Previously different architectures were using random sources of
differing strength and cost to decide the random kstack offset. A number
of architectures (loongarch, powerpc, s390, x86) were using their
timestamp counter, at whatever the frequency happened to be. Other
arches (arm64, riscv) were using entropy from the crng via
get_random_u16().

There have been concerns that in some cases the timestamp counters may
be too weak, because they can be easily guessed or influenced by user
space. And get_random_u16() has been shown to be too costly for the
level of protection kstack offset randomization provides.

So let's use a common, architecture-agnostic source of entropy; a
per-cpu prng, seeded at boot-time from the crng. This has a few
benefits:

  - We can remove choose_random_kstack_offset(); That was only there to
    try to make the timestamp counter value a bit harder to influence
    from user space.

  - The architecture code is simplified. All it has to do now is call
    add_random_kstack_offset() in the syscall path.

  - The strength of the randomness can be reasoned about independently
    of the architecture.

  - Arches previously using get_random_u16() now have much faster
    syscall paths, see below results.

There have been some claims that a prng may be less strong than the
timestamp counter if not regularly reseeded. But the prng has a period
of about 2^113. So as long as the prng state remains secret, it should
not be possible to guess. If the prng state can be accessed, we have
bigger problems.

Additionally, we are only consuming 6 bits to randomize the stack, so
there are only 64 possible random offsets. I assert that it would be
trivial for an attacker to brute force by repeating their attack and
waiting for the random stack offset to be the desired one. The prng
approach seems entirely proportional to this level of protection.

Performance data are provided below. The baseline is v6.18 with rndstack
on for each respective arch. (I)/(R) indicate statistically significant
improvement/regression. arm64 platform is AWS Graviton3 (m7g.metal).
x86_64 platform is AWS Sapphire Rapids (m7i.24xlarge):

+-----------------+--------------+---------------+---------------+
| Benchmark       | Result Class | per-task-prng | per-task-prng |
|                 |              | arm64 (metal) |   x86_64 (VM) |
+=================+==============+===============+===============+
| syscall/getpid  | mean (ns)    |    (I) -9.50% |   (I) -17.65% |
|                 | p99 (ns)     |   (I) -59.24% |   (I) -24.41% |
|                 | p99.9 (ns)   |   (I) -59.52% |   (I) -28.52% |
+-----------------+--------------+---------------+---------------+
| syscall/getppid | mean (ns)    |    (I) -9.52% |   (I) -19.24% |
|                 | p99 (ns)     |   (I) -59.25% |   (I) -25.03% |
|                 | p99.9 (ns)   |   (I) -59.50% |   (I) -28.17% |
+-----------------+--------------+---------------+---------------+
| syscall/invalid | mean (ns)    |   (I) -10.31% |   (I) -18.56% |
|                 | p99 (ns)     |   (I) -60.79% |   (I) -20.06% |
|                 | p99.9 (ns)   |   (I) -61.04% |   (I) -25.04% |
+-----------------+--------------+---------------+---------------+

I tested an earlier version of this change on x86 bare metal and it
showed a smaller but still significant improvement. The bare metal
system wasn't available this time around so testing was done in a VM
instance. I'm guessing the cost of rdtsc is higher for VMs.

Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
---
 arch/Kconfig                         |  5 ++-
 arch/arm64/kernel/syscall.c          | 11 ------
 arch/loongarch/kernel/syscall.c      | 11 ------
 arch/powerpc/kernel/syscall.c        | 12 -------
 arch/riscv/kernel/traps.c            | 12 -------
 arch/s390/include/asm/entry-common.h |  8 -----
 arch/x86/include/asm/entry-common.h  | 12 -------
 include/linux/randomize_kstack.h     | 52 +++++++++-------------------
 include/linux/sched.h                |  4 ---
 init/main.c                          |  8 +++++
 kernel/fork.c                        |  1 -
 11 files changed, 27 insertions(+), 109 deletions(-)

diff --git a/arch/Kconfig b/arch/Kconfig
index 31220f512b16..8591fe7b4ac1 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -1516,9 +1516,8 @@ config HAVE_ARCH_RANDOMIZE_KSTACK_OFFSET
 	def_bool n
 	help
 	  An arch should select this symbol if it can support kernel stack
-	  offset randomization with calls to add_random_kstack_offset()
-	  during syscall entry and choose_random_kstack_offset() during
-	  syscall exit. Careful removal of -fstack-protector-strong and
+	  offset randomization with a call to add_random_kstack_offset()
+	  during syscall entry. Careful removal of -fstack-protector-strong and
 	  -fstack-protector should also be applied to the entry code and
 	  closely examined, as the artificial stack bump looks like an array
 	  to the compiler, so it will attempt to add canary checks regardless
diff --git a/arch/arm64/kernel/syscall.c b/arch/arm64/kernel/syscall.c
index c062badd1a56..358ddfbf1401 100644
--- a/arch/arm64/kernel/syscall.c
+++ b/arch/arm64/kernel/syscall.c
@@ -52,17 +52,6 @@ static void invoke_syscall(struct pt_regs *regs, unsigned int scno,
 	}
 
 	syscall_set_return_value(current, regs, 0, ret);
-
-	/*
-	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
-	 * bits. The actual entropy will be further reduced by the compiler
-	 * when applying stack alignment constraints: the AAPCS mandates a
-	 * 16-byte aligned SP at function boundaries, which will remove the
-	 * 4 low bits from any entropy chosen here.
-	 *
-	 * The resulting 6 bits of entropy is seen in SP[9:4].
-	 */
-	choose_random_kstack_offset(get_random_u16());
 }
 
 static inline bool has_syscall_work(unsigned long flags)
diff --git a/arch/loongarch/kernel/syscall.c b/arch/loongarch/kernel/syscall.c
index 1249d82c1cd0..85da7e050d97 100644
--- a/arch/loongarch/kernel/syscall.c
+++ b/arch/loongarch/kernel/syscall.c
@@ -79,16 +79,5 @@ void noinstr __no_stack_protector do_syscall(struct pt_regs *regs)
 					   regs->regs[7], regs->regs[8], regs->regs[9]);
 	}
 
-	/*
-	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
-	 * bits. The actual entropy will be further reduced by the compiler
-	 * when applying stack alignment constraints: 16-bytes (i.e. 4-bits)
-	 * aligned, which will remove the 4 low bits from any entropy chosen
-	 * here.
-	 *
-	 * The resulting 6 bits of entropy is seen in SP[9:4].
-	 */
-	choose_random_kstack_offset(get_cycles());
-
 	syscall_exit_to_user_mode(regs);
 }
diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
index be159ad4b77b..b3d8b0f9823b 100644
--- a/arch/powerpc/kernel/syscall.c
+++ b/arch/powerpc/kernel/syscall.c
@@ -173,17 +173,5 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
 	}
 #endif
 
-	/*
-	 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
-	 * so the maximum stack offset is 1k bytes (10 bits).
-	 *
-	 * The actual entropy will be further reduced by the compiler when
-	 * applying stack alignment constraints: the powerpc architecture
-	 * may have two kinds of stack alignment (16-bytes and 8-bytes).
-	 *
-	 * So the resulting 6 or 7 bits of entropy is seen in SP[9:4] or SP[9:3].
-	 */
-	choose_random_kstack_offset(mftb());
-
 	return ret;
 }
diff --git a/arch/riscv/kernel/traps.c b/arch/riscv/kernel/traps.c
index 80230de167de..79b285bdfd1a 100644
--- a/arch/riscv/kernel/traps.c
+++ b/arch/riscv/kernel/traps.c
@@ -342,18 +342,6 @@ void do_trap_ecall_u(struct pt_regs *regs)
 		if (syscall >= 0 && syscall < NR_syscalls)
 			syscall_handler(regs, syscall);
 
