On 08.01.2014, at 11:25, Paul Mackerras <paulus@xxxxxxxxx> wrote:
> This adds saving of the transactional memory (TM) checkpointed state
> on guest entry and exit. We only do this if we see that the guest has
> an active transaction.
>
> It also adds emulation of the TM state changes when delivering IRQs
> into the guest. According to the architecture, if we are
> transactional when an IRQ occurs the TM state is changed to suspended,
> otherwise it's left unchanged.
>
> Signed-off-by: Michael Neuling <mikey@xxxxxxxxxxx>
> Signed-off-by: Paul Mackerras <paulus@xxxxxxxxx>
> ---
> arch/powerpc/kvm/book3s_64_mmu_hv.c | 9 +-
> arch/powerpc/kvm/book3s_hv_rmhandlers.S | 237 +++++++++++++++++++-------------
> 2 files changed, 146 insertions(+), 100 deletions(-)
>
> diff --git a/arch/powerpc/kvm/book3s_64_mmu_hv.c b/arch/powerpc/kvm/book3s_64_mmu_hv.c
> index 79e992d..ef971a5 100644
> --- a/arch/powerpc/kvm/book3s_64_mmu_hv.c
> +++ b/arch/powerpc/kvm/book3s_64_mmu_hv.c
> @@ -262,7 +262,14 @@ int kvmppc_mmu_hv_init(void)
>
> static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
> {
> - kvmppc_set_msr(vcpu, vcpu->arch.intr_msr);
> + unsigned long msr = vcpu->arch.intr_msr;
> +
> + /* If transactional, change to suspend mode on IRQ delivery */
> + if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
> + msr |= MSR_TS_S;
> + else
> + msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
> + kvmppc_set_msr(vcpu, msr);
> }
>
> /*
> diff --git a/arch/powerpc/kvm/book3s_hv_rmhandlers.S b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
> index 298d1755..f7bf681 100644
> --- a/arch/powerpc/kvm/book3s_hv_rmhandlers.S
> +++ b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
> @@ -28,6 +28,9 @@
> #include <asm/exception-64s.h>
> #include <asm/kvm_book3s_asm.h>
> #include <asm/mmu-hash64.h>
> +#include <uapi/asm/tm.h>
You shouldn't include uapi/ headers from kernel space. Please make this an #include for asm/tm.h and have that header check for __ASSEMBLY__.
> +
> +#define VCPU_GPRS_TM(reg) (((reg) * ULONG_SIZE) + VCPU_GPR_TM)
>
> #ifdef __LITTLE_ENDIAN__
> #error Need to fix lppaca and SLB shadow accesses in little endian mode
> @@ -597,6 +600,115 @@ BEGIN_FTR_SECTION
> END_FTR_SECTION_NESTED(CPU_FTR_ARCH_206, CPU_FTR_ARCH_206, 89)
> END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
>
> +#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
> +BEGIN_FTR_SECTION
> + b 1f
> +END_FTR_SECTION_IFCLR(CPU_FTR_TM)
> +
> + /* Turn on TM/FP/VSX/VMX so we can restore them. */
> + mfmsr r5
> + li r6, MSR_TM >> 32
> + sldi r6, r6, 32
> + or r5, r5, r6
> + ori r5, r5, MSR_FP
> +BEGIN_FTR_SECTION
> + oris r5, r5, MSR_VEC@h
> +END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
> +BEGIN_FTR_SECTION
> + oris r5, r5, MSR_VSX@h
> +END_FTR_SECTION_IFSET(CPU_FTR_VSX)
I thought we don't support CPUs with TM but no ALTIVEC or VSX?
> + mtmsrd r5
> +
> + /*
> + * The user may change these outside of a transaction, so they must
> + * always be context switched.
> + */
> + ld r5, VCPU_TFHAR(r4)
> + ld r6, VCPU_TFIAR(r4)
> + ld r7, VCPU_TEXASR(r4)
> + mtspr SPRN_TFHAR, r5
> + mtspr SPRN_TFIAR, r6
> + mtspr SPRN_TEXASR, r7
> +
> + ld r5, VCPU_MSR(r4)
> + rldicl. r5, r5, 64 - MSR_TS_S_LG, 62
> + beq 1f /* TM not active in guest */
> +
> + /*
> + * We need to load up the checkpointed state for the guest.
