blob: db7cf8727e1cfacc1e7f6ea2ce6e72622d36a2b9 [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>
Paul Gortmaker7a2463d2016-07-13 20:18:59 -040026#include <linux/export.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#include <linux/kexec.h>
Nicolai Stange8574df12018-07-29 12:15:33 +020038#include <linux/slab.h>
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +020039
Jeremy Fitzhardinge1ccbf532009-10-06 15:11:14 -070040#include <xen/xen.h>
Stefano Stabellini0ec53ec2012-09-14 13:37:32 +000041#include <xen/events.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070042#include <xen/interface/xen.h>
Jeremy Fitzhardingeecbf29c2008-12-16 12:37:07 -080043#include <xen/interface/version.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070044#include <xen/interface/physdev.h>
45#include <xen/interface/vcpu.h>
Sheng Yangbee6ab52010-05-14 12:39:33 +010046#include <xen/interface/memory.h>
Jan Beulichf221b042015-01-13 07:40:05 +000047#include <xen/interface/nmi.h>
Liu, Jinsongcef12ee2012-06-07 19:56:51 +080048#include <xen/interface/xen-mca.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070049#include <xen/features.h>
50#include <xen/page.h>
Sheng Yang38e20b02010-05-14 12:40:51 +010051#include <xen/hvm.h>
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -070052#include <xen/hvc-console.h>
Konrad Rzeszutek Wilk211063d2011-12-08 17:32:23 +080053#include <xen/acpi.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070054
55#include <asm/paravirt.h>
Ingo Molnar7b6aa332009-02-17 13:58:15 +010056#include <asm/apic.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070057#include <asm/page.h>
Alex Nixonb5401a92010-03-18 16:31:34 -040058#include <asm/xen/pci.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070059#include <asm/xen/hypercall.h>
60#include <asm/xen/hypervisor.h>
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +020061#include <asm/xen/cpuid.h>
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -070062#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>
Stefano Stabellinia314e3e2015-10-22 16:20:46 +000078#include <asm/cpu.h>
Greg Kroah-Hartman10d15c42018-06-03 13:37:03 +020079#include <asm/unwind_hints.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
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +0200122/* Linux <-> Xen vCPU id mapping */
Vitaly Kuznetsov55467de2016-07-29 11:06:48 +0200123DEFINE_PER_CPU(uint32_t, xen_vcpu_id);
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +0200124EXPORT_PER_CPU_SYMBOL(xen_vcpu_id);
125
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -0700126enum xen_domain_type xen_domain_type = XEN_NATIVE;
127EXPORT_SYMBOL_GPL(xen_domain_type);
128
Ian Campbell7e775062010-09-30 12:37:26 +0100129unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
130EXPORT_SYMBOL(machine_to_phys_mapping);
Jan Beulichccbcdf72011-08-16 15:07:41 +0100131unsigned long machine_to_phys_nr;
132EXPORT_SYMBOL(machine_to_phys_nr);
Ian Campbell7e775062010-09-30 12:37:26 +0100133
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700134struct start_info *xen_start_info;
135EXPORT_SYMBOL_GPL(xen_start_info);
136
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +0100137struct shared_info xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700138
Jeremy Fitzhardinge38341432009-02-02 13:55:31 -0800139void *xen_initial_gdt;
140
Sheng Yangbee6ab52010-05-14 12:39:33 +0100141RESERVE_BRK(shared_info_page_brk, PAGE_SIZE);
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -0400142__read_mostly int xen_have_vector_callback;
143EXPORT_SYMBOL_GPL(xen_have_vector_callback);
Sheng Yangbee6ab52010-05-14 12:39:33 +0100144
Boris Ostrovsky4d737042016-09-07 13:19:00 -0400145static int xen_cpu_up_prepare(unsigned int cpu);
146static int xen_cpu_up_online(unsigned int cpu);
147static int xen_cpu_dead(unsigned int cpu);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700148
149/*
150 * Point at some empty memory to start with. We map the real shared_info
151 * page as soon as fixmap is up and running.
152 */
Olaf Hering4648da72012-07-17 11:57:41 +0200153struct shared_info *HYPERVISOR_shared_info = &xen_dummy_shared_info;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700154
155/*
156 * Flag to determine whether vcpu info placement is available on all
157 * VCPUs. We assume it is to start with, and then set it to zero on
158 * the first failure. This is because it can succeed on some VCPUs
159 * and not others, since it can involve hypervisor memory allocation,
160 * or because the guest failed to guarantee all the appropriate
161 * constraints on all VCPUs (ie buffer can't cross a page boundary).
162 *
163 * Note that any particular CPU may be using a placed vcpu structure,
164 * but we can only optimise if the all are.
165 *
166 * 0: not available, 1: available
167 */
Jeremy Fitzhardingee4d04072009-02-02 13:55:54 -0800168static int have_vcpu_info_placement = 1;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700169
David Vrabel1c32cdc2012-07-09 11:39:08 +0100170struct tls_descs {
171 struct desc_struct desc[3];
172};
173
174/*
175 * Updating the 3 TLS descriptors in the GDT on every task switch is
176 * surprisingly expensive so we avoid updating them if they haven't
177 * changed. Since Xen writes different descriptors than the one
178 * passed in the update_descriptor hypercall we keep shadow copies to
179 * compare against.
180 */
181static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc);
182
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700183static void clamp_max_cpus(void)
184{
185#ifdef CONFIG_SMP
186 if (setup_max_cpus > MAX_VIRT_CPUS)
187 setup_max_cpus = MAX_VIRT_CPUS;
188#endif
189}
190
Vitaly Kuznetsovee42d662016-06-30 17:56:43 +0200191void xen_vcpu_setup(int cpu)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700192{
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700193 struct vcpu_register_vcpu_info info;
194 int err;
195 struct vcpu_info *vcpup;
196
Jeremy Fitzhardingea0d695c2008-05-26 23:31:21 +0100197 BUG_ON(HYPERVISOR_shared_info == &xen_dummy_shared_info);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700198
Konrad Rzeszutek Wilk7f1fc262013-05-05 09:30:09 -0400199 /*
200 * This path is called twice on PVHVM - first during bootup via
201 * smp_init -> xen_hvm_cpu_notify, and then if the VCPU is being
202 * hotplugged: cpu_up -> xen_hvm_cpu_notify.
203 * As we can only do the VCPUOP_register_vcpu_info once lets
204 * not over-write its result.
205 *
206 * For PV it is called during restore (xen_vcpu_restore) and bootup
207 * (xen_setup_vcpu_info_placement). The hotplug mechanism does not
208 * use this function.
209 */
210 if (xen_hvm_domain()) {
211 if (per_cpu(xen_vcpu, cpu) == &per_cpu(xen_vcpu_info, cpu))
212 return;
213 }
Vitaly Kuznetsove15a8622016-06-30 17:56:39 +0200214 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
215 per_cpu(xen_vcpu, cpu) =
216 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700217
218 if (!have_vcpu_info_placement) {
219 if (cpu >= MAX_VIRT_CPUS)
220 clamp_max_cpus();
221 return;
222 }
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700223
224 vcpup = &per_cpu(xen_vcpu_info, cpu);
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800225 info.mfn = arbitrary_virt_to_mfn(vcpup);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700226 info.offset = offset_in_page(vcpup);
227
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700228 /* Check to see if the hypervisor will put the vcpu_info
229 structure where we want it, which allows direct access via
Konrad Rzeszutek Wilka5209962013-05-05 08:51:47 -0400230 a percpu-variable.
231 N.B. This hypercall can _only_ be called once per CPU. Subsequent
232 calls will error out with -EINVAL. This is due to the fact that
233 hypervisor has no unregister variant and this hypercall does not
234 allow to over-write info.mfn and info.offset.
235 */
Vitaly Kuznetsovad5475f2016-06-30 17:56:38 +0200236 err = HYPERVISOR_vcpu_op(VCPUOP_register_vcpu_info, xen_vcpu_nr(cpu),
237 &info);
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700238
239 if (err) {
240 printk(KERN_DEBUG "register_vcpu_info failed: err=%d\n", err);
241 have_vcpu_info_placement = 0;
Mukesh Rathorc06ee782010-07-19 10:25:08 -0700242 clamp_max_cpus();
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700243 } else {
244 /* This cpu is using the registered vcpu info, even if
245 later ones fail to. */
246 per_cpu(xen_vcpu, cpu) = vcpup;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -0700247 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700248}
249
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100250/*
251 * On restore, set the vcpu placement up again.
252 * If it fails, then we're in a bad state, since
253 * we can't back out from using it...
254 */
255void xen_vcpu_restore(void)
256{
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800257 int cpu;
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100258
Wei Liu9d328a92012-12-13 15:33:05 +0000259 for_each_possible_cpu(cpu) {
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800260 bool other_cpu = (cpu != smp_processor_id());
Vitaly Kuznetsovad5475f2016-06-30 17:56:38 +0200261 bool is_up = HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu),
262 NULL);
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100263
Wei Liu9d328a92012-12-13 15:33:05 +0000264 if (other_cpu && is_up &&
Vitaly Kuznetsovad5475f2016-06-30 17:56:38 +0200265 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL))
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800266 BUG();
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100267
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800268 xen_setup_runstate_info(cpu);
Ian Campbellbe012922009-11-21 08:35:55 +0800269
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800270 if (have_vcpu_info_placement)
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100271 xen_vcpu_setup(cpu);
272
Wei Liu9d328a92012-12-13 15:33:05 +0000273 if (other_cpu && is_up &&
Vitaly Kuznetsovad5475f2016-06-30 17:56:38 +0200274 HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL))
Jeremy Fitzhardinge3905bb22009-11-21 08:46:29 +0800275 BUG();
Jeremy Fitzhardinge9c7a7942008-05-31 01:33:02 +0100276 }
277}
278
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700279static void __init xen_banner(void)
280{
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700281 unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL);
282 struct xen_extraversion extra;
283 HYPERVISOR_xen_version(XENVER_extraversion, &extra);
284
Mukesh Rathord285d682013-12-13 12:45:31 -0500285 pr_info("Booting paravirtualized kernel %son %s\n",
286 xen_feature(XENFEAT_auto_translated_physmap) ?