-		/*
-		 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
-		 * so the maximum stack offset is 1k bytes (10 bits).
-		 *
-		 * The actual entropy will be further reduced by the compiler when
-		 * applying stack alignment constraints: 16-byte (i.e. 4-bit) aligned
-		 * for RV32I or RV64I.
-		 *
-		 * The resulting 6 bits of entropy is seen in SP[9:4].
-		 */
-		choose_random_kstack_offset(get_random_u16());
-
 		syscall_exit_to_user_mode(regs);
 	} else {
 		irqentry_state_t state = irqentry_nmi_enter(regs);
diff --git a/arch/s390/include/asm/entry-common.h b/arch/s390/include/asm/entry-common.h
index 979af986a8fe..35450a485323 100644
--- a/arch/s390/include/asm/entry-common.h
+++ b/arch/s390/include/asm/entry-common.h
@@ -51,14 +51,6 @@ static __always_inline void arch_exit_to_user_mode(void)
 
 #define arch_exit_to_user_mode arch_exit_to_user_mode
 
-static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
-						  unsigned long ti_work)
-{
-	choose_random_kstack_offset(get_tod_clock_fast());
-}
-
-#define arch_exit_to_user_mode_prepare arch_exit_to_user_mode_prepare
-
 static __always_inline bool arch_in_rcu_eqs(void)
 {
 	if (IS_ENABLED(CONFIG_KVM))
diff --git a/arch/x86/include/asm/entry-common.h b/arch/x86/include/asm/entry-common.h
index ce3eb6d5fdf9..7535131c711b 100644
--- a/arch/x86/include/asm/entry-common.h
+++ b/arch/x86/include/asm/entry-common.h
@@ -82,18 +82,6 @@ static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
 	current_thread_info()->status &= ~(TS_COMPAT | TS_I386_REGS_POKED);
 #endif
 
-	/*
-	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
-	 * bits. The actual entropy will be further reduced by the compiler
-	 * when applying stack alignment constraints (see cc_stack_align4/8 in
-	 * arch/x86/Makefile), which will remove the 3 (x86_64) or 2 (ia32)
-	 * low bits from any entropy chosen here.
-	 *
-	 * Therefore, final stack offset entropy will be 7 (x86_64) or
-	 * 8 (ia32) bits.
-	 */
-	choose_random_kstack_offset(rdtsc());
-
 	/* Avoid unnecessary reads of 'x86_ibpb_exit_to_user' */
 	if (cpu_feature_enabled(X86_FEATURE_IBPB_EXIT_TO_USER) &&
 	    this_cpu_read(x86_ibpb_exit_to_user)) {
diff --git a/include/linux/randomize_kstack.h b/include/linux/randomize_kstack.h
index 5d3916ca747c..024fc20e7762 100644
--- a/include/linux/randomize_kstack.h
+++ b/include/linux/randomize_kstack.h
@@ -6,6 +6,7 @@
 #include <linux/kernel.h>
 #include <linux/jump_label.h>
 #include <linux/percpu-defs.h>
+#include <linux/prandom.h>
 
 DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
 			 randomize_kstack_offset);
@@ -45,9 +46,22 @@ DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
 #define KSTACK_OFFSET_MAX(x)	((x) & 0b1111111100)
 #endif
 
+DECLARE_PER_CPU(struct rnd_state, kstack_rnd_state);
+
+static __always_inline u32 get_kstack_offset(void)
+{
+	struct rnd_state *state;
+	u32 rnd;
+
+	state = &get_cpu_var(kstack_rnd_state);
+	rnd = prandom_u32_state(state);
+	put_cpu_var(kstack_rnd_state);
+
+	return rnd;
+}
+
 /**
- * add_random_kstack_offset - Increase stack utilization by previously
- *			      chosen random offset
+ * add_random_kstack_offset - Increase stack utilization by a random offset.
  *
  * This should be used in the syscall entry path after user registers have been
  * stored to the stack. Preemption may be enabled. For testing the resulting
@@ -56,47 +70,15 @@ DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
 #define add_random_kstack_offset() do {					\
 	if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,	\
 				&randomize_kstack_offset)) {		\
-		u32 offset = current->kstack_offset;			\
+		u32 offset = get_kstack_offset();			\
 		u8 *ptr = __kstack_alloca(KSTACK_OFFSET_MAX(offset));	\
 		/* Keep allocation even after "ptr" loses scope. */	\
 		asm volatile("" :: "r"(ptr) : "memory");		\
 	}								\
 } while (0)
 
-/**
- * choose_random_kstack_offset - Choose the random offset for the next
- *				 add_random_kstack_offset()
- *
- * This should only be used during syscall exit. Preemption may be enabled. This
- * position in the syscall flow is done to frustrate attacks from userspace
- * attempting to learn the next offset:
- * - Maximize the timing uncertainty visible from userspace: if the
- *   offset is chosen at syscall entry, userspace has much more control
- *   over the timing between choosing offsets. "How long will we be in
- *   kernel mode?" tends to be more difficult to predict than "how long
- *   will we be in user mode?"
- * - Reduce the lifetime of the new offset sitting in memory during
- *   kernel mode execution. Exposure of "thread-local" memory content
- *   (e.g. current, percpu, etc) tends to be easier than arbitrary
- *   location memory exposure.
- */
-#define choose_random_kstack_offset(rand) do {				\
-	if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,	\
-				&randomize_kstack_offset)) {		\
-		u32 offset = current->kstack_offset;			\
-		offset = ror32(offset, 5) ^ (rand);			\
-		current->kstack_offset = offset;			\
-	}								\
-} while (0)
-
-static inline void random_kstack_task_init(struct task_struct *tsk)
-{
-	tsk->kstack_offset = 0;
-}
 #else /* CONFIG_RANDOMIZE_KSTACK_OFFSET */
 #define add_random_kstack_offset()		do { } while (0)
-#define choose_random_kstack_offset(rand)	do { } while (0)
-#define random_kstack_task_init(tsk)		do { } while (0)
 #endif /* CONFIG_RANDOMIZE_KSTACK_OFFSET */
 
 #endif
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 9e0080ed1484..d395f2810fac 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1591,10 +1591,6 @@ struct task_struct {
 	unsigned long			prev_lowest_stack;
 #endif
 
-#ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
-	u32				kstack_offset;
-#endif
-
 #ifdef CONFIG_X86_MCE
 	void __user			*mce_vaddr;
 	__u64				mce_kflags;
diff --git a/init/main.c b/init/main.c
index 27fcbbde933e..8626e048095a 100644
--- a/init/main.c
+++ b/init/main.c
@@ -830,6 +830,14 @@ static inline void initcall_debug_enable(void)
 #ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
 DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
 			   randomize_kstack_offset);
+DEFINE_PER_CPU(struct rnd_state, kstack_rnd_state);
+
+static int __init random_kstack_init(void)
+{
+	prandom_seed_full_state(&kstack_rnd_state);
+	return 0;
+}
+late_initcall(random_kstack_init);
 
 static int __init early_randomize_kstack_offset(char *buf)
 {
diff --git a/kernel/fork.c b/kernel/fork.c
index b061e1edbc43..68d9766288fd 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -2232,7 +2232,6 @@ __latent_entropy struct task_struct *copy_process(
 	if (retval)
 		goto bad_fork_cleanup_io;
 
-	random_kstack_task_init(p);
 	stackleak_task_init(p);
 
 	if (pid != &init_struct_pid) {
-- 
2.43.0

Hi Ryan,

I have a couple of comments below, but those are largely for posterity.