> + * We need to do this early as it will blow away any GPRs, VSRs and
> + * some SPRs.
> + */
> +
> + mr r31, r4
> + addi r3, r31, VCPU_FPRS_TM
> + bl .load_fp_state
> +BEGIN_FTR_SECTION
> + addi r3, r31, VCPU_VRS_TM
> + bl .load_vr_state
> +END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
> + mr r4, r31
> + lwz r7, VCPU_VRSAVE_TM(r4)
> + mtspr SPRN_VRSAVE, r7
> +
> + ld r5, VCPU_LR_TM(r4)
> + lwz r6, VCPU_CR_TM(r4)
> + ld r7, VCPU_CTR_TM(r4)
> + ld r8, VCPU_AMR_TM(r4)
> + ld r9, VCPU_TAR_TM(r4)
> + mtlr r5
> + mtcr r6
> + mtctr r7
> + mtspr SPRN_AMR, r8
> + mtspr SPRN_TAR, r9
> +
> + /*
> + * Load up PPR and DSCR values but don't put them in the actual SPRs
> + * till the last moment to avoid running with userspace PPR and DSCR for
> + * too long.
> + */
> + ld r29, VCPU_DSCR_TM(r4)
> + ld r30, VCPU_PPR_TM(r4)
> +
> + std r2, PACATMSCRATCH(r13) /* Save TOC */
> +
> + /* Clear the MSR RI since r1, r13 are all going to be foobar. */
> + li r5, 0
> + mtmsrd r5, 1
> +
> + /* Load GPRs r0-r28 */
> + reg = 0
> + .rept 29
> + ld reg, VCPU_GPRS_TM(reg)(r31)
> + reg = reg + 1
> + .endr
> +
> + mtspr SPRN_DSCR, r29
> + mtspr SPRN_PPR, r30
> +
> + /* Load final GPRs */
> + ld 29, VCPU_GPRS_TM(29)(r31)
> + ld 30, VCPU_GPRS_TM(30)(r31)
> + ld 31, VCPU_GPRS_TM(31)(r31)
> +
> + /* TM checkpointed state is now setup. All GPRs are now volatile. */
> + TRECHKPT
Where do we fault to if we get a transaction abort now? We can get one because we're not in suspended TM state, right?
> +
> + /* Now let's get back the state we need. */
> + HMT_MEDIUM
> + GET_PACA(r13)
> + ld r29, HSTATE_DSCR(r13)
> + mtspr SPRN_DSCR, r29
> + ld r4, HSTATE_KVM_VCPU(r13)
> + ld r1, HSTATE_HOST_R1(r13)
> + ld r2, PACATMSCRATCH(r13)
> +
> + /* Set the MSR RI since we have our registers back. */
> + li r5, MSR_RI
> + mtmsrd r5, 1
> +1:
> +#endif
> +
> /* Load guest PMU registers */
> /* R4 is live here (vcpu pointer) */
> li r3, 1
> @@ -704,14 +816,6 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
> ld r6, VCPU_VTB(r4)
> mtspr SPRN_IC, r5
> mtspr SPRN_VTB, r6
> -#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
> - ld r5, VCPU_TFHAR(r4)
> - ld r6, VCPU_TFIAR(r4)
> - ld r7, VCPU_TEXASR(r4)
> - mtspr SPRN_TFHAR, r5
> - mtspr SPRN_TFIAR, r6
> - mtspr SPRN_TEXASR, r7
> -#endif
> ld r8, VCPU_EBBHR(r4)
> mtspr SPRN_EBBHR, r8
> ld r5, VCPU_EBBRR(r4)
> @@ -817,7 +921,8 @@ END_FTR_SECTION_IFSET(CPU_FTR_ARCH_206)
> 12: mtspr SPRN_SRR0, r10
> mr r10,r0
> mtspr SPRN_SRR1, r11
> - ld r11, VCPU_INTR_MSR(r4)
> + mr r9, r4
> + bl kvmppc_msr_interrupt
> 5:
>
> /*
> @@ -1103,12 +1208,6 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_201)
> BEGIN_FTR_SECTION
> b 8f