287 "with PVH extensions " : "", pv_info.name);
Jeremy Fitzhardinge95c7c232008-07-15 13:42:34 -0700288 printk(KERN_INFO "Xen version: %d.%d%s%s\n",
289 version >> 16, version & 0xffff, extra.extraversion,
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -0700290 xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : "");
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700291}
Konrad Rzeszutek Wilk394b40f2012-11-27 11:39:40 -0500292/* Check if running on Xen version (major, minor) or later */
293bool
294xen_running_on_version_or_later(unsigned int major, unsigned int minor)
295{
296 unsigned int version;
297
298 if (!xen_domain())
299 return false;
300
301 version = HYPERVISOR_xen_version(XENVER_version, NULL);
302 if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) ||
303 ((version >> 16) > major))
304 return true;
305 return false;
306}
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700307
Andre Przywara5e626252012-05-29 13:07:31 +0200308#define CPUID_THERM_POWER_LEAF 6
309#define APERFMPERF_PRESENT 0
310
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800311static __read_mostly unsigned int cpuid_leaf1_edx_mask = ~0;
312static __read_mostly unsigned int cpuid_leaf1_ecx_mask = ~0;
313
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500314static __read_mostly unsigned int cpuid_leaf1_ecx_set_mask;
315static __read_mostly unsigned int cpuid_leaf5_ecx_val;
316static __read_mostly unsigned int cpuid_leaf5_edx_val;
317
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100318static void xen_cpuid(unsigned int *ax, unsigned int *bx,
319 unsigned int *cx, unsigned int *dx)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700320{
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700321 unsigned maskebx = ~0;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800322 unsigned maskecx = ~0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700323 unsigned maskedx = ~0;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500324 unsigned setecx = 0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700325 /*
326 * Mask out inconvenient features, to try and disable as many
327 * unsupported kernel subsystems as possible.
328 */
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700329 switch (*ax) {
330 case 1:
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800331 maskecx = cpuid_leaf1_ecx_mask;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500332 setecx = cpuid_leaf1_ecx_set_mask;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800333 maskedx = cpuid_leaf1_edx_mask;
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700334 break;
335
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500336 case CPUID_MWAIT_LEAF:
337 /* Synthesize the values.. */
338 *ax = 0;
339 *bx = 0;
340 *cx = cpuid_leaf5_ecx_val;
341 *dx = cpuid_leaf5_edx_val;
342 return;
343
Andre Przywara5e626252012-05-29 13:07:31 +0200344 case CPUID_THERM_POWER_LEAF:
345 /* Disabling APERFMPERF for kernel usage */
346 maskecx = ~(1 << APERFMPERF_PRESENT);
347 break;
348
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700349 case 0xb:
350 /* Suppress extended topology stuff */
351 maskebx = 0;
352 break;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800353 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700354
355 asm(XEN_EMULATE_PREFIX "cpuid"
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100356 : "=a" (*ax),
357 "=b" (*bx),
358 "=c" (*cx),
359 "=d" (*dx)
360 : "0" (*ax), "2" (*cx));
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800361
Jeremy Fitzhardinge82d64692009-10-22 16:41:15 -0700362 *bx &= maskebx;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800363 *cx &= maskecx;
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500364 *cx |= setecx;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100365 *dx &= maskedx;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700366}
Josh Poimboeuf983bb6d2016-02-28 22:22:36 -0600367STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700368
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500369static bool __init xen_check_mwait(void)
370{
Liu, Jinsonge3aa4e62012-11-08 05:44:28 +0000371#ifdef CONFIG_ACPI
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500372 struct xen_platform_op op = {
373 .cmd = XENPF_set_processor_pminfo,
374 .u.set_pminfo.id = -1,
375 .u.set_pminfo.type = XEN_PM_PDC,
376 };
377 uint32_t buf[3];
378 unsigned int ax, bx, cx, dx;
379 unsigned int mwait_mask;
380
381 /* We need to determine whether it is OK to expose the MWAIT
382 * capability to the kernel to harvest deeper than C3 states from ACPI
383 * _CST using the processor_harvest_xen.c module. For this to work, we
384 * need to gather the MWAIT_LEAF values (which the cstate.c code
385 * checks against). The hypervisor won't expose the MWAIT flag because
386 * it would break backwards compatibility; so we will find out directly
387 * from the hardware and hypercall.
388 */
389 if (!xen_initial_domain())
390 return false;
391
Liu, Jinsonge3aa4e62012-11-08 05:44:28 +0000392 /*
393 * When running under platform earlier than Xen4.2, do not expose
394 * mwait, to avoid the risk of loading native acpi pad driver
395 */
396 if (!xen_running_on_version_or_later(4, 2))
397 return false;
398
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500399 ax = 1;
400 cx = 0;
401
402 native_cpuid(&ax, &bx, &cx, &dx);
403
404 mwait_mask = (1 << (X86_FEATURE_EST % 32)) |
405 (1 << (X86_FEATURE_MWAIT % 32));
406
407 if ((cx & mwait_mask) != mwait_mask)
408 return false;
409
410 /* We need to emulate the MWAIT_LEAF and for that we need both
411 * ecx and edx. The hypercall provides only partial information.
412 */
413
414 ax = CPUID_MWAIT_LEAF;
415 bx = 0;
416 cx = 0;
417 dx = 0;
418
419 native_cpuid(&ax, &bx, &cx, &dx);
420
421 /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so,
422 * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3.
423 */
424 buf[0] = ACPI_PDC_REVISION_ID;
425 buf[1] = 1;
426 buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP);
427
428 set_xen_guest_handle(op.u.set_pminfo.pdc, buf);
429
Stefano Stabellinicfafae92015-11-23 10:36:12 +0000430 if ((HYPERVISOR_platform_op(&op) == 0) &&
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500431 (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) {
432 cpuid_leaf5_ecx_val = cx;
433 cpuid_leaf5_edx_val = dx;
434 }
435 return true;
436#else
437 return false;
438#endif
439}
Daniel Kiperad3062a2011-05-04 20:15:18 +0200440static void __init xen_init_cpuid_mask(void)
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800441{
442 unsigned int ax, bx, cx, dx;
Shan Haitao947ccf92010-11-09 11:43:36 -0800443 unsigned int xsave_mask;
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800444
445 cpuid_leaf1_edx_mask =
Liu, Jinsongcef12ee2012-06-07 19:56:51 +0800446 ~((1 << X86_FEATURE_MTRR) | /* disable MTRR */
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800447 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
448
Andy Lutomirskib52f9372017-06-29 08:53:21 -0700449 /*
450 * Xen PV would need some work to support PCID: CR3 handling as well
451 * as xen_flush_tlb_others() would need updating.
452 */
453 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_PCID % 32)); /* disable PCID */
454
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800455 if (!xen_initial_domain())
456 cpuid_leaf1_edx_mask &=
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400457 ~((1 << X86_FEATURE_ACPI)); /* disable ACPI */
Zhenzhong Duan4ea9b9a2013-03-29 09:14:00 +0800458
459 cpuid_leaf1_ecx_mask &= ~(1 << (X86_FEATURE_X2APIC % 32));
460
Shan Haitao947ccf92010-11-09 11:43:36 -0800461 ax = 1;
Konrad Rzeszutek Wilk5e287832011-09-29 13:06:42 -0400462 cx = 0;
Mukesh Rathord285d682013-12-13 12:45:31 -0500463 cpuid(1, &ax, &bx, &cx, &dx);
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800464
Shan Haitao947ccf92010-11-09 11:43:36 -0800465 xsave_mask =
466 (1 << (X86_FEATURE_XSAVE % 32)) |
467 (1 << (X86_FEATURE_OSXSAVE % 32));
468
469 /* Xen will set CR4.OSXSAVE if supported and not disabled by force */
470 if ((cx & xsave_mask) != xsave_mask)
471 cpuid_leaf1_ecx_mask &= ~xsave_mask; /* disable XSAVE & OSXSAVE */
Konrad Rzeszutek Wilk73c154c2012-02-13 22:26:32 -0500472 if (xen_check_mwait())
473 cpuid_leaf1_ecx_set_mask = (1 << (X86_FEATURE_MWAIT % 32));
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -0800474}
475
Juergen Grossbb70de12018-05-30 13:09:57 +0200476static void __init xen_init_capabilities(void)
477{
Juergen Grossc43b4ff2018-05-30 13:09:58 +0200478 if (xen_pv_domain())
Juergen Grossbb70de12018-05-30 13:09:57 +0200479 setup_force_cpu_cap(X86_FEATURE_XENPV);
Juergen Grossbb70de12018-05-30 13:09:57 +0200480}
481
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700482static void xen_set_debugreg(int reg, unsigned long val)
483{
484 HYPERVISOR_set_debugreg(reg, val);
485}
486
487static unsigned long xen_get_debugreg(int reg)
488{
489 return HYPERVISOR_get_debugreg(reg);
490}
491
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800492static void xen_end_context_switch(struct task_struct *next)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700493{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700494 xen_mc_flush();
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800495 paravirt_end_context_switch(next);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700496}
497
498static unsigned long xen_store_tr(void)
499{
500 return 0;
501}
502
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700503/*
Jeremy Fitzhardingecef43bf2008-07-28 13:33:44 -0700504 * Set the page permissions for a particular virtual address. If the
505 * address is a vmalloc mapping (or other non-linear mapping), then
506 * find the linear mapping of the page and also set its protections to
507 * match.
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700508 */
509static void set_aliased_prot(void *v, pgprot_t prot)
510{
511 int level;
512 pte_t *ptep;
513 pte_t pte;
514 unsigned long pfn;
515 struct page *page;
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700516 unsigned char dummy;
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700517
518 ptep = lookup_address((unsigned long)v, &level);
519 BUG_ON(ptep == NULL);
520
521 pfn = pte_pfn(*ptep);
522 page = pfn_to_page(pfn);
523
524 pte = pfn_pte(pfn, prot);
525
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700526 /*
527 * Careful: update_va_mapping() will fail if the virtual address
528 * we're poking isn't populated in the page tables. We don't
529 * need to worry about the direct map (that's always in the page
530 * tables), but we need to be careful about vmap space. In
531 * particular, the top level page table can lazily propagate
532 * entries between processes, so if we've switched mms since we
533 * vmapped the target in the first place, we might not have the
534 * top-level page table entry populated.
535 *
536 * We disable preemption because we want the same mm active when
537 * we probe the target and when we issue the hypercall. We'll
538 * have the same nominal mm, but if we're a kernel thread, lazy
539 * mm dropping could change our pgd.
540 *
541 * Out of an abundance of caution, this uses __get_user() to fault
542 * in the target address just in case there's some obscure case
543 * in which the target address isn't readable.
544 */
545
546 preempt_disable();
547
Andy Lutomirski99158f12016-05-24 15:48:38 -0700548 probe_kernel_read(&dummy, v, 1);
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700549
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700550 if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0))
551 BUG();
552
553 if (!PageHighMem(page)) {
554 void *av = __va(PFN_PHYS(pfn));
555
556 if (av != v)
557 if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0))
558 BUG();
559 } else
560 kmap_flush_unused();
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700561
562 preempt_enable();
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700563}
564
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700565static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries)
566{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700567 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700568 int i;
569
Andy Lutomirskiaa1acff2015-07-30 14:31:31 -0700570 /*
571 * We need to mark the all aliases of the LDT pages RO. We
572 * don't need to call vm_flush_aliases(), though, since that's
573 * only responsible for flushing aliases out the TLBs, not the
574 * page tables, and Xen will flush the TLB for us if needed.
575 *
576 * To avoid confusing future readers: none of this is necessary
577 * to load the LDT. The hypervisor only checks this when the
578 * LDT is faulted in due to subsequent descriptor access.
