Nested Virtualization: hap-on-hap
PATCH 12/12 Nested Virtualization: hap-on-hap
Signature · 7 lines
-- ---to satisfy European Law for business letters: Advanced Micro Devices GmbH Einsteinring 24, 85609 Dornach b. Muenchen Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen Registergericht Muenchen, HRB Nr. 43632
# HG changeset patch
# User cegger
# Date 1292857188 -3600
Implement Nested-on-Nested.
This allows the guest to run nested guest with hap enabled.
xen/arch/x86/hvm/hvm.c +53 -6
@@ -1149,12 +1149,40 @@ void hvm_inject_exception(unsigned int t hvm_funcs.inject_exception (trapnr, errcode, cr2); } -bool_t hvm_hap_nested_page_fault (unsigned long gfn) +int hvm_hap_nested_page_fault (paddr_t gpa,struct cpu_user_regs *regs) {p2m_type_t p2mt;mfn_t mfn;struct vcpu *v = current;struct p2m_domain *p2m =p2m_get_hostp2m (v->domain); +unsigned long gfn = gpa >>PAGE_SHIFT ; +int rv; + +/* On Nested Virtualization, walk the guest page table. +* If this succeeds, all is fine. +* If this fails, inject a nested page fault into the guest. +*/ +if (nestedhvm_enabled (v->domain) + &&nestedhvm_vcpu_in_guestmode (v) + &&nestedhvm_paging_mode_hap (v) ) + { +/* The vcpu is in guest mode and the l1 guest +* uses hap. That means 'gpa' is in l2 guest +* physical address space. +* Fix the nested p2m or inject nested page fault +* into l1 guest if not fixable. The algorithm is +* the same as for shadow paging. +*/ + rv =nestedhvm_hap_nested_page_fault (v, gpa); +switch (rv) { +case NESTEDHVM_PAGEFAULT_DONE : +return 1 ; +case NESTEDHVM_PAGEFAULT_ERROR : +return 0 ; +case NESTEDHVM_PAGEFAULT_INJECT : +return -1 ; + } + } mfn =gfn_to_mfn_guest (p2m, gfn, &p2mt);
@@ -1257,6 +1285,15 @@ int hvm_set_efer(uint64_t value)return X86EMUL_EXCEPTION ; } +if (nestedhvm_enabled (v->domain) && cpu_has_svm && + ((value &EFER_SVME ) ==0 ) && + ((value ^ v->arch.hvm_vcpu.guest_efer) &EFER_SVME ) ) + { +/* Cleared EFER.SVME: Flush all nestedp2m tables */ +p2m_flush_nestedp2m (v->domain); +nestedhvm_vcpu_reset (v); + } + value |= v->arch.hvm_vcpu.guest_efer &EFER_LMA ; v->arch.hvm_vcpu.guest_efer = value;hvm_update_guest_efer (v);
@@ -1407,8 +1444,12 @@ int hvm_set_cr0(unsigned long value) v->arch.hvm_vcpu.guest_cr[0 ] = value;hvm_update_guest_cr (v,0 ); -if ( (value ^ old_value) &X86_CR0_PG ) -paging_update_paging_modes (v); +if ( (value ^ old_value) &X86_CR0_PG ) { +if ( !nestedhvm_vmswitch_in_progress (v) &&nestedhvm_vcpu_in_guestmode (v) ) +paging_update_nestedmode (v); +else +paging_update_paging_modes (v); + }return X86EMUL_OKAY ;
@@ -1475,8 +1516,12 @@ int hvm_set_cr4(unsigned long value)hvm_update_guest_cr (v,4 );/* Modifying CR4.{PSE,PAE,PGE} invalidates all TLB entries, inc. Global. */ -if ( (old_cr ^ value) & (X86_CR4_PSE |X86_CR4_PGE |X86_CR4_PAE ) ) -paging_update_paging_modes (v); +if ( (old_cr ^ value) & (X86_CR4_PSE |X86_CR4_PGE |X86_CR4_PAE ) ) { +if ( !nestedhvm_vmswitch_in_progress (v) &&nestedhvm_vcpu_in_guestmode (v) ) +paging_update_nestedmode (v); +else +paging_update_paging_modes (v); + }return X86EMUL_OKAY ;