On Fri, Jan 02, 2026 at 01:11:54PM +0000, Ryan Roberts wrote:
> Previously different architectures were using random sources of
> differing strength and cost to decide the random kstack offset. A number
> of architectures (loongarch, powerpc, s390, x86) were using their
> timestamp counter, at whatever the frequency happened to be. Other
> arches (arm64, riscv) were using entropy from the crng via
> get_random_u16().
> 
> There have been concerns that in some cases the timestamp counters may
> be too weak, because they can be easily guessed or influenced by user
> space. And get_random_u16() has been shown to be too costly for the
> level of protection kstack offset randomization provides.
> 
> So let's use a common, architecture-agnostic source of entropy; a
> per-cpu prng, seeded at boot-time from the crng. This has a few
> benefits:
> 
>   - We can remove choose_random_kstack_offset(); That was only there to
>     try to make the timestamp counter value a bit harder to influence
>     from user space.

It *might* be worth mentioning that this gets rid of some redundant work
on s390 and x86. Before this patch, those architectures called
choose_random_kstack_offset() under arch_exit_to_user_mode_prepare(),
which also called for exception returns to userspace which were *not*
syscalls (e.g. regular interrupts). Getting rid of
choose_random_kstack_offset() avoids a small amount of redundant work
for the non-syscall cases.

>   - The architecture code is simplified. All it has to do now is call
>     add_random_kstack_offset() in the syscall path.
> 
>   - The strength of the randomness can be reasoned about independently
>     of the architecture.
> 
>   - Arches previously using get_random_u16() now have much faster
>     syscall paths, see below results.
> 
> There have been some claims that a prng may be less strong than the
> timestamp counter if not regularly reseeded. But the prng has a period
> of about 2^113. So as long as the prng state remains secret, it should
> not be possible to guess. If the prng state can be accessed, we have
> bigger problems.
> 
> Additionally, we are only consuming 6 bits to randomize the stack, so
> there are only 64 possible random offsets. I assert that it would be
> trivial for an attacker to brute force by repeating their attack and
> waiting for the random stack offset to be the desired one. The prng
> approach seems entirely proportional to this level of protection.

FWIW, I agree with all of the above rationale.

> Performance data are provided below. The baseline is v6.18 with rndstack
> on for each respective arch. (I)/(R) indicate statistically significant
> improvement/regression. arm64 platform is AWS Graviton3 (m7g.metal).
> x86_64 platform is AWS Sapphire Rapids (m7i.24xlarge):
> 
> +-----------------+--------------+---------------+---------------+
> | Benchmark       | Result Class | per-task-prng | per-task-prng |
> |                 |              | arm64 (metal) |   x86_64 (VM) |
> +=================+==============+===============+===============+
> | syscall/getpid  | mean (ns)    |    (I) -9.50% |   (I) -17.65% |
> |                 | p99 (ns)     |   (I) -59.24% |   (I) -24.41% |
> |                 | p99.9 (ns)   |   (I) -59.52% |   (I) -28.52% |
> +-----------------+--------------+---------------+---------------+
> | syscall/getppid | mean (ns)    |    (I) -9.52% |   (I) -19.24% |
> |                 | p99 (ns)     |   (I) -59.25% |   (I) -25.03% |
> |                 | p99.9 (ns)   |   (I) -59.50% |   (I) -28.17% |
> +-----------------+--------------+---------------+---------------+
> | syscall/invalid | mean (ns)    |   (I) -10.31% |   (I) -18.56% |
> |                 | p99 (ns)     |   (I) -60.79% |   (I) -20.06% |
> |                 | p99.9 (ns)   |   (I) -61.04% |   (I) -25.04% |
> +-----------------+--------------+---------------+---------------+
> 
> I tested an earlier version of this change on x86 bare metal and it
> showed a smaller but still significant improvement. The bare metal
> system wasn't available this time around so testing was done in a VM
> instance. I'm guessing the cost of rdtsc is higher for VMs.
> 
> Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>

Acked-by: Mark Rutland <mark.rutland@arm.com>

Mark.