579 */
580
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700581 for(i = 0; i < entries; i += entries_per_page)
582 set_aliased_prot(ldt + i, PAGE_KERNEL_RO);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700583}
584
585static void xen_free_ldt(struct desc_struct *ldt, unsigned entries)
586{
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700587 const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE;
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700588 int i;
589
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700590 for(i = 0; i < entries; i += entries_per_page)
591 set_aliased_prot(ldt + i, PAGE_KERNEL);
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -0700592}
593
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700594static void xen_set_ldt(const void *addr, unsigned entries)
595{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700596 struct mmuext_op *op;
597 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
598
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800599 trace_xen_cpu_set_ldt(addr, entries);
600
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700601 op = mcs.args;
602 op->cmd = MMUEXT_SET_LDT;
Jan Beulich4dbf7af2008-01-30 13:33:14 +0100603 op->arg1.linear_addr = (unsigned long)addr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700604 op->arg2.nr_ents = entries;
605
606 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
607
608 xen_mc_issue(PARAVIRT_LAZY_CPU);
609}
610
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100611static void xen_load_gdt(const struct desc_ptr *dtr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700612{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700613 unsigned long va = dtr->address;
614 unsigned int size = dtr->size + 1;
Amitoj Kaur Chawla585423c2016-06-29 20:30:38 +0530615 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
Jeremy Fitzhardinge3ce5fa72009-03-04 15:26:00 -0800616 unsigned long frames[pages];
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700617 int f;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700618
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700619 /*
620 * A GDT can be up to 64k in size, which corresponds to 8192
621 * 8-byte entries, or 16 4k pages..
622 */
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700623
624 BUG_ON(size > 65536);
625 BUG_ON(va & ~PAGE_MASK);
626
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700627 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800628 int level;
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700629 pte_t *ptep;
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800630 unsigned long pfn, mfn;
631 void *virt;
632
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700633 /*
634 * The GDT is per-cpu and is in the percpu data area.
635 * That can be virtually mapped, so we need to do a
636 * page-walk to get the underlying MFN for the
637 * hypercall. The page can also be in the kernel's
638 * linear range, so we need to RO that mapping too.
639 */
640 ptep = lookup_address(va, &level);
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800641 BUG_ON(ptep == NULL);
642
643 pfn = pte_pfn(*ptep);
644 mfn = pfn_to_mfn(pfn);
645 virt = __va(PFN_PHYS(pfn));
646
647 frames[f] = mfn;
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800648
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700649 make_lowmem_page_readonly((void *)va);
Jeremy Fitzhardinge6ed6bf422009-03-04 13:02:18 -0800650 make_lowmem_page_readonly(virt);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700651 }
652
Jeremy Fitzhardinge3ce5fa72009-03-04 15:26:00 -0800653 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
654 BUG();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700655}
656
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700657/*
658 * load_gdt for early boot, when the gdt is only mapped once
659 */
Daniel Kiperad3062a2011-05-04 20:15:18 +0200660static void __init xen_load_gdt_boot(const struct desc_ptr *dtr)
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700661{
662 unsigned long va = dtr->address;
663 unsigned int size = dtr->size + 1;
Amitoj Kaur Chawla585423c2016-06-29 20:30:38 +0530664 unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE);
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700665 unsigned long frames[pages];
666 int f;
667
668 /*
669 * A GDT can be up to 64k in size, which corresponds to 8192
670 * 8-byte entries, or 16 4k pages..
671 */
672
673 BUG_ON(size > 65536);
674 BUG_ON(va & ~PAGE_MASK);
675
676 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
677 pte_t pte;
678 unsigned long pfn, mfn;
679
680 pfn = virt_to_pfn(va);
681 mfn = pfn_to_mfn(pfn);
682
683 pte = pfn_pte(pfn, PAGE_KERNEL_RO);
684
685 if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0))
686 BUG();
687
688 frames[f] = mfn;
689 }
690
691 if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct)))
692 BUG();
693}
694
David Vrabel59290362012-07-09 11:39:07 +0100695static inline bool desc_equal(const struct desc_struct *d1,
696 const struct desc_struct *d2)
697{
698 return d1->a == d2->a && d1->b == d2->b;
699}
700
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700701static void load_TLS_descriptor(struct thread_struct *t,
702 unsigned int cpu, unsigned int i)
703{
David Vrabel1c32cdc2012-07-09 11:39:08 +0100704 struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i];
705 struct desc_struct *gdt;
706 xmaddr_t maddr;
707 struct multicall_space mc;
708
709 if (desc_equal(shadow, &t->tls_array[i]))
710 return;
711
712 *shadow = t->tls_array[i];
713
714 gdt = get_cpu_gdt_table(cpu);
715 maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
716 mc = __xen_mc_entry(0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700717
718 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
719}
720
721static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
722{
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700723 /*
Tejun Heoccbeed32009-02-09 22:17:40 +0900724 * XXX sleazy hack: If we're being called in a lazy-cpu zone
725 * and lazy gs handling is enabled, it means we're in a
726 * context switch, and %gs has just been saved. This means we
727 * can zero it out to prevent faults on exit from the
728 * hypervisor if the next process has no %gs. Either way, it
729 * has been saved, and the new value will get loaded properly.
730 * This will go away as soon as Xen has been modified to not
731 * save/restore %gs for normal hypercalls.
Eduardo Habkost8a954082008-07-08 15:07:10 -0700732 *
733 * On x86_64, this hack is not used for %gs, because gs points
734 * to KERNEL_GS_BASE (and uses it for PDA references), so we
735 * must not zero %gs on x86_64
736 *
737 * For x86_64, we need to zero %fs, otherwise we may get an
738 * exception between the new %fs descriptor being loaded and
739 * %fs being effectively cleared at __switch_to().
Jeremy Fitzhardinge8b84ad92007-07-17 18:37:06 -0700740 */
Eduardo Habkost8a954082008-07-08 15:07:10 -0700741 if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) {
742#ifdef CONFIG_X86_32
Tejun Heoccbeed32009-02-09 22:17:40 +0900743 lazy_load_gs(0);
Eduardo Habkost8a954082008-07-08 15:07:10 -0700744#else
745 loadsegment(fs, 0);
746#endif
747 }
748
749 xen_mc_batch();
750
751 load_TLS_descriptor(t, cpu, 0);
752 load_TLS_descriptor(t, cpu, 1);
753 load_TLS_descriptor(t, cpu, 2);
754
755 xen_mc_issue(PARAVIRT_LAZY_CPU);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700756}
757
Eduardo Habkosta8fc1082008-07-08 15:07:05 -0700758#ifdef CONFIG_X86_64
759static void xen_load_gs_index(unsigned int idx)
760{
761 if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx))
762 BUG();
763}
764#endif
765
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700766static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100767 const void *ptr)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700768{
Jeremy Fitzhardingecef43bf2008-07-28 13:33:44 -0700769 xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]);
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100770 u64 entry = *(u64 *)ptr;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700771
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800772 trace_xen_cpu_write_ldt_entry(dt, entrynum, entry);
773
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700774 preempt_disable();
775
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700776 xen_mc_flush();
777 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
778 BUG();
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700779
780 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700781}
782
Eduardo Habkoste176d362008-07-08 15:06:59 -0700783static int cvt_gate_to_trap(int vector, const gate_desc *val,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700784 struct trap_info *info)
785{
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700786 unsigned long addr;
787
Jeremy Fitzhardinge6d02c422009-03-29 22:57:15 -0700788 if (val->type != GATE_TRAP && val->type != GATE_INTERRUPT)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700789 return 0;
790
791 info->vector = vector;
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700792
793 addr = gate_offset(*val);
794#ifdef CONFIG_X86_64
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700795 /*
796 * Look for known traps using IST, and substitute them
797 * appropriately. The debugger ones are the only ones we care
Liu, Jinsong05e36002012-06-07 20:10:37 +0800798 * about. Xen will handle faults like double_fault,
799 * so we should never see them. Warn if
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700800 * there's an unexpected IST-using fault handler.
801 */
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700802 if (addr == (unsigned long)debug)
803 addr = (unsigned long)xen_debug;
804 else if (addr == (unsigned long)int3)
805 addr = (unsigned long)xen_int3;
806 else if (addr == (unsigned long)stack_segment)
807 addr = (unsigned long)xen_stack_segment;
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400808 else if (addr == (unsigned long)double_fault) {
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700809 /* Don't need to handle these */
810 return 0;
811#ifdef CONFIG_X86_MCE
812 } else if (addr == (unsigned long)machine_check) {
Liu, Jinsong05e36002012-06-07 20:10:37 +0800813 /*
814 * when xen hypervisor inject vMCE to guest,
815 * use native mce handler to handle it
816 */
817 ;
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700818#endif
Konrad Rzeszutek Wilk6efa20e2013-07-19 11:51:31 -0400819 } else if (addr == (unsigned long)nmi)
820 /*
821 * Use the native version as well.
822 */
823 ;
824 else {
Jeremy Fitzhardingeb80119b2009-04-24 00:22:08 -0700825 /* Some other trap using IST? */
826 if (WARN_ON(val->ist != 0))
827 return 0;
828 }
Jeremy Fitzhardinge6cac5a92009-03-29 19:56:29 -0700829#endif /* CONFIG_X86_64 */
830 info->address = addr;
831
Eduardo Habkoste176d362008-07-08 15:06:59 -0700832 info->cs = gate_segment(*val);
833 info->flags = val->dpl;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700834 /* interrupt gates clear IF */
Jeremy Fitzhardinge6d02c422009-03-29 22:57:15 -0700835 if (val->type == GATE_INTERRUPT)
836 info->flags |= 1 << 2;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700837
838 return 1;
839}
840
841/* Locations of each CPU's IDT */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100842static DEFINE_PER_CPU(struct desc_ptr, idt_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700843
844/* Set an IDT entry. If the entry is part of the current IDT, then
845 also update Xen. */
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100846static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700847{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700848 unsigned long p = (unsigned long)&dt[entrynum];
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700849 unsigned long start, end;
850
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800851 trace_xen_cpu_write_idt_entry(dt, entrynum, g);
852
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700853 preempt_disable();
854
Christoph Lameter780f36d2010-12-06 11:16:29 -0600855 start = __this_cpu_read(idt_desc.address);
856 end = start + __this_cpu_read(idt_desc.size) + 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700857
858 xen_mc_flush();
859
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100860 native_write_idt_entry(dt, entrynum, g);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700861
862 if (p >= start && (p + 8) <= end) {
863 struct trap_info info[2];
864
865 info[1].address = 0;
866
Eduardo Habkoste176d362008-07-08 15:06:59 -0700867 if (cvt_gate_to_trap(entrynum, g, &info[0]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700868 if (HYPERVISOR_set_trap_table(info))
869 BUG();
870 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700871
872 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700873}
874
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100875static void xen_convert_trap_info(const struct desc_ptr *desc,
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700876 struct trap_info *traps)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700877{
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700878 unsigned in, out, count;
879
Eduardo Habkoste176d362008-07-08 15:06:59 -0700880 count = (desc->size+1) / sizeof(gate_desc);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700881 BUG_ON(count > 256);
882
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700883 for (in = out = 0; in < count; in++) {
Eduardo Habkoste176d362008-07-08 15:06:59 -0700884 gate_desc *entry = (gate_desc*)(desc->address) + in;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700885
Eduardo Habkoste176d362008-07-08 15:06:59 -0700886 if (cvt_gate_to_trap(in, entry, &traps[out]))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700887 out++;
888 }
889 traps[out].address = 0;
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700890}
891
892void xen_copy_trap_info(struct trap_info *traps)
893{
Christoph Lameter89cbc762014-08-17 12:30:40 -0500894 const struct desc_ptr *desc = this_cpu_ptr(&idt_desc);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700895
896 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700897}
898
899/* Load a new IDT into Xen. In principle this can be per-CPU, so we
900 hold a spinlock to protect the static traps[] array (static because
901 it avoids allocation, and saves stack space). */
Glauber de Oliveira Costa6b68f012008-01-30 13:31:12 +0100902static void xen_load_idt(const struct desc_ptr *desc)
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700903{
904 static DEFINE_SPINLOCK(lock);
905 static struct trap_info traps[257];
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700906
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800907 trace_xen_cpu_load_idt(desc);
908
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700909 spin_lock(&lock);
910
Christoph Lameter89cbc762014-08-17 12:30:40 -0500911 memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc));
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700912
Jeremy Fitzhardingef87e4ca2007-07-17 18:37:06 -0700913 xen_convert_trap_info(desc, traps);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700914
915 xen_mc_flush();
916 if (HYPERVISOR_set_trap_table(traps))
917 BUG();
918
919 spin_unlock(&lock);
920}
921
922/* Write a GDT descriptor entry. Ignore LDT descriptors, since
923 they're handled differently. */
924static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100925 const void *desc, int type)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700926{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800927 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
928
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700929 preempt_disable();
930
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100931 switch (type) {
932 case DESC_LDT:
933 case DESC_TSS:
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700934 /* ignore */
935 break;
936
937 default: {
Jeremy Fitzhardinge9976b392009-02-27 09:19:26 -0800938 xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700939
940 xen_mc_flush();
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100941 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700942 BUG();
943 }
944
945 }
Jeremy Fitzhardingef120f132007-07-17 18:37:06 -0700946
947 preempt_enable();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700948}
949
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700950/*
951 * Version of write_gdt_entry for use at early boot-time needed to
952 * update an entry as simply as possible.
