Re: [PATCH 10/11] scs: generic scs code updated to leverage hw assisted shadow stack

[Deepak Gupta <debug@rivosinc.com>]

On Fri, Jul 25, 2025 at 05:06:17PM +0000, Edgecombe, Rick P wrote:
>On Thu, 2025-07-24 at 16:37 -0700, Deepak Gupta wrote:
>> If shadow stack have memory protections from underlying cpu, use those
>> protections. arches can define PAGE_KERNEL_SHADOWSTACK to vmalloc such shadow
>> stack pages. Hw assisted shadow stack pages grow downwards like regular
>> stack. Clang based software shadow call stack grows low to high address.
>> Thus this patch addresses some of those needs due to opposite direction
>> of shadow stack. Furthermore, hw shadow stack can't be memset because memset
>> uses normal stores. Lastly to store magic word at base of shadow stack, arch
>> specific shadow stack store has to be performed.
>>
>> Signed-off-by: Deepak Gupta <debug@rivosinc.com>
>> ---
>>  include/linux/scs.h | 26 +++++++++++++++++++++++++-
>>  kernel/scs.c        | 38 +++++++++++++++++++++++++++++++++++---
>>  2 files changed, 60 insertions(+), 4 deletions(-)
>>
>> diff --git a/include/linux/scs.h b/include/linux/scs.h
>> index 4ab5bdc898cf..6ceee07c2d1a 100644
>> --- a/include/linux/scs.h
>> +++ b/include/linux/scs.h
>> @@ -12,6 +12,7 @@
>>  #include <linux/poison.h>
>>  #include <linux/sched.h>
>>  #include <linux/sizes.h>
>> +#include <asm/scs.h>
>>
>>  #ifdef CONFIG_SHADOW_CALL_STACK
>>
>> @@ -37,22 +38,45 @@ static inline void scs_task_reset(struct task_struct *tsk)
>>  	 * Reset the shadow stack to the base address in case the task
>>  	 * is reused.
>>  	 */
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	task_scs_sp(tsk) = task_scs(tsk) + SCS_SIZE;
>> +#else
>>  	task_scs_sp(tsk) = task_scs(tsk);
>> +#endif
>>  }
>>
>>  static inline unsigned long *__scs_magic(void *s)
>>  {
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	return (unsigned long *)(s);
>> +#else
>>  	return (unsigned long *)(s + SCS_SIZE) - 1;
>> +#endif
>>  }
>>
>>  static inline bool task_scs_end_corrupted(struct task_struct *tsk)
>>  {
>>  	unsigned long *magic = __scs_magic(task_scs(tsk));
>> -	unsigned long sz = task_scs_sp(tsk) - task_scs(tsk);
>> +	unsigned long sz;
>> +
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	sz = (task_scs(tsk) + SCS_SIZE) - task_scs_sp(tsk);
>> +#else
>> +	sz = task_scs_sp(tsk) - task_scs(tsk);
>> +#endif
>>
>>  	return sz >= SCS_SIZE - 1 || READ_ONCE_NOCHECK(*magic) != SCS_END_MAGIC;
>>  }
>>
>> +static inline void __scs_store_magic(unsigned long *s, unsigned long magic_val)
>> +{
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	arch_scs_store(s, magic_val);
>> +#else
>> +	*__scs_magic(s) = magic_val;
>> +#endif
>> +}
>> +
>>  DECLARE_STATIC_KEY_FALSE(dynamic_scs_enabled);
>>
>>  static inline bool scs_is_dynamic(void)
>> diff --git a/kernel/scs.c b/kernel/scs.c
>> index d7809affe740..5910c0a8eabd 100644
>> --- a/kernel/scs.c
>> +++ b/kernel/scs.c
>> @@ -11,6 +11,7 @@
>>  #include <linux/scs.h>
>>  #include <linux/vmalloc.h>
>>  #include <linux/vmstat.h>
>> +#include <asm-generic/set_memory.h>
>>
>>  #ifdef CONFIG_DYNAMIC_SCS
>>  DEFINE_STATIC_KEY_FALSE(dynamic_scs_enabled);
>> @@ -32,19 +33,31 @@ static void *__scs_alloc(int node)
>>  {
>>  	int i;
>>  	void *s;
>> +	pgprot_t prot = PAGE_KERNEL;
>> +
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	prot = PAGE_KERNEL_SHADOWSTACK;
>> +#endif
>>
>>  	for (i = 0; i < NR_CACHED_SCS; i++) {
>>  		s = this_cpu_xchg(scs_cache[i], NULL);
>>  		if (s) {
>>  			s = kasan_unpoison_vmalloc(s, SCS_SIZE,
>>  						   KASAN_VMALLOC_PROT_NORMAL);
>> +/*
>> + * If software shadow stack, its safe to memset. Else memset is not
>> + * possible on hw protected shadow stack. memset constitutes stores and
>> + * stores to shadow stack memory are disallowed and will fault.
>> + */
>> +#ifndef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>>  			memset(s, 0, SCS_SIZE);
>> +#endif
>>  			goto out;
>>  		}
>>  	}
>>
>>  	s = __vmalloc_node_range(SCS_SIZE, 1, VMALLOC_START, VMALLOC_END,
>> -				    GFP_SCS, PAGE_KERNEL, 0, node,
>> +				    GFP_SCS, prot, 0, node,
>>  				    __builtin_return_address(0));
>
>This doesn't update the direct map alias I think. Do you want to protect it?

