public inbox for linux-coco@lists.linux.dev
 help / color / mirror / Atom feed
From: Sean Christopherson <seanjc@google.com>
To: Thomas Gleixner <tglx@kernel.org>, Ingo Molnar <mingo@redhat.com>,
	Borislav Petkov <bp@alien8.de>,
	 Dave Hansen <dave.hansen@linux.intel.com>,
	x86@kernel.org,  Kiryl Shutsemau <kas@kernel.org>,
	Sean Christopherson <seanjc@google.com>,
	Paolo Bonzini <pbonzini@redhat.com>
Cc: linux-kernel@vger.kernel.org, linux-coco@lists.linux.dev,
	 kvm@vger.kernel.org, Kai Huang <kai.huang@intel.com>,
	 Rick Edgecombe <rick.p.edgecombe@intel.com>,
	Yan Zhao <yan.y.zhao@intel.com>,
	 Vishal Annapurve <vannapurve@google.com>,
	Ackerley Tng <ackerleytng@google.com>,
	 Sagi Shahar <sagis@google.com>,
	Binbin Wu <binbin.wu@linux.intel.com>,
	 Xiaoyao Li <xiaoyao.li@intel.com>,
	Isaku Yamahata <isaku.yamahata@intel.com>
Subject: [RFC PATCH v5 15/45] x86/virt/tdx: Improve PAMT refcounts allocation for sparse memory
Date: Wed, 28 Jan 2026 17:14:47 -0800	[thread overview]
Message-ID: <20260129011517.3545883-16-seanjc@google.com> (raw)
In-Reply-To: <20260129011517.3545883-1-seanjc@google.com>

From: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

init_pamt_metadata() allocates PAMT refcounts for all physical memory up
to max_pfn. It might be suboptimal if the physical memory layout is
discontinuous and has large holes.

The refcount allocation vmalloc allocation. This is necessary to support a
large allocation size. The virtually contiguous property also makes it
easy to find a specific 2MB range’s refcount since it can simply be
indexed.

Since vmalloc mappings support remapping during normal kernel runtime,
switch to an approach that only populates refcount pages for the vmalloc
mapping when there is actually memory for that range. This means any holes
in the physical address space won’t use actual physical memory.

The validity of this memory optimization is based on a couple assumptions:
1. Physical holes in the ram layout are commonly large enough for it to be
   worth it.
2. An alternative approach that looks the refcounts via some more layered
   data structure wouldn’t overly complicate the lookups. Or at least
   more than the complexity of managing the vmalloc mapping.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
[Add feedback, update log]
Signed-off-by: Rick Edgecombe <rick.p.edgecombe@intel.com>
Tested-by: Sagi Shahar <sagis@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 arch/x86/virt/vmx/tdx/tdx.c | 122 ++++++++++++++++++++++++++++++++++--
 1 file changed, 118 insertions(+), 4 deletions(-)

diff --git a/arch/x86/virt/vmx/tdx/tdx.c b/arch/x86/virt/vmx/tdx/tdx.c
index db48bf2ce601..f6e80aba5895 100644
--- a/arch/x86/virt/vmx/tdx/tdx.c
+++ b/arch/x86/virt/vmx/tdx/tdx.c
@@ -195,30 +195,135 @@ int tdx_cpu_enable(void)
 }
 EXPORT_SYMBOL_FOR_KVM(tdx_cpu_enable);
 
