From patchwork Wed Apr 5 15:40:06 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Marc Zyngier X-Patchwork-Id: 13202102 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 3F33EC77B60 for ; Wed, 5 Apr 2023 15:47:02 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S239025AbjDEPrB (ORCPT ); Wed, 5 Apr 2023 11:47:01 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:57562 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S239063AbjDEPqz (ORCPT ); Wed, 5 Apr 2023 11:46:55 -0400 Received: from dfw.source.kernel.org (dfw.source.kernel.org [IPv6:2604:1380:4641:c500::1]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 8DCDE40DA for ; Wed, 5 Apr 2023 08:46:27 -0700 (PDT) Received: from smtp.kernel.org (relay.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dfw.source.kernel.org (Postfix) with ESMTPS id 2C96E63E1F for ; Wed, 5 Apr 2023 15:46:27 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 91AF5C433D2; Wed, 5 Apr 2023 15:46:26 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1680709586; bh=C+eYkxFuOlWI1J0Ai+7uKAkwuQz44DowGsncqCBujTA=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=QLwyPTCVUNwX1ra03YiaNfDc3ORLFLTEcXljf4JoPppIdHgp51m5yr5k/J/snkYPs Q9sg49L4y1IX06258+nn1DhGHU9h8P4TDKQtXvRWIsfjgbLUadfQaU3/jK/AE20pqM CfxXSqErplFDkmhWuyBP3YdhURLDIcpO9IuxO7FdDCVBzKxqVz1ER/E3Bm9R2ejd7x yQiG0zQ2GSQ86CxcMDX7GVoCufAqhdiLZqrVKgbIXUJGTqBvNLU5AUc20oJZwerTP9 bzenpExdBzgY8olp6GhsqpX0T8lxiBQGF71fXF4qDwg/J7zqgp/58hxzLF5yGl5LdP oZrztHeZqQZKg== Received: from sofa.misterjones.org ([185.219.108.64] helo=valley-girl.lan) by disco-boy.misterjones.org with esmtpsa (TLS1.3) tls TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384 (Exim 4.95) (envelope-from ) id 1pk5Fw-0062PV-71; Wed, 05 Apr 2023 16:41:04 +0100 From: Marc Zyngier To: kvmarm@lists.linux.dev, kvm@vger.kernel.org, linux-arm-kernel@lists.infradead.org Cc: Alexandru Elisei , Andre Przywara , Chase Conklin , Christoffer Dall , Ganapatrao Kulkarni , Darren Hart , Jintack Lim , Russell King , Miguel Luis , James Morse , Suzuki K Poulose , Oliver Upton , Zenghui Yu Subject: [PATCH v9 48/50] KVM: arm64: nv: Fast-track EL1 TLBIs for VHE guests Date: Wed, 5 Apr 2023 16:40:06 +0100 Message-Id: <20230405154008.3552854-49-maz@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20230405154008.3552854-1-maz@kernel.org> References: <20230405154008.3552854-1-maz@kernel.org> MIME-Version: 1.0 X-SA-Exim-Connect-IP: 185.219.108.64 X-SA-Exim-Rcpt-To: kvmarm@lists.linux.dev, kvm@vger.kernel.org, linux-arm-kernel@lists.infradead.org, alexandru.elisei@arm.com, andre.przywara@arm.com, chase.conklin@arm.com, christoffer.dall@arm.com, gankulkarni@os.amperecomputing.com, darren@os.amperecomputing.com, jintack@cs.columbia.edu, rmk+kernel@armlinux.org.uk, miguel.luis@oracle.com, james.morse@arm.com, suzuki.poulose@arm.com, oliver.upton@linux.dev, yuzenghui@huawei.com X-SA-Exim-Mail-From: maz@kernel.org X-SA-Exim-Scanned: No (on disco-boy.misterjones.org); SAEximRunCond expanded to false Precedence: bulk List-ID: X-Mailing-List: kvm@vger.kernel.org Due to the way ARMv8.4-NV suppresses traps when accessing EL2 system registers, we can't track when the guest changes its HCR_EL2.TGE setting. This means we always trap EL1 TLBIs, even if they don't affect any guest. This obviously has a huge impact on performance, as we handle TLBI traps as a normal exit, and a normal VHE host issues thousands of TLBIs when booting (and quite a few when running userspace). A cheap way to reduce the overhead is to handle the limited case of {E2H,TGE}=={1,1} as a guest fixup, as we already have the right mmu configuration in place. Just execute the decoded instruction right away and return to the guest. Signed-off-by: Marc Zyngier --- arch/arm64/kvm/hyp/vhe/switch.c | 43 ++++++++++++++++++++++++++++++++- arch/arm64/kvm/hyp/vhe/tlb.c | 6 +++-- arch/arm64/kvm/sys_regs.c | 12 --------- 3 files changed, 46 insertions(+), 15 deletions(-) diff --git a/arch/arm64/kvm/hyp/vhe/switch.c b/arch/arm64/kvm/hyp/vhe/switch.c index 238c6613cf47..a3555b90d9e1 100644 --- a/arch/arm64/kvm/hyp/vhe/switch.c +++ b/arch/arm64/kvm/hyp/vhe/switch.c @@ -168,6 +168,47 @@ void deactivate_traps_vhe_put(struct kvm_vcpu *vcpu) __deactivate_traps_common(vcpu); } +static bool kvm_hyp_handle_tlbi_el1(struct kvm_vcpu *vcpu, u64 *exit_code) +{ + u32 instr; + u64 val; + + /* + * Ideally, we would never trap on EL1 TLB invalidations when the + * guest's HCR_EL2.{E2H,TGE} == {1,1}. But "thanks" to ARMv8.4, we + * don't trap writes to HCR_EL2, meaning that we can't track + * changes to the virtual TGE bit. So we leave HCR_EL2.TTLB set on + * the host. Oopsie... + * + * In order to speed-up EL1 TLBIs from the vEL2 guest when TGE is + * set, try and handle these invalidation as quickly as possible, + * without fully exiting (unless this needs forwarding). + */ + if (!vcpu_has_nv2(vcpu) || + !vcpu_is_el2(vcpu) || + (__vcpu_sys_reg(vcpu, HCR_EL2) & (HCR_E2H | HCR_TGE)) != (HCR_E2H | HCR_TGE)) + return false; + + instr = esr_sys64_to_sysreg(kvm_vcpu_get_esr(vcpu)); + if (sys_reg_Op0(instr) != TLBI_Op0 || + sys_reg_Op1(instr) != TLBI_Op1_EL1) + return false; + + val = vcpu_get_reg(vcpu, kvm_vcpu_sys_get_rt(vcpu)); + __kvm_tlb_el1_instr(NULL, val, instr); + __kvm_skip_instr(vcpu); + + return true; +} + +static bool kvm_hyp_handle_sysreg_vhe(struct kvm_vcpu *vcpu, u64 *exit_code) +{ + if (kvm_hyp_handle_tlbi_el1(vcpu, exit_code)) + return true; + + return kvm_hyp_handle_sysreg(vcpu, exit_code); +} + static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code) { struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt; @@ -216,7 +257,7 @@ static bool kvm_hyp_handle_eret(struct kvm_vcpu *vcpu, u64 *exit_code) static const exit_handler_fn hyp_exit_handlers[] = { [0 ... ESR_ELx_EC_MAX] = NULL, [ESR_ELx_EC_CP15_32] = kvm_hyp_handle_cp15_32, - [ESR_ELx_EC_SYS64] = kvm_hyp_handle_sysreg, + [ESR_ELx_EC_SYS64] = kvm_hyp_handle_sysreg_vhe, [ESR_ELx_EC_SVE] = kvm_hyp_handle_fpsimd, [ESR_ELx_EC_FP_ASIMD] = kvm_hyp_handle_fpsimd, [ESR_ELx_EC_IABT_LOW] = kvm_hyp_handle_iabt_low, diff --git a/arch/arm64/kvm/hyp/vhe/tlb.c b/arch/arm64/kvm/hyp/vhe/tlb.c index c4389db4cc22..beb162468c0b 100644 --- a/arch/arm64/kvm/hyp/vhe/tlb.c +++ b/arch/arm64/kvm/hyp/vhe/tlb.c @@ -201,7 +201,8 @@ void __kvm_tlb_el1_instr(struct kvm_s2_mmu *mmu, u64 val, u64 sys_encoding) dsb(ishst); /* Switch to requested VMID */ - __tlb_switch_to_guest(mmu, &cxt); + if (mmu) + __tlb_switch_to_guest(mmu, &cxt); /* * Execute the same instruction as the guest hypervisor did, @@ -240,5 +241,6 @@ void __kvm_tlb_el1_instr(struct kvm_s2_mmu *mmu, u64 val, u64 sys_encoding) dsb(ish); isb(); - __tlb_switch_to_host(&cxt); + if (mmu) + __tlb_switch_to_host(&cxt); } diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c index 4a1c4a97f702..a671629fb366 100644 --- a/arch/arm64/kvm/sys_regs.c +++ b/arch/arm64/kvm/sys_regs.c @@ -3113,18 +3113,6 @@ static bool handle_tlbi_el1(struct kvm_vcpu *vcpu, struct sys_reg_params *p, WARN_ON(!vcpu_is_el2(vcpu)); - if ((__vcpu_sys_reg(vcpu, HCR_EL2) & (HCR_E2H | HCR_TGE)) == (HCR_E2H | HCR_TGE)) { - mutex_lock(&vcpu->kvm->lock); - /* - * ARMv8.4-NV allows the guest to change TGE behind - * our back, so we always trap EL1 TLBIs from vEL2... - */ - __kvm_tlb_el1_instr(&vcpu->kvm->arch.mmu, p->regval, sys_encoding); - mutex_unlock(&vcpu->kvm->lock); - - return true; - } - kvm_s2_mmu_iterate_by_vmid(vcpu->kvm, get_vmid(vttbr), &(union tlbi_info) { .va = {