Yes any alternate address mapping which is writeable is a problem and dilutes
the mechanism. How do I go about updating direct map ? (I pretty new to linux
kernel and have limited understanding on which kernel api's to use here to unmap
direct map)

>
>>
>>  out:
>> @@ -59,7 +72,7 @@ void *scs_alloc(int node)
>>  	if (!s)
>>  		return NULL;
>>
>> -	*__scs_magic(s) = SCS_END_MAGIC;
>> +	__scs_store_magic(__scs_magic(s), SCS_END_MAGIC);
>>
>>  	/*
>>  	 * Poison the allocation to catch unintentional accesses to
>> @@ -87,6 +100,16 @@ void scs_free(void *s)
>>  			return;
>>
>>  	kasan_unpoison_vmalloc(s, SCS_SIZE, KASAN_VMALLOC_PROT_NORMAL);
>> +	/*
>> +	 * Hardware protected shadow stack is not writeable by regular stores
>> +	 * Thus adding this back to free list will raise faults by vmalloc
>> +	 * It needs to be writeable again. It's good sanity as well because
>> +	 * then it can't be inadvertently accesses and if done, it will fault.
>> +	 */
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	set_memory_rw((unsigned long)s, (SCS_SIZE/PAGE_SIZE));
>
>Above you don't update the direct map permissions. So I don't think you need
>this. vmalloc should flush the permissioned mapping before re-using it with the
>lazy cleanup scheme.

If I didn't do this, I was getting a page fault on this vmalloc address. It directly
uses first 8 bytes to add it into some list and that was the location of fault.

>
>> +#endif
>> +
>
>I was thinking someday when we get to this for CET we would protect the direct
>map, and so would need some pool of shadow stacks because flushing the TLB for
>every thread alloc/free would likely be too impactful.

Yes pool would be useful per-cpu.

>
>
>>  	vfree_atomic(s);
>>  }
>>
>> @@ -96,6 +119,9 @@ static int scs_cleanup(unsigned int cpu)
>>  	void **cache = per_cpu_ptr(scs_cache, cpu);
>>
>>  	for (i = 0; i < NR_CACHED_SCS; i++) {
>
>Oh! There is a cache, but the size is only 2.

Yes.
In next iteration, I would likely increase the size of the cache if
CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK=y.

>
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +		set_memory_rw((unsigned long)cache[i], (SCS_SIZE/PAGE_SIZE));
>> +#endif
>>  		vfree(cache[i]);
>>  		cache[i] = NULL;
>>  	}
>> @@ -122,7 +148,13 @@ int scs_prepare(struct task_struct *tsk, int node)
>>  	if (!s)
>>  		return -ENOMEM;
>>
>> -	task_scs(tsk) = task_scs_sp(tsk) = s;
>> +	task_scs(tsk) = s;
>> +#ifdef CONFIG_ARCH_HAS_KERNEL_SHADOW_STACK
>> +	task_scs_sp(tsk) = s + SCS_SIZE;
>> +#else
>> +	task_scs_sp(tsk) = s;
>> +#endif
>> +
>>  	return 0;
>>  }
>>
>>
>