@@ -38,6 +38,16 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
vcpu->arch.hcr = HCR_GUEST_MASK;
}
+static inline unsigned long vcpu_get_hcr(struct kvm_vcpu *vcpu)
+{
+ return vcpu->arch.hcr;
+}
+
+static inline void vcpu_set_hcr(struct kvm_vcpu *vcpu, unsigned long hcr)
+{
+ vcpu->arch.hcr = hcr;
+}
+
static inline bool vcpu_mode_is_32bit(struct kvm_vcpu *vcpu)
{
return 1;
@@ -125,9 +125,6 @@ struct kvm_vcpu_arch {
* Anything that is not used directly from assembly code goes
* here.
*/
- /* dcache set/way operation pending */
- int last_pcpu;
- cpumask_t require_dcache_flush;
/* Don't run the guest on this vcpu */
bool pause;
@@ -190,7 +190,8 @@ static inline void coherent_cache_guest_page(struct kvm_vcpu *vcpu, hva_t hva,
#define kvm_virt_to_phys(x) virt_to_idmap((unsigned long)(x))
-void stage2_flush_vm(struct kvm *kvm);
+void kvm_set_way_flush(struct kvm_vcpu *vcpu);
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
#endif /* !__ASSEMBLY__ */
@@ -281,15 +281,6 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
vcpu->cpu = cpu;
vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
- /*
- * Check whether this vcpu requires the cache to be flushed on
- * this physical CPU. This is a consequence of doing dcache
- * operations by set/way on this vcpu. We do it here to be in
- * a non-preemptible section.
- */
- if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
- flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
-
kvm_arm_set_running_vcpu(vcpu);
}
@@ -541,7 +532,6 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
vcpu->mode = OUTSIDE_GUEST_MODE;
- vcpu->arch.last_pcpu = smp_processor_id();
kvm_guest_exit();
trace_kvm_exit(*vcpu_pc(vcpu));
/*
@@ -189,44 +189,17 @@ static bool access_l2ectlr(struct kvm_vcpu *vcpu,
return true;
}
-/* See note at ARM ARM B1.14.4 */
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ */
static bool access_dcsw(struct kvm_vcpu *vcpu,
const struct coproc_params *p,
const struct coproc_reg *r)
{
- unsigned long val;
- int cpu;
-
if (!p->is_write)
return read_from_write_only(vcpu, p);
- cpu = get_cpu();
-
- cpumask_setall(&vcpu->arch.require_dcache_flush);
- cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
-
- /* If we were already preempted, take the long way around */
- if (cpu != vcpu->arch.last_pcpu) {
- flush_cache_all();
- goto done;
- }
-
- val = *vcpu_reg(vcpu, p->Rt1);
-
- switch (p->CRm) {
- case 6: /* Upgrade DCISW to DCCISW, as per HCR.SWIO */
- case 14: /* DCCISW */
- asm volatile("mcr p15, 0, %0, c7, c14, 2" : : "r" (val));
- break;
-
- case 10: /* DCCSW */
- asm volatile("mcr p15, 0, %0, c7, c10, 2" : : "r" (val));
- break;
- }
-
-done:
- put_cpu();
-
+ kvm_set_way_flush(vcpu);
return true;
}
@@ -258,13 +231,11 @@ bool access_sctlr(struct kvm_vcpu *vcpu,
const struct coproc_params *p,
const struct coproc_reg *r)
{
- access_vm_reg(vcpu, p, r);
+ bool was_enabled = vcpu_has_cache_enabled(vcpu);
- if (vcpu_has_cache_enabled(vcpu)) { /* MMU+Caches enabled? */
- vcpu->arch.hcr &= ~HCR_TVM;
- stage2_flush_vm(vcpu->kvm);
- }
+ access_vm_reg(vcpu, p, r);
+ kvm_toggle_cache(vcpu, was_enabled);
return true;
}
@@ -278,7 +278,7 @@ static void stage2_flush_memslot(struct kvm *kvm,
* Go through the stage 2 page tables and invalidate any cache lines
* backing memory already mapped to the VM.
