@@ -39,6 +39,7 @@ enum emulation_result {
extern int __kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu);
extern char kvmppc_handlers_start[];
extern unsigned long kvmppc_handler_len;
+extern void kvmppc_handler_highmem(void);
extern void kvmppc_dump_vcpu(struct kvm_vcpu *vcpu);
extern int kvmppc_handle_load(struct kvm_run *run, struct kvm_vcpu *vcpu,
new file mode 100644
@@ -0,0 +1,392 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ * Copyright SUSE Linux Products GmbH 2009
+ *
+ * Authors: Alexander Graf <agraf@suse.de>
+ */
+
+#include <asm/ppc_asm.h>
+#include <asm/kvm_asm.h>
+#include <asm/reg.h>
+#include <asm/page.h>
+#include <asm/asm-offsets.h>
+#include <asm/exception-64s.h>
+
+#define KVMPPC_HANDLE_EXIT .kvmppc_handle_exit
+#define ULONG_SIZE 8
+#define VCPU_GPR(n) (VCPU_GPRS + (n * ULONG_SIZE))
+
+.macro mfpaca tmp_reg, src_reg, offset, vcpu_reg
+ ld \tmp_reg, (PACA_EXMC+\offset)(r13)
+ std \tmp_reg, VCPU_GPR(\src_reg)(\vcpu_reg)
+.endm
+
+.macro DISABLE_INTERRUPTS
+ mfmsr r0
+ rldicl r0,r0,48,1
+ rotldi r0,r0,16
+ mtmsrd r0,1
+.endm
+
+/*****************************************************************************
+ * *
+ * Guest entry / exit code that is in kernel module memory (highmem) *
+ * *
+ ****************************************************************************/
+
+/* Registers:
+ * r3: kvm_run pointer
+ * r4: vcpu pointer
+ */
+_GLOBAL(__kvmppc_vcpu_entry)
+
+kvm_start_entry:
+ /* Write correct stack frame */
+ mflr r0
+ std r0,16(r1)
+
+ /* Save host state to the stack */
+ stdu r1, -SWITCH_FRAME_SIZE(r1)
+
+ /* Save r3 (kvm_run) and r4 (vcpu) */
+ SAVE_2GPRS(3, r1)
+
+ /* Save non-volatile registers (r14 - r31) */
+ SAVE_NVGPRS(r1)
+
+ /* Save LR */
+ mflr r14
+ std r14, _LINK(r1)
+
+/* XXX optimize non-volatile loading away */
+kvm_start_lightweight:
+
+ DISABLE_INTERRUPTS
+
+ /* Save R1/R2 in the PACA */
+ std r1, PACAR1(r13)
+ std r2, (PACA_EXMC+EX_SRR0)(r13)
+ ld r3, VCPU_HIGHMEM_HANDLER(r4)
+ std r3, PACASAVEDMSR(r13)
+
+ /* Load non-volatile guest state from the vcpu */
+ ld r14, VCPU_GPR(r14)(r4)
+ ld r15, VCPU_GPR(r15)(r4)
+ ld r16, VCPU_GPR(r16)(r4)
+ ld r17, VCPU_GPR(r17)(r4)
+ ld r18, VCPU_GPR(r18)(r4)
+ ld r19, VCPU_GPR(r19)(r4)
+ ld r20, VCPU_GPR(r20)(r4)
+ ld r21, VCPU_GPR(r21)(r4)
+ ld r22, VCPU_GPR(r22)(r4)
+ ld r23, VCPU_GPR(r23)(r4)
+ ld r24, VCPU_GPR(r24)(r4)
+ ld r25, VCPU_GPR(r25)(r4)
+ ld r26, VCPU_GPR(r26)(r4)
+ ld r27, VCPU_GPR(r27)(r4)
+ ld r28, VCPU_GPR(r28)(r4)
+ ld r29, VCPU_GPR(r29)(r4)
+ ld r30, VCPU_GPR(r30)(r4)
+ ld r31, VCPU_GPR(r31)(r4)
+
+ ld r9, VCPU_PC(r4) /* r9 = vcpu->arch.pc */
+ ld r10, VCPU_SHADOW_MSR(r4) /* r10 = vcpu->arch.shadow_msr */
+
+ ld r3, VCPU_TRAMPOLINE_ENTER(r4)
+ mtsrr0 r3
+
+ LOAD_REG_IMMEDIATE(r3, MSR_KERNEL & ~(MSR_IR | MSR_DR))
+ mtsrr1 r3
+
+ /* Load guest state in the respective registers */
+ lwz r3, VCPU_CR(r4) /* r3 = vcpu->arch.cr */
+ stw r3, (PACA_EXMC + EX_CCR)(r13)
+
+ ld r3, VCPU_CTR(r4) /* r3 = vcpu->arch.ctr */
+ mtctr r3 /* CTR = r3 */
+
+ ld r3, VCPU_LR(r4) /* r3 = vcpu->arch.lr */
+ mtlr r3 /* LR = r3 */
+
+ ld r3, VCPU_XER(r4) /* r3 = vcpu->arch.xer */
+ std r3, (PACA_EXMC + EX_R3)(r13)
+
+ /* Some guests may need to have dcbz set to 32 byte length.