953 */
Daniel Kiperad3062a2011-05-04 20:15:18 +0200954static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry,
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700955 const void *desc, int type)
956{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800957 trace_xen_cpu_write_gdt_entry(dt, entry, desc, type);
958
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -0700959 switch (type) {
960 case DESC_LDT:
961 case DESC_TSS:
962 /* ignore */
963 break;
964
965 default: {
966 xmaddr_t maddr = virt_to_machine(&dt[entry]);
967
968 if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc))
969 dt[entry] = *(struct desc_struct *)desc;
970 }
971
972 }
973}
974
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100975static void xen_load_sp0(struct tss_struct *tss,
Jeremy Fitzhardingea05d2eb2008-07-27 08:45:02 -0700976 struct thread_struct *thread)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700977{
Jeremy Fitzhardingeab78f7a2010-12-17 15:54:28 -0800978 struct multicall_space mcs;
979
980 mcs = xen_mc_entry(0);
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100981 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700982 xen_mc_issue(PARAVIRT_LAZY_CPU);
Andy Lutomirski8ef46a62015-03-05 19:19:02 -0800983 tss->x86_tss.sp0 = thread->sp0;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700984}
985
Andy Lutomirskib7a584592016-03-16 14:14:21 -0700986void xen_set_iopl_mask(unsigned mask)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -0700987{
988 struct physdev_set_iopl set_iopl;
989
990 /* Force the change at ring 0. */
991 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
992 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
993}
994
995static void xen_io_delay(void)
996{
997}
998
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +0100999static void xen_clts(void)
1000{
1001 struct multicall_space mcs;
1002
1003 mcs = xen_mc_entry(0);
1004
1005 MULTI_fpu_taskswitch(mcs.mc, 0);
1006
1007 xen_mc_issue(PARAVIRT_LAZY_CPU);
1008}
1009
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001010static DEFINE_PER_CPU(unsigned long, xen_cr0_value);
1011
1012static unsigned long xen_read_cr0(void)
1013{
Alex Shi2113f462012-01-13 23:53:35 +08001014 unsigned long cr0 = this_cpu_read(xen_cr0_value);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001015
1016 if (unlikely(cr0 == 0)) {
1017 cr0 = native_read_cr0();
Alex Shi2113f462012-01-13 23:53:35 +08001018 this_cpu_write(xen_cr0_value, cr0);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001019 }
1020
1021 return cr0;
1022}
1023
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001024static void xen_write_cr0(unsigned long cr0)
1025{
1026 struct multicall_space mcs;
1027
Alex Shi2113f462012-01-13 23:53:35 +08001028 this_cpu_write(xen_cr0_value, cr0);
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001029
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001030 /* Only pay attention to cr0.TS; everything else is
1031 ignored. */
1032 mcs = xen_mc_entry(0);
1033
1034 MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0);
1035
1036 xen_mc_issue(PARAVIRT_LAZY_CPU);
1037}
1038
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001039static void xen_write_cr4(unsigned long cr4)
1040{
Boris Ostrovsky3375d822015-08-10 16:34:38 -04001041 cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE);
Jeremy Fitzhardinge2956a352008-05-26 23:31:04 +01001042
1043 native_write_cr4(cr4);
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001044}
Konrad Rzeszutek Wilk1a7bbda2012-10-10 13:25:48 -04001045#ifdef CONFIG_X86_64
1046static inline unsigned long xen_read_cr8(void)
1047{
1048 return 0;
1049}
1050static inline void xen_write_cr8(unsigned long val)
1051{
1052 BUG_ON(val);
1053}
1054#endif
Boris Ostrovsky31795b472015-02-11 14:39:18 -05001055
1056static u64 xen_read_msr_safe(unsigned int msr, int *err)
1057{
1058 u64 val;
1059
Boris Ostrovsky6b08cd62015-08-10 16:34:36 -04001060 if (pmu_msr_read(msr, &val, err))
1061 return val;
1062
Boris Ostrovsky31795b472015-02-11 14:39:18 -05001063 val = native_read_msr_safe(msr, err);
1064 switch (msr) {
1065 case MSR_IA32_APICBASE:
1066#ifdef CONFIG_X86_X2APIC
1067 if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31))))
1068#endif
1069 val &= ~X2APIC_ENABLE;
1070 break;
1071 }
1072 return val;
1073}
1074
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001075static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high)
1076{
1077 int ret;
1078
1079 ret = 0;
1080
Tejf63c2f22008-12-16 11:56:06 -08001081 switch (msr) {
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001082#ifdef CONFIG_X86_64
1083 unsigned which;
1084 u64 base;
1085
1086 case MSR_FS_BASE: which = SEGBASE_FS; goto set;
1087 case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set;
1088 case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set;
1089
1090 set:
1091 base = ((u64)high << 32) | low;
1092 if (HYPERVISOR_set_segment_base(which, base) != 0)
H. Peter Anvin0cc02132009-08-31 14:23:29 -07001093 ret = -EIO;
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001094 break;
1095#endif
Jeremy Fitzhardinged89961e2008-07-24 13:48:58 -07001096
1097 case MSR_STAR:
1098 case MSR_CSTAR:
1099 case MSR_LSTAR:
1100 case MSR_SYSCALL_MASK:
1101 case MSR_IA32_SYSENTER_CS:
1102 case MSR_IA32_SYSENTER_ESP:
1103 case MSR_IA32_SYSENTER_EIP:
1104 /* Fast syscall setup is all done in hypercalls, so
1105 these are all ignored. Stub them out here to stop
1106 Xen console noise. */
Boris Ostrovsky2ecf91b2015-09-21 09:09:28 -04001107 break;
Jeremy Fitzhardinge41f2e472010-03-30 11:47:40 -07001108
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001109 default:
Boris Ostrovsky6b08cd62015-08-10 16:34:36 -04001110 if (!pmu_msr_write(msr, low, high, &ret))
1111 ret = native_write_msr_safe(msr, low, high);
Jeremy Fitzhardinge11539682008-07-08 15:07:16 -07001112 }
1113
1114 return ret;
1115}
1116
Andy Lutomirskidd2f4a02016-04-02 07:01:38 -07001117static u64 xen_read_msr(unsigned int msr)
1118{
1119 /*
1120 * This will silently swallow a #GP from RDMSR. It may be worth
1121 * changing that.
1122 */
1123 int err;
1124
1125 return xen_read_msr_safe(msr, &err);
1126}
1127
1128static void xen_write_msr(unsigned int msr, unsigned low, unsigned high)
1129{
1130 /*
1131 * This will silently swallow a #GP from WRMSR. It may be worth
1132 * changing that.