*/
-void stage2_flush_vm(struct kvm *kvm)
+static void stage2_flush_vm(struct kvm *kvm)
{
struct kvm_memslots *slots;
struct kvm_memory_slot *memslot;
@@ -1411,3 +1411,71 @@ void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
unmap_stage2_range(kvm, gpa, size);
spin_unlock(&kvm->mmu_lock);
}
+
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ *
+ * Main problems:
+ * - S/W ops are local to a CPU (not broadcast)
+ * - We have line migration behind our back (speculation)
+ * - System caches don't support S/W at all (damn!)
+ *
+ * In the face of the above, the best we can do is to try and convert
+ * S/W ops to VA ops. Because the guest is not allowed to infer the
+ * S/W to PA mapping, it can only use S/W to nuke the whole cache,
+ * which is a rather good thing for us.
+ *
+ * Also, it is only used when turning caches on/off ("The expected
+ * usage of the cache maintenance instructions that operate by set/way
+ * is associated with the cache maintenance instructions associated
+ * with the powerdown and powerup of caches, if this is required by
+ * the implementation.").
+ *
+ * We use the following policy:
+ *
+ * - If we trap a S/W operation, we enable VM trapping to detect
+ * caches being turned on/off.
+ *
+ * - If the caches have already been turned off when doing the S/W op,
+ * we nuke the whole VM cache.
+ *
+ * - We flush the cache on both caches being turned on and off.
+ *
+ * - Once the caches are enabled, we stop trapping VM ops.
+ */
+void kvm_set_way_flush(struct kvm_vcpu *vcpu)
+{
+ unsigned long hcr;
+
+ /*
+ * If this is the first time we do a S/W operation
+ * (i.e. HCR_TVM not set) and the caches are already disabled,
+ * flush the whole memory.
+ *
+ * Otherwise, just enable trapping, and wait for the MMU +
+ * Caches to be turned off. At that point, we'll be able to
+ * actually clean the caches.
+ */
+ hcr = vcpu_get_hcr(vcpu);
+ if (!(hcr & HCR_TVM) && !vcpu_has_cache_enabled(vcpu))
+ stage2_flush_vm(vcpu->kvm);
+
+ vcpu_set_hcr(vcpu, hcr | HCR_TVM);
+}
+
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
+{
+ bool now_enabled = vcpu_has_cache_enabled(vcpu);
+
+ /*
+ * If switching the MMU+caches on, need to invalidate the caches.
+ * If switching it off, need to clean the caches.
+ * Clean + invalidate does the trick always.
+ */
+ if (now_enabled != was_enabled)
+ stage2_flush_vm(vcpu->kvm);
+
+ /* Caches are now on, stop trapping VM ops (until a S/W op) */
+ if (now_enabled)
+ vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM);
+}
@@ -45,6 +45,16 @@ static inline void vcpu_reset_hcr(struct kvm_vcpu *vcpu)
vcpu->arch.hcr_el2 &= ~HCR_RW;
}
+static inline unsigned long vcpu_get_hcr(struct kvm_vcpu *vcpu)
+{
+ return vcpu->arch.hcr_el2;
+}
+
+static inline void vcpu_set_hcr(struct kvm_vcpu *vcpu, unsigned long hcr)
+{
+ vcpu->arch.hcr_el2 = hcr;
+}
+
static inline unsigned long *vcpu_pc(const struct kvm_vcpu *vcpu)
{
return (unsigned long *)&vcpu_gp_regs(vcpu)->regs.pc;
@@ -116,9 +116,6 @@ struct kvm_vcpu_arch {
* Anything that is not used directly from assembly code goes
* here.
