From mboxrd@z Thu Jan 1 00:00:00 1970 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 bombadil.infradead.org (bombadil.infradead.org [198.137.202.133]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id A4716C3DA6E for ; Wed, 3 Jan 2024 11:10:57 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.infradead.org; s=bombadil.20210309; h=Sender: Content-Transfer-Encoding:Content-Type:List-Subscribe:List-Help:List-Post: List-Archive:List-Unsubscribe:List-Id:In-Reply-To:MIME-Version:References: Message-ID:Subject:Cc:To:From:Date:Reply-To:Content-ID:Content-Description: Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID: List-Owner; bh=VdAy6FM2ZunaGf9a6dqdyN4oTPpX7OhW6s4TSQZXUu8=; b=2DeI5qrRLT/a+m KdjSsc0i3bx9b8lKtdtCa9qWKtNjzw4rJE3GZGTHmnz3x2mcWe1E5eaykqCJcJtweCFg5n9jeDLpP oUWjLYqThK7nBJF/sB67qKPsBW0gzyh1EYYXPIz+7UKsedhK2YiG+9iLx0wMPhTS/5CTrHfmnUjJJ Y+G5gs8K4z2n+NCOEdR5I1KjqlGbBDXJwaxPVqJ/Ub8s12vrG+lZ00z+wQqPUeGUN+nuRW1mWb2Sn 80JVfuK7Tqnq1SyJpTV27l5o0CxsgRNpCRudQSMLC3Zw5f5acVzlwZGmP7ioGLjmr6KSyKfSeEaAU 9/5Cs9GOYvn0imupzwTg==; Received: from localhost ([::1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.96 #2 (Red Hat Linux)) id 1rKz92-00ATlF-0M; Wed, 03 Jan 2024 11:10:44 +0000 Received: from ams.source.kernel.org ([145.40.68.75]) by bombadil.infradead.org with esmtps (Exim 4.96 #2 (Red Hat Linux)) id 1rKz8y-00ATkc-11 for linux-riscv@lists.infradead.org; Wed, 03 Jan 2024 11:10:42 +0000 Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by ams.source.kernel.org (Postfix) with ESMTP id 16EEAB8109A; Wed, 3 Jan 2024 11:10:38 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 49CB0C433C7; Wed, 3 Jan 2024 11:10:36 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704280237; bh=vLoZYtFxzJ2+K06nFXKmnrlfGz1y0ecWX2UzGR9hwI0=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=ojFX72Sr1iuIEWNfq3+EibKDiSiBnB0uCNBpToafuraSESQqXMB+/6mIiVmfCbYMJ MXgdHSsj8YdOK7mgWpGy9zLQKvERKCKJYAmEsPlTzWR0ba/BKoy6zfQY44+Hp3n7mV zWPJFNAPPrPuafzz/gevLPHy9sMqcJFPqmoQpFBI5ZOm4qXmwC2VO9+VHTdxlPY8v7 U0Qg1BTheOHUKbHAzHtTlYMau0be/H8DUIWofJ4txtePgtJDooJLY04zGDh7lpXaTo EuRrbNBl4hadcG2hE4SX9IvGcXrc5NCQjfL8eJS8lEgNZAjolL3JWsL/Rp/QjXVTHr t/Pt1u9GaJWAw== Date: Wed, 3 Jan 2024 18:57:55 +0800 From: Jisheng Zhang To: Alexandre Ghiti Cc: Paul Walmsley , Palmer Dabbelt , Albert Ou , linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Message-ID: References: <20240102141851.105144-1-alexghiti@rivosinc.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <20240102141851.105144-1-alexghiti@rivosinc.com> X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20240103_031040_696888_9D582C60 X-CRM114-Status: GOOD ( 30.49 ) X-BeenThere: linux-riscv@lists.infradead.org X-Mailman-Version: 2.1.34 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Sender: "linux-riscv" Errors-To: linux-riscv-bounces+linux-riscv=archiver.kernel.org@lists.infradead.org On Tue, Jan 02, 2024 at 03:18:51PM +0100, Alexandre Ghiti wrote: > Allow to defer the flushing of the TLB when unmapping pges, which allows > to reduce the numbers of IPI and the number of sfence.vma. > > The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support > batched/deferred tlb shootdown during page reclamation/migration") shows > good performance improvement and perf reports an important decrease in > time spent flushing the tlb (results come from qemu): Hi Alex, I tried this micro benchmark with your patch on T-HEAD TH1520 platform, I didn't see any performance improvement for the micro benchmark. Per myunderstanding, the micro benchmark is special case for arm64 because in a normal tlb flush flow, below sequence is