KVM: MMU: unify slots_lock usage
Marcelo Tosatti [Sat, 29 Mar 2008 23:17:59 +0000 (20:17 -0300)]
Unify slots_lock acquision around vcpu_run(). This is simpler and less
error-prone.

Also fix some callsites that were not grabbing the lock properly.

[avi: drop slots_lock while in guest mode to avoid holding the lock
      for indefinite periods]

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@qumranet.com>

arch/x86/kvm/mmu.c
arch/x86/kvm/paging_tmpl.h
arch/x86/kvm/vmx.c
arch/x86/kvm/x86.c
include/asm-x86/kvm_host.h

index 6fc3421..c563283 100644 (file)
@@ -1204,8 +1204,6 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
 
        struct page *page;
 
-       down_read(&vcpu->kvm->slots_lock);
-
        down_read(&current->mm->mmap_sem);
        if (is_largepage_backed(vcpu, gfn & ~(KVM_PAGES_PER_HPAGE-1))) {
                gfn &= ~(KVM_PAGES_PER_HPAGE-1);
@@ -1218,7 +1216,6 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
        /* mmio */
        if (is_error_page(page)) {
                kvm_release_page_clean(page);
-               up_read(&vcpu->kvm->slots_lock);
                return 1;
        }
 
@@ -1228,7 +1225,6 @@ static int nonpaging_map(struct kvm_vcpu *vcpu, gva_t v, int write, gfn_t gfn)
                         PT32E_ROOT_LEVEL);
        spin_unlock(&vcpu->kvm->mmu_lock);
 
-       up_read(&vcpu->kvm->slots_lock);
 
        return r;
 }
@@ -1376,9 +1372,9 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
                largepage = 1;
        }
        page = gfn_to_page(vcpu->kvm, gfn);
+       up_read(&current->mm->mmap_sem);
        if (is_error_page(page)) {
                kvm_release_page_clean(page);
-               up_read(&current->mm->mmap_sem);
                return 1;
        }
        spin_lock(&vcpu->kvm->mmu_lock);
@@ -1386,7 +1382,6 @@ static int tdp_page_fault(struct kvm_vcpu *vcpu, gva_t gpa,
        r = __direct_map(vcpu, gpa, error_code & PFERR_WRITE_MASK,
                         largepage, gfn, page, TDP_ROOT_LEVEL);
        spin_unlock(&vcpu->kvm->mmu_lock);
-       up_read(&current->mm->mmap_sem);
 
        return r;
 }
@@ -1808,9 +1803,7 @@ int kvm_mmu_unprotect_page_virt(struct kvm_vcpu *vcpu, gva_t gva)
        gpa_t gpa;
        int r;
 
-       down_read(&vcpu->kvm->slots_lock);
        gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, gva);
-       up_read(&vcpu->kvm->slots_lock);
 
        spin_lock(&vcpu->kvm->mmu_lock);
        r = kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT);
@@ -2063,7 +2056,7 @@ static int kvm_pv_mmu_write(struct kvm_vcpu *vcpu,
        if (r)
                return r;
 
-       if (!__emulator_write_phys(vcpu, addr, &value, bytes))
+       if (!emulator_write_phys(vcpu, addr, &value, bytes))
                return -EFAULT;
 
        return 1;
@@ -2127,7 +2120,6 @@ int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
        int r;
        struct kvm_pv_mmu_op_buffer buffer;
 
-       down_read(&vcpu->kvm->slots_lock);
        down_read(&current->mm->mmap_sem);
 
        buffer.ptr = buffer.buf;
@@ -2150,7 +2142,6 @@ int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
 out:
        *ret = buffer.processed;
        up_read(&current->mm->mmap_sem);
-       up_read(&vcpu->kvm->slots_lock);
        return r;
 }
 
index e9ae5db..57d872a 100644 (file)
@@ -388,7 +388,6 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
        if (r)
                return r;
 