+ *
+ * Usually we ensure that by patching the guest's instructions
+ * to trap on dcbz and emulate it in the hypervisor.
+ *
+ * If we can, we should tell the CPU to use 32 byte dcbz though,
+ * because that's a lot faster.
+ */
+
+ ld r3, VCPU_HFLAGS(r4)
+ rldicl. r3, r3, 0, 63 /* CR = ((r3 & 1) == 0) */
+ beq no_dcbz32_on
+
+ mfspr r3,SPRN_HID5
+ ori r3, r3, 0x80 /* XXX HID5_dcbz32 = 0x80 */
+ mtspr SPRN_HID5,r3
+
+no_dcbz32_on:
+ /* Load guest GPRs */
+
+ ld r3, VCPU_GPR(r9)(r4)
+ std r3, (PACA_EXMC + EX_R9)(r13)
+ ld r3, VCPU_GPR(r10)(r4)
+ std r3, (PACA_EXMC + EX_R10)(r13)
+ ld r3, VCPU_GPR(r11)(r4)
+ std r3, (PACA_EXMC + EX_R11)(r13)
+ ld r3, VCPU_GPR(r12)(r4)
+ std r3, (PACA_EXMC + EX_R12)(r13)
+ ld r3, VCPU_GPR(r13)(r4)
+ std r3, (PACA_EXMC + EX_R13)(r13)
+
+ ld r0, VCPU_GPR(r0)(r4)
+ ld r1, VCPU_GPR(r1)(r4)
+ ld r2, VCPU_GPR(r2)(r4)
+ ld r3, VCPU_GPR(r3)(r4)
+ ld r5, VCPU_GPR(r5)(r4)
+ ld r6, VCPU_GPR(r6)(r4)
+ ld r7, VCPU_GPR(r7)(r4)
+ ld r8, VCPU_GPR(r8)(r4)
+ ld r4, VCPU_GPR(r4)(r4)
+
+ /* This sets the Magic value for the trampoline */
+
+ li r11, 1
+ stb r11, PACA_KVM_IN_GUEST(r13)
+
+ /* Jump to SLB patching handlder and into our guest */
+ RFI
+
+/*
+ * This is the handler in module memory. It gets jumped at from the
+ * lowmem trampoline code, so it's basically the guest exit code.
+ *
+ */
+
+.global kvmppc_handler_highmem
+kvmppc_handler_highmem:
+
+ /*
+ * Register usage at this point:
+ *
+ * R00 = guest R13
+ * R01 = host R1
+ * R02 = host R2
+ * R10 = guest PC
+ * R11 = guest MSR
+ * R12 = exit handler id
+ * R13 = PACA
+ * PACA.exmc.R9 = guest R1
+ * PACA.exmc.R10 = guest R10
+ * PACA.exmc.R11 = guest R11
+ * PACA.exmc.R12 = guest R12
+ * PACA.exmc.R13 = guest R2
+ * PACA.exmc.DAR = guest DAR
+ * PACA.exmc.DSISR = guest DSISR
+ * PACA.exmc.LR = guest instruction
+ * PACA.exmc.CCR = guest CR
+ * PACA.exmc.SRR0 = guest R0
+ *
+ */
+
+ std r3, (PACA_EXMC+EX_R3)(r13)
+
+ /* save the exit id in R3 */
+ mr r3, r12
+
+ /* R12 = vcpu */
+ ld r12, GPR4(r1)
+
+ /* Now save the guest state */
+
+ std r0, VCPU_GPR(r13)(r12)
+ std r4, VCPU_GPR(r4)(r12)
+ std r5, VCPU_GPR(r5)(r12)
+ std r6, VCPU_GPR(r6)(r12)
+ std r7, VCPU_GPR(r7)(r12)
+ std r8, VCPU_GPR(r8)(r12)
+ std r9, VCPU_GPR(r9)(r12)
+
+ /* get registers from PACA */
+ mfpaca r5, r0, EX_SRR0, r12
+ mfpaca r5, r3, EX_R3, r12
+ mfpaca r5, r1, EX_R9, r12
+ mfpaca r5, r10, EX_R10, r12
+ mfpaca r5, r11, EX_R11, r12
+ mfpaca r5, r12, EX_R12, r12
+ mfpaca r5, r2, EX_R13, r12
+
+ lwz r5, (PACA_EXMC+EX_LR)(r13)
+ stw r5, VCPU_LAST_INST(r12)
+
+ lwz r5, (PACA_EXMC+EX_CCR)(r13)
+ stw r5, VCPU_CR(r12)
+
+ ld r5, VCPU_HFLAGS(r12)
+ rldicl. r5, r5, 0, 63 /* CR = ((r5 & 1) == 0) */
+ beq no_dcbz32_off
+
+ mfspr r5,SPRN_HID5
+ rldimi r5,r5,6,56
+ mtspr SPRN_HID5,r5
+
+no_dcbz32_off:
+
+ /* XXX maybe skip on lightweight? */
+ std r14, VCPU_GPR(r14)(r12)
+ std r15, VCPU_GPR(r15)(r12)
+ std r16, VCPU_GPR(r16)(r12)
+ std r17, VCPU_GPR(r17)(r12)
+ std r18, VCPU_GPR(r18)(r12)