> ---
>  arch/Kconfig                         |  5 ++-
>  arch/arm64/kernel/syscall.c          | 11 ------
>  arch/loongarch/kernel/syscall.c      | 11 ------
>  arch/powerpc/kernel/syscall.c        | 12 -------
>  arch/riscv/kernel/traps.c            | 12 -------
>  arch/s390/include/asm/entry-common.h |  8 -----
>  arch/x86/include/asm/entry-common.h  | 12 -------
>  include/linux/randomize_kstack.h     | 52 +++++++++-------------------
>  include/linux/sched.h                |  4 ---
>  init/main.c                          |  8 +++++
>  kernel/fork.c                        |  1 -
>  11 files changed, 27 insertions(+), 109 deletions(-)
> 
> diff --git a/arch/Kconfig b/arch/Kconfig
> index 31220f512b16..8591fe7b4ac1 100644
> --- a/arch/Kconfig
> +++ b/arch/Kconfig
> @@ -1516,9 +1516,8 @@ config HAVE_ARCH_RANDOMIZE_KSTACK_OFFSET
>  	def_bool n
>  	help
>  	  An arch should select this symbol if it can support kernel stack
> -	  offset randomization with calls to add_random_kstack_offset()
> -	  during syscall entry and choose_random_kstack_offset() during
> -	  syscall exit. Careful removal of -fstack-protector-strong and
> +	  offset randomization with a call to add_random_kstack_offset()
> +	  during syscall entry. Careful removal of -fstack-protector-strong and
>  	  -fstack-protector should also be applied to the entry code and
>  	  closely examined, as the artificial stack bump looks like an array
>  	  to the compiler, so it will attempt to add canary checks regardless
> diff --git a/arch/arm64/kernel/syscall.c b/arch/arm64/kernel/syscall.c
> index c062badd1a56..358ddfbf1401 100644
> --- a/arch/arm64/kernel/syscall.c
> +++ b/arch/arm64/kernel/syscall.c
> @@ -52,17 +52,6 @@ static void invoke_syscall(struct pt_regs *regs, unsigned int scno,
>  	}
>  
>  	syscall_set_return_value(current, regs, 0, ret);
> -
> -	/*
> -	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
> -	 * bits. The actual entropy will be further reduced by the compiler
> -	 * when applying stack alignment constraints: the AAPCS mandates a
> -	 * 16-byte aligned SP at function boundaries, which will remove the
> -	 * 4 low bits from any entropy chosen here.
> -	 *
> -	 * The resulting 6 bits of entropy is seen in SP[9:4].
> -	 */
> -	choose_random_kstack_offset(get_random_u16());
>  }
>  
>  static inline bool has_syscall_work(unsigned long flags)
> diff --git a/arch/loongarch/kernel/syscall.c b/arch/loongarch/kernel/syscall.c
> index 1249d82c1cd0..85da7e050d97 100644
> --- a/arch/loongarch/kernel/syscall.c
> +++ b/arch/loongarch/kernel/syscall.c
> @@ -79,16 +79,5 @@ void noinstr __no_stack_protector do_syscall(struct pt_regs *regs)
>  					   regs->regs[7], regs->regs[8], regs->regs[9]);
>  	}
>  
> -	/*
> -	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
> -	 * bits. The actual entropy will be further reduced by the compiler
> -	 * when applying stack alignment constraints: 16-bytes (i.e. 4-bits)
> -	 * aligned, which will remove the 4 low bits from any entropy chosen
> -	 * here.
> -	 *
> -	 * The resulting 6 bits of entropy is seen in SP[9:4].
> -	 */
> -	choose_random_kstack_offset(get_cycles());
> -
>  	syscall_exit_to_user_mode(regs);
>  }
> diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
> index be159ad4b77b..b3d8b0f9823b 100644
> --- a/arch/powerpc/kernel/syscall.c
> +++ b/arch/powerpc/kernel/syscall.c
> @@ -173,17 +173,5 @@ notrace long system_call_exception(struct pt_regs *regs, unsigned long r0)
>  	}
>  #endif
>  
> -	/*
> -	 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
> -	 * so the maximum stack offset is 1k bytes (10 bits).
> -	 *
> -	 * The actual entropy will be further reduced by the compiler when
> -	 * applying stack alignment constraints: the powerpc architecture
> -	 * may have two kinds of stack alignment (16-bytes and 8-bytes).
> -	 *
> -	 * So the resulting 6 or 7 bits of entropy is seen in SP[9:4] or SP[9:3].
> -	 */
> -	choose_random_kstack_offset(mftb());
> -
>  	return ret;
>  }
> diff --git a/arch/riscv/kernel/traps.c b/arch/riscv/kernel/traps.c
> index 80230de167de..79b285bdfd1a 100644
> --- a/arch/riscv/kernel/traps.c
> +++ b/arch/riscv/kernel/traps.c
> @@ -342,18 +342,6 @@ void do_trap_ecall_u(struct pt_regs *regs)
>  		if (syscall >= 0 && syscall < NR_syscalls)
>  			syscall_handler(regs, syscall);
>  
> -		/*
> -		 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
> -		 * so the maximum stack offset is 1k bytes (10 bits).
> -		 *
> -		 * The actual entropy will be further reduced by the compiler when
> -		 * applying stack alignment constraints: 16-byte (i.e. 4-bit) aligned
> -		 * for RV32I or RV64I.
> -		 *
> -		 * The resulting 6 bits of entropy is seen in SP[9:4].
> -		 */
> -		choose_random_kstack_offset(get_random_u16());
> -
>  		syscall_exit_to_user_mode(regs);
>  	} else {
>  		irqentry_state_t state = irqentry_nmi_enter(regs);
> diff --git a/arch/s390/include/asm/entry-common.h b/arch/s390/include/asm/entry-common.h
> index 979af986a8fe..35450a485323 100644
> --- a/arch/s390/include/asm/entry-common.h
> +++ b/arch/s390/include/asm/entry-common.h
> @@ -51,14 +51,6 @@ static __always_inline void arch_exit_to_user_mode(void)
>  
>  #define arch_exit_to_user_mode arch_exit_to_user_mode
>  
> -static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
> -						  unsigned long ti_work)
> -{
> -	choose_random_kstack_offset(get_tod_clock_fast());
> -}
> -
> -#define arch_exit_to_user_mode_prepare arch_exit_to_user_mode_prepare
> -
>  static __always_inline bool arch_in_rcu_eqs(void)
>  {
>  	if (IS_ENABLED(CONFIG_KVM))
> diff --git a/arch/x86/include/asm/entry-common.h b/arch/x86/include/asm/entry-common.h
> index ce3eb6d5fdf9..7535131c711b 100644
> --- a/arch/x86/include/asm/entry-common.h
> +++ b/arch/x86/include/asm/entry-common.h
> @@ -82,18 +82,6 @@ static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
>  	current_thread_info()->status &= ~(TS_COMPAT | TS_I386_REGS_POKED);
>  #endif
>  
> -	/*
> -	 * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
> -	 * bits. The actual entropy will be further reduced by the compiler
> -	 * when applying stack alignment constraints (see cc_stack_align4/8 in
> -	 * arch/x86/Makefile), which will remove the 3 (x86_64) or 2 (ia32)
> -	 * low bits from any entropy chosen here.
> -	 *
> -	 * Therefore, final stack offset entropy will be 7 (x86_64) or
> -	 * 8 (ia32) bits.
> -	 */
> -	choose_random_kstack_offset(rdtsc());
> -
>  	/* Avoid unnecessary reads of 'x86_ibpb_exit_to_user' */
>  	if (cpu_feature_enabled(X86_FEATURE_IBPB_EXIT_TO_USER) &&
>  	    this_cpu_read(x86_ibpb_exit_to_user)) {
> diff --git a/include/linux/randomize_kstack.h b/include/linux/randomize_kstack.h
> index 5d3916ca747c..024fc20e7762 100644
> --- a/include/linux/randomize_kstack.h
> +++ b/include/linux/randomize_kstack.h
> @@ -6,6 +6,7 @@
>  #include <linux/kernel.h>
>  #include <linux/jump_label.h>
>  #include <linux/percpu-defs.h>
> +#include <linux/prandom.h>
>  
>  DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
>  			 randomize_kstack_offset);
> @@ -45,9 +46,22 @@ DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
>  #define KSTACK_OFFSET_MAX(x)	((x) & 0b1111111100)
>  #endif
>  
> +DECLARE_PER_CPU(struct rnd_state, kstack_rnd_state);
> +
> +static __always_inline u32 get_kstack_offset(void)
> +{
> +	struct rnd_state *state;
> +	u32 rnd;
> +
> +	state = &get_cpu_var(kstack_rnd_state);
> +	rnd = prandom_u32_state(state);
> +	put_cpu_var(kstack_rnd_state);
> +
> +	return rnd;
> +}
> +
>  /**
> - * add_random_kstack_offset - Increase stack utilization by previously
> - *			      chosen random offset
> + * add_random_kstack_offset - Increase stack utilization by a random offset.
>   *
>   * This should be used in the syscall entry path after user registers have been
>   * stored to the stack. Preemption may be enabled. For testing the resulting
> @@ -56,47 +70,15 @@ DECLARE_STATIC_KEY_MAYBE(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
>  #define add_random_kstack_offset() do {					\
>  	if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,	\
>  				&randomize_kstack_offset)) {		\
> -		u32 offset = current->kstack_offset;			\
> +		u32 offset = get_kstack_offset();			\
>  		u8 *ptr = __kstack_alloca(KSTACK_OFFSET_MAX(offset));	\
>  		/* Keep allocation even after "ptr" loses scope. */	\
>  		asm volatile("" :: "r"(ptr) : "memory");		\
>  	}								\
>  } while (0)
>  
> -/**
> - * choose_random_kstack_offset - Choose the random offset for the next
> - *				 add_random_kstack_offset()
> - *
> - * This should only be used during syscall exit. Preemption may be enabled. This
> - * position in the syscall flow is done to frustrate attacks from userspace
> - * attempting to learn the next offset:
> - * - Maximize the timing uncertainty visible from userspace: if the
> - *   offset is chosen at syscall entry, userspace has much more control
> - *   over the timing between choosing offsets. "How long will we be in
> - *   kernel mode?" tends to be more difficult to predict than "how long
> - *   will we be in user mode?"
> - * - Reduce the lifetime of the new offset sitting in memory during
> - *   kernel mode execution. Exposure of "thread-local" memory content
> - *   (e.g. current, percpu, etc) tends to be easier than arbitrary
> - *   location memory exposure.
> - */
> -#define choose_random_kstack_offset(rand) do {				\
> -	if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,	\
> -				&randomize_kstack_offset)) {		\
> -		u32 offset = current->kstack_offset;			\
> -		offset = ror32(offset, 5) ^ (rand);			\
> -		current->kstack_offset = offset;			\
> -	}								\
> -} while (0)
> -
> -static inline void random_kstack_task_init(struct task_struct *tsk)
> -{
> -	tsk->kstack_offset = 0;
> -}
>  #else /* CONFIG_RANDOMIZE_KSTACK_OFFSET */
>  #define add_random_kstack_offset()		do { } while (0)
> -#define choose_random_kstack_offset(rand)	do { } while (0)
> -#define random_kstack_task_init(tsk)		do { } while (0)
>  #endif /* CONFIG_RANDOMIZE_KSTACK_OFFSET */
>  
>  #endif
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 9e0080ed1484..d395f2810fac 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -1591,10 +1591,6 @@ struct task_struct {
>  	unsigned long			prev_lowest_stack;
>  #endif
>  
> -#ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
> -	u32				kstack_offset;
> -#endif
> -
>  #ifdef CONFIG_X86_MCE
>  	void __user			*mce_vaddr;
>  	__u64				mce_kflags;
> diff --git a/init/main.c b/init/main.c
> index 27fcbbde933e..8626e048095a 100644
> --- a/init/main.c
> +++ b/init/main.c
> @@ -830,6 +830,14 @@ static inline void initcall_debug_enable(void)
>  #ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
>  DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
>  			   randomize_kstack_offset);
> +DEFINE_PER_CPU(struct rnd_state, kstack_rnd_state);
> +
> +static int __init random_kstack_init(void)
> +{
> +	prandom_seed_full_state(&kstack_rnd_state);
> +	return 0;
> +}
> +late_initcall(random_kstack_init);
>  
>  static int __init early_randomize_kstack_offset(char *buf)
>  {
> diff --git a/kernel/fork.c b/kernel/fork.c
> index b061e1edbc43..68d9766288fd 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -2232,7 +2232,6 @@ __latent_entropy struct task_struct *copy_process(
>  	if (retval)
>  		goto bad_fork_cleanup_io;
>  
> -	random_kstack_task_init(p);
>  	stackleak_task_init(p);
>  
>  	if (pid != &init_struct_pid) {
> -- 
> 2.43.0
>