@@ -1989,7 +2034,7 @@ static enum hvm_copy_result __hvm_copy( void *buf, paddr_t addr, int size, unsigned int flags, uint32_t pfec) { struct vcpu *curr = current; -struct p2m_domain *p2m =p2m_get_hostp2m (curr->domain); +struct p2m_domain *p2m;unsigned long gfn, mfn;p2m_type_t p2mt;char *p;
@@ -2002,6 +2047,8 @@ static enum hvm_copy_result __hvm_copy(if (in_atomic () )return HVMCOPY_unhandleable; + p2m =p2m_get_hostp2m (curr->domain); +while ( todo >0 ) { count =min_t (int,PAGE_SIZE - (addr & ~PAGE_MASK ), todo);
xen/arch/x86/hvm/nestedhvm.c +31 -0
@@ -20,6 +20,7 @@#include <asm/msr.h> #include <asm/hvm/support.h> /* for HVM_DELIVER_NO_ERROR_CODE */ #include <asm/hvm/hvm.h> +#include <asm/p2m.h> /* for struct p2m_domain */ #include <asm/hvm/nestedhvm.h> #include <asm/event.h> /* for local_event_delivery_(en|dis)able */ #include <asm/paging.h> /* for paging_mode_hap() */
@@ -100,13 +101,43 @@ nestedhvm_vcpu_destroy(struct vcpu *v)void nestedhvm_vcpu_enter_guestmode (struct vcpu *v) { +struct p2m_domain *p2m;vcpu_nestedhvm (v).nv_guestmode =1 ; + + p2m =vcpu_nestedhvm (v).nv_p2m; +if (p2m ==NULL ) +/* p2m has either been invalidated or not yet assigned. */ +return ; + +cpu_set (v->processor, p2m->p2m_dirty_cpumask); }void nestedhvm_vcpu_exit_guestmode (struct vcpu *v) { +struct p2m_domain *p2m;vcpu_nestedhvm (v).nv_guestmode =0 ; + + p2m =vcpu_nestedhvm (v).nv_p2m; +if (p2m ==NULL ) +/* p2m has either been invalidated or not yet assigned. */ +return ; + +cpu_clear (v->processor, p2m->p2m_dirty_cpumask); +} + +bool_t +nestedhvm_is_n2 (struct vcpu *v) +{ +if (!nestedhvm_enabled (v->domain) + ||nestedhvm_vmswitch_in_progress (v) + || !nestedhvm_paging_mode_hap (v)) +return 0 ; + +if (nestedhvm_vcpu_in_guestmode (v)) +return 1 ; + +return 0 ; }/* Common shadow IO Permission bitmap */
xen/arch/x86/hvm/svm/nestedsvm.c +5 -0
@@ -26,6 +26,7 @@#include <asm/hvm/svm/svmdebug.h> #include <asm/paging.h> /* paging_mode_hap */ #include <asm/event.h> /* for local_event_delivery_(en|dis)able */ +#include <asm/p2m.h> /* p2m_get_pagetable, p2m_get_nestedp2m */ static void nestedsvm_vcpu_clgi (struct vcpu *v)
@@ -456,6 +457,10 @@ static int nsvm_vmcb_prepare4vmrun(struc/* Nested paging mode */ if (nestedhvm_paging_mode_hap (v)) {/* host nested paging + guest nested paging. */ + host_vmcb->_np_enable =1 ; + + host_vmcb->_h_cr3 = +pagetable_get_paddr (p2m_get_pagetable (p2m_get_nestedp2m (v, ns_vmcb->_h_cr3)));/* hvm_set_cr3() below sets v->arch.hvm_vcpu.guest_cr[3] for us. */ rc =hvm_set_cr3 (ns_vmcb->_cr3);
xen/arch/x86/hvm/svm/svm.c +25 -8
