From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-alma10-1.taild15c8.ts.net [100.103.45.18]) (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 10A8046F489; Tue, 21 Jul 2026 16:24:43 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=100.103.45.18 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1784651085; cv=none; b=TXP5L9ManSYbhIuSG8lyxaUcv9ekbbgLTJfnA7Dq2U0mAdFMjUqHNvDX+bPlmH0w7LKkBLYrSeQPHk+4tZWERy6OE/4VUWjQDWjR0Zi9uJbUcbFV2WQqhMxDnLU4SXvmWf/SMrA+ADrHoe0Ixb5YeElXOqoPR7Gl4a9VWGnhpwI= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1784651085; c=relaxed/simple; bh=UtFEd4bDoNqGjRjg/yG8UKvLcV9RwvxmUMaNCsLgmfE=; h=From:Date:Subject:MIME-Version:Content-Type:Message-Id:References: In-Reply-To:To:Cc; b=YEu81yyRuhMs6z2TzcscQzDLDr5An9e1vN/+QwEiWMa/Mg2BcllF3Ygt9zqDno2Ph6TAC5HzqnFKc36l40otxspDCVviem9BsuR3xExWPZEbfUIBhm+ax9CY4gGP5ZvoSchbEK3+oZB48shYkDA+m5BCzAIOlzLxi6Cm+NXYyGU= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=mduyV/4o; arc=none smtp.client-ip=100.103.45.18 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="mduyV/4o" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 1032F1F00A3A; Tue, 21 Jul 2026 16:24:36 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel.org; s=k20260515; t=1784651083; bh=nokpUTK/BQEU2xPd8dil79b/pTUs86asyupxsitnFKA=; h=From:Date:Subject:References:In-Reply-To:To:Cc; b=mduyV/4oIfEViBoL6lriq5H4veut2axJNwNq2s8j+Qo/kcJQY2HEyGHhleoOmd1Ba KqLtqTyZvyOcUE/AJKvW6Fvi+sOLhFoZYGyhft2SjmZUALxgwD5DGq9OsXZr8lT1+t Z1NUg49kmKteh+uVkht2fdwI7NtwumvsJCfpA9k8YL5d/G4pidA3LDhwZQisFKj0WT bJlJEAeJYy76RZjM+xZVDhcI3yxkOWWVkJLQMTUtiPeaY8q902iyg65RZ21W/j9hWZ plSCPQAYE8Bab5QU90e8Xoo9DivS7xzxi6Pg0rG3OxFpbwABrQidREgO0VdToD8Had qxbQhX+cRlHKQ== From: "Mike Rapoport (Microsoft)" Date: Tue, 21 Jul 2026 19:23:32 +0300 Subject: [PATCH RFC 09/18] x86/mm/pat: use a single locked lookup for PTE verification Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit Message-Id: <20260721-generic-set-memory-v0-1-v1-9-2c1fc62306b3@kernel.org> References: <20260721-generic-set-memory-v0-1-v1-0-2c1fc62306b3@kernel.org> In-Reply-To: <20260721-generic-set-memory-v0-1-v1-0-2c1fc62306b3@kernel.org> To: linux-mm@kvack.org Cc: Albert Ou , Andrew Morton , Andy Lutomirski , Borislav Petkov , Brendan Jackman , Catalin Marinas , Dave Hansen , David Hildenbrand , Huacai Chen , Ingo Molnar , Ingo Molnar , Kevin Brodsky , "Liam R. Howlett" , Lorenzo Stoakes , Michal Hocko , Mike Rapoport , Palmer Dabbelt , Paul Walmsley , Peter Zijlstra , "H. Peter Anvin" , Suren Baghdasaryan , Thomas Gleixner , Vlastimil Babka , Will Deacon , linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, linux-riscv@lists.infradead.org, loongarch@lists.linux.dev, x86@kernel.org X-Mailer: b4 0.16-dev Currently both __should_split_large_page() and __split_large_page() verify that the page table entry wasn't changed by another CPU while pgd_lock was released. pgd_lock must be released between __should_split_large_page() and __split_large_page() because split of a large page requires allocation of a new page table page. Preallocating the page table page before checking if the large page should be split allows running both __should_split_large_page() and __split_large_page() inside the same critical section protected by pgd_lock and allows dropping the duplicated lookup. Move allocation of the page table before the call to should_split_large_page() and pull out race-checking lookup into cpa_handle_large_page(). Assisted-by: Copilot:claude-opus-4.8 Signed-off-by: Mike Rapoport (Microsoft) --- arch/x86/mm/pat/set_memory.c | 89 ++++++++++++++++++++------------------------ 1 file changed, 40 insertions(+), 49 deletions(-) diff --git a/arch/x86/mm/pat/set_memory.c b/arch/x86/mm/pat/set_memory.c index c2a72cd8b832..ef206ba6e254 100644 --- a/arch/x86/mm/pat/set_memory.c +++ b/arch/x86/mm/pat/set_memory.c @@ -60,6 +60,9 @@ struct cpa_split_data { unsigned long address; pte_t *kpte; struct ptdesc *ptdesc; + enum