From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from out-171.mta0.migadu.com (out-171.mta0.migadu.com [91.218.175.171]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 5E30A175A6E for ; Thu, 4 Jun 2026 02:04:49 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=91.218.175.171 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1780538691; cv=none; b=Uyt04hmPE5QY8Zfwy6sXA2z4mIC+PJ6ggbh2mtcrPeFpU+Uq3sJB/E593lpcJBAZ6LljQyC+qyBELyKxJEMLTFQzhwoUOEGWLjOWKJOEBVQ2ID9B8tk81fbcRLdsWGqN4gGUc7lt0HsDfTQe3XsoaBve+aFPbX7kU/HLwZ5jj94= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1780538691; c=relaxed/simple; bh=TXmUXWcu0twRoRTW9JRkbyZu0Iuxms8v42tQCFs6w+c=; h=Message-ID:Date:MIME-Version:Subject:To:Cc:References:From: In-Reply-To:Content-Type; b=DLXIYWyHDlIaK7ojKeC8CXyGvxDdc2Ghf5fxQ1VvDDv/fY8UDY5hVVX3gHc4wqWXmI0Cvt2he3FpAwt/dHuunyMTSbtiTpkETyZl+VCmGCc6mhG0gaAhoLWq0/SIf6h+dmz9ooT7OLyFRfTcR8JXZIgtn6Lz2K8tHbGavDXlQ+U= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.dev; spf=pass smtp.mailfrom=linux.dev; dkim=pass (1024-bit key) header.d=linux.dev header.i=@linux.dev header.b=pvHYfi0Q; arc=none smtp.client-ip=91.218.175.171 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.dev Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=linux.dev Authentication-Results: smtp.subspace.kernel.org; dkim=pass (1024-bit key) header.d=linux.dev header.i=@linux.dev header.b="pvHYfi0Q" Message-ID: DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1780538686; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=EXRcOFASlBgpHHUVWdbAPeNZ/dLtG8AjFSQfXCxwvmI=; b=pvHYfi0QlnC1XF8xzxauyx9BvSGGVXEobFJilJt3YeUfOcAnwOBNNRUDfCKGdBuWFB+SQE 2DdH3o13DP240xm65Pq8Qs7Anrxseux+S6uLk0EGY7oMJTKx3l+5M/+Y8FKD11Cvy76t1Y 1D/+res3j7HMMjQT0zdgr8riRT5KjgM= Date: Thu, 4 Jun 2026 10:04:31 +0800 Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Subject: Re: [PATCH v5 14/14] mm: thp: always enable mTHP support Content-Language: en-US To: Luiz Capitulino , Nico Pache Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org, david@kernel.org, baolin.wang@linux.alibaba.com, ziy@nvidia.com, corbet@lwn.net, tsbogend@alpha.franken.de, maddy@linux.ibm.com, mpe@ellerman.id.au, agordeev@linux.ibm.com, gerald.schaefer@linux.ibm.com, hca@linux.ibm.com, gor@linux.ibm.com, x86@kernel.org, tglx@kernel.org, mingo@redhat.com, bp@alien8.de, ira.weiny@intel.com, hughd@google.com, dave.hansen@linux.intel.com, djbw@kernel.org, vishal.l.verma@intel.com, dave.jiang@intel.com, akpm@linux-foundation.org, yintirui@huawei.com, ljs@kernel.org References: <8781a9a0f115705ee11884ed3184b65a1ce39923.1780066530.git.luizcap@redhat.com> <20260603081207.1375-1-lance.yang@linux.dev> <710692d6-27e7-434e-a308-e717793d4fce@redhat.com> X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. From: Lance Yang In-Reply-To: <710692d6-27e7-434e-a308-e717793d4fce@redhat.com> Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit X-Migadu-Flow: FLOW_OUT On 2026/6/4 01:40, Luiz Capitulino wrote: > On 2026-06-03 04:12, Lance Yang wrote: >> Would it makes sense to call khugepaged_enter_vma() for anon mTHP faults >> as well? >> >> vm_fault_t do_huge_pmd_anonymous_page(struct vm_fault *vmf) >> { >> [...] >>     if (!thp_vma_suitable_order(vma, haddr, PMD_ORDER)) >>         return VM_FAULT_FALLBACK; >> [...] >>     khugepaged_enter_vma(vma, vma->vm_flags); >> [...] >> } >> >> Without PMD leaves, do_huge_pmd_anonymous_page() is not reached. Apart >> from MADV_HUGEPAGE, AFAIK, the mm has no chance to enter khugepaged >> from the fault side :) > > Hey Lance, Hey > Wouldn't we also skip do_huge_pmd_anonymous_page() page today if > PMD-sized THPs are disabled in sysfs but mTHP is enabled? Right. I got mixed up here :P > In any case, by reading through khugepaged_enter_vma() I see that > khugepaged only supports PMD-size for now. So, shouldn't this case be > addressed in Nico's series [1] ? Yep. Anyway, not a big deal. No need to pile more onto Nico's series (enough rounds already ... ). Best left as a follow-up after it lands :) > [1] https://lore.kernel.org/linux-mm/20260522150009.121603-1- > npache@redhat.com/ Cheers, Lance >> >> Cheers, Lance >> >> On Fri, May 29, 2026 at 10:55:32AM -0400, Luiz Capitulino wrote: >>> If PMD-sized pages are not supported on an architecture (ie. the >>> arch implements arch_has_pmd_leaves() and it returns false) then the >>> current code disables all THP, including mTHP. >>> >>> This commit fixes this by allowing mTHP to be always enabled for all >>> archs. When PMD-sized pages are not supported, its sysfs entry won't be >>> created and their mapping will be disallowed at page-fault time. >>> >>> Similarly, this commit implements the following changes for shmem in >>> shmem_allowable_huge_orders(): >>> >>> - Drop the pgtable_has_pmd_leaves() check so that mTHP sizes are >>>    considered >>> - Filter out PMD and PUD orders from allowable orders when >>>    PMD-sized pages are not supported by the CPU >>> >>> Signed-off-by: Luiz Capitulino >>> --- >>> mm/huge_memory.c | 25 ++++++++++++++++++++----- >>> mm/shmem.c       | 14 +++++++++----- >>> 2 files changed, 29 insertions(+), 10 deletions(-) >>> >>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c >>> index 32254febe097..059901a8c6cb 100644 >>> --- a/mm/huge_memory.c >>> +++ b/mm/huge_memory.c >>> @@ -126,6 +126,15 @@ unsigned long __thp_vma_allowable_orders(struct >>> vm_area_struct *vma, >>>     else >>>         supported_orders = THP_ORDERS_ALL_FILE_DEFAULT; >>> >>> +    if (!pgtable_has_pmd_leaves()) { >>> +        /* >>> +         * If the CPU does not support PMD leaves, assume for >>> +         * now that it does not support PUD leaves and disable >>> +         * both folio orders. >>> +         */ >>> +        supported_orders &= ~(BIT(PMD_ORDER) | BIT(PUD_ORDER)); >>> +    } >>> + >>>     orders &= supported_orders; >>>     if (!orders) >>>         return 0; >>> @@ -133,7 +142,7 @@ unsigned long __thp_vma_allowable_orders(struct >>> vm_area_struct *vma, >>>     if (!vma->vm_mm)        /* vdso */ >>>         return 0; >>> >>> -    if (!pgtable_has_pmd_leaves() || vma_thp_disabled(vma, vm_flags, >>> forced_collapse)) >>> +    if (vma_thp_disabled(vma, vm_flags, forced_collapse)) >>>         return 0; >>> >>>     /* khugepaged doesn't collapse DAX vma, but page fault is fine. */ >>> @@ -848,7 +857,7 @@ static int __init hugepage_init_sysfs(struct >>> kobject **hugepage_kobj) >>>      * disable all other sizes. powerpc's PMD_ORDER isn't a compile-time >>>      * constant so we have to do this here. >>>      */ >>> -    if (!anon_orders_configured) >>> +    if (!anon_orders_configured && pgtable_has_pmd_leaves()) >>>         huge_anon_orders_inherit = BIT(PMD_ORDER); >>> >>>     *hugepage_kobj = kobject_create_and_add("transparent_hugepage", >>> mm_kobj); >>> @@ -870,6 +879,15 @@ static int __init hugepage_init_sysfs(struct >>> kobject **hugepage_kobj) >>>     } >>> >>>     orders = THP_ORDERS_ALL_ANON | THP_ORDERS_ALL_FILE_DEFAULT; >>> +    if (!pgtable_has_pmd_leaves()) { >>> +        /* >>> +         * If the CPU does not support PMD leaves, assume for >>> +         * now that it does not support PUD leaves and disable >>> +         * both folio orders. >>> +         */ >>> +        orders &= ~(BIT(PMD_ORDER) | BIT(PUD_ORDER)); >>> +    } >>> + >>>     order = highest_order(orders); >>>     while (orders) { >>>         thpsize = thpsize_create(order, *hugepage_kobj); >>> @@ -969,9 +987,6 @@ static int __init hugepage_init(void) >>>     int err; >>>     struct kobject *hugepage_kobj; >>> >>> -    if (!pgtable_has_pmd_leaves()) >>> -        return -EINVAL; >>> - >>>     /* >>>      * hugepages can't be allocated by the buddy allocator >>>      */ >>> diff --git a/mm/shmem.c b/mm/shmem.c >>> index 079e299ea789..c15dffd0eb41 100644 >>> --- a/mm/shmem.c >>> +++ b/mm/shmem.c >>> @@ -1844,16 +1844,19 @@ unsigned long >>> shmem_allowable_huge_orders(struct inode *inode, >>>     unsigned long mask = READ_ONCE(huge_shmem_orders_always); >>>     unsigned long within_size_orders = >>> READ_ONCE(huge_shmem_orders_within_size); >>>     vm_flags_t vm_flags = vma ? vma->vm_flags : 0; >>> -    unsigned int global_orders; >>> +    unsigned int global_orders, disabled_orders = 0; >>> >>> -    if (!pgtable_has_pmd_leaves() || (vma && vma_thp_disabled(vma, >>> vm_flags, shmem_huge_force))) >>> +    if (vma && vma_thp_disabled(vma, vm_flags, shmem_huge_force)) >>>         return 0; >>> >>> +    if (!pgtable_has_pmd_leaves()) >>> +        disabled_orders = BIT(PMD_ORDER); >>> + >>>     global_orders = shmem_huge_global_enabled(inode, index, write_end, >>>                           shmem_huge_force, vma, vm_flags); >>>     /* Tmpfs huge pages allocation */ >>>     if (!vma || !vma_is_anon_shmem(vma)) >>> -        return global_orders; >>> +        return global_orders & ~disabled_orders; >>> >>>     /* >>>      * Following the 'deny' semantics of the top level, force the huge >>> @@ -1867,7 +1870,7 @@ unsigned long >>> shmem_allowable_huge_orders(struct inode *inode, >>>      * means non-PMD sized THP can not override 'huge' mount option now. >>>      */ >>>     if (shmem_huge == SHMEM_HUGE_FORCE) >>> -        return READ_ONCE(huge_shmem_orders_inherit); >>> +        return READ_ONCE(huge_shmem_orders_inherit) & ~disabled_orders; >>> >>>     /* Allow mTHP that will be fully within i_size. */ >>>     mask |= shmem_get_orders_within_size(inode, within_size_orders, >>> index, 0); >>> @@ -1878,6 +1881,7 @@ unsigned long >>> shmem_allowable_huge_orders(struct inode *inode, >>>     if (global_orders > 0) >>>         mask |= READ_ONCE(huge_shmem_orders_inherit); >>> >>> +    mask &= ~disabled_orders; >>>     return THP_ORDERS_ALL_FILE_DEFAULT & mask; >>> } >>> >>> @@ -5461,7 +5465,7 @@ void __init shmem_init(void) >>>      * Default to setting PMD-sized THP to inherit the global setting >>> and >>>      * disable all other multi-size THPs. >>>      */ >>> -    if (!shmem_orders_configured) >>> +    if (!shmem_orders_configured && pgtable_has_pmd_leaves()) >>>         huge_shmem_orders_inherit = BIT(HPAGE_PMD_ORDER); >>> #endif >>>     return; >>> -- >>> 2.54.0 >>> >>> >> >