From patchwork Mon Mar 10 13:22:21 2025 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xu Lu X-Patchwork-Id: 14010133 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 24D57C35FF2 for ; Mon, 10 Mar 2025 14:49:14 +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:MIME-Version:References:In-Reply-To: Message-Id:Date:Subject:Cc:To:From:Reply-To:Content-ID:Content-Description: Resent-Date:Resent-From:Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID: List-Owner; bh=mXyCCGtFCqs8uYxL9FvBgUoRzSU+T8vT3SKQ3B+D3u4=; b=1MBzYg/4q8rc2D k9it9jnMagL/s0oq0qaRQmPP1+Vjvgg+LaHPn80wQqOIXm3J0PUycrKI2F8+wIb8Oy4woonqyJ87d wRlH8JRlltiy4rSZzPxq9daV34rLo/DW+txvUNcRy69Hl2Hr3KBwEEr7srygFcQ3Vu1A195EBbq03 k7MzPuYUQK1Bk7qDtTvdbEBDdAnMgqeWuwkfLOq21Q5QJtP7fwb3VAK7j7SiD/XQmJhM8X/eLpUse YT1DVd+woWXJQ1AjKp5uPOQuB7hSIFTusL9lMf4YzWhbONLYq7eQ/86cCX3ZbsMzPBBmsc8kVg2SV XTpmE2HhLY0JS84Xv+/w==; Received: from localhost ([::1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.98 #2 (Red Hat Linux)) id 1treRH-000000030Ll-3c2Q; Mon, 10 Mar 2025 14:49:07 +0000 Received: from mail-pj1-x1029.google.com ([2607:f8b0:4864:20::1029]) by bombadil.infradead.org with esmtps (Exim 4.98 #2 (Red Hat Linux)) id 1trd5d-00000002mRh-3K7D for linux-riscv@lists.infradead.org; Mon, 10 Mar 2025 13:22:42 +0000 Received: by mail-pj1-x1029.google.com with SMTP id 98e67ed59e1d1-2ff69365e1dso6079049a91.3 for ; Mon, 10 Mar 2025 06:22:41 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=bytedance.com; s=google; t=1741612961; x=1742217761; darn=lists.infradead.org; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:from:to:cc:subject:date :message-id:reply-to; bh=CbZEX0tRaUwuzIIhwgz0oK3rdSaEKh+GGhpQM1uwbPo=; b=VbnCCU3vmeLSMboB/x6H2Xz0sa0LHVTU8zn0w8E/vMBp0PWQEBovKLc9oa6Rm0Q6KF fRX/Fi37BIdMf/oO0A59PZ1/CFeF0gDzW1VUou76A4Z35Of6tSF/Qk4Ky4aRUo3cI7ot EWIR+jXYDsVxmc1D45tnNVgyAJVDytDCwSZwO4y00mAexz44bNRtQZC3KL6b/HIATbIS WSY8reUM90zr/LREjcuwC75V/regPeP8/EVGeR4XFuxLg2i3uCxHxhmmE9YhF5zpLVlT n1p8IpvvK6C6xCi6tykpY7NIZUI0EozeYq3fWEtKe/QLLcfE/bsjhqEfyl4OholH0pil mOLA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1741612961; x=1742217761; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=CbZEX0tRaUwuzIIhwgz0oK3rdSaEKh+GGhpQM1uwbPo=; b=w2LpgaUd+zt9YzXBent4g9P4gVUgF5sGDDD6kY7IWaS/ljneNj1tDy9cAS+2ezczR6 K4DE0mGiQHR9CXhsT/ArasPXx4bnxC0LWkFnaU6oiFTFgVLUfvor4M+baHrjUmsrp2jX tfAVxRZwujVSy8MdoEALLVCxBPk8AL5RaqJduS3auB/hHhUD034fAG+LiFDoTgo5Hbl9 9OuGL3E0685jeCf0YI7wb2Yn7J1E5cdQh563RZ5/il1KuyRKgRGMxDjMeWHGHiDctMGL 34qWGw5BzWhjiVO69yVDvPRCmBe7M12hmcV6X4eF6zOpCZhLUczazRtYHN3FekdIDBRu j2lQ== X-Forwarded-Encrypted: i=1; AJvYcCX4kUmLx6qJQeAjkd4Vbnmt7zzCHbXFCFZSPtYol3IvP0nbfEhhUFNRGCeQ7gyreZ0yA9PxMiVxYMAqRg==@lists.infradead.org X-Gm-Message-State: AOJu0YyazobZZjQC3u6Y4oqDmzi9Ti5T9oNAC+PFZjmf42W00AWlQz7I Y1jtWPkgBWhvtam9Tf+fVURv0Amr9Ts33GAMOGFGf7KaTkaMnUyHn8QGm2aYbCQ= X-Gm-Gg: ASbGncv3bQeM9HKnmiTCQnAS3rx+rrUhbVApuUSUvPRMFwrHRrka1/Qh8hTcfHSd9nR LkvVIOsMS/rSqXNROynaNCjKeHzOoOPdn7BSJAv7Zop8C+bngMs0uRNvaWO9e2sF9ulM4c/a8up