[RFC 1/4] powerpc/64s: Add linear_map_use_base_page command line option
flat view
From: "Ritesh Harjani (IBM)" <ritesh.list@gmail.com>
Date: 2026-10-10 12:22:05
Also in:
lkml
Subsystem:
documentation, linux for powerpc (32-bit and 64-bit), the rest · Maintainers:
Jonathan Corbet, Madhavan Srinivasan, Linus Torvalds
The radix linear map is built with 1G and 2M mappings wherever it can. That is the right default for TLB reach, but it also means a single page cannot be pulled out of the linear map later. There is no support for splitting a live leaf, so memfd_secret() has nothing to work with. debug_pagealloc=on already forces PAGE_SIZE mappings, but that turns on a lot of allocator debugging that nobody wants just for secretmem. Add linear_map_use_base_page so the linear map can be built out of base pages on its own. debug_pagealloc=on still implies it. The extra page tables are roughly 3% of RAM (for 64K pagesize) and it can also cause extra TLB pressure, so leave the option off by default. Only radix honours this for now (later we will implement this for Hash as well). Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com> --- .../admin-guide/kernel-parameters.txt | 10 +++++++++ arch/powerpc/include/asm/book3s/64/mmu.h | 2 ++ arch/powerpc/mm/book3s64/pgtable.c | 22 +++++++++++++++++++ arch/powerpc/mm/book3s64/radix_pgtable.c | 2 +- 4 files changed, 35 insertions(+), 1 deletion(-)
diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 33cd30996e47..382d18c6e8e8 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt@@ -3563,6 +3563,16 @@ Kernel parameters If there are multiple matching configurations changing the same attribute, the last one is used. + linear_map_use_base_page[=<bool>] + [PPC(Radix), EARLY] + Build the kernel linear mapping out of PAGE_SIZE + mappings only, instead of using 2M/1G mappings where + possible. This allows individual pages to be removed + from the linear mapping at runtime, which is required + by memfd_secret(2). It costs roughly 3% extra memory for + page tables and increases kernel TLB pressure, so it is + off by default. debug_pagealloc=on implies this. + liveupdate= [KNL,EARLY] Format: <bool> Enable Live Update Orchestrator (LUO).
diff --git a/arch/powerpc/include/asm/book3s/64/mmu.h b/arch/powerpc/include/asm/book3s/64/mmu.h
index 48631365b48c..3739e5677312 100644
--- a/arch/powerpc/include/asm/book3s/64/mmu.h
+++ b/arch/powerpc/include/asm/book3s/64/mmu.h@@ -73,6 +73,8 @@ extern unsigned int mmu_base_pid; extern unsigned long __ro_after_init memory_block_size; +bool linear_map_uses_base_pagesize(void); + #define PRTB_SIZE_SHIFT (mmu_pid_bits + 4) #define PRTB_ENTRIES (1ul << mmu_pid_bits)
diff --git a/arch/powerpc/mm/book3s64/pgtable.c b/arch/powerpc/mm/book3s64/pgtable.c
index 85ab6723c8f2..1a007902f307 100644
--- a/arch/powerpc/mm/book3s64/pgtable.c
+++ b/arch/powerpc/mm/book3s64/pgtable.c@@ -48,6 +48,28 @@ static int __init parse_kfence_early_init(char *arg) early_param("kfence.sample_interval", parse_kfence_early_init); #endif +static bool __ro_after_init linear_map_use_base_page; + +static int __init parse_linear_map_use_base_page(char *p) +{ + bool val; + + if (!p) + val = true; + else if (kstrtobool(p, &val)) + return -EINVAL; + + linear_map_use_base_page = val; + + return 0; +} +early_param("linear_map_use_base_page", parse_linear_map_use_base_page); + +bool linear_map_uses_base_pagesize(void) +{ + return debug_pagealloc_enabled() || linear_map_use_base_page; +} + #ifdef CONFIG_TRANSPARENT_HUGEPAGE /* * This is called when relaxing access to a hugepage. It's also called in the page
diff --git a/arch/powerpc/mm/book3s64/radix_pgtable.c b/arch/powerpc/mm/book3s64/radix_pgtable.c
index cf692b2b5f7b..2c4cb6d6ab6e 100644
--- a/arch/powerpc/mm/book3s64/radix_pgtable.c
+++ b/arch/powerpc/mm/book3s64/radix_pgtable.c@@ -308,7 +308,7 @@ static int __meminit create_physical_mapping(unsigned long start, if (mapping_sz_limit < max_mapping_size) max_mapping_size = mapping_sz_limit; - if (debug_pagealloc_enabled()) + if (linear_map_uses_base_pagesize()) max_mapping_size = PAGE_SIZE; start = ALIGN(start, PAGE_SIZE);
--
2.39.5