingec2419b42011-05-31 10:50:10 -04001681 const struct dom0_vga_console_info *info =
1682 (void *)((char *)xen_start_info +
1683 xen_start_info->console.dom0.info_off);
Konrad Rzeszutek Wilkffb8b232012-09-21 12:30:35 -04001684 struct xen_platform_op op = {
1685 .cmd = XENPF_firmware_info,
1686 .interface_version = XENPF_INTERFACE_VERSION,
1687 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1688 };
Jeremy Fitzhardingec2419b42011-05-31 10:50:10 -04001689
1690 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1691 xen_start_info->console.domU.mfn = 0;
1692 xen_start_info->console.domU.evtchn = 0;
1693
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001694 if (HYPERVISOR_platform_op(&op) == 0)
Konrad Rzeszutek Wilkffb8b232012-09-21 12:30:35 -04001695 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1696
Chris Wright5d990b62009-12-04 12:15:21 -08001697 /* Make sure ACS will be enabled */
1698 pci_request_acs();
Konrad Rzeszutek Wilk211063d2011-12-08 17:32:23 +08001699
1700 xen_acpi_sleep_register();
Konrad Rzeszutek Wilkbd499402012-09-19 08:30:55 -04001701
1702 /* Avoid searching for BIOS MP tables */
1703 x86_init.mpparse.find_smp_config = x86_init_noop;
1704 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
David Vrabel96f28bc2013-04-03 17:31:50 +01001705
1706 xen_boot_params_init_edd();
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001707 }
David Vrabel76a8df72012-05-04 14:29:46 +01001708#ifdef CONFIG_PCI
1709 /* PCI BIOS service won't work from a PV guest. */
1710 pci_probe &= ~PCI_PROBE_BIOS;
1711#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001712 xen_raw_console_write("about to get started...\n");
1713
Jeremy Fitzhardinge499d19b2009-11-24 09:38:25 -08001714 xen_setup_runstate_info(0);
1715
Daniel Kiperc7341d62014-07-12 23:09:48 +02001716 xen_efi_init();
Daniel Kiperbe81c8a2014-06-30 19:53:02 +02001717
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001718 /* Start the world */
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001719#ifdef CONFIG_X86_32
Yinghai Luf0d43102008-05-29 12:56:36 -07001720 i386_start_kernel();
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001721#else
Boris Ostrovsky5054daa2015-02-23 11:01:00 -05001722 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001723 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001724#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001725}
Sheng Yangbee6ab52010-05-14 12:39:33 +01001726
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001727void __ref xen_hvm_init_shared_info(void)
Olaf Hering9d02b432012-11-01 22:02:30 +01001728{
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001729 int cpu;
Olaf Hering9d02b432012-11-01 22:02:30 +01001730 struct xen_add_to_physmap xatp;
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001731 static struct shared_info *shared_info_page = 0;
Olaf Hering9d02b432012-11-01 22:02:30 +01001732
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001733 if (!shared_info_page)
1734 shared_info_page = (struct shared_info *)
1735 extend_brk(PAGE_SIZE, PAGE_SIZE);
Sheng Yangbee6ab52010-05-14 12:39:33 +01001736 xatp.domid = DOMID_SELF;
1737 xatp.idx = 0;
1738 xatp.space = XENMAPSPACE_shared_info;
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001739 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001740 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1741 BUG();
1742
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001743 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001744
Stefano Stabellini016b6f52010-05-14 12:45:07 +01001745 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1746 * page, we use it in the event channel upcall and in some pvclock
1747 * related functions. We don't need the vcpu_info placement
1748 * optimizations because we don't use any pv_mmu or pv_irq op on
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001749 * HVM.
1750 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1751 * online but xen_hvm_init_shared_info is run at resume time too and
1752 * in that case multiple vcpus might be online. */
1753 for_each_online_cpu(cpu) {
Konrad Rzeszutek Wilkd5b17db2013-05-05 09:24:07 -04001754 /* Leave it to be NULL. */
1755 if (cpu >= MAX_VIRT_CPUS)
1756 continue;
Stefano Stabellini016b6f52010-05-14 12:45:07 +01001757 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
Olaf Hering9d02b432012-11-01 22:02:30 +01001758 }
Olaf Hering9d02b432012-11-01 22:02:30 +01001759}
1760
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001761#ifdef CONFIG_XEN_PVHVM
Olaf Hering4ff2d062012-07-17 11:59:15 +02001762static void __init init_hvm_pv_info(void)
1763{
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001764 int major, minor;
Konrad Rzeszutek Wilk5eb65be2013-02-14 21:29:27 -05001765 uint32_t eax, ebx, ecx, edx, pages, msr, base;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001766 u64 pfn;
1767
1768 base = xen_cpuid_base();
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001769 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1770
1771 major = eax >> 16;
1772 minor = eax & 0xffff;
1773 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1774
Sheng Yangbee6ab52010-05-14 12:39:33 +01001775 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1776
1777 pfn = __pa(hypercall_page);
1778 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1779