+/* Find PAMT refcount for a given physical address */
+static atomic_t *tdx_find_pamt_refcount(unsigned long pfn)
+{
+	/* Find which PMD a PFN is in. */
+	unsigned long index = pfn >> (PMD_SHIFT - PAGE_SHIFT);
+
+	return &pamt_refcounts[index];
+}
+
+/* Map a page into the PAMT refcount vmalloc region */
+static int pamt_refcount_populate(pte_t *pte, unsigned long addr, void *data)
+{
+	struct page *page;
+	pte_t entry;
+
+	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (!page)
+		return -ENOMEM;
+
+	entry = mk_pte(page, PAGE_KERNEL);
+
+	spin_lock(&init_mm.page_table_lock);
+	/*
+	 * PAMT refcount populations can overlap due to rounding of the
+	 * start/end pfn. Make sure the PAMT range is only populated once.
+	 */
+	if (pte_none(ptep_get(pte)))
+		set_pte_at(&init_mm, addr, pte, entry);
+	else
+		__free_page(page);
+	spin_unlock(&init_mm.page_table_lock);
+
+	return 0;
+}
+
 /*
- * Allocate PAMT reference counters for all physical memory.
+ * Allocate PAMT reference counters for the given PFN range.
  *
  * It consumes 2MiB for every 1TiB of physical memory.
  */
+static int alloc_pamt_refcount(unsigned long start_pfn, unsigned long end_pfn)
+{
+	unsigned long refcount_first, refcount_last;
+	unsigned long mapping_start, mapping_end;
+
+	if (!tdx_supports_dynamic_pamt(&tdx_sysinfo))
+		return 0;
+
+	/*
+	 * 'start_pfn' is inclusive and 'end_pfn' is exclusive. Find the
+	 * range of refcounts the pfn range will need.
+	 */
+	refcount_first = (unsigned long)tdx_find_pamt_refcount(start_pfn);
+	refcount_last   = (unsigned long)tdx_find_pamt_refcount(end_pfn - 1);
+
+	/*
+	 * Calculate the page aligned range that includes the refcounts. The
+	 * teardown logic needs to handle potentially overlapping refcount
+	 * mappings resulting from the alignments.
+	 */
+	mapping_start = round_down(refcount_first, PAGE_SIZE);
+	mapping_end   = round_up(refcount_last + sizeof(*pamt_refcounts), PAGE_SIZE);
+
+
+	return apply_to_page_range(&init_mm, mapping_start, mapping_end - mapping_start,
+				   pamt_refcount_populate, NULL);
+}
+
+/*
+ * Reserve vmalloc range for PAMT reference counters. It covers all physical
+ * address space up to max_pfn. It is going to be populated from
+ * build_tdx_memlist() only for present memory that available for TDX use.
+ *
+ * It reserves 2MiB of virtual address space for every 1TiB of physical memory.
+ */
 static int init_pamt_metadata(void)
 {
-	size_t size = DIV_ROUND_UP(max_pfn, PTRS_PER_PTE) * sizeof(*pamt_refcounts);
+	struct vm_struct *area;
+	size_t size;
 
 	if (!tdx_supports_dynamic_pamt(&tdx_sysinfo))
 		return 0;
 
-	pamt_refcounts = __vmalloc(size, GFP_KERNEL | __GFP_ZERO);
-	if (!pamt_refcounts)
+	size = DIV_ROUND_UP(max_pfn, PTRS_PER_PTE) * sizeof(*pamt_refcounts);
+
+	area = get_vm_area(size, VM_SPARSE);
+	if (!area)
 		return -ENOMEM;
 
+	pamt_refcounts = area->addr;
 	return 0;
 }
 
+/* Unmap a page from the PAMT refcount vmalloc region */
+static int pamt_refcount_depopulate(pte_t *pte, unsigned long addr, void *data)
+{
+	struct page *page;
+	pte_t entry;
+
+	spin_lock(&init_mm.page_table_lock);
+
+	entry = ptep_get(pte);
+	/* refcount allocation is sparse, may not be populated */
+	if (!pte_none(entry)) {
+		pte_clear(&init_mm, addr, pte);
+		page = pte_page(entry);
+		__free_page(page);
+	}
+
+	spin_unlock(&init_mm.page_table_lock);
+
+	return 0;
+}
+
+/* Unmap all PAMT refcount pages and free vmalloc range */
 static void free_pamt_metadata(void)
 {
+	size_t size;
+
 	if (!tdx_supports_dynamic_pamt(&tdx_sysinfo))
 		return;
 
