Skip to contentSkip to conversation

LoongArch: Move KASLR to EFI stub to avoid initrd overlap

PATCH Superseded WANG Rui 2 participants 8 messages Apr 27

Subsystems (2)

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

WANG Rui

author
#0 lore raw

cover LoongArch: Move KASLR to EFI stub to avoid initrd overlap

This series addresses a potential overlap issue between the kernel image and the initrd when KASLR is enabled.

In the normal boot flow, the bootloader is responsible for loading both vmlinux and the initrd, and it can guarantee that the two do not overlap in memory. However, this assumption only holds as long as neither image changes its location afterwards.

The in-kernel KASLR implementation breaks that assumption. When the initrd is placed close to the kernel image, randomizing the kernel location at runtime may move it into the initrd region, leading to memory corruption early during boot.

To fix this, this series moves the KASLR logic out of the kernel proper and into the EFI stub. With this change, the final placement of both the kernel image and the initrd is determined by the EFI memory allocator. This ensures that the two allocations are coordinated and cannot overlap.

Functionally, the kernel still supports KASLR as before, but the randomization now happens before the kernel is entered, rather than during early kernel relocation.

WANG Rui (3):
LoongArch: Allow rdtime_h() and rdtime_l() in 64-bit builds efi/loongarch: Randomize kernel preferred address for KASLR
LoongArch: Skip relocation-time KASLR if it has already been applied

 arch/loongarch/Kconfig                   |  2 +-
 arch/loongarch/include/asm/efi.h         |  4 +++-
 arch/loongarch/include/asm/loongarch.h   |  6 ------
 arch/loongarch/kernel/relocate.c         |  4 ++++
 drivers/firmware/efi/libstub/loongarch.c | 16 ++++++++++++++++
 5 files changed, 24 insertions(+), 8 deletions(-)
Signature · 1 line
2.54.0
author
#1 lore raw

PATCH 1/3 LoongArch: Allow rdtime_h() and rdtime_l() in 64-bit builds

Remove the CONFIG_32BIT guard around rdtime_h() and rdtime_l() so they can also be used in 64-bit builds.

---
 arch/loongarch/include/asm/loongarch.h | 6 ------
 1 file changed, 6 deletions(-)
arch/loongarch/include/asm/loongarch.h +0 -6
index 2a6bc99177d8..bb2da9c58411 100644
@@ -1248,8 +1248,6 @@
 
 #ifndef __ASSEMBLER__
 
-#ifdef CONFIG_32BIT
-
 static __always_inline u32 rdtime_h(void)
 {
 	u32 val = 0;
@@ -1274,8 +1272,6 @@ static __always_inline u32 rdtime_l(void)
 	return val;
 }
 
-#else
-
 static __always_inline u64 rdtime_d(void)
 {
 	u64 val = 0;
@@ -1288,8 +1284,6 @@ static __always_inline u64 rdtime_d(void)
 	return val;
 }
 
-#endif
-
 static inline unsigned int get_csr_cpuid(void)
 {
 	return csr_read32(LOONGARCH_CSR_CPUID);
Signature · 1 line
2.54.0
author
#2 lore raw

PATCH 2/3 efi/loongarch: Randomize kernel preferred address for KASLR

In reply to WANG Rui

Introduce efi_get_kimg_kaslr_address() to compute the preferred kernel image address dynamically when CONFIG_RANDOMIZE_BASE is enabled. The function derives a random offset using EFI-provided randomness combined with the timer value, and constrains it within CONFIG_RANDOMIZE_BASE_MAX_OFFSET.

Update EFI_KIMG_PREFERRED_ADDRESS to call this helper so that the EFI stub can select a randomized load address when KASLR is active, while preserving the original base address behavior when KASLR is disabled or nokaslr is specified.

---
 arch/loongarch/Kconfig                   |  2 +-
 arch/loongarch/include/asm/efi.h         |  4 +++-
 drivers/firmware/efi/libstub/loongarch.c | 16 ++++++++++++++++
 3 files changed, 20 insertions(+), 2 deletions(-)
arch/loongarch/Kconfig +1 -1
index 3b042dbb2c41..a5afb70f73d9 100644
@@ -730,7 +730,7 @@ config RANDOMIZE_BASE
 config RANDOMIZE_BASE_MAX_OFFSET
 	hex "Maximum KASLR offset" if EXPERT
 	depends on RANDOMIZE_BASE
-	range 0x0 0x10000000
+	range 0x20000 0x10000000
 	default "0x01000000"
 	help
 	  When KASLR is active, this provides the maximum offset that will
arch/loongarch/include/asm/efi.h +3 -1
index eddc8e79b3fa..f831320efd41 100644
@@ -30,6 +30,8 @@ static inline unsigned long efi_get_kimg_min_align(void)
 	return SZ_2M;
 }
 
-#define EFI_KIMG_PREFERRED_ADDRESS	PHYSADDR(VMLINUX_LOAD_ADDRESS)
+unsigned long efi_get_kimg_kaslr_address(void);
+
+#define EFI_KIMG_PREFERRED_ADDRESS	efi_get_kimg_kaslr_address()
 
 #endif /* _ASM_LOONGARCH_EFI_H */
drivers/firmware/efi/libstub/loongarch.c +16 -0
index 9825f5218137..df67ef8c68ab 100644
@@ -38,6 +38,22 @@ static efi_status_t exit_boot_func(struct efi_boot_memmap *map, void *priv)
 	return EFI_SUCCESS;
 }
 
+unsigned long efi_get_kimg_kaslr_address(void)
+{
+	unsigned int random_offset = 0;
+
+#ifdef CONFIG_RANDOMIZE_BASE
+	if (!efi_nokaslr) {
+		efi_get_random_bytes(sizeof(random_offset), (u8 *)&random_offset);
+		random_offset ^= (rdtime_l() << 16);
+		random_offset &= (CONFIG_RANDOMIZE_BASE_MAX_OFFSET - SZ_64K - 1);
+		random_offset = ALIGN(random_offset, SZ_64K) + SZ_64K;
+	}
+#endif
+
+	return PHYSADDR(VMLINUX_LOAD_ADDRESS) + random_offset;
+}
+
 unsigned long __weak kernel_entry_address(unsigned long kernel_addr,
 		efi_loaded_image_t *image)
 {
Signature · 1 line
2.54.0
maintainer
#3 lore raw

Re: [PATCH 2/3] efi/loongarch: Randomize kernel preferred address for KASLR

Hi, Rui,

Quoted 24 lines · WANG Rui

Introduce efi_get_kimg_kaslr_address() to compute the preferred kernel image address dynamically when CONFIG_RANDOMIZE_BASE is enabled. The function derives a random offset using EFI-provided randomness combined with the timer value, and constrains it within CONFIG_RANDOMIZE_BASE_MAX_OFFSET.

Update EFI_KIMG_PREFERRED_ADDRESS to call this helper so that the EFI stub can select a randomized load address when KASLR is active, while preserving the original base address behavior when KASLR is disabled or nokaslr is specified.

---
 arch/loongarch/Kconfig                   |  2 +-
 arch/loongarch/include/asm/efi.h         |  4 +++-
 drivers/firmware/efi/libstub/loongarch.c | 16 ++++++++++++++++
 3 files changed, 20 insertions(+), 2 deletions(-)
arch/loongarch/Kconfig +0 -0
index 3b042dbb2c41..a5afb70f73d9 100644

arch/loongarch/Kconfig config RANDOMIZE_BASE · line 733

Quoted 5 lines
arch/loongarch/Kconfig +1 -1
@@ -730,7 +730,7 @@ config RANDOMIZE_BASE
 config RANDOMIZE_BASE_MAX_OFFSET
        hex "Maximum KASLR offset" if EXPERT
        depends on RANDOMIZE_BASE
-       range 0x0 0x10000000
+       range 0x20000 0x10000000

Why modify this?

Quoted 8 lines
arch/loongarch/Kconfig +0 -0
        default "0x01000000"
        help
          When KASLR is active, this provides the maximum offset that will
arch/loongarch/include/asm/efi.h +0 -0
index eddc8e79b3fa..f831320efd41 100644

arch/loongarch/include/asm/efi.h static inline unsigned long efi_get_kimg_min_align(void) · line 35

Quoted 7 lines
arch/loongarch/include/asm/efi.h +3 -1
@@ -30,6 +30,8 @@ static inline unsigned long efi_get_kimg_min_align(void)
        return SZ_2M;
 }
 
-#define EFI_KIMG_PREFERRED_ADDRESS     PHYSADDR(VMLINUX_LOAD_ADDRESS)
+unsigned long efi_get_kimg_kaslr_address(void);
+
+#define EFI_KIMG_PREFERRED_ADDRESS     efi_get_kimg_kaslr_address()

Can we reuse something in drivers/firmware/efi/libstub/kaslr.c?

Quoted 8 lines
arch/loongarch/include/asm/efi.h +0 -0
 #endif /* _ASM_LOONGARCH_EFI_H */
drivers/firmware/efi/libstub/loongarch.c +0 -0
index 9825f5218137..df67ef8c68ab 100644
@@ -38,6 +38,22 @@ static efi_status_t exit_boot_func(struct efi_boot_memmap *map, void *priv)
        return EFI_SUCCESS;

drivers/firmware/efi/libstub/loongarch.c static efi_status_t exit_boot_func(struct efi_boot_memmap *map, void *priv) · line 48

Quoted 10 lines
drivers/firmware/efi/libstub/loongarch.c +8 -0
 }
 
+unsigned long efi_get_kimg_kaslr_address(void)
+{
+       unsigned int random_offset = 0;
+
+#ifdef CONFIG_RANDOMIZE_BASE
+       if (!efi_nokaslr) {
+               efi_get_random_bytes(sizeof(random_offset), (u8 *)&random_offset);
+               random_offset ^= (rdtime_l() << 16);

Use get_cycles() then the first patch can be dropped.

Huacai

Quoted 14 lines
drivers/firmware/efi/libstub/loongarch.c +8 -1
+               random_offset &= (CONFIG_RANDOMIZE_BASE_MAX_OFFSET - SZ_64K - 1);
+               random_offset = ALIGN(random_offset, SZ_64K) + SZ_64K;
+       }
+#endif
+
+       return PHYSADDR(VMLINUX_LOAD_ADDRESS) + random_offset;
+}
+
 unsigned long __weak kernel_entry_address(unsigned long kernel_addr,
                efi_loaded_image_t *image)
 {
--

2.54.0

author
#4 lore raw

Re: [PATCH 2/3] efi/loongarch: Randomize kernel preferred address for KASLR

Hi Huacai,

Quoted 23 lines · Huacai Chen

Hi, Rui,

Quoted 19 lines · WANG Rui

Introduce efi_get_kimg_kaslr_address() to compute the preferred kernel image address dynamically when CONFIG_RANDOMIZE_BASE is enabled. The function derives a random offset using EFI-provided randomness combined with the timer value, and constrains it within CONFIG_RANDOMIZE_BASE_MAX_OFFSET.

Update EFI_KIMG_PREFERRED_ADDRESS to call this helper so that the EFI stub can select a randomized load address when KASLR is active, while preserving the original base address behavior when KASLR is disabled or nokaslr is specified.

---
 arch/loongarch/Kconfig                   |  2 +-
 arch/loongarch/include/asm/efi.h         |  4 +++-
 drivers/firmware/efi/libstub/loongarch.c | 16 ++++++++++++++++
 3 files changed, 20 insertions(+), 2 deletions(-)
Quoted 11 lines

arch/loongarch/Kconfig config RANDOMIZE_BASE · line 733

Quoted 10 lines
arch/loongarch/Kconfig +1 -1
index 3b042dbb2c41..a5afb70f73d9 100644
@@ -730,7 +730,7 @@ config RANDOMIZE_BASE
 config RANDOMIZE_BASE_MAX_OFFSET
        hex "Maximum KASLR offset" if EXPERT
        depends on RANDOMIZE_BASE
-       range 0x0 0x10000000
+       range 0x20000 0x10000000

Why modify this?

The reason for changeing the lower bound is to simplify the random_offset calculation, making sure that (CONFIG_RANDOMIZE_BASE_MAX_OFFSET - SZ_64K - 1) is always >= 0.

Quoted 12 lines
Quoted 12 lines
arch/loongarch/Kconfig +0 -0
        default "0x01000000"
        help
          When KASLR is active, this provides the maximum offset that will
arch/loongarch/include/asm/efi.h +0 -0
index eddc8e79b3fa..f831320efd41 100644
@@ -30,6 +30,8 @@ static inline unsigned long efi_get_kimg_min_align(void)
        return SZ_2M;
 }
 
Quoted 5 lines

arch/loongarch/include/asm/efi.h static inline unsigned long efi_get_kimg_min_align(void) · line 35

Quoted 4 lines
arch/loongarch/include/asm/efi.h +3 -1
-#define EFI_KIMG_PREFERRED_ADDRESS     PHYSADDR(VMLINUX_LOAD_ADDRESS)
+unsigned long efi_get_kimg_kaslr_address(void);
+
+#define EFI_KIMG_PREFERRED_ADDRESS     efi_get_kimg_kaslr_address()

Can we reuse something in drivers/firmware/efi/libstub/kaslr.c?

This aligns with the kernel's built-in KASLR, ensuring the random range stays within CONFIG_RANDOMIZE_BASE_MAX_OFFSET. This is also why we didn't reuse kaslr.c.

Quoted 11 lines
Quoted 11 lines
arch/loongarch/include/asm/efi.h +0 -0
 #endif /* _ASM_LOONGARCH_EFI_H */
drivers/firmware/efi/libstub/loongarch.c +0 -0
index 9825f5218137..df67ef8c68ab 100644
@@ -38,6 +38,22 @@ static efi_status_t exit_boot_func(struct efi_boot_memmap *map, void *priv)
        return EFI_SUCCESS;
 }
 
Quoted 9 lines

drivers/firmware/efi/libstub/loongarch.c static efi_status_t exit_boot_func(struct efi_boot_memmap *map, void *priv) · line 48

Quoted 8 lines
drivers/firmware/efi/libstub/loongarch.c +8 -0
+unsigned long efi_get_kimg_kaslr_address(void)
+{
+       unsigned int random_offset = 0;
+
+#ifdef CONFIG_RANDOMIZE_BASE
+       if (!efi_nokaslr) {
+               efi_get_random_bytes(sizeof(random_offset), (u8 *)&random_offset);
+               random_offset ^= (rdtime_l() << 16);

Use get_cycles() then the first patch can be dropped.

I'll use random_get_entropy() here.

Thanks,
Rui

Quoted 17 lines

Huacai

Quoted 14 lines
drivers/firmware/efi/libstub/loongarch.c +8 -1
+               random_offset &= (CONFIG_RANDOMIZE_BASE_MAX_OFFSET - SZ_64K - 1);
+               random_offset = ALIGN(random_offset, SZ_64K) + SZ_64K;
+       }
+#endif
+
+       return PHYSADDR(VMLINUX_LOAD_ADDRESS) + random_offset;
+}
+
 unsigned long __weak kernel_entry_address(unsigned long kernel_addr,
                efi_loaded_image_t *image)
 {
--

2.54.0

author
#5 lore raw

PATCH 3/3 LoongArch: Skip relocation-time KASLR if it has already been applied

In reply to WANG Rui

When the kernel is relocated during early boot, a randomized load address may already have been selected and applied. In this case, performing KASLR again in relocate.c is unnecessary.

---
 arch/loongarch/kernel/relocate.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/loongarch/kernel/relocate.c +4 -0
index 16f6a9b39659..c36604a81d08 100644
@@ -139,6 +139,10 @@ static inline __init bool kaslr_disabled(void)
 	char *str;
 	const char *builtin_cmdline = CONFIG_CMDLINE;
 
+	/* KASLR is performed during early boot. */
+	if (kaslr_offset())
+		return true;
+
 	str = strstr(builtin_cmdline, "nokaslr");
 	if (str == builtin_cmdline || (str > builtin_cmdline && *(str - 1) == ' ')) {
 		pr_info(KASLR_DISABLED_MESSAGE, "\'nokaslr\'", "built-in");
Signature · 1 line
2.54.0
maintainer
#6 lore raw

Re: [PATCH 3/3] LoongArch: Skip relocation-time KASLR if it has already been applied

Hi, Rui,

Quoted 15 lines · WANG Rui

When the kernel is relocated during early boot, a randomized load address may already have been selected and applied. In this case, performing KASLR again in relocate.c is unnecessary.

---
 arch/loongarch/kernel/relocate.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/loongarch/kernel/relocate.c +0 -0
index 16f6a9b39659..c36604a81d08 100644

arch/loongarch/kernel/relocate.c static inline __init bool kaslr_disabled(void) · line 144

Quoted 6 lines
arch/loongarch/kernel/relocate.c +3 -0
@@ -139,6 +139,10 @@ static inline __init bool kaslr_disabled(void)
        char *str;
        const char *builtin_cmdline = CONFIG_CMDLINE;
 
+       /* KASLR is performed during early boot. */
+       if (kaslr_offset())
+               return true;

Maybe the "kexec_file" case is also handled by this?

Huacai

Quoted 7 lines
arch/loongarch/kernel/relocate.c +1 -1
+
        str = strstr(builtin_cmdline, "nokaslr");
        if (str == builtin_cmdline || (str > builtin_cmdline && *(str - 1) == ' ')) {
                pr_info(KASLR_DISABLED_MESSAGE, "\'nokaslr\'", "built-in");
--

2.54.0

author
#7 lore raw

Re: [PATCH 3/3] LoongArch: Skip relocation-time KASLR if it has already been applied

Hi Huacai,

Quoted 22 lines · Huacai Chen

Hi, Rui,

Quoted 18 lines · WANG Rui

When the kernel is relocated during early boot, a randomized load address may already have been selected and applied. In this case, performing KASLR again in relocate.c is unnecessary.

---
 arch/loongarch/kernel/relocate.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/loongarch/kernel/relocate.c +0 -0
index 16f6a9b39659..c36604a81d08 100644
@@ -139,6 +139,10 @@ static inline __init bool kaslr_disabled(void)
        char *str;
        const char *builtin_cmdline = CONFIG_CMDLINE;
 
Quoted 4 lines

arch/loongarch/kernel/relocate.c static inline __init bool kaslr_disabled(void) · line 144

Quoted 3 lines
arch/loongarch/kernel/relocate.c +3 -0
+       /* KASLR is performed during early boot. */
+       if (kaslr_offset())
+               return true;

Maybe the "kexec_file" case is also handled by this?

Yeah, it looks like kexec_file doesn't load the kernel image to VMLINUX_LOAD_ADDRESS, so it will disable KASLR here. This aligns with our expectation of disabling KASLR for this case.

Thanks,
Rui

Quoted 10 lines

Huacai

Quoted 7 lines
arch/loongarch/kernel/relocate.c +1 -1
+
        str = strstr(builtin_cmdline, "nokaslr");
        if (str == builtin_cmdline || (str > builtin_cmdline && *(str - 1) == ' ')) {
                pr_info(KASLR_DISABLED_MESSAGE, "\'nokaslr\'", "built-in");
--

2.54.0