2 * Core of Xen paravirt_ops implementation.
4 * This file contains the xen_paravirt_ops structure itself, and the
6 * - privileged instructions
11 * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007
14 #include <linux/kernel.h>
15 #include <linux/init.h>
16 #include <linux/smp.h>
17 #include <linux/preempt.h>
18 #include <linux/percpu.h>
19 #include <linux/delay.h>
20 #include <linux/start_kernel.h>
21 #include <linux/sched.h>
22 #include <linux/bootmem.h>
23 #include <linux/module.h>
25 #include <xen/interface/xen.h>
26 #include <xen/interface/physdev.h>
27 #include <xen/interface/vcpu.h>
28 #include <xen/features.h>
31 #include <asm/paravirt.h>
33 #include <asm/xen/hypercall.h>
34 #include <asm/xen/hypervisor.h>
35 #include <asm/fixmap.h>
36 #include <asm/processor.h>
37 #include <asm/setup.h>
39 #include <asm/pgtable.h>
42 #include "multicalls.h"
44 EXPORT_SYMBOL_GPL(hypercall_page);
46 DEFINE_PER_CPU(enum paravirt_lazy_mode, xen_lazy_mode);
48 DEFINE_PER_CPU(struct vcpu_info *, xen_vcpu);
49 DEFINE_PER_CPU(struct vcpu_info, xen_vcpu_info);
50 DEFINE_PER_CPU(unsigned long, xen_cr3);
52 struct start_info *xen_start_info;
53 EXPORT_SYMBOL_GPL(xen_start_info);
55 static void xen_vcpu_setup(int cpu)
57 per_cpu(xen_vcpu, cpu) = &HYPERVISOR_shared_info->vcpu_info[cpu];
60 static void __init xen_banner(void)
62 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
64 printk(KERN_INFO "Hypervisor signature: %s\n", xen_start_info->magic);
67 static void xen_cpuid(unsigned int *eax, unsigned int *ebx,
68 unsigned int *ecx, unsigned int *edx)
70 unsigned maskedx = ~0;
73 * Mask out inconvenient features, to try and disable as many
74 * unsupported kernel subsystems as possible.
77 maskedx = ~((1 << X86_FEATURE_APIC) | /* disable APIC */
78 (1 << X86_FEATURE_ACPI) | /* disable ACPI */
79 (1 << X86_FEATURE_ACC)); /* thermal monitoring */
81 asm(XEN_EMULATE_PREFIX "cpuid"
86 : "0" (*eax), "2" (*ecx));
90 static void xen_set_debugreg(int reg, unsigned long val)
92 HYPERVISOR_set_debugreg(reg, val);
95 static unsigned long xen_get_debugreg(int reg)
97 return HYPERVISOR_get_debugreg(reg);
100 static unsigned long xen_save_fl(void)
102 struct vcpu_info *vcpu;
106 vcpu = x86_read_percpu(xen_vcpu);
107 /* flag has opposite sense of mask */
108 flags = !vcpu->evtchn_upcall_mask;
111 /* convert to IF type flag
115 return (-flags) & X86_EFLAGS_IF;
118 static void xen_restore_fl(unsigned long flags)
120 struct vcpu_info *vcpu;
124 /* convert from IF type flag */
125 flags = !(flags & X86_EFLAGS_IF);
126 vcpu = x86_read_percpu(xen_vcpu);
127 vcpu->evtchn_upcall_mask = flags;
130 /* Unmask then check (avoid races). We're only protecting
131 against updates by this CPU, so there's no need for
132 anything stronger. */
135 if (unlikely(vcpu->evtchn_upcall_pending))
136 force_evtchn_callback();
139 preempt_enable_no_resched();
142 static void xen_irq_disable(void)
144 struct vcpu_info *vcpu;
146 vcpu = x86_read_percpu(xen_vcpu);
147 vcpu->evtchn_upcall_mask = 1;
148 preempt_enable_no_resched();
151 static void xen_irq_enable(void)
153 struct vcpu_info *vcpu;
156 vcpu = x86_read_percpu(xen_vcpu);
157 vcpu->evtchn_upcall_mask = 0;
159 /* Unmask then check (avoid races). We're only protecting