1133 */
1134 xen_write_msr_safe(msr, low, high);
1135}
1136
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +01001137void xen_setup_shared_info(void)
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001138{
1139 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001140 set_fixmap(FIX_PARAVIRT_BOOTMAP,
1141 xen_start_info->shared_info);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001142
Jeremy Fitzhardinge15664f92008-07-08 15:06:47 -07001143 HYPERVISOR_shared_info =
1144 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001145 } else
1146 HYPERVISOR_shared_info =
1147 (struct shared_info *)__va(xen_start_info->shared_info);
1148
1149#ifndef CONFIG_SMP
1150 /* In UP this is as good a place as any to set up shared info */
1151 xen_setup_vcpu_info_placement();
1152#endif
Jeremy Fitzhardinged5edbc12008-05-26 23:31:22 +01001153
1154 xen_setup_mfn_list_list();
Jeremy Fitzhardinge2e8fe712008-03-17 16:36:53 -07001155}
1156
Rusty Russell5f054e32012-03-29 15:38:31 +10301157/* This is called once we have the cpu_possible_mask */
Jeremy Fitzhardinge0e913982008-05-26 23:31:27 +01001158void xen_setup_vcpu_info_placement(void)
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001159{
1160 int cpu;
1161
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +02001162 for_each_possible_cpu(cpu) {
1163 /* Set up direct vCPU id mapping for PV guests. */
1164 per_cpu(xen_vcpu_id, cpu) = cpu;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001165 xen_vcpu_setup(cpu);
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +02001166 }
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001167
1168 /* xen_vcpu_setup managed to place the vcpu_info within the
Mukesh Rathor27713742013-12-11 15:36:51 -05001169 * percpu area for all cpus, so make use of it. Note that for
1170 * PVH we want to use native IRQ mechanism. */
1171 if (have_vcpu_info_placement && !xen_pvh_domain()) {
Jeremy Fitzhardingeecb93d12009-01-28 14:35:05 -08001172 pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct);
1173 pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct);
1174 pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct);
1175 pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001176 pv_mmu_ops.read_cr2 = xen_read_cr2_direct;
Jeremy Fitzhardinge60223a32007-07-17 18:37:07 -07001177 }
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001178}
1179
Andi Kleenab144f52007-08-10 22:31:03 +02001180static unsigned xen_patch(u8 type, u16 clobbers, void *insnbuf,
1181 unsigned long addr, unsigned len)
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001182{
1183 char *start, *end, *reloc;
1184 unsigned ret;
1185
1186 start = end = reloc = NULL;
1187
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001188#define SITE(op, x) \
1189 case PARAVIRT_PATCH(op.x): \
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001190 if (have_vcpu_info_placement) { \
1191 start = (char *)xen_##x##_direct; \
1192 end = xen_##x##_direct_end; \
1193 reloc = xen_##x##_direct_reloc; \
1194 } \
1195 goto patch_site
1196
1197 switch (type) {
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001198 SITE(pv_irq_ops, irq_enable);
1199 SITE(pv_irq_ops, irq_disable);
1200 SITE(pv_irq_ops, save_fl);
1201 SITE(pv_irq_ops, restore_fl);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001202#undef SITE
1203
1204 patch_site:
1205 if (start == NULL || (end-start) > len)
1206 goto default_patch;
1207
Andi Kleenab144f52007-08-10 22:31:03 +02001208 ret = paravirt_patch_insns(insnbuf, len, start, end);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001209
1210 /* Note: because reloc is assigned from something that
1211 appears to be an array, gcc assumes it's non-null,
1212 but doesn't know its relationship with start and
1213 end. */
1214 if (reloc > start && reloc < end) {
1215 int reloc_off = reloc - start;
Andi Kleenab144f52007-08-10 22:31:03 +02001216 long *relocp = (long *)(insnbuf + reloc_off);
1217 long delta = start - (char *)addr;
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001218
1219 *relocp += delta;
1220 }
1221 break;
1222
1223 default_patch:
1224 default:
Andi Kleenab144f52007-08-10 22:31:03 +02001225 ret = paravirt_patch_default(type, clobbers, insnbuf,
1226 addr, len);
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001227 break;
1228 }
1229
1230 return ret;
1231}
1232
Daniel Kiperad3062a2011-05-04 20:15:18 +02001233static const struct pv_info xen_info __initconst = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001234 .shared_kernel_pmd = 0,
1235
Andy Lutomirski318f5a22011-08-03 09:31:53 -04001236#ifdef CONFIG_X86_64
1237 .extra_user_64bit_cs = FLAT_USER_CS64,
1238#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001239 .name = "Xen",
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001240};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001241
Daniel Kiperad3062a2011-05-04 20:15:18 +02001242static const struct pv_init_ops xen_init_ops __initconst = {
Jeremy Fitzhardinge64876732007-07-17 18:37:07 -07001243 .patch = xen_patch,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001244};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001245
Daniel Kiperad3062a2011-05-04 20:15:18 +02001246static const struct pv_cpu_ops xen_cpu_ops __initconst = {
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001247 .cpuid = xen_cpuid,
1248
1249 .set_debugreg = xen_set_debugreg,
1250 .get_debugreg = xen_get_debugreg,
1251
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001252 .clts = xen_clts,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001253
Jeremy Fitzhardingea789ed52009-04-24 00:26:50 -07001254 .read_cr0 = xen_read_cr0,
Jeremy Fitzhardinge7b1333a2008-05-26 23:31:01 +01001255 .write_cr0 = xen_write_cr0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001256
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001257 .read_cr4 = native_read_cr4,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001258 .write_cr4 = xen_write_cr4,
1259
Konrad Rzeszutek Wilk1a7bbda2012-10-10 13:25:48 -04001260#ifdef CONFIG_X86_64
1261 .read_cr8 = xen_read_cr8,
1262 .write_cr8 = xen_write_cr8,
1263#endif
1264
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001265 .wbinvd = native_wbinvd,
1266
Andy Lutomirskidd2f4a02016-04-02 07:01:38 -07001267 .read_msr = xen_read_msr,
1268 .write_msr = xen_write_msr,
1269
Andy Lutomirskic2ee03b2016-04-02 07:01:36 -07001270 .read_msr_safe = xen_read_msr_safe,
1271 .write_msr_safe = xen_write_msr_safe,
Konrad Rzeszutek Wilk1ab46fd2012-05-30 18:23:56 -04001272
Boris Ostrovsky65d0cf02015-08-10 16:34:34 -04001273 .read_pmc = xen_read_pmc,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001274
Jeremy Fitzhardinge81e103f2008-04-17 17:40:51 +02001275 .iret = xen_iret,
Jeremy Fitzhardinge6fcac6d2008-07-08 15:07:14 -07001276#ifdef CONFIG_X86_64
Jeremy Fitzhardinge6fcac6d2008-07-08 15:07:14 -07001277 .usergs_sysret64 = xen_sysret64,
1278#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001279
1280 .load_tr_desc = paravirt_nop,
1281 .set_ldt = xen_set_ldt,
1282 .load_gdt = xen_load_gdt,
1283 .load_idt = xen_load_idt,
1284 .load_tls = xen_load_tls,
Eduardo Habkosta8fc1082008-07-08 15:07:05 -07001285#ifdef CONFIG_X86_64
1286 .load_gs_index = xen_load_gs_index,
1287#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001288
Jeremy Fitzhardinge38ffbe62008-07-23 14:21:18 -07001289 .alloc_ldt = xen_alloc_ldt,
1290 .free_ldt = xen_free_ldt,
1291
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001292 .store_idt = native_store_idt,
1293 .store_tr = xen_store_tr,
1294
1295 .write_ldt_entry = xen_write_ldt_entry,
1296 .write_gdt_entry = xen_write_gdt_entry,
1297 .write_idt_entry = xen_write_idt_entry,
H. Peter Anvinfaca6222008-01-30 13:31:02 +01001298 .load_sp0 = xen_load_sp0,
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001299
1300 .set_iopl_mask = xen_set_iopl_mask,
1301 .io_delay = xen_io_delay,
1302
Jeremy Fitzhardinge952d1d72008-07-08 15:07:01 -07001303 /* Xen takes care of %gs when switching to usermode for us */
1304 .swapgs = paravirt_nop,
1305
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -08001306 .start_context_switch = paravirt_start_context_switch,
1307 .end_context_switch = xen_end_context_switch,
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001308};
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001309
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001310static void xen_reboot(int reason)
1311{
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001312 struct sched_shutdown r = { .reason = reason };
Boris Ostrovsky65d0cf02015-08-10 16:34:34 -04001313 int cpu;
1314
1315 for_each_online_cpu(cpu)
1316 xen_pmu_finish(cpu);
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001317
Jeremy Fitzhardinge349c7092008-05-26 23:31:02 +01001318 if (HYPERVISOR_sched_op(SCHEDOP_shutdown, &r))
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001319 BUG();
1320}
1321
1322static void xen_restart(char *msg)
1323{
1324 xen_reboot(SHUTDOWN_reboot);
1325}
1326
1327static void xen_emergency_restart(void)
1328{
1329 xen_reboot(SHUTDOWN_reboot);
1330}
1331
1332static void xen_machine_halt(void)
1333{
1334 xen_reboot(SHUTDOWN_poweroff);
1335}
1336
Tom Goetzb2abe502011-05-16 15:06:26 -04001337static void xen_machine_power_off(void)
1338{
1339 if (pm_power_off)
1340 pm_power_off();
1341 xen_reboot(SHUTDOWN_poweroff);
1342}
1343
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001344static void xen_crash_shutdown(struct pt_regs *regs)
1345{
1346 xen_reboot(SHUTDOWN_crash);
1347}
1348
Donald Dutilef09f6d12010-07-15 14:56:49 -04001349static int
1350xen_panic_event(struct notifier_block *this, unsigned long event, void *ptr)
1351{
Petr Tesarikc0253112016-08-02 14:06:19 -07001352 if (!kexec_crash_loaded())
1353 xen_reboot(SHUTDOWN_crash);
Donald Dutilef09f6d12010-07-15 14:56:49 -04001354 return NOTIFY_DONE;
1355}
1356
1357static struct notifier_block xen_panic_block = {
1358 .notifier_call= xen_panic_event,
Radim Krčmářbc5eb202014-05-13 18:56:25 +02001359 .priority = INT_MIN
Donald Dutilef09f6d12010-07-15 14:56:49 -04001360};
1361
1362int xen_panic_handler_init(void)
1363{
1364 atomic_notifier_chain_register(&panic_notifier_list, &xen_panic_block);
1365 return 0;
1366}
1367
Daniel Kiperad3062a2011-05-04 20:15:18 +02001368static const struct machine_ops xen_machine_ops __initconst = {
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001369 .restart = xen_restart,
1370 .halt = xen_machine_halt,
Tom Goetzb2abe502011-05-16 15:06:26 -04001371 .power_off = xen_machine_power_off,
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001372 .shutdown = xen_machine_halt,
1373 .crash_shutdown = xen_crash_shutdown,
1374 .emergency_restart = xen_emergency_restart,
1375};
1376
Jan Beulichf221b042015-01-13 07:40:05 +00001377static unsigned char xen_get_nmi_reason(void)
1378{
1379 unsigned char reason = 0;
1380
1381 /* Construct a value which looks like it came from port 0x61. */
1382 if (test_bit(_XEN_NMIREASON_io_error,
1383 &HYPERVISOR_shared_info->arch.nmi_reason))
1384 reason |= NMI_REASON_IOCHK;
1385 if (test_bit(_XEN_NMIREASON_pci_serr,
1386 &HYPERVISOR_shared_info->arch.nmi_reason))
1387 reason |= NMI_REASON_SERR;
1388
1389 return reason;
1390}
1391
David Vrabel96f28bc2013-04-03 17:31:50 +01001392static void __init xen_boot_params_init_edd(void)
1393{
1394#if IS_ENABLED(CONFIG_EDD)
1395 struct xen_platform_op op;
1396 struct edd_info *edd_info;
1397 u32 *mbr_signature;
1398 unsigned nr;
1399 int ret;
1400
1401 edd_info = boot_params.eddbuf;
1402 mbr_signature = boot_params.edd_mbr_sig_buffer;
1403
1404 op.cmd = XENPF_firmware_info;
1405
1406 op.u.firmware_info.type = XEN_FW_DISK_INFO;
1407 for (nr = 0; nr < EDDMAXNR; nr++) {
1408 struct edd_info *info = edd_info + nr;
1409
1410 op.u.firmware_info.index = nr;
1411 info->params.length = sizeof(info->params);
1412 set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params,
1413 &info->params);
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001414 ret = HYPERVISOR_platform_op(&op);
David Vrabel96f28bc2013-04-03 17:31:50 +01001415 if (ret)
1416 break;
1417
1418#define C(x) info->x = op.u.firmware_info.u.disk_info.x
1419 C(device);
1420 C(version);
1421 C(interface_support);
1422 C(legacy_max_cylinder);
1423 C(legacy_max_head);
1424 C(legacy_sectors_per_track);
1425#undef C
1426 }
1427 boot_params.eddbuf_entries = nr;
1428
1429 op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE;
1430 for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) {
1431 op.u.firmware_info.index = nr;
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001432 ret = HYPERVISOR_platform_op(&op);
David Vrabel96f28bc2013-04-03 17:31:50 +01001433 if (ret)
1434 break;
1435 mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature;
1436 }
1437 boot_params.edd_mbr_sig_buf_entries = nr;
1438#endif
1439}
1440
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001441/*
1442 * Set up the GDT and segment registers for -fstack-protector. Until
1443 * we do this, we have to be careful not to call any stack-protected
1444 * function, which is most of the kernel.