On Fri,  2 Jan 2026 13:11:54 +0000
Ryan Roberts <ryan.roberts@arm.com> wrote:

> Previously different architectures were using random sources of
> differing strength and cost to decide the random kstack offset. A number
> of architectures (loongarch, powerpc, s390, x86) were using their
> timestamp counter, at whatever the frequency happened to be. Other
> arches (arm64, riscv) were using entropy from the crng via
> get_random_u16().
> 
> There have been concerns that in some cases the timestamp counters may
> be too weak, because they can be easily guessed or influenced by user
> space. And get_random_u16() has been shown to be too costly for the
> level of protection kstack offset randomization provides.
> 
> So let's use a common, architecture-agnostic source of entropy; a
> per-cpu prng, seeded at boot-time from the crng. This has a few
> benefits:
> 
>   - We can remove choose_random_kstack_offset(); That was only there to
>     try to make the timestamp counter value a bit harder to influence
>     from user space.
> 
>   - The architecture code is simplified. All it has to do now is call
>     add_random_kstack_offset() in the syscall path.
> 
>   - The strength of the randomness can be reasoned about independently
>     of the architecture.
> 
>   - Arches previously using get_random_u16() now have much faster
>     syscall paths, see below results.
> 
> There have been some claims that a prng may be less strong than the
> timestamp counter if not regularly reseeded. But the prng has a period
> of about 2^113. So as long as the prng state remains secret, it should
> not be possible to guess. If the prng state can be accessed, we have
> bigger problems.

If you have 128 bits of output from consecutive outputs I think you
can trivially determine the full state using (almost) 'school boy' maths
that could be done on pencil and paper.
(Most of the work only has to be done once.)

The underlying problem is that the TAUSWORTHE() transformation is 'linear'
So that TAUSWORTHE(x ^ y) == TAUSWORTHE(x) ^ TAUSWORTHE(y).
(This is true of a LFSR/CRC and TOUSWORTH() is doing some subset of CRCs.)
This means that each output bit is the 'xor' of some of the input bits.
The four new 'state' values are just xor of the the bits of the old ones.
The final xor of the four states gives a 32bit value with each bit just
an xor of some of the 128 state bits.
Get four consecutive 32 bit values and you can solve the 128 simultaneous
equations (by trivial substitution) and get the initial state.
The solution gives you the 128 128bit constants for:
	u128 state = 0;
	u128 val = 'value returned from 4 calls';
	for (int i = 0; i < 128; i++)
		state |= parity(const128[i] ^ val) << i;
You done need all 32bits, just accumulate 128 bits.  
So if you can get the 5bit stack offset from 26 system calls you know the
value that will be used for all the subsequent calls.

Simply changing the final line to use + not ^ makes the output non-linear
and solving the equations a lot harder.

I might sit down tomorrow and see if I can actually code it...

	David

On Sun, 4 Jan 2026 23:01:36 +0000
David Laight <david.laight.linux@gmail.com> wrote:

> On Fri,  2 Jan 2026 13:11:54 +0000
> Ryan Roberts <ryan.roberts@arm.com> wrote:
> 
> > Previously different architectures were using random sources of
> > differing strength and cost to decide the random kstack offset. A number
> > of architectures (loongarch, powerpc, s390, x86) were using their
> > timestamp counter, at whatever the frequency happened to be. Other
> > arches (arm64, riscv) were using entropy from the crng via
> > get_random_u16().
> > 
> > There have been concerns that in some cases the timestamp counters may
> > be too weak, because they can be easily guessed or influenced by user
> > space. And get_random_u16() has been shown to be too costly for the
> > level of protection kstack offset randomization provides.
> > 
> > So let's use a common, architecture-agnostic source of entropy; a
> > per-cpu prng, seeded at boot-time from the crng. This has a few
> > benefits:
> > 
> >   - We can remove choose_random_kstack_offset(); That was only there to
> >     try to make the timestamp counter value a bit harder to influence
> >     from user space.
> > 
> >   - The architecture code is simplified. All it has to do now is call
> >     add_random_kstack_offset() in the syscall path.
> > 
> >   - The strength of the randomness can be reasoned about independently
> >     of the architecture.
> > 
> >   - Arches previously using get_random_u16() now have much faster
> >     syscall paths, see below results.
> > 
> > There have been some claims that a prng may be less strong than the
> > timestamp counter if not regularly reseeded. But the prng has a period
> > of about 2^113. So as long as the prng state remains secret, it should
> > not be possible to guess. If the prng state can be accessed, we have
> > bigger problems.  
> 
> If you have 128 bits of output from consecutive outputs I think you
> can trivially determine the full state using (almost) 'school boy' maths
> that could be done on pencil and paper.
> (Most of the work only has to be done once.)
> 
> The underlying problem is that the TAUSWORTHE() transformation is 'linear'
> So that TAUSWORTHE(x ^ y) == TAUSWORTHE(x) ^ TAUSWORTHE(y).
> (This is true of a LFSR/CRC and TOUSWORTH() is doing some subset of CRCs.)
> This means that each output bit is the 'xor' of some of the input bits.
> The four new 'state' values are just xor of the the bits of the old ones.
> The final xor of the four states gives a 32bit value with each bit just
> an xor of some of the 128 state bits.
> Get four consecutive 32 bit values and you can solve the 128 simultaneous
> equations (by trivial substitution) and get the initial state.
> The solution gives you the 128 128bit constants for:
> 	u128 state = 0;
> 	u128 val = 'value returned from 4 calls';
> 	for (int i = 0; i < 128; i++)
> 		state |= parity(const128[i] ^ val) << i;
> You don't need all 32bits, just accumulate 128 bits.  
> So if you can get the 5bit stack offset from 26 system calls you know the
> value that will be used for all the subsequent calls.

Some of the state bits don't get used, so you only need 123 bits.
The stack offset is 6 bits - so you need the values from 19 calls.

> Simply changing the final line to use + not ^ makes the output non-linear
> and solving the equations a lot harder.
> 
> I might sit down tomorrow and see if I can actually code it...