-       down_read(&vcpu->kvm->slots_lock);
        /*
         * Look up the shadow pte for the faulting address.
         */
@@ -402,7 +401,6 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
                pgprintk("%s: guest page fault\n", __func__);
                inject_page_fault(vcpu, addr, walker.error_code);
                vcpu->arch.last_pt_write_count = 0; /* reset fork detector */
-               up_read(&vcpu->kvm->slots_lock);
                return 0;
        }
 
@@ -422,7 +420,6 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
        if (is_error_page(page)) {
                pgprintk("gfn %x is mmio\n", walker.gfn);
                kvm_release_page_clean(page);
-               up_read(&vcpu->kvm->slots_lock);
                return 1;
        }
 
@@ -440,7 +437,6 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
        ++vcpu->stat.pf_fixed;
        kvm_mmu_audit(vcpu, "post page fault (fixed)");
        spin_unlock(&vcpu->kvm->mmu_lock);
-       up_read(&vcpu->kvm->slots_lock);
 
        return write_pt;
 }
index 87eee7a..6249810 100644 (file)
@@ -1505,7 +1505,6 @@ static int init_rmode_tss(struct kvm *kvm)
        int ret = 0;
        int r;
 
-       down_read(&kvm->slots_lock);
        r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
        if (r < 0)
                goto out;
@@ -1528,7 +1527,6 @@ static int init_rmode_tss(struct kvm *kvm)
 
        ret = 1;
 out:
-       up_read(&kvm->slots_lock);
        return ret;
 }
 
@@ -1730,6 +1728,7 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
        u64 msr;
        int ret;
 
+       down_read(&vcpu->kvm->slots_lock);
        if (!init_rmode_tss(vmx->vcpu.kvm)) {
                ret = -ENOMEM;
                goto out;
@@ -1833,9 +1832,10 @@ static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
 
        vpid_sync_vcpu_all(vmx);
 
-       return 0;
+       ret = 0;
 
 out:
+       up_read(&vcpu->kvm->slots_lock);
        return ret;
 }
 
index 32d9100..e6a38bf 100644 (file)
@@ -201,7 +201,6 @@ int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
        int ret;
        u64 pdpte[ARRAY_SIZE(vcpu->arch.pdptrs)];
 
-       down_read(&vcpu->kvm->slots_lock);
        ret = kvm_read_guest_page(vcpu->kvm, pdpt_gfn, pdpte,
                                  offset * sizeof(u64), sizeof(pdpte));
        if (ret < 0) {
@@ -218,7 +217,6 @@ int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
 
        memcpy(vcpu->arch.pdptrs, pdpte, sizeof(vcpu->arch.pdptrs));
 out:
-       up_read(&vcpu->kvm->slots_lock);
 
        return ret;
 }
@@ -233,13 +231,11 @@ static bool pdptrs_changed(struct kvm_vcpu *vcpu)
        if (is_long_mode(vcpu) || !is_pae(vcpu))
                return false;
 
-       down_read(&vcpu->kvm->slots_lock);
        r = kvm_read_guest(vcpu->kvm, vcpu->arch.cr3 & ~31u, pdpte, sizeof(pdpte));
        if (r < 0)
                goto out;
        changed = memcmp(pdpte, vcpu->arch.pdptrs, sizeof(pdpte)) != 0;
 out:
-       up_read(&vcpu->kvm->slots_lock);
 
        return changed;
 }
@@ -377,7 +373,6 @@ void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
                 */
        }
 
-       down_read(&vcpu->kvm->slots_lock);
        /*
         * Does the new cr3 value map to physical memory? (Note, we
         * catch an invalid cr3 even in real-mode, because it would
@@ -393,7 +388,6 @@ void kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
                vcpu->arch.cr3 = cr3;
                vcpu->arch.mmu.new_cr3(vcpu);
        }
-       up_read(&vcpu->kvm->slots_lock);
 }
 EXPORT_SYMBOL_GPL(kvm_set_cr3);
 