@@ -1004,14 +1004,16 @@ struct hvm_function_table * __init startreturn &svm_function_table; } -static void svm_do_nested_pgfault (paddr_t gpa) +static void svm_do_nested_pgfault (struct vcpu *v, +struct cpu_user_regs *regs,paddr_t gpa) { +int ret;unsigned long gfn = gpa >>PAGE_SHIFT ;mfn_t mfn;p2m_type_t p2mt; -struct p2m_domain *p2m; +struct p2m_domain *p2m =NULL ; - p2m =p2m_get_hostp2m (current->domain); + ret =hvm_hap_nested_page_fault (gpa, regs);if ( tb_init_done ) {
@@ -1022,6 +1024,7 @@ static void svm_do_nested_pgfault(paddr_uint32_t p2mt; } _d; + p2m =p2m_get_p2m (v); _d.gpa = gpa; _d.qualification =0 ; _d.mfn =mfn_x (gfn_to_mfn_query (p2m, gfn, &_d.p2mt));
@@ -1029,14 +1032,26 @@ static void svm_do_nested_pgfault(paddr___trace_var (TRC_HVM_NPF ,0 ,sizeof (_d), &_d); } -if (hvm_hap_nested_page_fault (gfn) ) +switch (ret) { +case 0 : +break ; +case 1 :return ; +case -1 : +ASSERT (nestedhvm_enabled (v->domain) &&nestedhvm_vcpu_in_guestmode (v)); +/* inject #VMEXIT(NPF) into guest. */ +nestedsvm_vmexit_defer (v,VMEXIT_NPF , regs->error_code, gpa); +return ; + } +if ( p2m ==NULL ) + p2m =p2m_get_p2m (v);/* Everything else is an error. */ mfn =gfn_to_mfn_guest (p2m, gfn, &p2mt); -gdprintk (XENLOG_ERR ,"SVM violation gpa %#" PRIpaddr", mfn %#lx, type %i\n" , - gpa,mfn_x (mfn), p2mt); -domain_crash (current->domain); +gdprintk (XENLOG_ERR , +"SVM violation gpa %#" PRIpaddr", mfn %#lx, type %i\n" , + gpa,mfn_x (mfn), p2mt); +domain_crash (v->domain); }static void svm_fpu_dirty_intercept (void )
@@ -1630,6 +1645,8 @@ asmlinkage void svm_vmexit_handler(strucstruct vmcb_struct *ns_vmcb = nv->nv_vmcx;uint64_t exitinfo1, exitinfo2; +paging_update_nestedmode (v); +/* Write real exitinfo1 back into virtual vmcb. * nestedsvm_check_intercepts() expects to have the correct * exitinfo1 value there.
@@ -1919,7 +1936,7 @@ asmlinkage void svm_vmexit_handler(struccase VMEXIT_NPF :perfc_incra (svmexits,VMEXIT_NPF_PERFC ); regs->error_code = vmcb->exitinfo1; -svm_do_nested_pgfault (vmcb->exitinfo2); +svm_do_nested_pgfault (v, regs, vmcb->exitinfo2);break ;case VMEXIT_IRET : {
xen/arch/x86/hvm/vmx/vmx.c +1 -1
@@ -2080,7 +2080,7 @@ static void ept_handle_violation(unsigne }if ( (qualification &EPT_GLA_VALID ) && -hvm_hap_nested_page_fault (gfn) ) +hvm_hap_nested_page_fault (gpa,guest_cpu_user_regs ()) )return ;/* Everything else is an error. */
xen/arch/x86/mm/hap/Makefile +1 -0
@@ -3,6 +3,7 @@ obj-y += guest_walk_2level.o obj-y += guest_walk_3level.o obj-y += guest_walk_4level.o obj-y += p2m-ept.o +obj-y += nested_hap.o guest_levels = $(subst level,,$(filter %level,$(subst ., ,$(subst _, ,$(1))))) guest_walk_defns = -DGUEST_PAGING_LEVELS=$(call guest_levels,$(1))
xen/arch/x86/mm/hap/guest_walk.c +22 -5