Mukesh Rathor5840c842013-12-13 11:48:08 -05001445 *
1446 * Note, that it is __ref because the only caller of this after init
1447 * is PVH which is not going to use xen_load_gdt_boot or other
1448 * __init functions.
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001449 */
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001450static void __ref xen_setup_gdt(int cpu)
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001451{
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001452 if (xen_feature(XENFEAT_auto_translated_physmap)) {
1453#ifdef CONFIG_X86_64
1454 unsigned long dummy;
1455
Mukesh Rathor5840c842013-12-13 11:48:08 -05001456 load_percpu_segment(cpu); /* We need to access per-cpu area */
1457 switch_to_new_gdt(cpu); /* GDT and GS set */
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001458
1459 /* We are switching of the Xen provided GDT to our HVM mode
1460 * GDT. The new GDT has __KERNEL_CS with CS.L = 1
1461 * and we are jumping to reload it.
1462 */
Greg Kroah-Hartman10d15c42018-06-03 13:37:03 +02001463 asm volatile (UNWIND_HINT_SAVE
1464 "pushq %0\n"
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001465 "leaq 1f(%%rip),%0\n"
1466 "pushq %0\n"
1467 "lretq\n"
Greg Kroah-Hartman10d15c42018-06-03 13:37:03 +02001468 UNWIND_HINT_RESTORE
Mukesh Rathor8d656bb2013-12-13 13:03:37 -05001469 "1:\n"
1470 : "=&r" (dummy) : "0" (__KERNEL_CS));
1471
1472 /*
1473 * While not needed, we also set the %es, %ds, and %fs
1474 * to zero. We don't care about %ss as it is NULL.
1475 * Strictly speaking this is not needed as Xen zeros those
1476 * out (and also MSR_FS_BASE, MSR_GS_BASE, MSR_KERNEL_GS_BASE)
1477 *
1478 * Linux zeros them in cpu_init() and in secondary_startup_64
1479 * (for BSP).
1480 */
1481 loadsegment(es, 0);
1482 loadsegment(ds, 0);
1483 loadsegment(fs, 0);
1484#else
1485 /* PVH: TODO Implement. */
1486 BUG();
1487#endif
1488 return; /* PVH does not need any PV GDT ops. */
1489 }
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001490 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot;
1491 pv_cpu_ops.load_gdt = xen_load_gdt_boot;
1492
1493 setup_stack_canary_segment(0);
1494 switch_to_new_gdt(0);
1495
1496 pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry;
1497 pv_cpu_ops.load_gdt = xen_load_gdt;
1498}
1499
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001500#ifdef CONFIG_XEN_PVH
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001501/*
1502 * A PV guest starts with default flags that are not set for PVH, set them
1503 * here asap.
1504 */
1505static void xen_pvh_set_cr_flags(int cpu)
1506{
1507
1508 /* Some of these are setup in 'secondary_startup_64'. The others:
1509 * X86_CR0_TS, X86_CR0_PE, X86_CR0_ET are set by Xen for HVM guests
1510 * (which PVH shared codepaths), while X86_CR0_PG is for PVH. */
1511 write_cr0(read_cr0() | X86_CR0_MP | X86_CR0_NE | X86_CR0_WP | X86_CR0_AM);
Mukesh Rathorafca5012014-01-29 16:15:18 -08001512
1513 if (!cpu)
1514 return;
1515 /*
1516 * For BSP, PSE PGE are set in probe_page_size_mask(), for APs
Ingo Molnar21c4cd12015-04-26 14:27:17 +02001517 * set them here. For all, OSFXSR OSXMMEXCPT are set in fpu__init_cpu().
Mukesh Rathorafca5012014-01-29 16:15:18 -08001518 */
Borislav Petkov16bf9222016-03-29 17:42:03 +02001519 if (boot_cpu_has(X86_FEATURE_PSE))
Andy Lutomirski375074c2014-10-24 15:58:07 -07001520 cr4_set_bits_and_update_boot(X86_CR4_PSE);
Mukesh Rathorafca5012014-01-29 16:15:18 -08001521
Borislav Petkovc109bf92016-03-29 17:42:02 +02001522 if (boot_cpu_has(X86_FEATURE_PGE))
Andy Lutomirski375074c2014-10-24 15:58:07 -07001523 cr4_set_bits_and_update_boot(X86_CR4_PGE);
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001524}
1525
1526/*
1527 * Note, that it is ref - because the only caller of this after init
1528 * is PVH which is not going to use xen_load_gdt_boot or other
1529 * __init functions.
1530 */
1531void __ref xen_pvh_secondary_vcpu_init(int cpu)
1532{
1533 xen_setup_gdt(cpu);
1534 xen_pvh_set_cr_flags(cpu);
1535}
1536
Mukesh Rathord285d682013-12-13 12:45:31 -05001537static void __init xen_pvh_early_guest_init(void)
1538{
1539 if (!xen_feature(XENFEAT_auto_translated_physmap))
1540 return;
1541
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -04001542 if (!xen_feature(XENFEAT_hvm_callback_vector))
1543 return;
1544
1545 xen_have_vector_callback = 1;
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001546
1547 xen_pvh_early_cpu_init(0, false);
Roger Pau Monnec9f6e992014-01-20 09:20:07 -05001548 xen_pvh_set_cr_flags(0);
Mukesh Rathord285d682013-12-13 12:45:31 -05001549
1550#ifdef CONFIG_X86_32
1551 BUG(); /* PVH: Implement proper support. */
1552#endif
1553}
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001554#endif /* CONFIG_XEN_PVH */
Mukesh Rathord285d682013-12-13 12:45:31 -05001555
Luis R. Rodriguez8d152e72016-04-13 17:04:34 -07001556static void __init xen_dom0_set_legacy_features(void)
1557{
1558 x86_platform.legacy.rtc = 1;
1559}
1560
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001561static int xen_cpuhp_setup(void)
1562{
1563 int rc;
1564
1565 rc = cpuhp_setup_state_nocalls(CPUHP_XEN_PREPARE,
1566 "XEN_HVM_GUEST_PREPARE",
1567 xen_cpu_up_prepare, xen_cpu_dead);
1568 if (rc >= 0) {
1569 rc = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
1570 "XEN_HVM_GUEST_ONLINE",
1571 xen_cpu_up_online, NULL);
1572 if (rc < 0)
1573 cpuhp_remove_state_nocalls(CPUHP_XEN_PREPARE);
1574 }
1575
1576 return rc >= 0 ? 0 : rc;
1577}
1578
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001579/* First C function to be called on Xen boot */
Andi Kleen2605fc22014-05-02 00:44:37 +02001580asmlinkage __visible void __init xen_start_kernel(void)
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001581{
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001582 struct physdev_set_iopl set_iopl;
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001583 unsigned long initrd_start = 0;
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001584 int rc;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001585
1586 if (!xen_start_info)
1587 return;
1588
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07001589 xen_domain_type = XEN_PV_DOMAIN;
1590
Mukesh Rathord285d682013-12-13 12:45:31 -05001591 xen_setup_features();
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001592#ifdef CONFIG_XEN_PVH
Mukesh Rathord285d682013-12-13 12:45:31 -05001593 xen_pvh_early_guest_init();
Mukesh Rathora2ef5dc2014-09-10 16:36:06 -07001594#endif
Ian Campbell7e775062010-09-30 12:37:26 +01001595 xen_setup_machphys_mapping();
1596
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001597 /* Install Xen paravirt ops */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001598 pv_info = xen_info;
1599 pv_init_ops = xen_init_ops;
Jan Beulichf221b042015-01-13 07:40:05 +00001600 if (!xen_pvh_domain()) {
Mukesh Rathord285d682013-12-13 12:45:31 -05001601 pv_cpu_ops = xen_cpu_ops;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001602
Jan Beulichf221b042015-01-13 07:40:05 +00001603 x86_platform.get_nmi_reason = xen_get_nmi_reason;
1604 }
1605
David Vrabelabacaad2014-06-02 17:58:01 +01001606 if (xen_feature(XENFEAT_auto_translated_physmap))
1607 x86_init.resources.memory_setup = xen_auto_xlated_memory_setup;
1608 else
1609 x86_init.resources.memory_setup = xen_memory_setup;
Thomas Gleixner42bbdb42009-08-20 13:04:10 +02001610 x86_init.oem.arch_setup = xen_arch_setup;
Thomas Gleixner6f30c1a2009-08-20 13:19:57 +02001611 x86_init.oem.banner = xen_banner;
Thomas Gleixner845b3942009-08-19 15:37:03 +02001612
Stefano Stabellini409771d2010-05-14 12:48:19 +01001613 xen_init_time_ops();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001614
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001615 /*
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001616 * Set up some pagetable state before starting to set any ptes.
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001617 */
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001618
Jeremy Fitzhardinge973df352009-10-27 16:54:19 -07001619 xen_init_mmu_ops();
1620
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001621 /* Prevent unwanted bits from being set in PTEs. */
1622 __supported_pte_mask &= ~_PAGE_GLOBAL;
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001623
Ian Campbell817a8242010-02-26 17:16:00 +00001624 /*
1625 * Prevent page tables from being allocated in highmem, even
1626 * if CONFIG_HIGHPTE is enabled.
1627 */
1628 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
1629
Jeremy Fitzhardingeb75fe4e2009-09-21 13:34:06 -07001630 /* Work out if we support NX */
H. Peter Anvin4763ed42009-11-13 15:28:16 -08001631 x86_configure_nx();
Jeremy Fitzhardingeb75fe4e2009-09-21 13:34:06 -07001632
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001633 /* Get mfn list */
Konrad Rzeszutek Wilk696fd7c2013-12-15 12:37:46 -05001634 xen_build_dynamic_phys_to_machine();
Jeremy Fitzhardinge577eebe2009-08-27 12:46:35 -07001635
1636 /*
1637 * Set up kernel GDT and segment registers, mainly so that
1638 * -fstack-protector code can be executed.
1639 */
Mukesh Rathor5840c842013-12-13 11:48:08 -05001640 xen_setup_gdt(0);
Jeremy Fitzhardinge0d1edf42008-07-28 11:53:57 -07001641
Jeremy Fitzhardingece2eef32009-08-17 12:26:53 -07001642 xen_init_irq_ops();
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -08001643 xen_init_cpuid_mask();
Juergen Grossbb70de12018-05-30 13:09:57 +02001644 xen_init_capabilities();
Jeremy Fitzhardingee826fe12009-03-07 17:09:27 -08001645
Yinghai Lu94a8c3c2008-07-13 22:19:35 -07001646#ifdef CONFIG_X86_LOCAL_APIC
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001647 /*
Yinghai Lu94a8c3c2008-07-13 22:19:35 -07001648 * set up the basic apic ops.