@@ -503,7 +497,6 @@ static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 
        version++;
 
-       down_read(&kvm->slots_lock);
        kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 
        wc_ts = current_kernel_time();
@@ -515,7 +508,6 @@ static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 
        version++;
        kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
-       up_read(&kvm->slots_lock);
 }
 
 static void kvm_write_guest_time(struct kvm_vcpu *v)
@@ -609,10 +601,8 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
                vcpu->arch.hv_clock.tsc_shift = 22;
 
                down_read(&current->mm->mmap_sem);
-               down_read(&vcpu->kvm->slots_lock);
                vcpu->arch.time_page =
                                gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
-               up_read(&vcpu->kvm->slots_lock);
                up_read(&current->mm->mmap_sem);
 
                if (is_error_page(vcpu->arch.time_page)) {
@@ -715,9 +705,11 @@ static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
 
        vcpu_load(vcpu);
 
+       down_read(&vcpu->kvm->slots_lock);
        for (i = 0; i < msrs->nmsrs; ++i)
                if (do_msr(vcpu, entries[i].index, &entries[i].data))
                        break;
+       up_read(&vcpu->kvm->slots_lock);
 
        vcpu_put(vcpu);
 
@@ -1768,7 +1760,6 @@ int emulator_read_std(unsigned long addr,
        void *data = val;
        int r = X86EMUL_CONTINUE;
 
-       down_read(&vcpu->kvm->slots_lock);
        while (bytes) {
                gpa_t gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
                unsigned offset = addr & (PAGE_SIZE-1);
@@ -1790,7 +1781,6 @@ int emulator_read_std(unsigned long addr,
                addr += tocopy;
        }
 out:
-       up_read(&vcpu->kvm->slots_lock);
        return r;
 }
 EXPORT_SYMBOL_GPL(emulator_read_std);
@@ -1809,9 +1799,7 @@ static int emulator_read_emulated(unsigned long addr,
                return X86EMUL_CONTINUE;
        }
 
-       down_read(&vcpu->kvm->slots_lock);
        gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
-       up_read(&vcpu->kvm->slots_lock);
 
        /* For APIC access vmexit */
        if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
@@ -1844,7 +1832,7 @@ mmio:
        return X86EMUL_UNHANDLEABLE;
 }
 
-int __emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
+int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
                          const void *val, int bytes)
 {
        int ret;
@@ -1856,17 +1844,6 @@ int __emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
        return 1;
 }
 
-static int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
-                       const void *val, int bytes)
-{
-       int ret;
-
-       down_read(&vcpu->kvm->slots_lock);
-       ret =__emulator_write_phys(vcpu, gpa, val, bytes);
-       up_read(&vcpu->kvm->slots_lock);
-       return ret;
-}
-
 static int emulator_write_emulated_onepage(unsigned long addr,
                                           const void *val,
                                           unsigned int bytes,
@@ -1875,9 +1852,7 @@ static int emulator_write_emulated_onepage(unsigned long addr,
        struct kvm_io_device *mmio_dev;
        gpa_t                 gpa;
 
-       down_read(&vcpu->kvm->slots_lock);
        gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
-       up_read(&vcpu->kvm->slots_lock);
 
        if (gpa == UNMAPPED_GVA) {
                kvm_inject_page_fault(vcpu, addr, 2);
@@ -1954,7 +1929,6 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
                char *kaddr;
                u64 val;
 