kLGoaixzPfCAmTCGd6TCwkUbnwhVSicLHCObDIPdh2RBU+98xiJ3tTsEbeTrH9R2FGWMsIdZ8au YO8nVqoq6GfK1XIgse7rOdAgPt5zRfZp3DMt18SMLDFguPn4C2E9Z/WA/tE5zQjRL9IbOoEURHA G9mC5Q3TIi4mmB34PRsxIr3xpOMF4DgxFuKCB7vEoaHRMGHc/66oBMioNG1ozR5kZFruVNWe8fS Uj91oNEUNXgNPwxz4dS8GB8VtN6Q= X-Google-Smtp-Source: AGHT+IHCQte13YpgjRC8VlNMd1j/6bcecbJqZgEElVCIYtvZlpVlkeYhvsrIux781DK+TTsFHbqgBw== X-Received: by 2002:a17:90b:4c4a:b0:2ff:7ad4:77b1 with SMTP id 98e67ed59e1d1-2ff7ce4f260mr22306375a91.2.1741612960728; Mon, 10 Mar 2025 06:22:40 -0700 (PDT) Received: from J9GPGXL7NT.bytedance.net ([61.213.176.55]) by smtp.gmail.com with ESMTPSA id d9443c01a7336-224109e8617sm77318785ad.61.2025.03.10.06.22.37 (version=TLS1_3 cipher=TLS_CHACHA20_POLY1305_SHA256 bits=256/256); Mon, 10 Mar 2025 06:22:40 -0700 (PDT) From: Xu Lu To: akpm@linux-foundation.org, tjeznach@rivosinc.com, joro@8bytes.org, will@kernel.org, robin.murphy@arm.com Cc: lihangjing@bytedance.com, xieyongji@bytedance.com, linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org, Xu Lu Subject: [PATCH 3/4] iommu/riscv: Introduce IOMMU page table lock Date: Mon, 10 Mar 2025 21:22:21 +0800 Message-Id: <20250310132222.58378-4-luxu.kernel@bytedance.com> X-Mailer: git-send-email 2.39.5 (Apple Git-154) In-Reply-To: <20250310132222.58378-1-luxu.kernel@bytedance.com> References: <20250310132222.58378-1-luxu.kernel@bytedance.com> MIME-Version: 1.0 X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20250310_062241_832944_EF507CC3 X-CRM114-Status: GOOD ( 16.30 ) 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: , Sender: "linux-riscv" Errors-To: linux-riscv-bounces+linux-riscv=archiver.kernel.org@lists.infradead.org Introduce page table lock to address competition issues when modifying multiple PTEs, for example, when applying Svnapot. We use fine-grained page table locks to minimize lock contention. Signed-off-by: Xu Lu --- drivers/iommu/riscv/iommu.c | 126 ++++++++++++++++++++++++++++++------ 1 file changed, 108 insertions(+), 18 deletions(-) diff --git a/drivers/iommu/riscv/iommu.c b/drivers/iommu/riscv/iommu.c index f752096989a79..ffc474987a075 100644 --- a/drivers/iommu/riscv/iommu.c +++ b/drivers/iommu/riscv/iommu.c @@ -808,6 +808,7 @@ struct riscv_iommu_domain { struct iommu_domain domain; struct list_head bonds; spinlock_t lock; /* protect bonds list updates. */ + spinlock_t page_table_lock; /* protect page table updates. */ int pscid; bool amo_enabled; int numa_node; @@ -1086,8 +1087,80 @@ static void riscv_iommu_iotlb_sync(struct iommu_domain *iommu_domain, #define _io_pte_none(pte) (pte_val(pte) == 0) #define _io_pte_entry(pn, prot) (__pte((_PAGE_PFN_MASK & ((pn) << _PAGE_PFN_SHIFT)) | (prot))) +#define RISCV_IOMMU_PMD_LEVEL 1 + +static bool riscv_iommu_ptlock_init(struct ptdesc *ptdesc, int level) +{ + if (level <= RISCV_IOMMU_PMD_LEVEL) + return ptlock_init(ptdesc); + return true; +} + +static void riscv_iommu_ptlock_free(struct ptdesc *ptdesc, int level) +{ + if (level <= RISCV_IOMMU_PMD_LEVEL) + ptlock_free(ptdesc); +} + +static spinlock_t *riscv_iommu_ptlock(struct riscv_iommu_domain *domain, + pte_t *pte, int level) +{ + spinlock_t *ptl; + +#ifdef CONFIG_SPLIT_PTE_PTLOCKS + if (level <= RISCV_IOMMU_PMD_LEVEL) + ptl = ptlock_ptr(page_ptdesc(virt_to_page(pte))); + else +#endif + ptl = &domain->page_table_lock; + spin_lock(ptl); + + return ptl; +} + +static void *riscv_iommu_alloc_pagetable_node(int numa_node, gfp_t gfp, int level) +{ + struct ptdesc *ptdesc; + void *addr; + + addr = iommu_alloc_page_node(numa_node, gfp); + if (!addr) + return NULL; + + ptdesc = page_ptdesc(virt_to_page(addr)); + if (!riscv_iommu_ptlock_init(ptdesc, level)) { + iommu_free_page(addr); + addr = NULL; + } + + return addr; +} + +static void riscv_iommu_free_pagetable(void *addr, int level) +{ + struct ptdesc *ptdesc = page_ptdesc(virt_to_page(addr)); + + riscv_iommu_ptlock_free(ptdesc, level); + iommu_free_page(addr); +} + +static int pgsize_to_level(size_t pgsize) +{ + int level = RISCV_IOMMU_DC_FSC_IOSATP_MODE_SV57 - + RISCV_IOMMU_DC_FSC_IOSATP_MODE_SV39 + 2; + int shift = PAGE_SHIFT + PT_SHIFT * level; + + while (pgsize < ((size_t)1 << shift)) { + shift -= PT_SHIFT; + level--; + } + + return level; +} + static void riscv_iommu_pte_free(struct riscv_iommu_domain *domain, - pte_t pte, struct list_head *freelist) + pte_t pte, int level, + struct list_head *freelist) { pte_t *ptr; int i; @@ -1102,10 +1175,11 @@ static void riscv_iommu_pte_free(struct riscv_iommu_domain *domain, pte = ptr[i]; if (!_io_pte_none(pte)) { ptr[i] = __pte(0); - riscv_iommu_pte_free(domain, pte, freelist); + riscv_iommu_pte_free(domain, pte, level - 1, freelist); } } + riscv_iommu_ptlock_free(page_ptdesc(virt_to_page(ptr)), level); if (freelist) list_add_tail(&virt_to_page(ptr)->lru, freelist); else @@ -1117,8 +1191,9 @@ static pte_t *riscv_iommu_pte_alloc(struct riscv_iommu_domain *domain, gfp_t gfp) { pte_t *ptr = domain->pgd_root; - pte_t pte, old; + pte_t pte; int level = domain->pgd_mode - RISCV_IOMMU_DC_FSC_IOSATP_MODE_SV39 + 2; + spinlock_t *ptl; void *addr; do { @@ -1146,15 +1221,21 @@ static pte_t *riscv_iommu_pte_alloc(struct riscv_iommu_domain *domain, * page table. This might race with other mappings, retry. */ if (_io_pte_none(pte)) { - addr = iommu_alloc_page_node(domain->numa_node, gfp); + addr = riscv_iommu_alloc_pagetable_node(domain->numa_node, gfp, + level - 1); if (!addr) return NULL; - old = pte; - pte = _io_pte_entry(virt_to_pfn(addr), _PAGE_TABLE); - if (cmpxchg_relaxed(ptr, old, pte) != old) { - iommu_free_page(addr); + + ptl = riscv_iommu_ptlock(domain, ptr, level); + pte = ptep_get(ptr); + if (!_io_pte_none(pte)) { + spin_unlock(ptl); + riscv_iommu_free_pagetable(addr, level - 1); goto pte_retry; } + pte = _io_pte_entry(virt_to_pfn(addr), _PAGE_TABLE); + set_pte(ptr, pte); + spin_unlock(ptl); } ptr = (pte_t *)pfn_to_virt(pte_pfn(pte)); } while (level-- > 0); @@ -1194,9 +1275,10 @@ static int riscv_iommu_map_pages(struct iommu_domain *iommu_domain, struct riscv_iommu_domain *domain = iommu_domain_to_riscv(iommu_domain); size_t size = 0; pte_t *ptr; - pte_t pte, old; + pte_t pte; unsigned long pte_prot; - int rc = 0; + int rc = 0, level; + spinlock_t *ptl; LIST_HEAD(freelist); if (!(prot & IOMMU_WRITE)) @@ -1213,11 +1295,12 @@ static int riscv_iommu_map_pages(struct iommu_domain *iommu_domain, break; } - old = ptep_get(ptr); + level = pgsize_to_level(pgsize); + ptl = riscv_iommu_ptlock(domain, ptr, level); + riscv_iommu_pte_free(domain, ptep_get(ptr), level, &freelist); pte = _io_pte_entry(phys_to_pfn(phys), pte_prot); set_pte(ptr, pte); - - riscv_iommu_pte_free(domain, old, &freelist); + spin_unlock(ptl); size += pgsize; iova += pgsize; @@ -1252,6 +1335,7 @@ static size_t riscv_iommu_unmap_pages(struct iommu_domain *iommu_domain, pte_t *ptr; size_t unmapped = 0; size_t pte_size; + spinlock_t *ptl; while (unmapped < size) { ptr = riscv_iommu_pte_fetch(domain, iova, &pte_size); @@ -1262,7 +1346,9 @@ static size_t riscv_iommu_unmap_pages(struct iommu_domain *iommu_domain, if (iova & (pte_size - 1)) return unmapped; + ptl = riscv_iommu_ptlock(domain, ptr, pgsize_to_level(pte_size)); set_pte(ptr, __pte(0)); + spin_unlock(ptl); iommu_iotlb_gather_add_page(&domain->domain, gather, iova, pte_size); @@ -1292,13 +1378,14 @@ static void riscv_iommu_free_paging_domain(struct iommu_domain *iommu_domain) { struct riscv_iommu_domain *domain = iommu_domain_to_riscv(iommu_domain); const unsigned long pfn = virt_to_pfn(domain->pgd_root); + int level = domain->pgd_mode - RISCV_IOMMU_DC_FSC_IOSATP_MODE_SV39 + 2; WARN_ON(!list_empty(&domain->bonds)); if ((int)domain->pscid > 0) ida_free(&riscv_iommu_pscids, domain->pscid); - riscv_iommu_pte_free(domain, _io_pte_entry(pfn, _PAGE_TABLE), NULL); + riscv_iommu_pte_free(domain, _io_pte_entry(pfn, _PAGE_TABLE), level, NULL); kfree(domain); } @@ -1359,7 +1446,7 @@ static struct iommu_domain *riscv_iommu_alloc_paging_domain(struct device *dev) struct riscv_iommu_device *iommu; unsigned int pgd_mode; dma_addr_t va_mask; - int va_bits; + int va_bits, level; iommu = dev_to_iommu(dev); if (iommu->caps & RISCV_IOMMU_CAPABILITIES_SV57) { @@ -1382,11 +1469,14 @@ static struct iommu_domain *riscv_iommu_alloc_paging_domain(struct device *dev) INIT_LIST_HEAD_RCU(&domain->bonds); spin_lock_init(&domain->lock); + spin_lock_init(&domain->page_table_lock); domain->numa_node = dev_to_node(iommu->dev); domain->amo_enabled = !!(iommu->caps & RISCV_IOMMU_CAPABILITIES_AMO_HWAD); domain->pgd_mode = pgd_mode; - domain->pgd_root = iommu_alloc_page_node(domain->numa_node, - GFP_KERNEL_ACCOUNT); + level = domain->pgd_mode - RISCV_IOMMU_DC_FSC_IOSATP_MODE_SV39 + 2; + domain->pgd_root = riscv_iommu_alloc_pagetable_node(domain->numa_node, + GFP_KERNEL_ACCOUNT, + level); if (!domain->pgd_root) { kfree(domain); return ERR_PTR(-ENOMEM); @@ -1395,7 +1485,7 @@ static struct iommu_domain *riscv_iommu_alloc_paging_domain(struct device *dev) domain->pscid = ida_alloc_range(&riscv_iommu_pscids, 1, RISCV_IOMMU_MAX_PSCID, GFP_KERNEL); if (domain->pscid < 0) { - iommu_free_page(domain->pgd_root); + riscv_iommu_free_pagetable(domain->pgd_root, level); kfree(domain); return ERR_PTR(-ENOMEM); }