+	size = max_pfn / PTRS_PER_PTE * sizeof(*pamt_refcounts);
+	size = round_up(size, PAGE_SIZE);
+
+	apply_to_existing_page_range(&init_mm,
+				     (unsigned long)pamt_refcounts,
+				     size, pamt_refcount_depopulate,
+				     NULL);
 	vfree(pamt_refcounts);
 	pamt_refcounts = NULL;
 }
@@ -289,10 +394,19 @@ static int build_tdx_memlist(struct list_head *tmb_list)
 		ret = add_tdx_memblock(tmb_list, start_pfn, end_pfn, nid);
 		if (ret)
 			goto err;
+
+		/* Allocated PAMT refcountes for the memblock */
+		ret = alloc_pamt_refcount(start_pfn, end_pfn);
+		if (ret)
+			goto err;
 	}
 
 	return 0;
 err:
+	/*
+	 * Only free TDX memory blocks here, PAMT refcount pages
+	 * will be freed in the init_tdx_module() error path.
+	 */
 	free_tdx_memlist(tmb_list);
 	return ret;
 }
-- 
2.53.0.rc1.217.geba53bf80e-goog


  parent reply	other threads:[~2026-01-29  1:15 UTC|newest]

Thread overview: 148+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-01-29  1:14 [RFC PATCH v5 00/45] TDX: Dynamic PAMT + S-EPT Hugepage Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 01/45] x86/tdx: Use pg_level in TDX APIs, not the TDX-Module's 0-based level Sean Christopherson
2026-01-29 17:37   ` Dave Hansen
2026-01-29  1:14 ` [RFC PATCH v5 02/45] KVM: x86/mmu: Update iter->old_spte if cmpxchg64 on mirror SPTE "fails" Sean Christopherson
2026-01-29 22:10   ` Edgecombe, Rick P
2026-01-29 22:23     ` Sean Christopherson
2026-01-29 22:48       ` Edgecombe, Rick P
2026-02-03  8:48   ` Yan Zhao
2026-02-03 10:30   ` Huang, Kai
2026-02-03 20:06     ` Sean Christopherson
2026-02-03 21:34       ` Huang, Kai
2026-01-29  1:14 ` [RFC PATCH v5 03/45] KVM: TDX: Account all non-transient page allocations for per-TD structures Sean Christopherson
2026-01-29 22:15   ` Edgecombe, Rick P
2026-02-03 10:36   ` Huang, Kai
2026-01-29  1:14 ` [RFC PATCH v5 04/45] KVM: x86: Make "external SPTE" ops that can fail RET0 static calls Sean Christopherson
2026-01-29 22:20   ` Edgecombe, Rick P
2026-01-30  1:28     ` Sean Christopherson
2026-01-30 17:32       ` Edgecombe, Rick P
2026-02-03 10:44         ` Huang, Kai
2026-02-04  1:16         ` Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 05/45] KVM: TDX: Drop kvm_x86_ops.link_external_spt(), use .set_external_spte() for all Sean Christopherson
2026-01-30 23:55   ` Edgecombe, Rick P
2026-02-03 10:19   ` Yan Zhao
2026-02-03 20:05     ` Sean Christopherson
2026-02-04  6:41       ` Yan Zhao
2026-02-05 23:14         ` Sean Christopherson
2026-02-06  2:27           ` Yan Zhao
2026-02-18 19:37       ` Edgecombe, Rick P
2026-02-20 17:36         ` Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 06/45] KVM: x86/mmu: Fold set_external_spte_present() into its sole caller Sean Christopherson
2026-02-04  7:38   ` Yan Zhao
2026-02-05 23:06     ` Sean Christopherson
2026-02-06  2:29       ` Yan Zhao
2026-01-29  1:14 ` [RFC PATCH v5 07/45] KVM: x86/mmu: Plumb the SPTE _pointer_ into the TDP MMU's handle_changed_spte() Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 08/45] KVM: x86/mmu: Propagate mirror SPTE removal to S-EPT in handle_changed_spte() Sean Christopherson
2026-02-04  9:06   ` Yan Zhao
2026-02-05  2:23     ` Sean Christopherson
2026-02-05  5:39       ` Yan Zhao
2026-02-05 22:33         ` Sean Christopherson
2026-02-06  2:17           ` Yan Zhao
2026-02-06 17:41             ` Sean Christopherson
2026-02-10 10:54               ` Yan Zhao
2026-02-10 19:52                 ` Sean Christopherson
2026-02-11  2:16                   ` Yan Zhao
2026-02-14  0:36                     ` Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 09/45] KVM: x86: Rework .free_external_spt() into .reclaim_external_sp() Sean Christopherson
2026-02-04  9:45   ` Yan Zhao
2026-02-05  7:04     ` Yan Zhao
2026-02-05 22:38       ` Sean Christopherson
2026-02-06  2:30         ` Yan Zhao