necessary: tlbi dsb while with BATCHED_UNMAP_TLB_FLUSH, the arm64 just does 'tlbi', leaving the dsb to the arch_tlbbatch_flush(). So the final result is several 'tlbi + dsb' sequence VS. several 'tlbi' instructions + only one dsb The performance improvement comes from the unnecessary dsb eliminations. Do you have suitable benchmark(s) for BATCHED_UNMAP_TLB_FLUSH on riscv? Thanks > > Before this patch: > > real 2m1.135s > user 0m0.980s > sys 2m0.096s > > 4.83% batch_tlb [kernel.kallsyms] [k] __flush_tlb_range > > After this patch: > > real 1m0.543s > user 0m1.059s > sys 0m59.489s > > 0.14% batch_tlb [kernel.kallsyms] [k] __flush_tlb_range > > Signed-off-by: Alexandre Ghiti > --- > arch/riscv/Kconfig | 1 + > arch/riscv/include/asm/tlbbatch.h | 15 +++++++ > arch/riscv/include/asm/tlbflush.h | 10 +++++ > arch/riscv/mm/tlbflush.c | 71 ++++++++++++++++++++++--------- > 4 files changed, 77 insertions(+), 20 deletions(-) > create mode 100644 arch/riscv/include/asm/tlbbatch.h > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig > index 7603bd8ab333..aa07bd43b138 100644 > --- a/arch/riscv/Kconfig > +++ b/arch/riscv/Kconfig > @@ -53,6 +53,7 @@ config RISCV > select ARCH_USE_MEMTEST > select ARCH_USE_QUEUED_RWLOCKS > select ARCH_USES_CFI_TRAPS if CFI_CLANG > + select ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH if SMP && MMU > select ARCH_WANT_DEFAULT_TOPDOWN_MMAP_LAYOUT if MMU > select ARCH_WANT_FRAME_POINTERS > select ARCH_WANT_GENERAL_HUGETLB if !RISCV_ISA_SVNAPOT > diff --git a/arch/riscv/include/asm/tlbbatch.h b/arch/riscv/include/asm/tlbbatch.h > new file mode 100644 > index 000000000000..46014f70b9da > --- /dev/null > +++ b/arch/riscv/include/asm/tlbbatch.h > @@ -0,0 +1,15 @@ > +/* SPDX-License-Identifier: GPL-2.0-only */ > +/* > + * Copyright (C) 2023 Rivos Inc. > + */ > + > +#ifndef _ASM_RISCV_TLBBATCH_H > +#define _ASM_RISCV_TLBBATCH_H > + > +#include > + > +struct arch_tlbflush_unmap_batch { > + struct cpumask cpumask; > +}; > + > +#endif /* _ASM_RISCV_TLBBATCH_H */ > diff --git a/arch/riscv/include/asm/tlbflush.h b/arch/riscv/include/asm/tlbflush.h > index 8f3418c5f172..f0b731ccc0c2 100644 > --- a/arch/riscv/include/asm/tlbflush.h > +++ b/arch/riscv/include/asm/tlbflush.h > @@ -46,6 +46,16 @@ void flush_tlb_kernel_range(unsigned long start, unsigned long end); > void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start, > unsigned long end); > #endif > + > +#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH > +bool arch_tlbbatch_should_defer(struct mm_struct *mm); > +void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch, > + struct mm_struct *mm, > + unsigned long uaddr); > +void arch_flush_tlb_batched_pending(struct mm_struct *mm); > +void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch); > +#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ > + > #else /* CONFIG_SMP && CONFIG_MMU */ > > #define flush_tlb_all() local_flush_tlb_all() > diff --git a/arch/riscv/mm/tlbflush.c b/arch/riscv/mm/tlbflush.c > index e6659d7368b3..bb623bca0a7d 100644 > --- a/arch/riscv/mm/tlbflush.c > +++ b/arch/riscv/mm/tlbflush.c > @@ -93,29 +93,23 @@ static void __ipi_flush_tlb_range_asid(void *info) > local_flush_tlb_range_asid(d->start, d->size, d->stride, d->asid); > } > > -static void __flush_tlb_range(struct mm_struct *mm, unsigned long start, > - unsigned long size, unsigned long stride) > +static void __flush_tlb_range(struct cpumask *cmask, unsigned long asid, > + unsigned long start, unsigned long size, > + unsigned long stride) > { > struct flush_tlb_range_data ftd; > - const struct cpumask *cmask; > - unsigned long asid = FLUSH_TLB_NO_ASID; > bool broadcast; > > - if (mm) { > - unsigned int cpuid; > + if (cpumask_empty(cmask)) > + return; > > - cmask = mm_cpumask(mm); > - if (cpumask_empty(cmask)) > - return; > + if (cmask != cpu_online_mask) { > + unsigned int cpuid; > > cpuid = get_cpu(); > /* check if the tlbflush needs to be sent to other CPUs */ > broadcast = cpumask_any_but(cmask, cpuid) < nr_cpu_ids; > - > - if (static_branch_unlikely(&use_asid_allocator)) > - asid = atomic_long_read(&mm->context.id) & asid_mask; > } else { > - cmask = cpu_online_mask; > broadcast = true; > } > > @@ -135,25 +129,34 @@ static void __flush_tlb_range(struct mm_struct *mm, unsigned long start, > local_flush_tlb_range_asid(start, size, stride, asid); > } > > - if (mm) > + if (cmask != cpu_online_mask) > put_cpu(); > } > > +static inline unsigned long get_mm_asid(struct mm_struct *mm) > +{ > + return static_branch_unlikely(&use_asid_allocator) ? > + atomic_long_read(&mm->context.id) & asid_mask : FLUSH_TLB_NO_ASID; > +} > + > void flush_tlb_mm(struct mm_struct *mm) > { > - __flush_tlb_range(mm, 0, FLUSH_TLB_MAX_SIZE, PAGE_SIZE); > + __flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm), > + 0, FLUSH_TLB_MAX_SIZE, PAGE_SIZE); > } > > void flush_tlb_mm_range(struct mm_struct *mm, > unsigned long start, unsigned long end, > unsigned int page_size) > { > - __flush_tlb_range(mm, start, end - start, page_size); > + __flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm), > + start, end - start, page_size); > } > > void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr) > { > - __flush_tlb_range(vma->vm_mm, addr, PAGE_SIZE, PAGE_SIZE); > + __flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm), > + addr, PAGE_SIZE, PAGE_SIZE); > } > > void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, > @@ -185,18 +188,46 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, > } > } > > - __flush_tlb_range(vma->vm_mm, start, end - start, stride_size); > + __flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm), > + start, end - start, stride_size); > } > > void flush_tlb_kernel_range(unsigned long start, unsigned long end) > { > - __flush_tlb_range(NULL, start, end - start, PAGE_SIZE); > + __flush_tlb_range((struct cpumask *)cpu_online_mask, FLUSH_TLB_NO_ASID, > + start, end - start, PAGE_SIZE); > } > > #ifdef CONFIG_TRANSPARENT_HUGEPAGE > void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start, > unsigned long end) > { > - __flush_tlb_range(vma->vm_mm, start, end - start, PMD_SIZE); > + __flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm), > + start, end - start, PMD_SIZE); > } > #endif > + > +#ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH > +bool arch_tlbbatch_should_defer(struct mm_struct *mm) > +{ > + return true; > +} > + > +void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch, > + struct mm_struct *mm, > + unsigned long uaddr) > +{ > + cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm)); > +} > + > +void arch_flush_tlb_batched_pending(struct mm_struct *mm) > +{ > + flush_tlb_mm(mm); > +} > + > +void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch) > +{ > + __flush_tlb_range(&batch->cpumask, FLUSH_TLB_NO_ASID, 0, > + FLUSH_TLB_MAX_SIZE, PAGE_SIZE); > +} > +#endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ > -- > 2.39.2 > > > _______________________________________________ > linux-riscv mailing list > linux-riscv@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-riscv _______________________________________________ linux-riscv mailing list linux-riscv@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-riscv