[<prev] [next>] [day] [month] [year] [list]
Message-ID: <20241108062556.4138380-1-chenhuacai@loongson.cn>
Date: Fri, 8 Nov 2024 14:25:56 +0800
From: Huacai Chen <chenhuacai@...ngson.cn>
To: Huacai Chen <chenhuacai@...nel.org>
Cc: loongarch@...ts.linux.dev,
Xuefeng Li <lixuefeng@...ngson.cn>,
Guo Ren <guoren@...nel.org>,
Xuerui Wang <kernel@...0n.name>,
Jiaxun Yang <jiaxun.yang@...goat.com>,
linux-kernel@...r.kernel.org,
loongson-kernel@...ts.loongnix.cn,
Huacai Chen <chenhuacai@...ngson.cn>
Subject: [PATCH V2] LoongArch: Disable KASAN if PGDIR_SIZE is too large for cpu_vabits
If PGDIR_SIZE is too large for cpu_vabits, KASAN_SHADOW_END will
overflow UINTPTR_MAX because KASAN_SHADOW_START/KASAN_SHADOW_END are
aligned up by PGDIR_SIZE. And then the overflowed KASAN_SHADOW_END looks
like a user space address.
For example, PGDIR_SIZE of CONFIG_4KB_4LEVEL is 2^39, which is too large
for Loongson-2K series whose cpu_vabits = 39.
Since CONFIG_4KB_4LEVEL is completely legal for CPUs with cpu_vabits <=
39, we just disable KASAN via early return in kasan_init(). Otherwise we
get a boot failure.
Moreover, we change KASAN_SHADOW_END from the first address after KASAN
shadow area to the last address in KASAN shadow area, in order to avoid
the end address exactly overflow to 0 (which is a legal case). We don't
need to worry about alignment because pgd_addr_end() can handle it.
Reviewed-by: Jiaxun Yang <jiaxun.yang@...goat.com>
Signed-off-by: Huacai Chen <chenhuacai@...ngson.cn>
---
V2: Change KASAN_SHADOW_END definition.
arch/loongarch/include/asm/kasan.h | 2 +-
arch/loongarch/mm/kasan_init.c | 15 +++++++++++++--
2 files changed, 14 insertions(+), 3 deletions(-)
diff --git a/arch/loongarch/include/asm/kasan.h b/arch/loongarch/include/asm/kasan.h
index c6bce5fbff57..cb74a47f620e 100644
--- a/arch/loongarch/include/asm/kasan.h
+++ b/arch/loongarch/include/asm/kasan.h
@@ -51,7 +51,7 @@
/* KAsan shadow memory start right after vmalloc. */
#define KASAN_SHADOW_START round_up(KFENCE_AREA_END, PGDIR_SIZE)
#define KASAN_SHADOW_SIZE (XKVRANGE_VC_SHADOW_END - XKPRANGE_CC_KASAN_OFFSET)
-#define KASAN_SHADOW_END round_up(KASAN_SHADOW_START + KASAN_SHADOW_SIZE, PGDIR_SIZE)
+#define KASAN_SHADOW_END (round_up(KASAN_SHADOW_START + KASAN_SHADOW_SIZE, PGDIR_SIZE) - 1)
#define XKPRANGE_CC_SHADOW_OFFSET (KASAN_SHADOW_START + XKPRANGE_CC_KASAN_OFFSET)
#define XKPRANGE_UC_SHADOW_OFFSET (KASAN_SHADOW_START + XKPRANGE_UC_KASAN_OFFSET)
diff --git a/arch/loongarch/mm/kasan_init.c b/arch/loongarch/mm/kasan_init.c
index 4a0d1880dd71..7277b7583e1b 100644
--- a/arch/loongarch/mm/kasan_init.c
+++ b/arch/loongarch/mm/kasan_init.c
@@ -238,7 +238,7 @@ static void __init kasan_map_populate(unsigned long start, unsigned long end,
asmlinkage void __init kasan_early_init(void)
{
BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_START, PGDIR_SIZE));
- BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_END, PGDIR_SIZE));
+ BUILD_BUG_ON(!IS_ALIGNED(KASAN_SHADOW_END + 1, PGDIR_SIZE));
}
static inline void kasan_set_pgd(pgd_t *pgdp, pgd_t pgdval)
@@ -253,7 +253,7 @@ static void __init clear_pgds(unsigned long start, unsigned long end)
* swapper_pg_dir. pgd_clear() can't be used
* here because it's nop on 2,3-level pagetable setups
*/
- for (; start < end; start += PGDIR_SIZE)
+ for (; start < end; start = pgd_addr_end(start, end))
kasan_set_pgd((pgd_t *)pgd_offset_k(start), __pgd(0));
}
@@ -262,6 +262,17 @@ void __init kasan_init(void)
u64 i;
phys_addr_t pa_start, pa_end;
+ /*
+ * If PGDIR_SIZE is too large for cpu_vabits, KASAN_SHADOW_END will
+ * overflow UINTPTR_MAX and then looks like a user space address.
+ * For example, PGDIR_SIZE of CONFIG_4KB_4LEVEL is 2^39, which is too
+ * large for Loongson-2K series whose cpu_vabits = 39.
+ */
+ if (KASAN_SHADOW_END < vm_map_base) {
+ pr_warn("PGDIR_SIZE too large for cpu_vabits, KernelAddressSanitizer disabled.\n");
+ return;
+ }
+
/*
* PGD was populated as invalid_pmd_table or invalid_pud_table
* in pagetable_init() which depends on how many levels of page
--
2.43.5
Powered by blists - more mailing lists