| Message ID | 20250101064928.389504-1-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | KVM: allow NULL writable argument to __kvm_faultin_pfn | expand |
On Wed, Jan 01, 2025, Paolo Bonzini wrote: > kvm_follow_pfn() is able to work with NULL in the .map_writable field > of the homonymous struct. But __kvm_faultin_pfn() rejects the combo > despite KVM for e500 trying to use it. Indeed .map_writable is not > particularly useful if the flags include FOLL_WRITE and readonly > guest memory is not supported, so add support to __kvm_faultin_pfn() > for this case. I would prefer to keep the sanity check to minimize the risk of a page fault handler not supporting opportunistic write mappings. e500 is definitely the odd one out here. What about adding a dedicated wrapper for getting a writable PFN? E.g. (untested) --- arch/powerpc/kvm/e500_mmu_host.c | 2 +- arch/x86/kvm/vmx/vmx.c | 3 +-- include/linux/kvm_host.h | 8 ++++++++ 3 files changed, 10 insertions(+), 3 deletions(-) diff --git a/arch/powerpc/kvm/e500_mmu_host.c b/arch/powerpc/kvm/e500_mmu_host.c index e5a145b578a4..2251bb30b8ec 100644 --- a/arch/powerpc/kvm/e500_mmu_host.c +++ b/arch/powerpc/kvm/e500_mmu_host.c @@ -444,7 +444,7 @@ static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500, if (likely(!pfnmap)) { tsize_pages = 1UL << (tsize + 10 - PAGE_SHIFT); - pfn = __kvm_faultin_pfn(slot, gfn, FOLL_WRITE, NULL, &page); + pfn = kvm_faultin_writable_pfn(slot, gfn, &page); if (is_error_noslot_pfn(pfn)) { if (printk_ratelimit()) pr_err("%s: real page not found for gfn %lx\n", diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c index 893366e53732..7012b583f2e8 100644 --- a/arch/x86/kvm/vmx/vmx.c +++ b/arch/x86/kvm/vmx/vmx.c @@ -6800,7 +6800,6 @@ void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu) struct page *refcounted_page; unsigned long mmu_seq; kvm_pfn_t pfn; - bool writable; /* Defer reload until vmcs01 is the current VMCS. */ if (is_guest_mode(vcpu)) { @@ -6836,7 +6835,7 @@ void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu) * controls the APIC-access page memslot, and only deletes the memslot * if APICv is permanently inhibited, i.e. the memslot won't reappear. */ - pfn = __kvm_faultin_pfn(slot, gfn, FOLL_WRITE, &writable, &refcounted_page); + pfn = kvm_faultin_writable_pfn(slot, gfn, &refcounted_page); if (is_error_noslot_pfn(pfn)) return; diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index c788d0bd952a..b0af7c7f99da 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -1287,6 +1287,14 @@ static inline kvm_pfn_t kvm_faultin_pfn(struct kvm_vcpu *vcpu, gfn_t gfn, write ? FOLL_WRITE : 0, writable, refcounted_page); } +static inline kvm_pfn_t kvm_faultin_writable_pfn(const struct kvm_memory_slot *slot, + gfn_t gfn, struct page **refcounted_page) +{ + bool writable; + + return __kvm_faultin_pfn(slot, gfn, FOLL_WRITE, &writable, refcounted_page); +} + int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, int len); int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len); base-commit: 2c3412e999738bfd60859c493ff47f5c268814a3 --
On Mon, Jan 06, 2025, Sean Christopherson wrote: > On Wed, Jan 01, 2025, Paolo Bonzini wrote: > > kvm_follow_pfn() is able to work with NULL in the .map_writable field > > of the homonymous struct. But __kvm_faultin_pfn() rejects the combo > > despite KVM for e500 trying to use it. Indeed .map_writable is not > > particularly useful if the flags include FOLL_WRITE and readonly > > guest memory is not supported, so add support to __kvm_faultin_pfn() > > for this case. > > I would prefer to keep the sanity check to minimize the risk of a page fault > handler not supporting opportunistic write mappings. e500 is definitely the > odd one out here. Per a quick chat at PUCK, Paolo is going to try and fix the e500 code to actually use the @writable param as it's intended.
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c index de2c11dae231..5177e56fdbd5 100644 --- a/virt/kvm/kvm_main.c +++ b/virt/kvm/kvm_main.c @@ -2975,10 +2975,11 @@ kvm_pfn_t __kvm_faultin_pfn(const struct kvm_memory_slot *slot, gfn_t gfn, .refcounted_page = refcounted_page, }; - if (WARN_ON_ONCE(!writable || !refcounted_page)) + if (WARN_ON_ONCE(!refcounted_page)) return KVM_PFN_ERR_FAULT; - *writable = false; + if (writable) + *writable = false; *refcounted_page = NULL; return kvm_follow_pfn(&kfp);
kvm_follow_pfn() is able to work with NULL in the .map_writable field of the homonymous struct. But __kvm_faultin_pfn() rejects the combo despite KVM for e500 trying to use it. Indeed .map_writable is not particularly useful if the flags include FOLL_WRITE and readonly guest memory is not supported, so add support to __kvm_faultin_pfn() for this case. Fixes: 1c7b627e9306 ("KVM: Add kvm_faultin_pfn() to specifically service guest page faults") Reported-by: Christian Zigotzky <chzigotzky@xenosoft.de> Tested-by: Christian Zigotzky <chzigotzky@xenosoft.de> Cc: linuxppc-dev@lists.ozlabs.org Cc: regressions@lists.linux.dev Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- virt/kvm/kvm_main.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)