> END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
> - /* Turn on TM so we can access TFHAR/TFIAR/TEXASR */
> - mfmsr r8
> - li r0, 1
> - rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
> - mtmsrd r8
> -
> /* Save POWER8-specific registers */
> mfspr r5, SPRN_IAMR
> mfspr r6, SPRN_PSPB
> @@ -1122,14 +1221,6 @@ END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
> std r5, VCPU_IC(r9)
> std r6, VCPU_VTB(r9)
> std r7, VCPU_TAR(r9)
> -#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
> - mfspr r5, SPRN_TFHAR
> - mfspr r6, SPRN_TFIAR
> - mfspr r7, SPRN_TEXASR
> - std r5, VCPU_TFHAR(r9)
> - std r6, VCPU_TFIAR(r9)
> - std r7, VCPU_TEXASR(r9)
> -#endif
> mfspr r8, SPRN_EBBHR
> std r8, VCPU_EBBHR(r9)
> mfspr r5, SPRN_EBBRR
> @@ -1504,76 +1595,6 @@ END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
> 1: addi r8,r8,16
> .endr
>
> -<<<<<<< HEAD
> -=======
Ah, so here you remove it again ;). Please run a bisectability test on the next patch set:
#!/bin/bash -e
for commit in $(git rev-list --reverse "$@"); do
echo "Commit $commit"
git checkout "$commit"
make -j80
done
> - /* Save DEC */
> - mfspr r5,SPRN_DEC
> - mftb r6
> - extsw r5,r5
> - add r5,r5,r6
> - std r5,VCPU_DEC_EXPIRES(r9)
> -
> -BEGIN_FTR_SECTION
> - b 8f
> -END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
> - /* Turn on TM so we can access TFHAR/TFIAR/TEXASR */
> - mfmsr r8
> - li r0, 1
> - rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
> - mtmsrd r8
> -
> - /* Save POWER8-specific registers */
> - mfspr r5, SPRN_IAMR
> - mfspr r6, SPRN_PSPB
> - mfspr r7, SPRN_FSCR
> - std r5, VCPU_IAMR(r9)
> - stw r6, VCPU_PSPB(r9)
> - std r7, VCPU_FSCR(r9)
> - mfspr r5, SPRN_IC
> - mfspr r6, SPRN_VTB
> - mfspr r7, SPRN_TAR
> - std r5, VCPU_IC(r9)
> - std r6, VCPU_VTB(r9)
> - std r7, VCPU_TAR(r9)
> -#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
> - mfspr r5, SPRN_TFHAR
> - mfspr r6, SPRN_TFIAR
> - mfspr r7, SPRN_TEXASR
> - std r5, VCPU_TFHAR(r9)
> - std r6, VCPU_TFIAR(r9)
> - std r7, VCPU_TEXASR(r9)
> -#endif
> - mfspr r8, SPRN_EBBHR
> - std r8, VCPU_EBBHR(r9)
> - mfspr r5, SPRN_EBBRR
> - mfspr r6, SPRN_BESCR
> - mfspr r7, SPRN_CSIGR
> - mfspr r8, SPRN_TACR
> - std r5, VCPU_EBBRR(r9)
> - std r6, VCPU_BESCR(r9)
> - std r7, VCPU_CSIGR(r9)
> - std r8, VCPU_TACR(r9)
> - mfspr r5, SPRN_TCSCR
> - mfspr r6, SPRN_ACOP
> - mfspr r7, SPRN_PID
> - mfspr r8, SPRN_WORT
> - std r5, VCPU_TCSCR(r9)
> - std r6, VCPU_ACOP(r9)
> - stw r7, VCPU_GUEST_PID(r9)
> - std r8, VCPU_WORT(r9)
> -8:
> -
> - /* Save and reset AMR and UAMOR before turning on the MMU */
> -BEGIN_FTR_SECTION
> - mfspr r5,SPRN_AMR
> - mfspr r6,SPRN_UAMOR
> - std r5,VCPU_AMR(r9)
> - std r6,VCPU_UAMOR(r9)
> - li r6,0