@@ -29,24 +29,32 @@#define _hap_gva_to_gfn (levels) hap_gva_to_gfn_##levels##_levels#define hap_gva_to_gfn (levels) _hap_gva_to_gfn(levels) +#define _hap_p2m_ga_to_gfn (levels) hap_p2m_ga_to_gfn_##levels##_levels +#define hap_p2m_ga_to_gfn (levels) _hap_p2m_ga_to_gfn(levels) +#if GUEST_PAGING_LEVELS <=CONFIG_PAGING_LEVELS #include <asm/guest_pt.h> #include <asm/p2m.h> unsigned long hap_gva_to_gfn (GUEST_PAGING_LEVELS )( -struct vcpu *v,unsigned long gva,uint32_t *pfec) +struct vcpu *v,struct p2m_domain *p2m,unsigned long gva,uint32_t *pfec) { -unsigned long cr3; +unsigned long cr3 = v->arch.hvm_vcpu.guest_cr[3 ]; +return hap_p2m_ga_to_gfn (GUEST_PAGING_LEVELS )(v, p2m, cr3, gva, pfec); +} + +unsigned long hap_p2m_ga_to_gfn (GUEST_PAGING_LEVELS )( +struct vcpu *v,struct p2m_domain *p2m,unsigned long cr3, +paddr_t ga,uint32_t *pfec) +{uint32_t missing;mfn_t top_mfn;void *top_map;p2m_type_t p2mt;walk_t gw; -struct p2m_domain *p2m =p2m_get_hostp2m (v->domain);/* Get the top-level table's MFN */ - cr3 = v->arch.hvm_vcpu.guest_cr[3 ]; top_mfn =gfn_to_mfn_unshare (p2m, cr3 >>PAGE_SHIFT , &p2mt,0 );if (p2m_is_paging (p2mt) ) {
@@ -72,7 +80,7 @@ unsigned long hap_gva_to_gfn(GUEST_PAGIN#if GUEST_PAGING_LEVELS ==3 top_map += (cr3 & ~(PAGE_MASK |31 ));#endif - missing =guest_walk_tables (v, p2m, gva, &gw, pfec[0 ], top_mfn, top_map); + missing =guest_walk_tables (v, p2m, ga, &gw, pfec[0 ], top_mfn, top_map);unmap_domain_page (top_map);/* Interpret the answer */
@@ -122,6 +130,15 @@ unsigned long hap_gva_to_gfn(GUEST_PAGIN return INVALID_GFN; } +unsigned long hap_p2m_ga_to_gfn (GUEST_PAGING_LEVELS )( +struct vcpu *v,struct p2m_domain *p2m,unsigned long cr3, +paddr_t ga,uint32_t *pfec) +{ +gdprintk (XENLOG_ERR , +"Guest paging level is greater than host paging level!\n" ); +domain_crash (v->domain); +return INVALID_GFN ; +}#endif
xen/arch/x86/mm/hap/hap.c +83 -12
199 lines not on this page · raw message
xen/arch/x86/mm/hap/nested_hap.c new file +231 -0
231 lines not on this page · raw message
xen/arch/x86/mm/hap/private.h +19 -3
@@ -23,11 +23,27 @@/********************************************/ /* GUEST TRANSLATION FUNCS */ /********************************************/ -unsigned long hap_gva_to_gfn_2_levels (struct vcpu *v,unsigned long gva, +unsigned long hap_gva_to_gfn_2_levels (struct vcpu *v, +struct p2m_domain *p2m, +unsigned long gva,uint32_t *pfec); -unsigned long hap_gva_to_gfn_3_levels (struct vcpu *v,unsigned long gva, +unsigned long hap_gva_to_gfn_3_levels (struct vcpu *v, +struct p2m_domain *p2m, +unsigned long gva,uint32_t *pfec); -unsigned long hap_gva_to_gfn_4_levels (struct vcpu *v,unsigned long gva, +unsigned long hap_gva_to_gfn_4_levels (struct vcpu *v, +struct p2m_domain *p2m, +unsigned long gva,uint32_t *pfec); +unsigned long hap_p2m_ga_to_gfn_2_levels (struct vcpu *v, +struct p2m_domain *p2m,unsigned long cr3, +paddr_t ga,uint32_t *pfec); +unsigned long hap_p2m_ga_to_gfn_3_levels (struct vcpu *v, +struct p2m_domain *p2m,unsigned long cr3, +paddr_t ga,uint32_t *pfec); +unsigned long hap_p2m_ga_to_gfn_4_levels (struct vcpu *v, +struct p2m_domain *p2m,unsigned long cr3, +paddr_t ga,uint32_t *pfec); +#endif /* __HAP_PRIVATE_H__ */
xen/arch/x86/mm/p2m.c +254 -33
434 lines not on this page · raw message
xen/arch/x86/mm/paging.c +43 -5
70 lines not on this page · raw message
xen/arch/x86/mm/shadow/multi.c +3 -2
@@ -3768,7 +3768,8 @@ sh_invlpg(struct vcpu *v, unsigned longstatic unsigned long -sh_gva_to_gfn (struct vcpu *v,unsigned long va,uint32_t *pfec) +sh_gva_to_gfn (struct vcpu *v,struct p2m_domain *p2m, +unsigned long va,uint32_t *pfec)/* Called to translate a guest virtual address to what the *guest* * pagetables would map it to. */ {
@@ -4820,7 +4821,7 @@ static mfn_t emulate_gva_to_mfn(struct vstruct p2m_domain *p2m =p2m_get_hostp2m (v->domain);/* Translate the VA to a GFN */ - gfn =sh_gva_to_gfn (v, vaddr, &pfec); + gfn =sh_gva_to_gfn (v, p2m, vaddr, &pfec);if ( gfn ==INVALID_GFN ) {if (is_hvm_vcpu (v) )
xen/include/asm-x86/domain.h +7 -0
@@ -210,6 +210,8 @@ struct paging_domain {struct paging_vcpu {/* Pointers to mode-specific entry points. */ const struct paging_mode *mode; +/* Nested Virtualization: paging mode of nested guest */ +const struct paging_mode *nestedmode;/* HVM guest: last emulate was to a pagetable */ unsigned int last_write_was_pt:1 ;/* HVM guest: last write emulation succeeds */
@@ -225,6 +227,7 @@ struct paging_vcpu {#define MAX_CPUID_INPUT 40typedef xen_domctl_cpuid_t cpuid_input_t ; +#define MAX_NESTEDP2M 10struct p2m_domain ;struct time_scale {int shift;
@@ -258,6 +261,10 @@ struct arch_domainstruct paging_domain paging;struct p2m_domain *p2m; +/* nestedhvm: translate l2 guest physical to host physical */ +struct p2m_domain *nested_p2m[MAX_NESTEDP2M ]; +spinlock_t nested_p2m_lock; +/* NB. protected by d->event_lock and by irq_desc[irq].lock */ int *irq_pirq;int *pirq_irq;
xen/include/asm-x86/hvm/hvm.h +1 -1
@@ -374,7 +374,7 @@ static inline void hvm_set_info_guest(stint hvm_debug_op (struct vcpu *v,int32_t op); -bool_t hvm_hap_nested_page_fault (unsigned long gfn); +int hvm_hap_nested_page_fault (paddr_t gpa,struct cpu_user_regs *regs);#define hvm_msr_tsc_aux (v) ({ \ struct domain *__d = (v)->domain; \
xen/include/asm-x86/hvm/nestedhvm.h +4 -2
@@ -61,7 +61,9 @@ void nestedhvm_vcpu_iomap_put(unsigned l#define nestedhvm_paging_mode_hap (v) (!!nhvm_vmcx_hap_enabled(v))#define nestedhvm_vmswitch_in_progress (v) \ (!!vcpu_nestedhvm((v)).nv_vmswitch_in_progress) -#define nestedhvm_vmcx_flushtlb (d) \ - flush_tlb_mask(&(d)->arch.hvm_domain.nh_dirty_cpumask) +#define nestedhvm_vmcx_flushtlb (p2m) \ + flush_tlb_mask(&((p2m)->p2m_dirty_cpumask)) + +bool_t nestedhvm_is_n2 (struct vcpu *v);#endif /* _HVM_NESTEDHVM_H */
xen/include/asm-x86/p2m.h +42 -0
80 lines not on this page · raw message
xen/include/asm-x86/paging.h +47 -28
149 lines not on this page · raw message
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel
Hi,
We still haven't resolved the issue of safely recycling a p2m table that is in use on another CPU. I'll quote my last email in its entirety here:
| At 16:27 +0000 on 20 Dec (1292862443), Christoph Egger wrote: | > > > An other vcpu is in VMRUN emulation after a nestedp2m is assigned. | > > > It will VMEXIT with a nested page fault. | > > | > > Why? | > | > Because the p2m is empty. The MMU can not do a page table walk. | > | > > > An other vcpu already running l2 guest. | > > > It will VMEXIT with a nested page fault immediately. | > > | > > Hmm. It will exit for the TLB shootdown IPI, but I think you need | to | > > clear vcpu_nestedhvm(v).nh_p2m on the other vcpu to make sure it | doesn't | > > re-enter with the p2m you've just recycled. | > | > The p2m is empty so I don't see a problem when it gets recycled. | | It's only empty very briefly. You've assigned it to a vcpu which is | about to take a nested fault and fill it with entries, right? | | What happens if the other vcpu is handling an SMI or executing a tight | loop of register arithmetic for a few thousand cycles? What stops it | seeing the new contents of the p2m?
One other issue I pointed out last time is still there in this patch:
Quoted 18 lines
xen/arch/x86/mm/hap/hap.c +9
-2@@ -835,38 +865,76 @@ static void hap_write_p2m_entry(struct vcpu *v, unsigned long gfn, l1_pgentry_t *p, mfn_t table_mfn, l1_pgentry_t new, unsigned int level) { + struct domain *d = v->domain;uint32_t old_flags; -hap_lock (v->domain); + old_flags =l1e_get_flags (*p); - old_flags =l1e_get_flags (*p); +/* We know always use the host p2m here, regardless if the vcpu +* is in host or guest mode. The vcpu can be in guest mode by +* a hypercall which passes a domain and chooses mostly the first +* vcpu. +* XXX This is the reason why this function can not be used re-used +* for updating the nestedp2m. Otherwise, hypercalls would randomly +* operate on host p2m and nested p2m.
xen/arch/x86/mm/hap/hap.c static void · line 889
Quoted 10 lines
xen/arch/x86/mm/hap/hap.c +10
-0+*/ +if (nestedhvm_enabled (d) ) { +mfn_t omfn =_mfn (l1e_get_pfn (*p)); +p2m_type_t op2mt =p2m_flags_to_type (old_flags); + +if (p2m_is_valid (op2mt) ) { +mfn_t nmfn =_mfn (l1e_get_pfn (new)); +p2m_type_t np2mt =p2m_flags_to_type (l1e_get_flags (new)); + +if (p2m_is_valid (np2mt) && (mfn_x (omfn) !=mfn_x (nmfn)) ) {
Checking that the mfns are the same isn't quite enough, as the permissions might have changed (e.g. log-dirty mode removing write access). You need to test for (new == *p) to be sure that you don't need to flush.
Cheers,
Tim.
Quoted 12 lines
xen/arch/x86/mm/hap/hap.c +12
-0+/* This GFN -> MFN is going to get removed. */ +/* XXX There is a more efficient way to do that +* but it works for now. +* Note, p2m_flush_nestedp2m calls hap_lock() internally. +*/ +p2m_flush_nestedp2m (d); + } + } + } + +hap_lock (d); +
Signature · 3 lines
Tim Deegan <[email protected]> Principal Software Engineer, Xen Platform Team Citrix Systems UK Ltd. (Company #02937203, SL9 0BG)