160 against updates by this CPU, so there's no need for
161 anything stronger. */
164 if (unlikely(vcpu->evtchn_upcall_pending))
165 force_evtchn_callback();
169 static void xen_safe_halt(void)
171 /* Blocking includes an implicit local_irq_enable(). */
172 if (HYPERVISOR_sched_op(SCHEDOP_block, 0) != 0)
176 static void xen_halt(void)
179 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
184 static void xen_set_lazy_mode(enum paravirt_lazy_mode mode)
187 case PARAVIRT_LAZY_NONE:
188 BUG_ON(x86_read_percpu(xen_lazy_mode) == PARAVIRT_LAZY_NONE);
191 case PARAVIRT_LAZY_MMU:
192 case PARAVIRT_LAZY_CPU:
193 BUG_ON(x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE);
196 case PARAVIRT_LAZY_FLUSH:
197 /* flush if necessary, but don't change state */
198 if (x86_read_percpu(xen_lazy_mode) != PARAVIRT_LAZY_NONE)
204 x86_write_percpu(xen_lazy_mode, mode);
207 static unsigned long xen_store_tr(void)
212 static void xen_set_ldt(const void *addr, unsigned entries)
214 unsigned long linear_addr = (unsigned long)addr;
215 struct mmuext_op *op;
216 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
219 op->cmd = MMUEXT_SET_LDT;
221 /* ldt my be vmalloced, use arbitrary_virt_to_machine */
223 maddr = arbitrary_virt_to_machine((unsigned long)addr);
224 linear_addr = (unsigned long)maddr.maddr;
226 op->arg1.linear_addr = linear_addr;
227 op->arg2.nr_ents = entries;
229 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
231 xen_mc_issue(PARAVIRT_LAZY_CPU);
234 static void xen_load_gdt(const struct Xgt_desc_struct *dtr)
236 unsigned long *frames;
237 unsigned long va = dtr->address;
238 unsigned int size = dtr->size + 1;
239 unsigned pages = (size + PAGE_SIZE - 1) / PAGE_SIZE;
241 struct multicall_space mcs;
243 /* A GDT can be up to 64k in size, which corresponds to 8192
244 8-byte entries, or 16 4k pages.. */
246 BUG_ON(size > 65536);
247 BUG_ON(va & ~PAGE_MASK);
249 mcs = xen_mc_entry(sizeof(*frames) * pages);
252 for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) {
253 frames[f] = virt_to_mfn(va);
254 make_lowmem_page_readonly((void *)va);
257 MULTI_set_gdt(mcs.mc, frames, size / sizeof(struct desc_struct));
259 xen_mc_issue(PARAVIRT_LAZY_CPU);
262 static void load_TLS_descriptor(struct thread_struct *t,
263 unsigned int cpu, unsigned int i)
265 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
266 xmaddr_t maddr = virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]);
267 struct multicall_space mc = __xen_mc_entry(0);
269 MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]);
272 static void xen_load_tls(struct thread_struct *t, unsigned int cpu)
276 load_TLS_descriptor(t, cpu, 0);
277 load_TLS_descriptor(t, cpu, 1);
278 load_TLS_descriptor(t, cpu, 2);
280 xen_mc_issue(PARAVIRT_LAZY_CPU);
283 static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum,
286 unsigned long lp = (unsigned long)&dt[entrynum];
287 xmaddr_t mach_lp = virt_to_machine(lp);
288 u64 entry = (u64)high << 32 | low;
291 if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry))
295 static int cvt_gate_to_trap(int vector, u32 low, u32 high,
296 struct trap_info *info)
300 type = (high >> 8) & 0x1f;
301 dpl = (high >> 13) & 3;
303 if (type != 0xf && type != 0xe)
306 info->vector = vector;