Finally done:

#include <stdio.h>
#include <unistd.h>
#include <fcntl.h>

typedef unsigned int u32;
typedef unsigned long long u64;
typedef unsigned __int128 u128;

struct rnd_state { u32 s1; u32 s2; u32 s3; u32 s4; };
u32 prandom_u32_state(struct rnd_state *state)
{
#define TAUSWORTHE(s, a, b, c, d) ((s & c) << d) ^ (((s << a) ^ s) >> b)
        state->s1 = TAUSWORTHE(state->s1,  6U, 13U, 4294967294U, 18U);
        state->s2 = TAUSWORTHE(state->s2,  2U, 27U, 4294967288U,  2U);
        state->s3 = TAUSWORTHE(state->s3, 13U, 21U, 4294967280U,  7U);
        state->s4 = TAUSWORTHE(state->s4,  3U, 12U, 4294967168U, 13U);

        return (state->s1 ^ state->s2 ^ state->s3 ^ state->s4);
}

#define X(n, hi, lo) [n] = (u128)0x##hi << 64 | 0x##lo
u128 map[128] = {
        X(  1, 23acb122e4a76, e206c3f6fe435cb6),
	...
        X(127, 00d3276d8a76a, e560d1975675be24) };

u128 parity_128(u128 v)                 
{                               
        return __builtin_parityll(v) ^ __builtin_parityll(v >> 64);
}

int main(int argc, char **argv)
{
        struct rnd_state s = {};
        u128 s0, v, r = 0;

        read(open("/dev/urandom", O_RDONLY), &s, sizeof s);
        // Remove low bits that get masked by the (s & c) term.
        s.s1 &= ~1; s.s2 &= ~7; s.s3 &= ~15; s.s4 &= ~127;
        s0 = (((u128)s.s4 << 32 | s.s3) << 32 | s.s2) << 32 | s.s1;
        v = prandom_u32_state(&s);
        v |= (u128)prandom_u32_state(&s) << 32;
        v |= (u128)prandom_u32_state(&s) << 64;
        v |= (u128)prandom_u32_state(&s) << 96;

        for (int n = 0; n < 128; n++)
                r |= parity_128(v & map[n]) << n;

        printf("%016llx%016llx\n", (u64)(s0 >> 64), (u64)s0);
        printf("values%s match\n", r == s0 ? "" : " do not");

        return r != s0;
}

I've trimmed the initialiser - it is very boring.
The code to create the initialiser is actually slightly smaller than it is.
Doable by hand provided you can do 128bit shift and xor without making
any mistakes.

I've just done a quick search through the kernel sources and haven't found
many uses of prandom_u32_state() outside of test code.
There is sched_rng() which uses a per-cpu rng to throw a 1024 sized die.
bpf also has a per-cpu one for 'unprivileged user space'.
net/sched/sch_netem.c seems to use one - mostly for packet loss generation.

Since the randomize_kstack code is now using a per-task rng (initialised
by clone?) that could be used instead of all the others provided they
are run when 'current' is valid.

But the existing prandom_u32_state() needs a big health warning that
four outputs leak the entire state.
That is fixable by changing the last line to:
        return state->s1 + state->s2 + state->s3 + state->s4;
That only affects the output value, the period is unchanged.

	David

On 07/01/2026 14:05, David Laight wrote:
> On Sun, 4 Jan 2026 23:01:36 +0000
> David Laight <david.laight.linux@gmail.com> wrote:
> 
>> On Fri,  2 Jan 2026 13:11:54 +0000
>> Ryan Roberts <ryan.roberts@arm.com> wrote:
>>
>>> Previously different architectures were using random sources of
>>> differing strength and cost to decide the random kstack offset. A number
>>> of architectures (loongarch, powerpc, s390, x86) were using their
>>> timestamp counter, at whatever the frequency happened to be. Other
>>> arches (arm64, riscv) were using entropy from the crng via
>>> get_random_u16().
>>>
>>> There have been concerns that in some cases the timestamp counters may
>>> be too weak, because they can be easily guessed or influenced by user
>>> space. And get_random_u16() has been shown to be too costly for the
>>> level of protection kstack offset randomization provides.
>>>
>>> So let's use a common, architecture-agnostic source of entropy; a
>>> per-cpu prng, seeded at boot-time from the crng. This has a few
>>> benefits:
>>>
>>>   - We can remove choose_random_kstack_offset(); That was only there to
>>>     try to make the timestamp counter value a bit harder to influence
>>>     from user space.
>>>
>>>   - The architecture code is simplified. All it has to do now is call
>>>     add_random_kstack_offset() in the syscall path.
>>>
>>>   - The strength of the randomness can be reasoned about independently
>>>     of the architecture.
>>>
>>>   - Arches previously using get_random_u16() now have much faster
>>>     syscall paths, see below results.
>>>
>>> There have been some claims that a prng may be less strong than the
>>> timestamp counter if not regularly reseeded. But the prng has a period
>>> of about 2^113. So as long as the prng state remains secret, it should
>>> not be possible to guess. If the prng state can be accessed, we have
>>> bigger problems.  
>>
>> If you have 128 bits of output from consecutive outputs I think you
>> can trivially determine the full state using (almost) 'school boy' maths
>> that could be done on pencil and paper.
>> (Most of the work only has to be done once.)
>>
>> The underlying problem is that the TAUSWORTHE() transformation is 'linear'
>> So that TAUSWORTHE(x ^ y) == TAUSWORTHE(x) ^ TAUSWORTHE(y).
>> (This is true of a LFSR/CRC and TOUSWORTH() is doing some subset of CRCs.)
>> This means that each output bit is the 'xor' of some of the input bits.
>> The four new 'state' values are just xor of the the bits of the old ones.
>> The final xor of the four states gives a 32bit value with each bit just
>> an xor of some of the 128 state bits.
>> Get four consecutive 32 bit values and you can solve the 128 simultaneous
>> equations (by trivial substitution) and get the initial state.
>> The solution gives you the 128 128bit constants for:
>> 	u128 state = 0;
>> 	u128 val = 'value returned from 4 calls';
>> 	for (int i = 0; i < 128; i++)
>> 		state |= parity(const128[i] ^ val) << i;
>> You don't need all 32bits, just accumulate 128 bits.  
>> So if you can get the 5bit stack offset from 26 system calls you know the
>> value that will be used for all the subsequent calls.
> 
> Some of the state bits don't get used, so you only need 123 bits.
> The stack offset is 6 bits - so you need the values from 19 calls.
> 
>> Simply changing the final line to use + not ^ makes the output non-linear
>> and solving the equations a lot harder.
>>
>> I might sit down tomorrow and see if I can actually code it...
> 
> Finally done:
> 
> #include <stdio.h>
> #include <unistd.h>
> #include <fcntl.h>
> 
> typedef unsigned int u32;
> typedef unsigned long long u64;
> typedef unsigned __int128 u128;
> 
> struct rnd_state { u32 s1; u32 s2; u32 s3; u32 s4; };
> u32 prandom_u32_state(struct rnd_state *state)
> {
> #define TAUSWORTHE(s, a, b, c, d) ((s & c) << d) ^ (((s << a) ^ s) >> b)
>         state->s1 = TAUSWORTHE(state->s1,  6U, 13U, 4294967294U, 18U);
>         state->s2 = TAUSWORTHE(state->s2,  2U, 27U, 4294967288U,  2U);
>         state->s3 = TAUSWORTHE(state->s3, 13U, 21U, 4294967280U,  7U);
>         state->s4 = TAUSWORTHE(state->s4,  3U, 12U, 4294967168U, 13U);
> 
>         return (state->s1 ^ state->s2 ^ state->s3 ^ state->s4);
> }
> 
> #define X(n, hi, lo) [n] = (u128)0x##hi << 64 | 0x##lo
> u128 map[128] = {
>         X(  1, 23acb122e4a76, e206c3f6fe435cb6),
> 	...
>         X(127, 00d3276d8a76a, e560d1975675be24) };
> 
> u128 parity_128(u128 v)                 
> {                               
>         return __builtin_parityll(v) ^ __builtin_parityll(v >> 64);
> }
> 
> int main(int argc, char **argv)
> {
>         struct rnd_state s = {};
>         u128 s0, v, r = 0;
> 
>         read(open("/dev/urandom", O_RDONLY), &s, sizeof s);
>         // Remove low bits that get masked by the (s & c) term.
>         s.s1 &= ~1; s.s2 &= ~7; s.s3 &= ~15; s.s4 &= ~127;
>         s0 = (((u128)s.s4 << 32 | s.s3) << 32 | s.s2) << 32 | s.s1;
>         v = prandom_u32_state(&s);
>         v |= (u128)prandom_u32_state(&s) << 32;
>         v |= (u128)prandom_u32_state(&s) << 64;
>         v |= (u128)prandom_u32_state(&s) << 96;
> 
>         for (int n = 0; n < 128; n++)
>                 r |= parity_128(v & map[n]) << n;
> 
>         printf("%016llx%016llx\n", (u64)(s0 >> 64), (u64)s0);
>         printf("values%s match\n", r == s0 ? "" : " do not");
> 
>         return r != s0;
> }
> 
> I've trimmed the initialiser - it is very boring.
> The code to create the initialiser is actually slightly smaller than it is.
> Doable by hand provided you can do 128bit shift and xor without making
> any mistakes.
> 
> I've just done a quick search through the kernel sources and haven't found
> many uses of prandom_u32_state() outside of test code.
> There is sched_rng() which uses a per-cpu rng to throw a 1024 sized die.
> bpf also has a per-cpu one for 'unprivileged user space'.
> net/sched/sch_netem.c seems to use one - mostly for packet loss generation.
> 
> Since the randomize_kstack code is now using a per-task rng (initialised
> by clone?) that could be used instead of all the others provided they
> are run when 'current' is valid.
> 
> But the existing prandom_u32_state() needs a big health warning that
> four outputs leak the entire state.
> That is fixable by changing the last line to:
>         return state->s1 + state->s2 + state->s3 + state->s4;
> That only affects the output value, the period is unchanged.

