arm64: kvm: allows kvm cpu hotplug
authorAKASHI Takahiro <takahiro.akashi@linaro.org>
Wed, 27 Apr 2016 16:47:05 +0000 (17:47 +0100)
committerAlex Shi <alex.shi@linaro.org>
Thu, 17 Nov 2016 01:49:39 +0000 (09:49 +0800)
The current kvm implementation on arm64 does cpu-specific initialization
at system boot, and has no way to gracefully shutdown a core in terms of
kvm. This prevents kexec from rebooting the system at EL2.

This patch adds a cpu tear-down function and also puts an existing cpu-init
code into a separate function, kvm_arch_hardware_disable() and
kvm_arch_hardware_enable() respectively.
We don't need the arm64 specific cpu hotplug hook any more.

Since this patch modifies common code between arm and arm64, one stub
definition, __cpu_reset_hyp_mode(), is added on arm side to avoid
compilation errors.

Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
[Rebase, added separate VHE init/exit path, changed resets use of
 kvm_call_hyp() to the __version, en/disabled hardware in init_subsystems(),
 added icache maintenance to __kvm_hyp_reset() and removed lr restore, removed
 guest-enter after teardown handling]
Signed-off-by: James Morse <james.morse@arm.com>
Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
(cherry picked from commit 67f6919766620e7ea7aab11a6a3470dc7b451359)
Signed-off-by: Alex Shi <alex.shi@linaro.org>
Conflicts:
arch/arm64/include/asm/kvm_host.h

arch/arm/include/asm/kvm_host.h
arch/arm/include/asm/kvm_mmu.h
arch/arm/kvm/arm.c
arch/arm/kvm/mmu.c
arch/arm64/include/asm/kvm_asm.h
arch/arm64/include/asm/kvm_host.h
arch/arm64/include/asm/kvm_mmu.h
arch/arm64/kvm/hyp-init.S
arch/arm64/kvm/reset.c

index 945bfa5e7752057a0161461ceda83843e98788fa..bedaf65c0ff96fa3f7d13938f59ed1faa9c94033 100644 (file)
@@ -218,6 +218,15 @@ static inline void __cpu_init_stage2(void)
 {
 }
 
+static inline void __cpu_reset_hyp_mode(phys_addr_t boot_pgd_ptr,
+                                       phys_addr_t phys_idmap_start)
+{
+       /*
+        * TODO
+        * kvm_call_reset(boot_pgd_ptr, phys_idmap_start);
+        */
+}
+
 static inline int kvm_arch_dev_ioctl_check_extension(long ext)
 {
        return 0;
@@ -230,7 +239,6 @@ void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot);
 
 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr);
 
-static inline void kvm_arch_hardware_disable(void) {}
 static inline void kvm_arch_hardware_unsetup(void) {}
 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
 static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}
index 9203c21b4673fd8a73b5f5797ed5190305b363ae..c7ba9a42e85720dcbf78dae2ef3698f8e7b0c851 100644 (file)
@@ -66,6 +66,7 @@ void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
 phys_addr_t kvm_mmu_get_httbr(void);
 phys_addr_t kvm_mmu_get_boot_httbr(void);
 phys_addr_t kvm_get_idmap_vector(void);
+phys_addr_t kvm_get_idmap_start(void);
 int kvm_mmu_init(void);
 void kvm_clear_hyp_idmap(void);
 
index 99b270300ab1f059810a367d4df30367a7959fa3..4cddf20cdb824394d5368bb5cd10a4eadff2d06c 100644 (file)
@@ -16,7 +16,6 @@
  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  */
 
-#include <linux/cpu.h>
 #include <linux/cpu_pm.h>
 #include <linux/errno.h>
 #include <linux/err.h>
@@ -65,6 +64,8 @@ static DEFINE_SPINLOCK(kvm_vmid_lock);
 
 static bool vgic_present;
 
+static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
+
 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
 {
        BUG_ON(preemptible());
@@ -89,11 +90,6 @@ struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
        return &kvm_arm_running_vcpu;
 }
 
-int kvm_arch_hardware_enable(void)
-{
-       return 0;
-}
-
 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
 {
        return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
@@ -964,11 +960,6 @@ long kvm_arch_vm_ioctl(struct file *filp,
        }
 }
 
-static void cpu_init_stage2(void *dummy)
-{
-       __cpu_init_stage2();
-}
-
 static void cpu_init_hyp_mode(void *dummy)
 {
        phys_addr_t boot_pgd_ptr;
@@ -996,43 +987,87 @@ static void cpu_hyp_reinit(void)
 {
        if (is_kernel_in_hyp_mode()) {
                /*
-                * cpu_init_stage2() is safe to call even if the PM
+                * __cpu_init_stage2() is safe to call even if the PM
                 * event was cancelled before the CPU was reset.
                 */
-               cpu_init_stage2(NULL);
+               __cpu_init_stage2();
        } else {
                if (__hyp_get_vectors() == hyp_default_vectors)
                        cpu_init_hyp_mode(NULL);
        }
 }
 