307 info->address = (high & 0xffff0000) | (low & 0x0000ffff);
308 info->cs = low >> 16;
310 /* interrupt gates clear IF */
317 /* Locations of each CPU's IDT */
318 static DEFINE_PER_CPU(struct Xgt_desc_struct, idt_desc);
320 /* Set an IDT entry. If the entry is part of the current IDT, then
322 static void xen_write_idt_entry(struct desc_struct *dt, int entrynum,
326 int cpu = smp_processor_id();
327 unsigned long p = (unsigned long)&dt[entrynum];
328 unsigned long start = per_cpu(idt_desc, cpu).address;
329 unsigned long end = start + per_cpu(idt_desc, cpu).size + 1;
333 write_dt_entry(dt, entrynum, low, high);
335 if (p >= start && (p + 8) <= end) {
336 struct trap_info info[2];
340 if (cvt_gate_to_trap(entrynum, low, high, &info[0]))
341 if (HYPERVISOR_set_trap_table(info))
346 /* Load a new IDT into Xen. In principle this can be per-CPU, so we
347 hold a spinlock to protect the static traps[] array (static because
348 it avoids allocation, and saves stack space). */
349 static void xen_load_idt(const struct Xgt_desc_struct *desc)
351 static DEFINE_SPINLOCK(lock);
352 static struct trap_info traps[257];
354 int cpu = smp_processor_id();
355 unsigned in, out, count;
357 per_cpu(idt_desc, cpu) = *desc;
359 count = (desc->size+1) / 8;
363 for (in = out = 0; in < count; in++) {
364 const u32 *entry = (u32 *)(desc->address + in * 8);
366 if (cvt_gate_to_trap(in, entry[0], entry[1], &traps[out]))
369 traps[out].address = 0;
372 if (HYPERVISOR_set_trap_table(traps))
378 /* Write a GDT descriptor entry. Ignore LDT descriptors, since
379 they're handled differently. */
380 static void xen_write_gdt_entry(struct desc_struct *dt, int entry,
383 switch ((high >> 8) & 0xff) {
390 xmaddr_t maddr = virt_to_machine(&dt[entry]);
391 u64 desc = (u64)high << 32 | low;
394 if (HYPERVISOR_update_descriptor(maddr.maddr, desc))
401 static void xen_load_esp0(struct tss_struct *tss,
402 struct thread_struct *thread)
404 struct multicall_space mcs = xen_mc_entry(0);
405 MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->esp0);
406 xen_mc_issue(PARAVIRT_LAZY_CPU);
409 static void xen_set_iopl_mask(unsigned mask)
411 struct physdev_set_iopl set_iopl;
413 /* Force the change at ring 0. */
414 set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3;
415 HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl);
418 static void xen_io_delay(void)
422 #ifdef CONFIG_X86_LOCAL_APIC
423 static unsigned long xen_apic_read(unsigned long reg)
429 static void xen_flush_tlb(void)
433 op.cmd = MMUEXT_TLB_FLUSH_LOCAL;
434 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
438 static void xen_flush_tlb_single(unsigned long addr)
442 op.cmd = MMUEXT_INVLPG_LOCAL;
443 op.arg1.linear_addr = addr & PAGE_MASK;
444 if (HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF))
448 static unsigned long xen_read_cr2(void)
450 return x86_read_percpu(xen_vcpu)->arch.cr2;
453 static void xen_write_cr4(unsigned long cr4)
455 /* never allow TSC to be disabled */
456 native_write_cr4(cr4 & ~X86_CR4_TSD);
460 * Page-directory addresses above 4GB do not fit into architectural %cr3.
461 * When accessing %cr3, or equivalent field in vcpu_guest_context, guests
462 * must use the following accessor macros to pack/unpack valid MFNs.