2026-01-29  1:14 ` [RFC PATCH v5 10/45] x86/tdx: Move all TDX error defines into <asm/shared/tdx_errno.h> Sean Christopherson
2026-01-29 18:13   ` Dave Hansen
2026-01-29  1:14 ` [RFC PATCH v5 11/45] x86/tdx: Add helpers to check return status codes Sean Christopherson
2026-01-29 18:58   ` Dave Hansen
2026-01-29 20:35     ` Sean Christopherson
2026-01-30  0:36       ` Edgecombe, Rick P
2026-02-03 20:32         ` Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 12/45] x86/virt/tdx: Simplify tdmr_get_pamt_sz() Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 13/45] x86/virt/tdx: Allocate page bitmap for Dynamic PAMT Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 14/45] x86/virt/tdx: Allocate reference counters for PAMT memory Sean Christopherson
2026-01-29  1:14 ` Sean Christopherson [this message]
2026-01-29  1:14 ` [RFC PATCH v5 16/45] x86/virt/tdx: Add tdx_alloc/free_control_page() helpers Sean Christopherson
2026-01-30  1:30   ` Sean Christopherson
2026-02-05  6:11   ` Yan Zhao
2026-02-05 22:35     ` Sean Christopherson
2026-02-06  2:32       ` Yan Zhao
2026-02-10 17:44   ` Dave Hansen
2026-02-10 22:15     ` Edgecombe, Rick P
2026-02-10 22:19       ` Dave Hansen
2026-02-10 22:46         ` Huang, Kai
2026-02-10 22:50           ` Dave Hansen
2026-02-10 23:02             ` Huang, Kai
2026-02-11  0:50     ` Edgecombe, Rick P
2026-01-29  1:14 ` [RFC PATCH v5 17/45] x86/virt/tdx: Optimize " Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 18/45] KVM: TDX: Allocate PAMT memory for TD and vCPU control structures Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 19/45] KVM: Allow owner of kvm_mmu_memory_cache to provide a custom page allocator Sean Christopherson
2026-02-03 10:56   ` Huang, Kai
2026-02-03 20:12     ` Sean Christopherson
2026-02-03 20:33       ` Edgecombe, Rick P
2026-02-03 21:17         ` Sean Christopherson
2026-02-03 21:29       ` Huang, Kai
2026-02-04  2:16         ` Sean Christopherson
2026-02-04  6:45           ` Huang, Kai
2026-01-29  1:14 ` [RFC PATCH v5 20/45] KVM: x86/mmu: Allocate/free S-EPT pages using tdx_{alloc,free}_control_page() Sean Christopherson
2026-02-03 11:16   ` Huang, Kai
2026-02-03 20:17     ` Sean Christopherson
2026-02-03 21:18       ` Huang, Kai
2026-02-06  9:48   ` Yan Zhao
2026-02-06 15:01     ` Sean Christopherson
2026-02-09  9:25       ` Yan Zhao
2026-02-09 23:20         ` Sean Christopherson
2026-02-10  8:30           ` Yan Zhao
2026-02-10  0:07         ` Dave Hansen
2026-02-10  1:40           ` Yan Zhao
2026-02-09 10:41       ` Huang, Kai
2026-02-09 22:44         ` Sean Christopherson
2026-02-10 10:54           ` Huang, Kai
2026-02-09 23:40       ` Dave Hansen
2026-02-10  0:03         ` Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 21/45] x86/tdx: Add APIs to support get/put of DPAMT entries from KVM, under spinlock Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 22/45] KVM: TDX: Get/put PAMT pages when (un)mapping private memory Sean Christopherson
2026-02-06 10:20   ` Yan Zhao
2026-02-06 16:03     ` Sean Christopherson
2026-02-06 19:27       ` Edgecombe, Rick P
2026-02-06 23:18         ` Sean Christopherson
2026-02-06 23:19           ` Edgecombe, Rick P
2026-02-09 10:33           ` Huang, Kai
2026-02-09 17:08             ` Edgecombe, Rick P
2026-02-09 21:05               ` Huang, Kai
2026-01-29  1:14 ` [RFC PATCH v5 23/45] x86/virt/tdx: Enable Dynamic PAMT Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 24/45] Documentation/x86: Add documentation for TDX's " Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 25/45] *** DO NOT MERGE *** x86/virt/tdx: Don't assume guest memory is backed by struct page Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 26/45] x86/virt/tdx: Enhance tdh_mem_page_aug() to support huge pages Sean Christopherson