1780 xen_setup_features();
1781
1782 pv_info.name = "Xen HVM";
1783
1784 xen_domain_type = XEN_HVM_DOMAIN;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001785}
1786
Paul Gortmaker148f9bb2013-06-18 18:23:59 -04001787static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
1788 void *hcpu)
Sheng Yang38e20b02010-05-14 12:40:51 +01001789{
1790 int cpu = (long)hcpu;
1791 switch (action) {
1792 case CPU_UP_PREPARE:
Zhenzhong Duan90d4f552011-10-27 22:28:59 -07001793 xen_vcpu_setup(cpu);
Konrad Rzeszutek Wilk7918c922013-04-16 15:18:00 -04001794 if (xen_have_vector_callback) {
Konrad Rzeszutek Wilk7918c922013-04-16 15:18:00 -04001795 if (xen_feature(XENFEAT_hvm_safe_pvclock))
1796 xen_setup_timer(cpu);
1797 }
Sheng Yang38e20b02010-05-14 12:40:51 +01001798 break;
1799 default:
1800 break;
1801 }
1802 return NOTIFY_OK;
1803}
1804
Paul Gortmaker148f9bb2013-06-18 18:23:59 -04001805static struct notifier_block xen_hvm_cpu_notifier = {
Sheng Yang38e20b02010-05-14 12:40:51 +01001806 .notifier_call = xen_hvm_cpu_notify,
1807};
1808
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +02001809#ifdef CONFIG_KEXEC_CORE
1810static void xen_hvm_shutdown(void)
1811{
1812 native_machine_shutdown();
1813 if (kexec_in_progress)
1814 xen_reboot(SHUTDOWN_soft_reset);
1815}
1816
1817static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1818{
1819 native_machine_crash_shutdown(regs);
1820 xen_reboot(SHUTDOWN_soft_reset);
1821}
1822#endif
1823
Sheng Yangbee6ab52010-05-14 12:39:33 +01001824static void __init xen_hvm_guest_init(void)
1825{
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001826 if (xen_pv_domain())
1827 return;
1828
Olaf Hering4ff2d062012-07-17 11:59:15 +02001829 init_hvm_pv_info();
Sheng Yangbee6ab52010-05-14 12:39:33 +01001830
Stefano Stabellini016b6f52010-05-14 12:45:07 +01001831 xen_hvm_init_shared_info();
Sheng Yang38e20b02010-05-14 12:40:51 +01001832
Vaughan Cao669b0ae2013-08-16 16:10:56 +08001833 xen_panic_handler_init();
1834
Sheng Yang38e20b02010-05-14 12:40:51 +01001835 if (xen_feature(XENFEAT_hvm_callback_vector))
1836 xen_have_vector_callback = 1;
Stefano Stabellini99bbb3a2010-12-02 17:55:10 +00001837 xen_hvm_smp_init();
Sheng Yang38e20b02010-05-14 12:40:51 +01001838 register_cpu_notifier(&xen_hvm_cpu_notifier);
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001839 xen_unplug_emulated_devices();
Sheng Yang38e20b02010-05-14 12:40:51 +01001840 x86_init.irqs.intr_init = xen_init_IRQ;
Stefano Stabellini409771d2010-05-14 12:48:19 +01001841 xen_hvm_init_time_ops();
Stefano Stabellini59151002010-06-17 14:22:52 +01001842 xen_hvm_init_mmu_ops();
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +02001843#ifdef CONFIG_KEXEC_CORE
1844 machine_ops.shutdown = xen_hvm_shutdown;
1845 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1846#endif
Sheng Yangbee6ab52010-05-14 12:39:33 +01001847}
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001848#endif
Sheng Yangbee6ab52010-05-14 12:39:33 +01001849
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001850static bool xen_nopv = false;
1851static __init int xen_parse_nopv(char *arg)
1852{
1853 xen_nopv = true;
1854 return 0;
1855}
1856early_param("xen_nopv", xen_parse_nopv);
1857
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001858static uint32_t __init xen_platform(void)
Sheng Yangbee6ab52010-05-14 12:39:33 +01001859{
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001860 if (xen_nopv)
1861 return 0;
1862
Jason Wang9df56f12013-07-25 16:54:35 +08001863 return xen_cpuid_base();
Sheng Yangbee6ab52010-05-14 12:39:33 +01001864}
1865
Sheng Yangd9b8ca82010-12-21 14:18:49 +08001866bool xen_hvm_need_lapic(void)
1867{
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001868 if (xen_nopv)
1869 return false;
Sheng Yangd9b8ca82010-12-21 14:18:49 +08001870 if (xen_pv_domain())
1871 return false;
1872 if (!xen_hvm_domain())
1873 return false;
1874 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
1875 return false;
1876 return true;
1877}
1878EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1879
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001880static void xen_set_cpu_features(struct cpuinfo_x86 *c)
1881{
Boris Ostrovsky91e2eea2015-11-19 16:55:45 -05001882 if (xen_pv_domain()) {
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001883 clear_cpu_bug(c, X86_BUG_SYSRET_SS_ATTRS);
Boris Ostrovsky91e2eea2015-11-19 16:55:45 -05001884 set_cpu_cap(c, X86_FEATURE_XENPV);
1885 }
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001886}
1887
1888const struct hypervisor_x86 x86_hyper_xen = {
1889 .name = "Xen",
1890 .detect = xen_platform,
1891#ifdef CONFIG_XEN_PVHVM
Sheng Yangbee6ab52010-05-14 12:39:33 +01001892 .init_platform = xen_hvm_guest_init,
Stefano Stabellinica65f9f2010-07-29 14:37:48 +01001893#endif
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001894 .x2apic_available = xen_x2apic_para_available,
1895 .set_cpu_features = xen_set_cpu_features,
1896};
1897EXPORT_SYMBOL(x86_hyper_xen);
Stefano Stabellinia314e3e2015-10-22 16:20:46 +00001898
1899#ifdef CONFIG_HOTPLUG_CPU
1900void xen_arch_register_cpu(int num)
1901{
1902 arch_register_cpu(num);
1903}
1904EXPORT_SYMBOL(xen_arch_register_cpu);
1905
1906void xen_arch_unregister_cpu(int num)
1907{
1908 arch_unregister_cpu(num);
1909}
1910EXPORT_SYMBOL(xen_arch_unregister_cpu);
1911#endif