*/
- /* dcache set/way operation pending */
- int last_pcpu;
- cpumask_t require_dcache_flush;
/* Don't run the guest */
bool pause;
@@ -260,7 +260,8 @@ static inline void coherent_cache_guest_page(struct kvm_vcpu *vcpu, hva_t hva,
#define kvm_virt_to_phys(x) __virt_to_phys((unsigned long)(x))
-void stage2_flush_vm(struct kvm *kvm);
+void kvm_set_way_flush(struct kvm_vcpu *vcpu);
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled);
#endif /* __ASSEMBLY__ */
#endif /* __ARM64_KVM_MMU_H__ */
@@ -69,56 +69,17 @@ static u32 get_ccsidr(u32 csselr)
return ccsidr;
}
-static void do_dc_cisw(u32 val)
-{
- asm volatile("dc cisw, %x0" : : "r" (val));
- dsb(ish);
-}
-
-static void do_dc_csw(u32 val)
-{
- asm volatile("dc csw, %x0" : : "r" (val));
- dsb(ish);
-}
-
-/* See note at ARM ARM B1.14.4 */
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ */
static bool access_dcsw(struct kvm_vcpu *vcpu,
const struct sys_reg_params *p,
const struct sys_reg_desc *r)
{
- unsigned long val;
- int cpu;
-
if (!p->is_write)
return read_from_write_only(vcpu, p);
- cpu = get_cpu();
-
- cpumask_setall(&vcpu->arch.require_dcache_flush);
- cpumask_clear_cpu(cpu, &vcpu->arch.require_dcache_flush);
-
- /* If we were already preempted, take the long way around */
- if (cpu != vcpu->arch.last_pcpu) {
- flush_cache_all();
- goto done;
- }
-
- val = *vcpu_reg(vcpu, p->Rt);
-
- switch (p->CRm) {
- case 6: /* Upgrade DCISW to DCCISW, as per HCR.SWIO */
- case 14: /* DCCISW */
- do_dc_cisw(val);
- break;
-
- case 10: /* DCCSW */
- do_dc_csw(val);
- break;
- }
-
-done:
- put_cpu();
-
+ kvm_set_way_flush(vcpu);
return true;
}
@@ -155,13 +116,11 @@ static bool access_sctlr(struct kvm_vcpu *vcpu,
const struct sys_reg_params *p,
const struct sys_reg_desc *r)
{
- access_vm_reg(vcpu, p, r);
+ bool was_enabled = vcpu_has_cache_enabled(vcpu);
- if (vcpu_has_cache_enabled(vcpu)) { /* MMU+Caches enabled? */
- vcpu->arch.hcr_el2 &= ~HCR_TVM;
- stage2_flush_vm(vcpu->kvm);
- }
+ access_vm_reg(vcpu, p, r);
+ kvm_toggle_cache(vcpu, was_enabled);
return true;
}
Trying to emulate the behaviour of set/way cache ops is fairly pointless, as there are too many ways we can end-up missing stuff. Also, there is some system caches out there that simply ignore set/way operations. So instead of trying to implement them, let's convert it to VA ops, and use them as a way to re-enable the trapping of VM ops. That way, we can detect the point when the MMU/caches are turned off, and do a full VM flush (which is what the guest was trying to do anyway). This allows a 32bit zImage to boot on the APM thingy, and will probably help bootloaders in general. Signed-off-by: Marc Zyngier <marc.zyngier@arm.com> --- arch/arm/include/asm/kvm_emulate.h | 10 ++++++ arch/arm/include/asm/kvm_host.h | 3 -- arch/arm/include/asm/kvm_mmu.h | 3 +- arch/arm/kvm/arm.c | 10 ------ arch/arm/kvm/coproc.c | 43 ++++------------------ arch/arm/kvm/mmu.c | 70 +++++++++++++++++++++++++++++++++++- arch/arm64/include/asm/kvm_emulate.h | 10 ++++++ arch/arm64/include/asm/kvm_host.h | 3 -- arch/arm64/include/asm/kvm_mmu.h | 3 +- arch/arm64/kvm/sys_regs.c | 55 ++++------------------------ 10 files changed, 107 insertions(+), 103 deletions(-)