2026-01-29  1:14 ` [RFC PATCH v5 27/45] x86/virt/tdx: Enhance tdh_phymem_page_wbinvd_hkid() to invalidate " Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 28/45] x86/virt/tdx: Extend "reset page" quirk to support " Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 29/45] x86/virt/tdx: Get/Put DPAMT page pair if and only if mapping size is 4KB Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 30/45] x86/virt/tdx: Add API to demote a 2MB mapping to 512 4KB mappings Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 31/45] KVM: x86/mmu: Prevent hugepage promotion for mirror roots in fault path Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 32/45] KVM: x86/mmu: Plumb the old_spte into kvm_x86_ops.set_external_spte() Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 33/45] KVM: TDX: Hoist tdx_sept_remove_private_spte() above set_private_spte() Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 34/45] KVM: TDX: Handle removal of leaf SPTEs in .set_private_spte() Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 35/45] KVM: TDX: Add helper to handle mapping leaf SPTE into S-EPT Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 36/45] KVM: TDX: Move S-EPT page demotion TODO to tdx_sept_set_private_spte() Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 37/45] KVM: x86/tdp_mmu: Alloc external_spt page for mirror page table splitting Sean Christopherson
2026-02-06 10:07   ` Yan Zhao
2026-02-06 16:09     ` Sean Christopherson
2026-02-11  9:49       ` Yan Zhao
2026-01-29  1:15 ` [RFC PATCH v5 38/45] KVM: x86/mmu: Add Dynamic PAMT support in TDP MMU for vCPU-induced page split Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 39/45] KVM: TDX: Add core support for splitting/demoting 2MiB S-EPT to 4KiB Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 40/45] KVM: x86: Introduce hugepage_set_guest_inhibit() Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 41/45] KVM: TDX: Honor the guest's accept level contained in an EPT violation Sean Christopherson
2026-01-29 15:32   ` Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 42/45] KVM: guest_memfd: Add helpers to get start/end gfns give gmem+slot+pgoff Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 43/45] *** DO NOT MERGE *** KVM: guest_memfd: Add pre-zap arch hook for shared<=>private conversion Sean Christopherson
2026-02-13  7:23   ` Huang, Kai
2026-01-29  1:15 ` [RFC PATCH v5 44/45] KVM: x86/mmu: Add support for splitting S-EPT hugepages on conversion Sean Christopherson
2026-01-29 15:39   ` Sean Christopherson
2026-02-11  8:43     ` Yan Zhao
2026-02-13 15:09       ` Sean Christopherson
2026-02-06 10:14   ` Yan Zhao
2026-02-06 14:46     ` Sean Christopherson
2026-01-29  1:15 ` [RFC PATCH v5 45/45] KVM: TDX: Turn on PG_LEVEL_2M Sean Christopherson
2026-01-29 17:13 ` [RFC PATCH v5 00/45] TDX: Dynamic PAMT + S-EPT Hugepage Konrad Rzeszutek Wilk
2026-01-29 17:17   ` Dave Hansen
2026-02-04 14:38   ` Sean Christopherson
2026-02-04 15:09     ` Dave Hansen
2026-02-05 15:53       ` Sean Christopherson
2026-02-05 16:01         ` Dave Hansen

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260129011517.3545883-16-seanjc@google.com \
    --to=seanjc@google.com \
    --cc=ackerleytng@google.com \
    --cc=binbin.wu@linux.intel.com \
    --cc=bp@alien8.de \
    --cc=dave.hansen@linux.intel.com \
    --cc=isaku.yamahata@intel.com \
    --cc=kai.huang@intel.com \
    --cc=kas@kernel.org \
    --cc=kvm@vger.kernel.org \
    --cc=linux-coco@lists.linux.dev \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mingo@redhat.com \
    --cc=pbonzini@redhat.com \
    --cc=rick.p.edgecombe@intel.com \
    --cc=sagis@google.com \
    --cc=tglx@kernel.org \
    --cc=vannapurve@google.com \
    --cc=x86@kernel.org \
    --cc=xiaoyao.li@intel.com \
    --cc=yan.y.zhao@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox