[PATCH v3 3/4] elf: align ET_DYN base for PTE coalescing and PMD mapping

[Usama Arif]

For PIE binaries (ET_DYN), the load address is randomized at PAGE_SIZE
granularity via arch_mmap_rnd(). On arm64 with 64K base pages, this
means the binary is 64K-aligned, but contpte mapping requires 2M
(CONT_PTE_SIZE) alignment.

Without proper virtual address alignment, readahead patches that
allocate large folios with aligned file offsets and physical addresses
cannot benefit from contpte mapping, as the contpte fold check in
contpte_set_ptes() requires the virtual address to be CONT_PTE_SIZE-
aligned.

Fix this by extending maximum_alignment() to consider folio alignment
at two tiers, matching the readahead allocation strategy:

- HPAGE_PMD_SIZE, so large folios can be PMD-mapped on
  architectures where PMD_SIZE is reasonable (e.g. 2M on x86-64
  and arm64 with 4K pages).

- exec_folio_order(), the minimum order for hardware TLB
  coalescing (e.g. arm64 contpte/HPA).

For each PT_LOAD segment, folio_alignment() tries both tiers and
returns the largest power-of-2 alignment that fits within the segment
size, with both p_vaddr and p_offset aligned to that size. This
ensures load_bias is folio-aligned so that file-offset-aligned folios
map to properly aligned virtual addresses, enabling hardware PTE
coalescing and PMD mappings for large folios.

The segment size check in folio_alignment() avoids reducing ASLR
entropy for small binaries that cannot benefit from large folio
alignment.

Signed-off-by: Usama Arif <usama.arif@xxxxxxxxx>
---
 fs/binfmt_elf.c | 50 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 16a56b6b3f6c..f84fae6daf23 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -488,6 +488,54 @@ static int elf_read(struct file *file, void *buf, size_t len, loff_t pos)
 	return 0;
 }
 
+/*
+ * Return the largest folio alignment for a PT_LOAD segment, so the
+ * hardware can coalesce PTEs (e.g. arm64 contpte) or use PMD mappings
+ * for large folios.
+ *
+ * Try PMD alignment so large folios can be PMD-mapped. Then try
+ * exec_folio_order() alignment for hardware TLB coalescing (e.g.
+ * arm64 contpte/HPA).
+ *
+ * Use the largest power-of-2 that fits within the segment size, capped
+ * by the target folio size.
+ * Only align when the segment's virtual address and file offset are
+ * already aligned to that size, as misalignment would prevent coalescing
+ * anyway.
+ *
+ * The segment size check avoids reducing ASLR entropy for small binaries
+ * that cannot benefit.
+ */
+static unsigned long folio_alignment(struct elf_phdr *cmd)
+{
+	unsigned long alignment = 0;
+	unsigned long seg_size;
+
+	if (!cmd->p_filesz)
+		return 0;
+
+	seg_size = rounddown_pow_of_two(cmd->p_filesz);
+
+	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
+		unsigned long size = min(seg_size, HPAGE_PMD_SIZE);
+
+		if (size > PAGE_SIZE &&
+		    IS_ALIGNED(cmd->p_vaddr | cmd->p_offset, size))
+			alignment = size;
+	}
+
+	if (!alignment && exec_folio_order()) {
+		unsigned long size = min(seg_size,
+					PAGE_SIZE << exec_folio_order());
+
+		if (size > PAGE_SIZE &&
+		    IS_ALIGNED(cmd->p_vaddr | cmd->p_offset, size))
+			alignment = size;
+	}
+
+	return alignment;
+}
+
 static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
 {
 	unsigned long alignment = 0;
@@ -501,6 +549,8 @@ static unsigned long maximum_alignment(struct elf_phdr *cmds, int nr)
 			if (!is_power_of_2(p_align))
 				continue;
 			alignment = max(alignment, p_align);
+			alignment = max(alignment,
+					folio_alignment(&cmds[i]));
 		}
 	}
 
-- 
2.52.0