pgtable_level level; + bool nx; + bool rw; }; enum cpa_warn { @@ -883,19 +886,12 @@ static int __should_split_large_page(struct cpa_data *cpa, { unsigned long numpages, pmask, psize, lpaddr, pfn, old_pfn; pgprot_t old_prot, new_prot, req_prot, chk_prot; + enum pgtable_level level = sd->level; unsigned long address = sd->address; - enum pgtable_level level; pte_t *kpte = sd->kpte; - pte_t new_pte, *tmp; - bool nx, rw; - - /* - * Check for races, another CPU might have split this page - * up already: - */ - tmp = _lookup_address_cpa(cpa, address, &level, &nx, &rw); - if (tmp != kpte) - return 1; + bool nx = sd->nx; + bool rw = sd->rw; + pte_t new_pte; switch (level) { case PGTABLE_LEVEL_PMD: @@ -1030,16 +1026,10 @@ static int __should_split_large_page(struct cpa_data *cpa, static int should_split_large_page(struct cpa_data *cpa, struct cpa_split_data *sd) { - int do_split; - if (cpa->force_split) return 1; - spin_lock(&pgd_lock); - do_split = __should_split_large_page(cpa, sd); - spin_unlock(&pgd_lock); - - return do_split; + return __should_split_large_page(cpa, sd); } static void split_set_pte(struct cpa_data *cpa, pte_t *pte, unsigned long pfn, @@ -1084,23 +1074,11 @@ static int __split_large_page(struct cpa_data *cpa, struct cpa_split_data *sd) struct ptdesc *ptdesc = sd->ptdesc; struct page *base = ptdesc_page(ptdesc); pte_t *pbase = (pte_t *)page_address(base); + enum pgtable_level level = sd->level; unsigned long address = sd->address; pte_t *kpte = sd->kpte; - unsigned int i, level; + unsigned int i; pgprot_t ref_prot; - bool nx, rw; - pte_t *tmp; - - spin_lock(&pgd_lock); - /* - * Check for races, another CPU might have split this page - * up for us already: - */ - tmp = _lookup_address_cpa(cpa, address, &level, &nx, &rw); - if (tmp != kpte) { - spin_unlock(&pgd_lock); - return 1; - } paravirt_alloc_pte(&init_mm, page_to_pfn(base)); @@ -1133,7 +1111,6 @@ static int __split_large_page(struct cpa_data *cpa, struct cpa_split_data *sd) break; default: - spin_unlock(&pgd_lock); return 1; } @@ -1181,23 +1158,13 @@ static int __split_large_page(struct cpa_data *cpa, struct cpa_split_data *sd) * just split large page entry. */ flush_tlb_all(); - spin_unlock(&pgd_lock); return 0; } static int split_large_page(struct cpa_data *cpa, struct cpa_split_data *sd) { - spin_unlock(&cpa_lock); - sd->ptdesc = pagetable_alloc(GFP_KERNEL, 0); - spin_lock(&cpa_lock); - if (!sd->ptdesc) - return -ENOMEM; - - if (__split_large_page(cpa, sd)) - pagetable_free(sd->ptdesc); - - return 0; + return __split_large_page(cpa, sd); } static int cpa_handle_large_page(struct cpa_data *cpa, pte_t *kpte, @@ -1207,13 +1174,31 @@ static int cpa_handle_large_page(struct cpa_data *cpa, pte_t *kpte, .address = address, .kpte = kpte, }; - int do_split, err; + int do_split, ret = 1; + pte_t *tmp; + + spin_unlock(&cpa_lock); + sd.ptdesc = pagetable_alloc(GFP_KERNEL, 0); + spin_lock(&cpa_lock); + if (!sd.ptdesc) + return -ENOMEM; + + spin_lock(&pgd_lock); + + /* + * Check for races, another CPU might have split this page + * up already: + */ + tmp = _lookup_address_cpa(cpa, address, &sd.level, &sd.nx, &sd.rw); + if (tmp != kpte) + goto out_free_ptdesc; /* * Check, whether we can keep the large page intact * and just change the pte: */ do_split = should_split_large_page(cpa, &sd); + ret = do_split; /* * When the range fits into the existing large page, no split is * required. @@ -1221,16 +1206,22 @@ static int cpa_handle_large_page(struct cpa_data *cpa, pte_t *kpte, * cpa->numpages and cpa->pfn. */ if (do_split <= 0) - return do_split; + goto out_free_ptdesc; /* * We have to split the large page: */ - err = split_large_page(cpa, &sd); - if (err) - return err; + ret = split_large_page(cpa, &sd); + if (ret) + goto out_free_ptdesc; + spin_unlock(&pgd_lock); return do_split; + +out_free_ptdesc: + spin_unlock(&pgd_lock); + pagetable_free(sd.ptdesc); + return ret; } static int collapse_pmd_page(pmd_t *pmd, unsigned long addr, -- 2.53.0