+ std r19, VCPU_GPR(r19)(r12)
+ std r20, VCPU_GPR(r20)(r12)
+ std r21, VCPU_GPR(r21)(r12)
+ std r22, VCPU_GPR(r22)(r12)
+ std r23, VCPU_GPR(r23)(r12)
+ std r24, VCPU_GPR(r24)(r12)
+ std r25, VCPU_GPR(r25)(r12)
+ std r26, VCPU_GPR(r26)(r12)
+ std r27, VCPU_GPR(r27)(r12)
+ std r28, VCPU_GPR(r28)(r12)
+ std r29, VCPU_GPR(r29)(r12)
+ std r30, VCPU_GPR(r30)(r12)
+ std r31, VCPU_GPR(r31)(r12)
+
+ /* Restore non-volatile host registers (r14 - r31) */
+ REST_NVGPRS(r1)
+
+ /* Save guest PC (R10) */
+ std r10, VCPU_PC(r12)
+
+ /* Save guest msr (R11) */
+ std r11, VCPU_SHADOW_MSR(r12)
+
+ /* Save guest CTR (in R12) */
+ mfctr r5
+ std r5, VCPU_CTR(r12)
+
+ /* Save guest LR */
+ mflr r5
+ std r5, VCPU_LR(r12)
+
+ /* Save guest XER */
+ mfxer r5
+ std r5, VCPU_XER(r12)
+
+ /* Save guest DAR */
+ ld r5, (PACA_EXMC+EX_DAR)(r13)
+ std r5, VCPU_FAULT_DEAR(r12)
+
+ /* Save guest DSISR */
+ lwz r5, (PACA_EXMC+EX_DSISR)(r13)
+ std r5, VCPU_FAULT_DSISR(r12)
+
+ /* Restore host msr -> SRR1 */
+ ld r7, VCPU_HOST_MSR(r12)
+ mtsrr1 r7
+
+ /* Restore host IP -> SRR0 */
+ ld r6, VCPU_HOST_RETIP(r12)
+ mtsrr0 r6
+
+ /*
+ * For some interrupts, we need to call the real Linux
+ * handler, so it can do work for us. This has to happen
+ * as if the interrupt arrived from the kernel though,
+ * so let's fake it here where most state is restored.
+ *
+ * Call Linux for hardware interrupts/decrementer
+ * r3 = address of interrupt handler (exit reason)
+ */
+
+ cmpwi r3, BOOK3S_INTERRUPT_EXTERNAL
+ beq call_linux_handler
+ cmpwi r3, BOOK3S_INTERRUPT_DECREMENTER
+ beq call_linux_handler
+
+ /* Back to Interruptable Mode! (goto kvm_return_point) */
+ RFI
+
+call_linux_handler:
+
+ /*
+ * If we land here we need to jump back to the handler we
+ * came from.
+ *
+ * We have a page that we can access from real mode, so let's
+ * jump back to that and use it as a trampoline to get back into the
+ * interrupt handler!
+ *
+ * R3 still contains the exit code,
+ * R6 VCPU_HOST_RETIP and
+ * R7 VCPU_HOST_MSR
+ */
+
+ mtlr r3
+
+ ld r5, VCPU_TRAMPOLINE_LOWMEM(r12)
+ mtsrr0 r5
+ LOAD_REG_IMMEDIATE(r5, MSR_KERNEL & ~(MSR_IR | MSR_DR))
+ mtsrr1 r5
+
+ RFI
+
+.global kvm_return_point
+kvm_return_point:
+
+ /* Jump back to lightweight entry if we're supposed to */
+ /* go back into the guest */
+ mr r5, r3
+ /* Restore r3 (kvm_run) and r4 (vcpu) */
+ REST_2GPRS(3, r1)
+ bl KVMPPC_HANDLE_EXIT
+
+#if 0 /* XXX get lightweight exits back */
+ cmpwi r3, RESUME_GUEST
+ bne kvm_exit_heavyweight
+
+ /* put VCPU and KVM_RUN back into place and roll again! */
+ REST_2GPRS(3, r1)
+ b kvm_start_lightweight
+
+kvm_exit_heavyweight:
+ /* Restore non-volatile host registers */
+ ld r14, _LINK(r1)
+ mtlr r14
+ REST_NVGPRS(r1)
+
+ addi r1, r1, SWITCH_FRAME_SIZE
+#else
+ ld r4, _LINK(r1)
+ mtlr r4
+
+ cmpwi r3, RESUME_GUEST
+ bne kvm_exit_heavyweight
+
+ REST_2GPRS(3, r1)
+
+ addi r1, r1, SWITCH_FRAME_SIZE
+
+ b kvm_start_entry
+
+kvm_exit_heavyweight:
+
+ addi r1, r1, SWITCH_FRAME_SIZE
+#endif
+
+ blr