Hi David,

This all seems interesting, but I'm not clear that it is a blocker for this
series. As I keep saying, we only use 6 bits for offset randmization so it is
trival to brute force, regardless of how easy it is to recover the prng state.

Perhaps we can decouple these 2 things and make them independent:

 - this series, which is motivated by speeding up syscalls on arm64; given 6
   bits is not hard to brute force, spending a lot of cycles calculating those
   bits is unjustified.

 - Your observation that that the current prng could be improved to make
   recoving it's state harder.

What do you think?

Thanks,
Ryan


> 
> 	David
> 
>

On Mon, 12 Jan 2026 12:26:26 +0000
Ryan Roberts <ryan.roberts@arm.com> wrote:

> On 07/01/2026 14:05, David Laight wrote:
> > On Sun, 4 Jan 2026 23:01:36 +0000
> > David Laight <david.laight.linux@gmail.com> wrote:
...
> > I've trimmed the initialiser - it is very boring.
> > The code to create the initialiser is actually slightly smaller than it is.
> > Doable by hand provided you can do 128bit shift and xor without making
> > any mistakes.
> > 
> > I've just done a quick search through the kernel sources and haven't found
> > many uses of prandom_u32_state() outside of test code.
> > There is sched_rng() which uses a per-cpu rng to throw a 1024 sized die.
> > bpf also has a per-cpu one for 'unprivileged user space'.
> > net/sched/sch_netem.c seems to use one - mostly for packet loss generation.
> > 
> > Since the randomize_kstack code is now using a per-task rng (initialised
> > by clone?) that could be used instead of all the others provided they
> > are run when 'current' is valid.
> > 
> > But the existing prandom_u32_state() needs a big health warning that
> > four outputs leak the entire state.
> > That is fixable by changing the last line to:
> >         return state->s1 + state->s2 + state->s3 + state->s4;
> > That only affects the output value, the period is unchanged.  
> 
> Hi David,
> 
> This all seems interesting, but I'm not clear that it is a blocker for this
> series. As I keep saying, we only use 6 bits for offset randmization so it is
> trival to brute force, regardless of how easy it is to recover the prng state.
> 
> Perhaps we can decouple these 2 things and make them independent:
> 
>  - this series, which is motivated by speeding up syscalls on arm64; given 6
>    bits is not hard to brute force, spending a lot of cycles calculating those
>    bits is unjustified.
> 
>  - Your observation that that the current prng could be improved to make
>    recoving it's state harder.
> 
> What do you think?

They are separate.
I should have a 'mostly written' patch series for prandom_u32_state().

If you unconditionally add a per-task prng there are a few places that could
use it instead of a per-cpu one.
It could be 'perturbed' during task switch - eg by:
	s->s1 = (s->s1 ^ something) | 2;
(The 2 stops the new value being 0 or 1, losing 1 bit wont be significant.)

This one is much nearer 'ready' and has an obvious impact.

	David

> 
> Thanks,
> Ryan
> 
> 
> > 
> > 	David
> > 
> >   
>

On 04/01/2026 23:01, David Laight wrote:
> On Fri,  2 Jan 2026 13:11:54 +0000
> Ryan Roberts <ryan.roberts@arm.com> wrote:
> 
>> Previously different architectures were using random sources of
>> differing strength and cost to decide the random kstack offset. A number
>> of architectures (loongarch, powerpc, s390, x86) were using their
>> timestamp counter, at whatever the frequency happened to be. Other
>> arches (arm64, riscv) were using entropy from the crng via
>> get_random_u16().
>>
>> There have been concerns that in some cases the timestamp counters may
>> be too weak, because they can be easily guessed or influenced by user
>> space. And get_random_u16() has been shown to be too costly for the
>> level of protection kstack offset randomization provides.
>>
>> So let's use a common, architecture-agnostic source of entropy; a
>> per-cpu prng, seeded at boot-time from the crng. This has a few
>> benefits:
>>
>>   - We can remove choose_random_kstack_offset(); That was only there to
>>     try to make the timestamp counter value a bit harder to influence
>>     from user space.
>>
>>   - The architecture code is simplified. All it has to do now is call
>>     add_random_kstack_offset() in the syscall path.
>>
>>   - The strength of the randomness can be reasoned about independently
>>     of the architecture.
>>
>>   - Arches previously using get_random_u16() now have much faster
>>     syscall paths, see below results.
>>
>> There have been some claims that a prng may be less strong than the
>> timestamp counter if not regularly reseeded. But the prng has a period
>> of about 2^113. So as long as the prng state remains secret, it should
>> not be possible to guess. If the prng state can be accessed, we have
>> bigger problems.
> 
> If you have 128 bits of output from consecutive outputs I think you
> can trivially determine the full state using (almost) 'school boy' maths
> that could be done on pencil and paper.
> (Most of the work only has to be done once.)
> 
> The underlying problem is that the TAUSWORTHE() transformation is 'linear'
> So that TAUSWORTHE(x ^ y) == TAUSWORTHE(x) ^ TAUSWORTHE(y).
> (This is true of a LFSR/CRC and TOUSWORTH() is doing some subset of CRCs.)
> This means that each output bit is the 'xor' of some of the input bits.
> The four new 'state' values are just xor of the the bits of the old ones.
> The final xor of the four states gives a 32bit value with each bit just
> an xor of some of the 128 state bits.
> Get four consecutive 32 bit values and you can solve the 128 simultaneous
> equations (by trivial substitution) and get the initial state.
> The solution gives you the 128 128bit constants for:
> 	u128 state = 0;
> 	u128 val = 'value returned from 4 calls';
> 	for (int i = 0; i < 128; i++)
> 		state |= parity(const128[i] ^ val) << i;

What is const128[] here?

