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 X-Spam-Level: X-Spam-Status: No, score=-6.8 required=3.0 tests=DKIM_INVALID,DKIM_SIGNED, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_HELO_NONE,SPF_PASS autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 2D6C7C34026 for ; Tue, 18 Feb 2020 10:54:55 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id D377822B48 for ; Tue, 18 Feb 2020 10:54:54 +0000 (UTC) Authentication-Results: mail.kernel.org; dkim=fail reason="signature verification failed" (1024-bit key) header.d=kernel.org header.i=@kernel.org header.b="nQZIkGPA" DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org D377822B48 Authentication-Results: mail.kernel.org; dmarc=fail (p=none dis=none) header.from=kernel.org Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=owner-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix) id 6A4D56B0003; Tue, 18 Feb 2020 05:54:54 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 655816B0006; Tue, 18 Feb 2020 05:54:54 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 543966B0007; Tue, 18 Feb 2020 05:54:54 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0061.hostedemail.com [216.40.44.61]) by kanga.kvack.org (Postfix) with ESMTP id 399906B0003 for ; Tue, 18 Feb 2020 05:54:54 -0500 (EST) Received: from smtpin23.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay03.hostedemail.com (Postfix) with ESMTP id CEF338248047 for ; Tue, 18 Feb 2020 10:54:53 +0000 (UTC) X-FDA: 76502940066.23.sand84_2aadfe7b0c052 X-HE-Tag: sand84_2aadfe7b0c052 X-Filterd-Recvd-Size: 12681 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by imf50.hostedemail.com (Postfix) with ESMTP for ; Tue, 18 Feb 2020 10:54:53 +0000 (UTC) Received: from hump (unknown [109.236.136.226]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPSA id 58260207FD; Tue, 18 Feb 2020 10:54:44 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=default; t=1582023292; bh=CMlZZWS7PO9lkPnhA08fc6Islr+OQCEeZuheLiBfVmU=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=nQZIkGPAmBAW2q8NANsj3kry2FjoyJoV6/tOeR6LNswBfh2J3GwBoJflFT5WH0M0E dfZCuTZ+znP94xdsEPbHpuBFhjabZfi5XCWgerZeTyyA+r8h1bx3I9KPRweYB3ciYM WYq+qWxM8vgm+Ut2efYgnUJmpLpvYONcSRjzEgV4= Date: Tue, 18 Feb 2020 12:54:40 +0200 From: Mike Rapoport To: Christophe Leroy Cc: linux-kernel@vger.kernel.org, Andrew Morton , Arnd Bergmann , Benjamin Herrenschmidt , Brian Cain , Catalin Marinas , Fenghua Yu , Geert Uytterhoeven , Guan Xuetao , James Morse , Jonas Bonn , Julien Thierry , Ley Foon Tan , Marc Zyngier , Michael Ellerman , Paul Mackerras , Rich Felker , Russell King , Stafford Horne , Stefan Kristiansson , Suzuki K Poulose , Tony Luck , Will Deacon , Yoshinori Sato , kvmarm@lists.cs.columbia.edu, kvm-ppc@vger.kernel.org, linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org, linux-hexagon@vger.kernel.org, linux-ia64@vger.kernel.org, linux-mm@kvack.org, linuxppc-dev@lists.ozlabs.org, linux-sh@vger.kernel.org, nios2-dev@lists.rocketboards.org, openrisc@lists.librecores.org, uclinux-h8-devel@lists.sourceforge.jp, Mike Rapoport Subject: Re: [PATCH v2 07/13] powerpc: add support for folded p4d page tables Message-ID: <20200218105440.GA1698@hump> References: <20200216081843.28670-1-rppt@kernel.org> <20200216081843.28670-8-rppt@kernel.org> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline In-Reply-To: Content-Transfer-Encoding: quoted-printable X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: On Sun, Feb 16, 2020 at 11:41:07AM +0100, Christophe Leroy wrote: >=20 >=20 > Le 16/02/2020 =E0 09:18, Mike Rapoport a =E9crit=A0: > > From: Mike Rapoport > >=20 > > Implement primitives necessary for the 4th level folding, add walks o= f p4d > > level where appropriate and replace 5level-fixup.h with pgtable-nop4d= .h. >=20 > I don't think it is worth adding all this additionnals walks of p4d, th= is > patch could be limited to changes like: >=20 > - pud =3D pud_offset(pgd, gpa); > + pud =3D pud_offset(p4d_offset(pgd, gpa), gpa); >=20 > The additionnal walks should be added through another patch the day pow= erpc > need them. Ok, I'll update the patch to reduce walking the p4d. =20 > See below for more comments. >=20 > >=20 > > Signed-off-by: Mike Rapoport > > Tested-by: Christophe Leroy # 8xx and 83xx > > --- ... > > diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powe= rpc/include/asm/book3s/64/pgtable.h > > index 201a69e6a355..ddddbafff0ab 100644 > > --- a/arch/powerpc/include/asm/book3s/64/pgtable.h > > +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h > > @@ -2,7 +2,7 @@ > > #ifndef _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ > > #define _ASM_POWERPC_BOOK3S_64_PGTABLE_H_ > > -#include > > +#include > > #ifndef __ASSEMBLY__ > > #include > > @@ -251,7 +251,7 @@ extern unsigned long __pmd_frag_size_shift; > > /* Bits to mask out from a PUD to get to the PMD page */ > > #define PUD_MASKED_BITS 0xc0000000000000ffUL > > /* Bits to mask out from a PGD to get to the PUD page */ > > -#define PGD_MASKED_BITS 0xc0000000000000ffUL > > +#define P4D_MASKED_BITS 0xc0000000000000ffUL > > /* > > * Used as an indicator for rcu callback functions > > @@ -949,54 +949,60 @@ static inline bool pud_access_permitted(pud_t p= ud, bool write) > > return pte_access_permitted(pud_pte(pud), write); > > } > > -#define pgd_write(pgd) pte_write(pgd_pte(pgd)) > > +#define __p4d_raw(x) ((p4d_t) { __pgd_raw(x) }) > > +static inline __be64 p4d_raw(p4d_t x) > > +{ > > + return pgd_raw(x.pgd); > > +} > > + >=20 > Shouldn't this be defined in asm/pgtable-be-types.h, just like other > __pxx_raw() ? Ideally yes, but this creates weird header file dependencies and untangli= ng them would generate way too much churn. =20 > > +#define p4d_write(p4d) pte_write(p4d_pte(p4d)) > > -static inline void pgd_clear(pgd_t *pgdp) > > +static inline void p4d_clear(p4d_t *p4dp) > > { > > - *pgdp =3D __pgd(0); > > + *p4dp =3D __p4d(0); > > } ... > > @@ -573,9 +596,15 @@ int kvmppc_create_pte(struct kvm *kvm, pgd_t *pg= table, pte_t pte, > > /* Traverse the guest's 2nd-level tree, allocate new levels needed= */ > > pgd =3D pgtable + pgd_index(gpa); > > - pud =3D NULL; > > + p4d =3D NULL; > > if (pgd_present(*pgd)) > > - pud =3D pud_offset(pgd, gpa); > > + p4d =3D p4d_offset(pgd, gpa); > > + else > > + new_p4d =3D p4d_alloc_one(kvm->mm, gpa); > > + > > + pud =3D NULL; > > + if (p4d_present(*p4d)) > > + pud =3D pud_offset(p4d, gpa); >=20 > Is it worth adding all this new code ? >=20 > My understanding is that the series objective is to get rid of > __ARCH_HAS_5LEVEL_HACK, to to add support for 5 levels to an architectu= re > that not need it (at least for now). > If we want to add support for 5 levels, it can be done later in another > patch. >=20 > Here I think your change could be limited to: >=20 > - pud =3D pud_offset(pgd, gpa); > + pud =3D pud_offset(p4d_offset(pgd, gpa), gpa); This won't work. Without __ARCH_USE_5LEVEL_HACK defined pgd_present() is hardwired to 1 and the actual check for the top level is performed with p4d_present(). The 'else' clause that allocates p4d will never be taken a= nd it could be removed, but I prefer to keep it for consistency. =20 > > else > > new_pud =3D pud_alloc_one(kvm->mm, gpa); > > @@ -597,12 +626,18 @@ int kvmppc_create_pte(struct kvm *kvm, pgd_t *p= gtable, pte_t pte, > > /* Now traverse again under the lock and change the tree */ > > ret =3D -ENOMEM; > > if (pgd_none(*pgd)) { > > + if (!new_p4d) > > + goto out_unlock; > > + pgd_populate(kvm->mm, pgd, new_p4d); > > + new_p4d =3D NULL; > > + } > > + if (p4d_none(*p4d)) { > > if (!new_pud) > > goto out_unlock; > > - pgd_populate(kvm->mm, pgd, new_pud); > > + p4d_populate(kvm->mm, p4d, new_pud); > > new_pud =3D NULL; > > } > > - pud =3D pud_offset(pgd, gpa); > > + pud =3D pud_offset(p4d, gpa); > > if (pud_is_leaf(*pud)) { > > unsigned long hgpa =3D gpa & PUD_MASK; > > @@ -1220,6 +1255,7 @@ static ssize_t debugfs_radix_read(struct file *= file, char __user *buf, > > pgd_t *pgt; > > struct kvm_nested_guest *nested; > > pgd_t pgd, *pgdp; > > + p4d_t p4d, *p4dp; > > pud_t pud, *pudp; > > pmd_t pmd, *pmdp; > > pte_t *ptep; > > @@ -1298,7 +1334,14 @@ static ssize_t debugfs_radix_read(struct file = *file, char __user *buf, > > continue; > > } > > - pudp =3D pud_offset(&pgd, gpa); > > + p4dp =3D p4d_offset(&pgd, gpa); > > + p4d =3D READ_ONCE(*p4dp); > > + if (!(p4d_val(p4d) & _PAGE_PRESENT)) { > > + gpa =3D (gpa & P4D_MASK) + P4D_SIZE; > > + continue; > > + } > > + > > + pudp =3D pud_offset(&p4d, gpa); >=20 > Same, here you are forcing a useless read with READ_ONCE(). >=20 > Your change could be limited to >=20 > - pudp =3D pud_offset(&pgd, gpa); > + pudp =3D pud_offset(p4d_offset(&pgd, gpa), gpa); Here again the actual check must be done against p4d rather than pgd. We could skip READ_ONCE() for pgd, but since it is a debugfs method I don't think it is more important than code consistency. =20 > This comment applies to many other places. I'll make another pass to see where we can take the shortcut and use=20 pudp =3D pud_offset(p4d_offset(...)) =20 > > pud =3D READ_ONCE(*pudp); > > if (!(pud_val(pud) & _PAGE_PRESENT)) { > > gpa =3D (gpa & PUD_MASK) + PUD_SIZE; > > diff --git a/arch/powerpc/lib/code-patching.c b/arch/powerpc/lib/code= -patching.c > > index 3345f039a876..7a59f6863cec 100644 > > --- a/arch/powerpc/lib/code-patching.c > > +++ b/arch/powerpc/lib/code-patching.c > > @@ -107,13 +107,18 @@ static inline int unmap_patch_area(unsigned lon= g addr) > > pte_t *ptep; > > pmd_t *pmdp; > > pud_t *pudp; > > + p4d_t *p4dp; > > pgd_t *pgdp; > > pgdp =3D pgd_offset_k(addr); > > if (unlikely(!pgdp)) > > return -EINVAL; > > - pudp =3D pud_offset(pgdp, addr); > > + p4dp =3D p4d_offset(pgdp, addr); > > + if (unlikely(!p4dp)) > > + return -EINVAL; > > + > > + pudp =3D pud_offset(p4dp, addr); > > if (unlikely(!pudp)) > > return -EINVAL; > > diff --git a/arch/powerpc/mm/book3s32/mmu.c b/arch/powerpc/mm/book3s3= 2/mmu.c > > index 0a1c65a2c565..b2fc3e71165c 100644 > > --- a/arch/powerpc/mm/book3s32/mmu.c > > +++ b/arch/powerpc/mm/book3s32/mmu.c > > @@ -312,7 +312,7 @@ void hash_preload(struct mm_struct *mm, unsigned = long ea) > > if (!Hash) > > return; > > - pmd =3D pmd_offset(pud_offset(pgd_offset(mm, ea), ea), ea); > > + pmd =3D pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, ea), ea), e= a), ea); >=20 > If we continue like this, in ten years this like is going to be many > kilometers long. >=20 > I think the above would be worth a generic helper. Agree. My plan was to first unify all the architectures and then start introducing the generic helpers, like e.g. pmd_offset_mm(). =20 > > if (!pmd_none(*pmd)) > > add_hash_page(mm->context.id, ea, pmd_val(*pmd)); > > } > > diff --git a/arch/powerpc/mm/book3s32/tlb.c b/arch/powerpc/mm/book3s3= 2/tlb.c > > index 2fcd321040ff..175bc33b41b7 100644 > > --- a/arch/powerpc/mm/book3s32/tlb.c > > +++ b/arch/powerpc/mm/book3s32/tlb.c > > @@ -87,7 +87,7 @@ static void flush_range(struct mm_struct *mm, unsig= ned long start, > > if (start >=3D end) > > return; > > end =3D (end - 1) | ~PAGE_MASK; > > - pmd =3D pmd_offset(pud_offset(pgd_offset(mm, start), start), start)= ; > > + pmd =3D pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, start), sta= rt), start), start); > > for (;;) { > > pmd_end =3D ((start + PGDIR_SIZE) & PGDIR_MASK) - 1; > > if (pmd_end > end) > > @@ -145,7 +145,7 @@ void flush_tlb_page(struct vm_area_struct *vma, u= nsigned long vmaddr) > > return; > > } > > mm =3D (vmaddr < TASK_SIZE)? vma->vm_mm: &init_mm; > > - pmd =3D pmd_offset(pud_offset(pgd_offset(mm, vmaddr), vmaddr), vmad= dr); > > + pmd =3D pmd_offset(pud_offset(p4d_offset(pgd_offset(mm, vmaddr), vm= addr), vmaddr), vmaddr); > > if (!pmd_none(*pmd)) > > flush_hash_pages(mm->context.id, vmaddr, pmd_val(*pmd), 1); > > } > > diff --git a/arch/powerpc/mm/book3s64/hash_pgtable.c b/arch/powerpc/m= m/book3s64/hash_pgtable.c > > index 64733b9cb20a..9cd15937e88a 100644 > > --- a/arch/powerpc/mm/book3s64/hash_pgtable.c > > +++ b/arch/powerpc/mm/book3s64/hash_pgtable.c > > @@ -148,6 +148,7 @@ void hash__vmemmap_remove_mapping(unsigned long s= tart, > > int hash__map_kernel_page(unsigned long ea, unsigned long pa, pgpro= t_t prot) > > { > > pgd_t *pgdp; > > + p4d_t *p4dp; > > pud_t *pudp; > > pmd_t *pmdp; > > pte_t *ptep; > > @@ -155,7 +156,8 @@ int hash__map_kernel_page(unsigned long ea, unsig= ned long pa, pgprot_t prot) > > BUILD_BUG_ON(TASK_SIZE_USER64 > H_PGTABLE_RANGE); > > if (slab_is_available()) { > > pgdp =3D pgd_offset_k(ea); > > - pudp =3D pud_alloc(&init_mm, pgdp, ea); > > + p4dp =3D p4d_offset(pgdp, ea); > > + pudp =3D pud_alloc(&init_mm, p4dp, ea); >=20 > Could be a single line, without a new var. >=20 > - pudp =3D pud_alloc(&init_mm, pgdp, ea); > + pudp =3D pud_alloc(&init_mm, p4d_offset(pgdp, ea), ea); >=20 >=20 > Same kind of comments as already done apply to the rest. >=20 > Christophe --=20 Sincerely yours, Mike.