-static int hyp_init_cpu_notify(struct notifier_block *self,
-                              unsigned long action, void *cpu)
+static void cpu_hyp_reset(void)
+{
+       phys_addr_t boot_pgd_ptr;
+       phys_addr_t phys_idmap_start;
+
+       if (!is_kernel_in_hyp_mode()) {
+               boot_pgd_ptr = kvm_mmu_get_boot_httbr();
+               phys_idmap_start = kvm_get_idmap_start();
+
+               __cpu_reset_hyp_mode(boot_pgd_ptr, phys_idmap_start);
+       }
+}
+
+static void _kvm_arch_hardware_enable(void *discard)
 {
-       switch (action) {
-       case CPU_STARTING:
-       case CPU_STARTING_FROZEN:
+       if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
                cpu_hyp_reinit();
+               __this_cpu_write(kvm_arm_hardware_enabled, 1);
        }
+}
+
+int kvm_arch_hardware_enable(void)
+{
+       _kvm_arch_hardware_enable(NULL);
+       return 0;
+}
 
-       return NOTIFY_OK;
+static void _kvm_arch_hardware_disable(void *discard)
+{
+       if (__this_cpu_read(kvm_arm_hardware_enabled)) {
+               cpu_hyp_reset();
+               __this_cpu_write(kvm_arm_hardware_enabled, 0);
+       }
 }
 
-static struct notifier_block hyp_init_cpu_nb = {
-       .notifier_call = hyp_init_cpu_notify,
-};
+void kvm_arch_hardware_disable(void)
+{
+       _kvm_arch_hardware_disable(NULL);
+}
 
 #ifdef CONFIG_CPU_PM
 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
                                    unsigned long cmd,
                                    void *v)
 {
-       if (cmd == CPU_PM_EXIT) {
-               cpu_hyp_reinit();
+       /*
+        * kvm_arm_hardware_enabled is left with its old value over
+        * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
+        * re-enable hyp.
+        */
+       switch (cmd) {
+       case CPU_PM_ENTER:
+               if (__this_cpu_read(kvm_arm_hardware_enabled))
+                       /*
+                        * don't update kvm_arm_hardware_enabled here
+                        * so that the hardware will be re-enabled
+                        * when we resume. See below.
+                        */
+                       cpu_hyp_reset();
+
+               return NOTIFY_OK;
+       case CPU_PM_EXIT:
+               if (__this_cpu_read(kvm_arm_hardware_enabled))
+                       /* The hardware was enabled before suspend. */
+                       cpu_hyp_reinit();
+
                return NOTIFY_OK;
-       }
 
-       return NOTIFY_DONE;
+       default:
+               return NOTIFY_DONE;
+       }
 }
 
 static struct notifier_block hyp_init_cpu_pm_nb = {
@@ -1067,18 +1102,12 @@ static int init_common_resources(void)
 
 static int init_subsystems(void)
 {
-       int err;
+       int err = 0;
 
        /*
-        * Register CPU Hotplug notifier
+        * Enable hardware so that subsystem initialisation can access EL2.
         */
-       cpu_notifier_register_begin();
-       err = __register_cpu_notifier(&hyp_init_cpu_nb);
-       cpu_notifier_register_done();
-       if (err) {
-               kvm_err("Cannot register KVM init CPU notifier (%d)\n", err);
-               return err;
-       }
+       on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
 
        /*
         * Register CPU lower-power notifier
@@ -1096,9 +1125,10 @@ static int init_subsystems(void)
        case -ENODEV:
        case -ENXIO:
                vgic_present = false;
+               err = 0;
                break;
        default:
-               return err;
+               goto out;
        }
 
        /*
@@ -1106,12 +1136,15 @@ static int init_subsystems(void)
         */
        err = kvm_timer_hyp_init();
        if (err)
-               return err;
+               goto out;
 
        kvm_perf_init();
        kvm_coproc_table_init();
 
-       return 0;
+out:
+       on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
+
+       return err;
 }
 
 static void teardown_hyp_mode(void)
@@ -1128,11 +1161,6 @@ static void teardown_hyp_mode(void)
 
 static int init_vhe_mode(void)
 {
-       /*
-        * Execute the init code on each CPU.
-        */
-       on_each_cpu(cpu_init_stage2, NULL, 1);
-
        /* set size of VMID supported by CPU */
        kvm_vmid_bits = kvm_get_vmid_bits();
        kvm_info("%d-bit VMID\n", kvm_vmid_bits);
@@ -1219,11 +1247,6 @@ static int init_hyp_mode(void)
                }
        }
 
-       /*
-        * Execute the init code on each CPU.
-        */
-       on_each_cpu(cpu_init_hyp_mode, NULL, 1);
-
 #ifndef CONFIG_HOTPLUG_CPU
        free_boot_hyp_pgd();
 #endif
index bf3697c1c1119be4f9f61e4e96ffa62e36cbcb38..767872411d97bbafddd9d4ee7c255db61a36c5c6 100644 (file)
@@ -1655,6 +1655,11 @@ phys_addr_t kvm_get_idmap_vector(void)
        return hyp_idmap_vector;
 }
 