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001649 */
Konrad Rzeszutek Wilkfeb44f12015-03-02 12:06:23 -05001650 xen_init_apic();
Suresh Siddhaad66dd32008-07-11 13:11:56 -07001651#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001652
Jeremy Fitzhardingee57778a2008-06-16 04:30:02 -07001653 if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) {
1654 pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start;
1655 pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit;
1656 }
1657
Jeremy Fitzhardingefefa6292007-07-17 18:37:07 -07001658 machine_ops = xen_machine_ops;
1659
Jeremy Fitzhardinge38341432009-02-02 13:55:31 -08001660 /*
1661 * The only reliable way to retain the initial address of the
1662 * percpu gdt_page is to remember it here, so we can go and
1663 * mark it RW later, when the initial percpu area is freed.
1664 */
1665 xen_initial_gdt = &per_cpu(gdt_page, 0);
Jeremy Fitzhardinge795f99b2009-01-30 17:47:54 +09001666
Jeremy Fitzhardingea9e70622008-07-08 15:06:41 -07001667 xen_smp_init();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001668
Ian Campbellc1f5db12010-12-03 09:54:03 +00001669#ifdef CONFIG_ACPI_NUMA
1670 /*
1671 * The pages we from Xen are not related to machine pages, so
1672 * any NUMA information the kernel tries to get from ACPI will
1673 * be meaningless. Prevent it from trying.
1674 */
1675 acpi_numa = -1;
1676#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001677 /* Don't do the full vcpu_info placement stuff until we have a
1678 possible map and a non-dummy shared_info. */
1679 per_cpu(xen_vcpu, 0) = &HYPERVISOR_shared_info->vcpu_info[0];
1680
Boris Ostrovsky565fdc62016-10-02 18:45:19 -04001681 WARN_ON(xen_cpuhp_setup());
1682
Jeremy Fitzhardinge55d80852009-02-25 09:42:25 -08001683 local_irq_disable();
Tejun Heo2ce802f2011-01-20 12:06:35 +01001684 early_boot_irqs_disabled = true;
Jeremy Fitzhardinge55d80852009-02-25 09:42:25 -08001685
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001686 xen_raw_console_write("mapping kernel into physical memory\n");
Juergen Gross6c2681c2015-07-17 06:51:34 +02001687 xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base,
1688 xen_start_info->nr_pages);
1689 xen_reserve_special_pages();
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001690
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001691 /* keep using Xen gdt for now; no urgent need to change it */
1692
Ian Campbelle68266b2010-01-13 10:16:08 +00001693#ifdef CONFIG_X86_32
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001694 pv_info.kernel_rpl = 1;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001695 if (xen_feature(XENFEAT_supervisor_mode_kernel))
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -07001696 pv_info.kernel_rpl = 0;
Ian Campbelle68266b2010-01-13 10:16:08 +00001697#else
1698 pv_info.kernel_rpl = 0;
1699#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001700 /* set the limit of our address space */
Jeremy Fitzhardingefb1d8402007-10-16 11:51:31 -07001701 xen_reserve_top();
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001702
Mukesh Rathord285d682013-12-13 12:45:31 -05001703 /* PVH: runs at default kernel iopl of 0 */
1704 if (!xen_pvh_domain()) {
1705 /*
1706 * We used to do this in xen_arch_setup, but that is too late
1707 * on AMD were early_cpu_init (run before ->arch_setup()) calls
1708 * early_amd_init which pokes 0xcf8 port.
1709 */
1710 set_iopl.iopl = 1;
1711 rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
1712 if (rc != 0)
1713 xen_raw_printk("physdev_op failed %d\n", rc);
1714 }
Konrad Rzeszutek Wilkec35a692010-11-16 12:09:59 -05001715
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001716#ifdef CONFIG_X86_32
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001717 /* set up basic CPUID stuff */
1718 cpu_detect(&new_cpu_data);
H. Peter Anvin60e019e2013-04-29 16:04:20 +02001719 set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU);
Jeremy Fitzhardinged560bc62009-08-25 12:53:02 -07001720 new_cpu_data.wp_works_ok = 1;
Huaitong Han16aaa532016-01-25 20:41:47 +01001721 new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1);
Jeremy Fitzhardinge7d087b62008-07-08 15:06:48 -07001722#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001723
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001724 if (xen_start_info->mod_start) {
1725 if (xen_start_info->flags & SIF_MOD_START_PFN)
1726 initrd_start = PFN_PHYS(xen_start_info->mod_start);
1727 else
1728 initrd_start = __pa(xen_start_info->mod_start);
1729 }
1730
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001731 /* Poke various useful things into boot_params */
H. Peter Anvin30c82642007-10-15 17:13:22 -07001732 boot_params.hdr.type_of_loader = (9 << 4) | 0;
Juergen Grossd1e9abd2014-09-17 06:12:36 +02001733 boot_params.hdr.ramdisk_image = initrd_start;
H. Peter Anvin30c82642007-10-15 17:13:22 -07001734 boot_params.hdr.ramdisk_size = xen_start_info->mod_len;
Jeremy Fitzhardingeb7c3c5c2008-07-08 15:07:07 -07001735 boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line);
Luis R. Rodriguezea179482016-04-13 17:04:32 -07001736 boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN;
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001737
Jeremy Fitzhardinge6e833582008-08-19 13:16:17 -07001738 if (!xen_initial_domain()) {
Jeremy Fitzhardinge83abc702008-05-26 23:31:12 +01001739 add_preferred_console("xenboot", 0, NULL);
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001740 add_preferred_console("tty", 0, NULL);
Markus Armbrusterb8c2d3d2008-02-27 14:56:35 +01001741 add_preferred_console("hvc", 0, NULL);
Alex Nixonb5401a92010-03-18 16:31:34 -04001742 if (pci_xen)
1743 x86_init.pci.arch_init = pci_xen_init;
Chris Wright5d990b62009-12-04 12:15:21 -08001744 } else {
Jeremy Fitzhardingec2419b42011-05-31 10:50:10 -04001745 const struct dom0_vga_console_info *info =
1746 (void *)((char *)xen_start_info +
1747 xen_start_info->console.dom0.info_off);
Konrad Rzeszutek Wilkffb8b232012-09-21 12:30:35 -04001748 struct xen_platform_op op = {
1749 .cmd = XENPF_firmware_info,
1750 .interface_version = XENPF_INTERFACE_VERSION,
1751 .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS,
1752 };
Jeremy Fitzhardingec2419b42011-05-31 10:50:10 -04001753
Luis R. Rodriguez8d152e72016-04-13 17:04:34 -07001754 x86_platform.set_legacy_features =
1755 xen_dom0_set_legacy_features;
Jeremy Fitzhardingec2419b42011-05-31 10:50:10 -04001756 xen_init_vga(info, xen_start_info->console.dom0.info_size);
1757 xen_start_info->console.domU.mfn = 0;
1758 xen_start_info->console.domU.evtchn = 0;
1759
Stefano Stabellinicfafae92015-11-23 10:36:12 +00001760 if (HYPERVISOR_platform_op(&op) == 0)
Konrad Rzeszutek Wilkffb8b232012-09-21 12:30:35 -04001761 boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags;
1762
Chris Wright5d990b62009-12-04 12:15:21 -08001763 /* Make sure ACS will be enabled */
1764 pci_request_acs();
Konrad Rzeszutek Wilk211063d2011-12-08 17:32:23 +08001765
1766 xen_acpi_sleep_register();
Konrad Rzeszutek Wilkbd499402012-09-19 08:30:55 -04001767
1768 /* Avoid searching for BIOS MP tables */
1769 x86_init.mpparse.find_smp_config = x86_init_noop;
1770 x86_init.mpparse.get_smp_config = x86_init_uint_noop;
David Vrabel96f28bc2013-04-03 17:31:50 +01001771
1772 xen_boot_params_init_edd();
Markus Armbruster9e124fe2008-05-26 23:31:07 +01001773 }
David Vrabel76a8df72012-05-04 14:29:46 +01001774#ifdef CONFIG_PCI
1775 /* PCI BIOS service won't work from a PV guest. */
1776 pci_probe &= ~PCI_PROBE_BIOS;
1777#endif
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001778 xen_raw_console_write("about to get started...\n");
1779
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +02001780 /* Let's presume PV guests always boot on vCPU with id 0. */
1781 per_cpu(xen_vcpu_id, 0) = 0;
1782
Jeremy Fitzhardinge499d19b2009-11-24 09:38:25 -08001783 xen_setup_runstate_info(0);
1784
Daniel Kiperc7341d62014-07-12 23:09:48 +02001785 xen_efi_init();
Daniel Kiperbe81c8a2014-06-30 19:53:02 +02001786
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001787 /* Start the world */
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001788#ifdef CONFIG_X86_32
Yinghai Luf0d43102008-05-29 12:56:36 -07001789 i386_start_kernel();
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001790#else
Boris Ostrovsky5054daa2015-02-23 11:01:00 -05001791 cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */
Jeremy Fitzhardinge084a2a42008-07-08 15:06:50 -07001792 x86_64_start_reservations((char *)__pa_symbol(&boot_params));
Jeremy Fitzhardingef5d36de2008-07-08 15:06:39 -07001793#endif
Jeremy Fitzhardinge5ead97c2007-07-17 18:37:04 -07001794}
Sheng Yangbee6ab52010-05-14 12:39:33 +01001795
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001796void __ref xen_hvm_init_shared_info(void)
Olaf Hering9d02b432012-11-01 22:02:30 +01001797{
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001798 int cpu;
Olaf Hering9d02b432012-11-01 22:02:30 +01001799 struct xen_add_to_physmap xatp;
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001800 static struct shared_info *shared_info_page = 0;
Olaf Hering9d02b432012-11-01 22:02:30 +01001801
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001802 if (!shared_info_page)
1803 shared_info_page = (struct shared_info *)
1804 extend_brk(PAGE_SIZE, PAGE_SIZE);
Sheng Yangbee6ab52010-05-14 12:39:33 +01001805 xatp.domid = DOMID_SELF;
1806 xatp.idx = 0;
1807 xatp.space = XENMAPSPACE_shared_info;
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001808 xatp.gpfn = __pa(shared_info_page) >> PAGE_SHIFT;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001809 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
1810 BUG();
1811
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001812 HYPERVISOR_shared_info = (struct shared_info *)shared_info_page;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001813
Stefano Stabellini016b6f52010-05-14 12:45:07 +01001814 /* xen_vcpu is a pointer to the vcpu_info struct in the shared_info
1815 * page, we use it in the event channel upcall and in some pvclock
1816 * related functions. We don't need the vcpu_info placement
1817 * optimizations because we don't use any pv_mmu or pv_irq op on
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001818 * HVM.