> You done need all 32bits, just accumulate 128 bits.  
> So if you can get the 5bit stack offset from 26 system calls you know the
> value that will be used for all the subsequent calls.

It's not immediately obvious to me how user space would do this, but I'll take
it on faith that it may be possible.

> 
> Simply changing the final line to use + not ^ makes the output non-linear
> and solving the equations a lot harder.

There has been pushback on introducing new primitives [1] but I don't think
that's a reason not to considder it.

[1] https://lore.kernel.org/all/aRyppb8PCxFKVphr@zx2c4.com/

> 
> I might sit down tomorrow and see if I can actually code it...

Thanks for the analysis! I look forward to seeing your conclusion... although
I'm not sure I'll be qualified to evaluate it mathematically.

FWIW, I previously had a go at various schemes using siphash to calculate some
random bits. I found it to be significantly slower than this prng. I've so far
taken the view that 6 bits of randomness is not much of a defence against brute
force so we really shouldn't be spending too many cycles to generate the bits.
If we can get to approach to work, I think that's best.

Thanks,
Ryan

> 
> 	David
> 
>

On Mon, 5 Jan 2026 11:05:18 +0000
Ryan Roberts <ryan.roberts@arm.com> wrote:

> On 04/01/2026 23:01, David Laight wrote:
> > On Fri,  2 Jan 2026 13:11:54 +0000
> > Ryan Roberts <ryan.roberts@arm.com> wrote:
> >   
> >> Previously different architectures were using random sources of
> >> differing strength and cost to decide the random kstack offset. A number
> >> of architectures (loongarch, powerpc, s390, x86) were using their
> >> timestamp counter, at whatever the frequency happened to be. Other
> >> arches (arm64, riscv) were using entropy from the crng via
> >> get_random_u16().
> >>
> >> There have been concerns that in some cases the timestamp counters may
> >> be too weak, because they can be easily guessed or influenced by user
> >> space. And get_random_u16() has been shown to be too costly for the
> >> level of protection kstack offset randomization provides.
> >>
> >> So let's use a common, architecture-agnostic source of entropy; a
> >> per-cpu prng, seeded at boot-time from the crng. This has a few
> >> benefits:
> >>
> >>   - We can remove choose_random_kstack_offset(); That was only there to
> >>     try to make the timestamp counter value a bit harder to influence
> >>     from user space.
> >>
> >>   - The architecture code is simplified. All it has to do now is call
> >>     add_random_kstack_offset() in the syscall path.
> >>
> >>   - The strength of the randomness can be reasoned about independently
> >>     of the architecture.
> >>
> >>   - Arches previously using get_random_u16() now have much faster
> >>     syscall paths, see below results.
> >>
> >> There have been some claims that a prng may be less strong than the
> >> timestamp counter if not regularly reseeded. But the prng has a period
> >> of about 2^113. So as long as the prng state remains secret, it should
> >> not be possible to guess. If the prng state can be accessed, we have
> >> bigger problems.  
> > 
> > If you have 128 bits of output from consecutive outputs I think you
> > can trivially determine the full state using (almost) 'school boy' maths
> > that could be done on pencil and paper.
> > (Most of the work only has to be done once.)
> > 
> > The underlying problem is that the TAUSWORTHE() transformation is 'linear'
> > So that TAUSWORTHE(x ^ y) == TAUSWORTHE(x) ^ TAUSWORTHE(y).
> > (This is true of a LFSR/CRC and TOUSWORTH() is doing some subset of CRCs.)
> > This means that each output bit is the 'xor' of some of the input bits.
> > The four new 'state' values are just xor of the the bits of the old ones.
> > The final xor of the four states gives a 32bit value with each bit just
> > an xor of some of the 128 state bits.
> > Get four consecutive 32 bit values and you can solve the 128 simultaneous
> > equations (by trivial substitution) and get the initial state.
> > The solution gives you the 128 128bit constants for:
> > 	u128 state = 0;
> > 	u128 val = 'value returned from 4 calls';
> > 	for (int i = 0; i < 128; i++)
> > 		state |= parity(const128[i] ^ val) << i;  
> 
> What is const128[] here?

Some values you prepared earlier :-)

> > You done need all 32bits, just accumulate 128 bits.  
> > So if you can get the 5bit stack offset from 26 system calls you know the
> > value that will be used for all the subsequent calls.  
> 
> It's not immediately obvious to me how user space would do this, but I'll take
> it on faith that it may be possible.

It shouldn't be possible, but anything that leaks a stack address would
give it away.
It is also pretty much why you care about the cycle length of the PRNG.
(If the length is short a rogue application can remember all the values.)

> > 
> > Simply changing the final line to use + not ^ makes the output non-linear
> > and solving the equations a lot harder.  
> 
> There has been pushback on introducing new primitives [1] but I don't think
> that's a reason not to considder it.

That is a more general issue with the PRNG.
ISTR it was true for the previous version that explicitly used four CRC.
Jason should know more about whether the xor are a good idea.

> 
> [1] https://lore.kernel.org/all/aRyppb8PCxFKVphr@zx2c4.com/
> 
> > 
> > I might sit down tomorrow and see if I can actually code it...  
> 
> Thanks for the analysis! I look forward to seeing your conclusion... although
> I'm not sure I'll be qualified to evaluate it mathematically.

I need to drag out the brian cells from when I learnt about CRC (actually
relating to burst error correction) over 40 years ago...
 
> FWIW, I previously had a go at various schemes using siphash to calculate some
> random bits. I found it to be significantly slower than this prng. I've so far
> taken the view that 6 bits of randomness is not much of a defence against brute
> force so we really shouldn't be spending too many cycles to generate the bits.
> If we can get to approach to work, I think that's best.

Indeed.
A single 32bit CRC using (crc + (crc >> 16)) & 0x3f could be 'good enough'.
Especially if the value is 'perturbed' during (say) context switch.
The '16' might need adjusting for the actual CRC, especially if TAUSWORTHE()
is used - you don't want the value to match one of the shifts it uses.

prandom_u32_state() is defined as:
#define TAUSWORTHE(s, a, b, c, d) ((s & c) << d) ^ (((s << a) ^ s) >> b)
	state->s1 = TAUSWORTHE(state->s1,  6U, 13U, 4294967294U, 18U);
	state->s2 = TAUSWORTHE(state->s2,  2U, 27U, 4294967288U,  2U);
	state->s3 = TAUSWORTHE(state->s3, 13U, 21U, 4294967280U,  7U);
	state->s4 = TAUSWORTHE(state->s4,  3U, 12U, 4294967168U, 13U);
	return (state->s1 ^ state->s2 ^ state->s3 ^ state->s4);
This is equivalent to:
#define TAUSWORTHE(s, a, b, c, d) ((s & ~c) << d) ^ (s >> a) ^ (s >> b)
	state->s1 = TAUSWORTHE(state->s1,  7, 13,   1, 18);
	state->s2 = TAUSWORTHE(state->s2, 25, 27,   7,  2);
	state->s3 = TAUSWORTHE(state->s3,  8, 21,  15,  7);
	state->s4 = TAUSWORTHE(state->s4,  9, 12, 127, 13);
which makes it clear that some low bits of each 's' get discarded reducing
the length of each CRC to (I think) 31, 29, 28 and 25.
Since 'b + d' matches the bits discarded by 'c', two of those shifts are
actually just a rotate, so there isn't really much 'bit stirring' going on.

By comparison CRC-16 (for hdlc comms like x25, isdn and ss7) reduces to:
u32 crc_step(u32 crc, u8 byte_val)
{
    u8 t = crc ^ byte_val;
    t = (t ^ t << 4);
    return crc >> 8 ^ t << 8 ^ t << 3 ^ t >> 4;
}
Much more 'stirring'.

	David