> - mtspr SPRN_AMR,r6
> -END_FTR_SECTION_IFSET(CPU_FTR_ARCH_206)
> -
> ->>>>>>> a65ae5a... KVM: PPC: Book3S HV: Add new state for transactional memory
> /* Unset guest mode */
> li r0, KVM_GUEST_MODE_NONE
> stb r0, HSTATE_IN_GUEST(r13)
> @@ -1627,7 +1648,7 @@ kvmppc_hdsi:
> mtspr SPRN_SRR0, r10
> mtspr SPRN_SRR1, r11
> li r10, BOOK3S_INTERRUPT_DATA_STORAGE
> - ld r11, VCPU_INTR_MSR(r9)
> + bl kvmppc_msr_interrupt
> fast_interrupt_c_return:
> 6: ld r7, VCPU_CTR(r9)
> lwz r8, VCPU_XER(r9)
> @@ -1696,7 +1717,7 @@ kvmppc_hisi:
> 1: mtspr SPRN_SRR0, r10
> mtspr SPRN_SRR1, r11
> li r10, BOOK3S_INTERRUPT_INST_STORAGE
> - ld r11, VCPU_INTR_MSR(r9)
> + bl kvmppc_msr_interrupt
> b fast_interrupt_c_return
>
> 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
> @@ -1739,7 +1760,7 @@ sc_1_fast_return:
> mtspr SPRN_SRR0,r10
> mtspr SPRN_SRR1,r11
> li r10, BOOK3S_INTERRUPT_SYSCALL
> - ld r11, VCPU_INTR_MSR(r9)
> + bl kvmppc_msr_interrupt
> mr r4,r9
> b fast_guest_return
>
> @@ -2067,7 +2088,7 @@ machine_check_realmode:
> beq mc_cont
> /* If not, deliver a machine check. SRR0/1 are already set */
> li r10, BOOK3S_INTERRUPT_MACHINE_CHECK
> - ld r11, VCPU_INTR_MSR(r9)
> + bl kvmppc_msr_interrupt
> b fast_interrupt_c_return
>
> /*
> @@ -2196,7 +2217,9 @@ BEGIN_FTR_SECTION
> END_FTR_SECTION_IFSET(CPU_FTR_VSX)
> #endif
> mtmsrd r8
> +BEGIN_FTR_SECTION
> isync
> +END_FTR_SECTION_IFSET(CPU_FTR_ARCH_201)
Why bother?
> addi r3,r3,VCPU_FPRS
> bl .store_fp_state
> #ifdef CONFIG_ALTIVEC
> @@ -2208,8 +2231,6 @@ END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
> mfspr r6,SPRN_VRSAVE
> stw r6,VCPU_VRSAVE(r3)
> mtlr r30
> - mtmsrd r5
> - isync
> blr
>
> /*
> @@ -2234,7 +2255,9 @@ BEGIN_FTR_SECTION
> END_FTR_SECTION_IFSET(CPU_FTR_VSX)
> #endif
> mtmsrd r8
> +BEGIN_FTR_SECTION
> isync
> +END_FTR_SECTION_IFSET(CPU_FTR_ARCH_201)
> addi r3,r4,VCPU_FPRS
> bl .load_fp_state
> #ifdef CONFIG_ALTIVEC
> @@ -2256,3 +2279,19 @@ END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
> */
> kvmppc_bad_host_intr:
> b .
> +
> +/*
> + * This mimics the MSR transition on IRQ delievery. The new guest MSR is taken
> + * from VCPU_INTR_MSR and is modified based on the required TM state changes.
> + * r11 has the guest MSR value
> + * r9 has a vcpu pointer
This needs to document the registers it touches.
Alex
> + */
> +kvmppc_msr_interrupt:
> + rldicl r0, r11, 64 - MSR_TS_S_LG, 62
> + cmpwi r0, 2 /* Check if we are in transactional state.. */
> + ld r11, VCPU_INTR_MSR(r9)
> + bne 1f
> + /* ... if transactional, change to suspended */
> + li r0, 1
> +1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
> + blr
> --
> 1.8.5.2
>
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