1819 * When xen_hvm_init_shared_info is run at boot time only vcpu 0 is
1820 * online but xen_hvm_init_shared_info is run at resume time too and
1821 * in that case multiple vcpus might be online. */
1822 for_each_online_cpu(cpu) {
Konrad Rzeszutek Wilkd5b17db2013-05-05 09:24:07 -04001823 /* Leave it to be NULL. */
Vitaly Kuznetsove15a8622016-06-30 17:56:39 +02001824 if (xen_vcpu_nr(cpu) >= MAX_VIRT_CPUS)
Konrad Rzeszutek Wilkd5b17db2013-05-05 09:24:07 -04001825 continue;
Vitaly Kuznetsove15a8622016-06-30 17:56:39 +02001826 per_cpu(xen_vcpu, cpu) =
1827 &HYPERVISOR_shared_info->vcpu_info[xen_vcpu_nr(cpu)];
Olaf Hering9d02b432012-11-01 22:02:30 +01001828 }
Olaf Hering9d02b432012-11-01 22:02:30 +01001829}
1830
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001831#ifdef CONFIG_XEN_PVHVM
Olaf Hering4ff2d062012-07-17 11:59:15 +02001832static void __init init_hvm_pv_info(void)
1833{
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001834 int major, minor;
Konrad Rzeszutek Wilk5eb65be2013-02-14 21:29:27 -05001835 uint32_t eax, ebx, ecx, edx, pages, msr, base;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001836 u64 pfn;
1837
1838 base = xen_cpuid_base();
Konrad Rzeszutek Wilke9daff22013-02-14 21:29:31 -05001839 cpuid(base + 1, &eax, &ebx, &ecx, &edx);
1840
1841 major = eax >> 16;
1842 minor = eax & 0xffff;
1843 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
1844
Sheng Yangbee6ab52010-05-14 12:39:33 +01001845 cpuid(base + 2, &pages, &msr, &ecx, &edx);
1846
1847 pfn = __pa(hypercall_page);
1848 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
1849
1850 xen_setup_features();
1851
Vitaly Kuznetsov88e957d2016-06-30 17:56:37 +02001852 cpuid(base + 4, &eax, &ebx, &ecx, &edx);
1853 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
1854 this_cpu_write(xen_vcpu_id, ebx);
1855 else
1856 this_cpu_write(xen_vcpu_id, smp_processor_id());
1857
Sheng Yangbee6ab52010-05-14 12:39:33 +01001858 pv_info.name = "Xen HVM";
1859
1860 xen_domain_type = XEN_HVM_DOMAIN;
Sheng Yangbee6ab52010-05-14 12:39:33 +01001861}
Arnd Bergmanncb5f7e72016-10-12 17:20:38 +02001862#endif
Sheng Yangbee6ab52010-05-14 12:39:33 +01001863
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001864static int xen_cpu_up_prepare(unsigned int cpu)
Sheng Yang38e20b02010-05-14 12:40:51 +01001865{
Boris Ostrovsky5fc509b2016-08-03 13:22:27 -04001866 int rc;
1867
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001868 if (xen_hvm_domain()) {
1869 /*
1870 * This can happen if CPU was offlined earlier and
1871 * offlining timed out in common_cpu_die().
1872 */
1873 if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
1874 xen_smp_intr_free(cpu);
1875 xen_uninit_lock_cpu(cpu);
Konrad Rzeszutek Wilk7918c922013-04-16 15:18:00 -04001876 }
Boris Ostrovsky5fc509b2016-08-03 13:22:27 -04001877
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +02001878 if (cpu_acpi_id(cpu) != U32_MAX)
1879 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
1880 else
1881 per_cpu(xen_vcpu_id, cpu) = cpu;
1882 xen_vcpu_setup(cpu);
Sheng Yang38e20b02010-05-14 12:40:51 +01001883 }
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001884
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -04001885 if (xen_pv_domain() ||
1886 (xen_have_vector_callback &&
1887 xen_feature(XENFEAT_hvm_safe_pvclock)))
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001888 xen_setup_timer(cpu);
1889
1890 rc = xen_smp_intr_init(cpu);
1891 if (rc) {
1892 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
1893 cpu, rc);
1894 return rc;
1895 }
1896 return 0;
Sheng Yang38e20b02010-05-14 12:40:51 +01001897}
1898
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001899static int xen_cpu_dead(unsigned int cpu)
1900{
1901 xen_smp_intr_free(cpu);
1902
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -04001903 if (xen_pv_domain() ||
1904 (xen_have_vector_callback &&
1905 xen_feature(XENFEAT_hvm_safe_pvclock)))
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001906 xen_teardown_timer(cpu);
1907
1908 return 0;
1909}
1910
1911static int xen_cpu_up_online(unsigned int cpu)
1912{
1913 xen_init_lock_cpu(cpu);
1914 return 0;
1915}
Sheng Yang38e20b02010-05-14 12:40:51 +01001916
Arnd Bergmanncb5f7e72016-10-12 17:20:38 +02001917#ifdef CONFIG_XEN_PVHVM
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +02001918#ifdef CONFIG_KEXEC_CORE
1919static void xen_hvm_shutdown(void)
1920{
1921 native_machine_shutdown();
1922 if (kexec_in_progress)
1923 xen_reboot(SHUTDOWN_soft_reset);
1924}
1925
1926static void xen_hvm_crash_shutdown(struct pt_regs *regs)
1927{
1928 native_machine_crash_shutdown(regs);
1929 xen_reboot(SHUTDOWN_soft_reset);
1930}
1931#endif
1932
Sheng Yangbee6ab52010-05-14 12:39:33 +01001933static void __init xen_hvm_guest_init(void)
1934{
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001935 if (xen_pv_domain())
1936 return;
1937
Olaf Hering4ff2d062012-07-17 11:59:15 +02001938 init_hvm_pv_info();
Sheng Yangbee6ab52010-05-14 12:39:33 +01001939
Stefano Stabellini016b6f52010-05-14 12:45:07 +01001940 xen_hvm_init_shared_info();
Sheng Yang38e20b02010-05-14 12:40:51 +01001941
Vaughan Cao669b0ae2013-08-16 16:10:56 +08001942 xen_panic_handler_init();
1943
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -04001944 if (xen_feature(XENFEAT_hvm_callback_vector))
1945 xen_have_vector_callback = 1;
Stefano Stabellini99bbb3a2010-12-02 17:55:10 +00001946 xen_hvm_smp_init();
Boris Ostrovsky4d737042016-09-07 13:19:00 -04001947 WARN_ON(xen_cpuhp_setup());
Stefano Stabellinic1c54132010-05-14 12:44:30 +01001948 xen_unplug_emulated_devices();
Sheng Yang38e20b02010-05-14 12:40:51 +01001949 x86_init.irqs.intr_init = xen_init_IRQ;
Stefano Stabellini409771d2010-05-14 12:48:19 +01001950 xen_hvm_init_time_ops();
Stefano Stabellini59151002010-06-17 14:22:52 +01001951 xen_hvm_init_mmu_ops();
Vitaly Kuznetsov0b34a162015-09-25 11:59:52 +02001952#ifdef CONFIG_KEXEC_CORE
1953 machine_ops.shutdown = xen_hvm_shutdown;
1954 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
1955#endif
Sheng Yangbee6ab52010-05-14 12:39:33 +01001956}
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001957#endif
Sheng Yangbee6ab52010-05-14 12:39:33 +01001958
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001959static bool xen_nopv = false;
1960static __init int xen_parse_nopv(char *arg)
1961{
1962 xen_nopv = true;
1963 return 0;
1964}
1965early_param("xen_nopv", xen_parse_nopv);
1966
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04001967static uint32_t __init xen_platform(void)
Sheng Yangbee6ab52010-05-14 12:39:33 +01001968{
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001969 if (xen_nopv)
1970 return 0;
1971
Jason Wang9df56f12013-07-25 16:54:35 +08001972 return xen_cpuid_base();
Sheng Yangbee6ab52010-05-14 12:39:33 +01001973}
1974
Sheng Yangd9b8ca82010-12-21 14:18:49 +08001975bool xen_hvm_need_lapic(void)
1976{
Konrad Rzeszutek Wilk8d693b92014-07-11 11:51:35 -04001977 if (xen_nopv)
1978 return false;
Sheng Yangd9b8ca82010-12-21 14:18:49 +08001979 if (xen_pv_domain())
1980 return false;
1981 if (!xen_hvm_domain())
1982 return false;
Boris Ostrovsky5d7ab832017-04-24 15:04:53 -04001983 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
Sheng Yangd9b8ca82010-12-21 14:18:49 +08001984 return false;
1985 return true;
1986}
1987EXPORT_SYMBOL_GPL(xen_hvm_need_lapic);
1988
Juergen Gross99bc6752016-08-29 08:48:45 +02001989static void xen_pin_vcpu(int cpu)
1990{
1991 static bool disable_pinning;
1992 struct sched_pin_override pin_override;
1993 int ret;
1994
1995 if (disable_pinning)
1996 return;
1997
1998 pin_override.pcpu = cpu;
1999 ret = HYPERVISOR_sched_op(SCHEDOP_pin_override, &pin_override);
2000
2001 /* Ignore errors when removing override. */
2002 if (cpu < 0)
2003 return;
2004
2005 switch (ret) {
2006 case -ENOSYS:
2007 pr_warn("Unable to pin on physical cpu %d. In case of problems consider vcpu pinning.\n",
2008 cpu);
2009 disable_pinning = true;
2010 break;
2011 case -EPERM:
2012 WARN(1, "Trying to pin vcpu without having privilege to do so\n");
2013 disable_pinning = true;
2014 break;
2015 case -EINVAL:
2016 case -EBUSY:
2017 pr_warn("Physical cpu %d not available for pinning. Check Xen cpu configuration.\n",
2018 cpu);
2019 break;
2020 case 0:
2021 break;
2022 default:
2023 WARN(1, "rc %d while trying to pin vcpu\n", ret);
2024 disable_pinning = true;
2025 }
2026}
2027
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04002028const struct hypervisor_x86 x86_hyper_xen = {
2029 .name = "Xen",
2030 .detect = xen_platform,
2031#ifdef CONFIG_XEN_PVHVM
Sheng Yangbee6ab52010-05-14 12:39:33 +01002032 .init_platform = xen_hvm_guest_init,
Stefano Stabellinica65f9f2010-07-29 14:37:48 +01002033#endif
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04002034 .x2apic_available = xen_x2apic_para_available,
Juergen Gross99bc6752016-08-29 08:48:45 +02002035 .pin_vcpu = xen_pin_vcpu,
Boris Ostrovskya71dbda2015-05-04 11:02:15 -04002036};
2037EXPORT_SYMBOL(x86_hyper_xen);
Stefano Stabellinia314e3e2015-10-22 16:20:46 +00002038
2039#ifdef CONFIG_HOTPLUG_CPU
2040void xen_arch_register_cpu(int num)
2041{
2042 arch_register_cpu(num);
2043}
2044EXPORT_SYMBOL(xen_arch_register_cpu);
2045
2046void xen_arch_unregister_cpu(int num)
2047{
2048 arch_unregister_cpu(num);
2049}
2050EXPORT_SYMBOL(xen_arch_unregister_cpu);
2051#endif