464 * Note that Xen is using the fact that the pagetable base is always
465 * page-aligned, and putting the 12 MSB of the address into the 12 LSB
468 #define xen_pfn_to_cr3(pfn) (((unsigned)(pfn) << 12) | ((unsigned)(pfn) >> 20))
469 #define xen_cr3_to_pfn(cr3) (((unsigned)(cr3) >> 12) | ((unsigned)(cr3) << 20))
471 static unsigned long xen_read_cr3(void)
473 return x86_read_percpu(xen_cr3);
476 static void xen_write_cr3(unsigned long cr3)
478 if (cr3 == x86_read_percpu(xen_cr3)) {
479 /* just a simple tlb flush */
484 x86_write_percpu(xen_cr3, cr3);
488 struct mmuext_op *op;
489 struct multicall_space mcs = xen_mc_entry(sizeof(*op));
490 unsigned long mfn = pfn_to_mfn(PFN_DOWN(cr3));
493 op->cmd = MMUEXT_NEW_BASEPTR;
496 MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF);
498 xen_mc_issue(PARAVIRT_LAZY_CPU);
502 static void xen_alloc_pt(struct mm_struct *mm, u32 pfn)
504 /* XXX pfn isn't necessarily a lowmem page */
505 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
508 static void xen_alloc_pd(u32 pfn)
510 make_lowmem_page_readonly(__va(PFN_PHYS(pfn)));
513 static void xen_release_pd(u32 pfn)
515 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
518 static void xen_release_pt(u32 pfn)
520 make_lowmem_page_readwrite(__va(PFN_PHYS(pfn)));
523 static void xen_alloc_pd_clone(u32 pfn, u32 clonepfn,
524 u32 start, u32 count)
529 static __init void xen_pagetable_setup_start(pgd_t *base)
531 pgd_t *xen_pgd = (pgd_t *)xen_start_info->pt_base;
535 * copy top-level of Xen-supplied pagetable into place. For
536 * !PAE we can use this as-is, but for PAE it is a stand-in
537 * while we copy the pmd pages.
539 memcpy(base, xen_pgd, PTRS_PER_PGD * sizeof(pgd_t));
541 if (PTRS_PER_PMD > 1) {
544 * For PAE, need to allocate new pmds, rather than
545 * share Xen's, since Xen doesn't like pmd's being
546 * shared between address spaces.
548 for (i = 0; i < PTRS_PER_PGD; i++) {
549 if (pgd_val_ma(xen_pgd[i]) & _PAGE_PRESENT) {
550 pmd_t *pmd = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
552 memcpy(pmd, (void *)pgd_page_vaddr(xen_pgd[i]),
555 xen_alloc_pd(PFN_DOWN(__pa(pmd)));
557 set_pgd(&base[i], __pgd(1 + __pa(pmd)));
563 /* make sure zero_page is mapped RO so we can use it in pagetables */
564 make_lowmem_page_readonly(empty_zero_page);
565 make_lowmem_page_readonly(base);
567 * Switch to new pagetable. This is done before
568 * pagetable_init has done anything so that the new pages
569 * added to the table can be prepared properly for Xen.
571 xen_write_cr3(__pa(base));
574 static __init void xen_pagetable_setup_done(pgd_t *base)
576 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
578 * Create a mapping for the shared info page.
579 * Should be set_fixmap(), but shared_info is a machine
580 * address with no corresponding pseudo-phys address.
583 set_pte_mfn(fix_to_virt(FIX_PARAVIRT_BOOTMAP),
584 PFN_DOWN(xen_start_info->shared_info),
588 HYPERVISOR_shared_info =
589 (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP);
592 HYPERVISOR_shared_info =
593 (struct shared_info *)__va(xen_start_info->shared_info);
599 xen_vcpu_setup(smp_processor_id());
602 static const struct paravirt_ops xen_paravirt_ops __initdata = {
603 .paravirt_enabled = 1,
604 .shared_kernel_pmd = 0,
607 .banner = xen_banner,
609 .patch = paravirt_patch_default,
611 .memory_setup = xen_memory_setup,
612 .arch_setup = xen_arch_setup,