-               down_read(&vcpu->kvm->slots_lock);
                gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, addr);
 
                if (gpa == UNMAPPED_GVA ||
@@ -1974,9 +1948,8 @@ static int emulator_cmpxchg_emulated(unsigned long addr,
                set_64bit((u64 *)(kaddr + offset_in_page(gpa)), val);
                kunmap_atomic(kaddr, KM_USER0);
                kvm_release_page_dirty(page);
-       emul_write:
-               up_read(&vcpu->kvm->slots_lock);
        }
+emul_write:
 #endif
 
        return emulator_write_emulated(addr, new, bytes, vcpu);
@@ -2368,10 +2341,8 @@ int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
                kvm_x86_ops->skip_emulated_instruction(vcpu);
 
        for (i = 0; i < nr_pages; ++i) {
-               down_read(&vcpu->kvm->slots_lock);
                page = gva_to_page(vcpu, address + i * PAGE_SIZE);
                vcpu->arch.pio.guest_pages[i] = page;
-               up_read(&vcpu->kvm->slots_lock);
                if (!page) {
                        kvm_inject_gp(vcpu, 0);
                        free_pio_guest_pages(vcpu);
@@ -2445,7 +2416,9 @@ int kvm_emulate_halt(struct kvm_vcpu *vcpu)
        ++vcpu->stat.halt_exits;
        if (irqchip_in_kernel(vcpu->kvm)) {
                vcpu->arch.mp_state = VCPU_MP_STATE_HALTED;
+               up_read(&vcpu->kvm->slots_lock);
                kvm_vcpu_block(vcpu);
+               down_read(&vcpu->kvm->slots_lock);
                if (vcpu->arch.mp_state != VCPU_MP_STATE_RUNNABLE)
                        return -EINTR;
                return 1;
@@ -2738,6 +2711,7 @@ static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
                vcpu->arch.mp_state = VCPU_MP_STATE_RUNNABLE;
        }
 
+       down_read(&vcpu->kvm->slots_lock);
        vapic_enter(vcpu);
 
 preempted:
@@ -2811,6 +2785,8 @@ again:
 
        kvm_lapic_sync_to_vapic(vcpu);
 
+       up_read(&vcpu->kvm->slots_lock);
+
        vcpu->guest_mode = 1;
        kvm_guest_enter();
 
@@ -2837,6 +2813,8 @@ again:
 
        preempt_enable();
 
+       down_read(&vcpu->kvm->slots_lock);
+
        /*
         * Profile KVM exit RIPs:
         */
@@ -2864,14 +2842,18 @@ again:
        }
 
 out:
+       up_read(&vcpu->kvm->slots_lock);
        if (r > 0) {
                kvm_resched(vcpu);
+               down_read(&vcpu->kvm->slots_lock);
                goto preempted;
        }
 
        post_kvm_run_save(vcpu, kvm_run);
 
+       down_read(&vcpu->kvm->slots_lock);
        vapic_exit(vcpu);
+       up_read(&vcpu->kvm->slots_lock);
 
        return r;
 }
@@ -2906,9 +2888,12 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
                memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
                vcpu->mmio_read_completed = 1;
                vcpu->mmio_needed = 0;
+
+               down_read(&vcpu->kvm->slots_lock);
                r = emulate_instruction(vcpu, kvm_run,
                                        vcpu->arch.mmio_fault_cr2, 0,
                                        EMULTYPE_NO_DECODE);
+               up_read(&vcpu->kvm->slots_lock);
                if (r == EMULATE_DO_MMIO) {
                        /*
                         * Read-modify-write.  Back to userspace.
@@ -3817,7 +3802,9 @@ fail:
 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 {
        kvm_free_lapic(vcpu);
+       down_read(&vcpu->kvm->slots_lock);
        kvm_mmu_destroy(vcpu);
+       up_read(&vcpu->kvm->slots_lock);
        free_page((unsigned long)vcpu->arch.pio_data);
 }
 
index 7b28cf9..2b081ed 100644 (file)
@@ -446,7 +446,7 @@ void kvm_mmu_change_mmu_pages(struct kvm *kvm, unsigned int kvm_nr_mmu_pages);
 
 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3);
 
-int __emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
+int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
                          const void *val, int bytes);
 int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
                  gpa_t addr, unsigned long *ret);