+phys_addr_t kvm_get_idmap_start(void)
+{
+       return hyp_idmap_start;
+}
+
 int kvm_mmu_init(void)
 {
        int err;
index edb51b81ea06888423adca9d930a72e183299635..fca51486cd3700ba7df7efa6c8c70602bcdd9996 100644 (file)
@@ -51,6 +51,7 @@ struct kvm_vcpu;
 
 extern char __kvm_hyp_init[];
 extern char __kvm_hyp_init_end[];
+extern char __kvm_hyp_reset[];
 
 extern char __kvm_hyp_vector[];
 
index bbdaa56c22244cfcc119ea403ab813424c8fed6c..3be7a7b52d809fa6eee7a3f1b4d2ec516b5697f4 100644 (file)
@@ -44,6 +44,7 @@
 int __attribute_const__ kvm_target_cpu(void);
 int kvm_reset_vcpu(struct kvm_vcpu *vcpu);
 int kvm_arch_dev_ioctl_check_extension(long ext);
+phys_addr_t kvm_hyp_reset_entry(void);
 
 struct kvm_arch {
        /* The VMID generation used for the virt. memory system */
@@ -330,7 +331,17 @@ static inline void __cpu_init_stage2(void)
 {
 }
 
-static inline void kvm_arch_hardware_disable(void) {}
+static inline void __cpu_reset_hyp_mode(phys_addr_t boot_pgd_ptr,
+                                       phys_addr_t phys_idmap_start)
+{
+       /*
+        * Call reset code, and switch back to stub hyp vectors.
+        * Uses __kvm_call_hyp() to avoid kaslr's kvm_ksym_ref() translation.
+        */
+       __kvm_call_hyp((void *)kvm_hyp_reset_entry(),
+                      boot_pgd_ptr, phys_idmap_start);
+}
+
 static inline void kvm_arch_hardware_unsetup(void) {}
 static inline void kvm_arch_sync_events(struct kvm *kvm) {}
 static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}
index 0bf8b4320a9154fda9e6fd3d999ef37a356bee85..342a5ac2f3da238da956eae467a37558a672c07a 100644 (file)
@@ -99,6 +99,7 @@ void kvm_mmu_free_memory_caches(struct kvm_vcpu *vcpu);
 phys_addr_t kvm_mmu_get_httbr(void);
 phys_addr_t kvm_mmu_get_boot_httbr(void);
 phys_addr_t kvm_get_idmap_vector(void);
+phys_addr_t kvm_get_idmap_start(void);
 int kvm_mmu_init(void);
 void kvm_clear_hyp_idmap(void);
 
index 034d152c3fbe7a6afce7f7ee827b1d6825b84cde..d87635e678b7edf2ea18a6a038d2db1dba802d8c 100644 (file)
@@ -152,6 +152,44 @@ merged:
        eret
 ENDPROC(__kvm_hyp_init)
 
+       /*
+        * x0: HYP boot pgd
+        * x1: HYP phys_idmap_start
+        */
+ENTRY(__kvm_hyp_reset)
+       /* We're in trampoline code in VA, switch back to boot page tables */
+       msr     ttbr0_el2, x0
+       isb
+
+       /* Ensure the PA branch doesn't find a stale tlb entry or stale code. */
+       ic      iallu
+       tlbi    alle2
+       dsb     sy
+       isb
+
+       /* Branch into PA space */
+       adr     x0, 1f
+       bfi     x1, x0, #0, #PAGE_SHIFT
+       br      x1
+
+       /* We're now in idmap, disable MMU */
+1:     mrs     x0, sctlr_el2
+       ldr     x1, =SCTLR_ELx_FLAGS
+       bic     x0, x0, x1              // Clear SCTL_M and etc
+       msr     sctlr_el2, x0
+       isb
+
+       /* Invalidate the old TLBs */
+       tlbi    alle2
+       dsb     sy
+
+       /* Install stub vectors */
+       adr_l   x0, __hyp_stub_vectors
+       msr     vbar_el2, x0
+
+       eret
+ENDPROC(__kvm_hyp_reset)
+
        .ltorg
 
        .popsection
index f34745cb3d236fe0a4731f8d02031f8ff764d69c..d6e155a212dc42fa9696879c710a395fc8be2779 100644 (file)
@@ -29,7 +29,9 @@
 #include <asm/cputype.h>
 #include <asm/ptrace.h>
 #include <asm/kvm_arm.h>
+#include <asm/kvm_asm.h>
 #include <asm/kvm_coproc.h>
+#include <asm/kvm_mmu.h>
 
 /*
  * ARMv8 Reset Values
@@ -123,3 +125,15 @@ int kvm_reset_vcpu(struct kvm_vcpu *vcpu)
        /* Reset timer */
        return kvm_timer_vcpu_reset(vcpu, cpu_vtimer_irq);
 }
+
+extern char __hyp_idmap_text_start[];
+
+phys_addr_t kvm_hyp_reset_entry(void)
+{
+       unsigned long offset;
+
+       offset = (unsigned long)__kvm_hyp_reset
+                - ((unsigned long)__hyp_idmap_text_start & PAGE_MASK);
+
+       return TRAMPOLINE_VA + offset;
+}