616 .set_debugreg = xen_set_debugreg,
617 .get_debugreg = xen_get_debugreg,
621 .read_cr0 = native_read_cr0,
622 .write_cr0 = native_write_cr0,
624 .read_cr2 = xen_read_cr2,
625 .write_cr2 = native_write_cr2,
627 .read_cr3 = xen_read_cr3,
628 .write_cr3 = xen_write_cr3,
630 .read_cr4 = native_read_cr4,
631 .read_cr4_safe = native_read_cr4_safe,
632 .write_cr4 = xen_write_cr4,
634 .save_fl = xen_save_fl,
635 .restore_fl = xen_restore_fl,
636 .irq_disable = xen_irq_disable,
637 .irq_enable = xen_irq_enable,
638 .safe_halt = xen_safe_halt,
640 .wbinvd = native_wbinvd,
642 .read_msr = native_read_msr_safe,
643 .write_msr = native_write_msr_safe,
644 .read_tsc = native_read_tsc,
645 .read_pmc = native_read_pmc,
647 .iret = (void *)&hypercall_page[__HYPERVISOR_iret],
648 .irq_enable_sysexit = NULL, /* never called */
650 .load_tr_desc = paravirt_nop,
651 .set_ldt = xen_set_ldt,
652 .load_gdt = xen_load_gdt,
653 .load_idt = xen_load_idt,
654 .load_tls = xen_load_tls,
656 .store_gdt = native_store_gdt,
657 .store_idt = native_store_idt,
658 .store_tr = xen_store_tr,
660 .write_ldt_entry = xen_write_ldt_entry,
661 .write_gdt_entry = xen_write_gdt_entry,
662 .write_idt_entry = xen_write_idt_entry,
663 .load_esp0 = xen_load_esp0,
665 .set_iopl_mask = xen_set_iopl_mask,
666 .io_delay = xen_io_delay,
668 #ifdef CONFIG_X86_LOCAL_APIC
669 .apic_write = paravirt_nop,
670 .apic_write_atomic = paravirt_nop,
671 .apic_read = xen_apic_read,
672 .setup_boot_clock = paravirt_nop,
673 .setup_secondary_clock = paravirt_nop,
674 .startup_ipi_hook = paravirt_nop,
677 .flush_tlb_user = xen_flush_tlb,
678 .flush_tlb_kernel = xen_flush_tlb,
679 .flush_tlb_single = xen_flush_tlb_single,
681 .pte_update = paravirt_nop,
682 .pte_update_defer = paravirt_nop,
684 .pagetable_setup_start = xen_pagetable_setup_start,
685 .pagetable_setup_done = xen_pagetable_setup_done,
687 .alloc_pt = xen_alloc_pt,
688 .alloc_pd = xen_alloc_pd,
689 .alloc_pd_clone = xen_alloc_pd_clone,
690 .release_pd = xen_release_pd,
691 .release_pt = xen_release_pt,
693 .set_lazy_mode = xen_set_lazy_mode,
696 /* First C function to be called on Xen boot */
697 asmlinkage void __init xen_start_kernel(void)
704 BUG_ON(memcmp(xen_start_info->magic, "xen-3.0", 7) != 0);
706 /* Install Xen paravirt ops */
707 paravirt_ops = xen_paravirt_ops;
709 xen_setup_features();
712 if (!xen_feature(XENFEAT_auto_translated_physmap))
713 phys_to_machine_mapping = (unsigned long *)xen_start_info->mfn_list;
715 pgd = (pgd_t *)xen_start_info->pt_base;
717 init_pg_tables_end = __pa(pgd) + xen_start_info->nr_pt_frames*PAGE_SIZE;
719 init_mm.pgd = pgd; /* use the Xen pagetables to start */
721 /* keep using Xen gdt for now; no urgent need to change it */
723 x86_write_percpu(xen_cr3, __pa(pgd));
726 paravirt_ops.kernel_rpl = 1;
727 if (xen_feature(XENFEAT_supervisor_mode_kernel))
728 paravirt_ops.kernel_rpl = 0;
730 /* set the limit of our address space */
731 reserve_top_address(-HYPERVISOR_VIRT_START + 2 * PAGE_SIZE);
733 /* set up basic CPUID stuff */
734 cpu_detect(&new_cpu_data);
735 new_cpu_data.hard_math = 1;
736 new_cpu_data.x86_capability[0] = cpuid_edx(1);
738 /* Poke various useful things into boot_params */
739 LOADER_TYPE = (9 << 4) | 0;
740 INITRD_START = xen_start_info->mod_start ? __pa(xen_start_info->mod_start) : 0;
741 INITRD_SIZE = xen_start_info->mod_len;
743 /* Start the world */