[PATCH v2 0/6] arm64: Enable UPROBES with GCS

[PATCH v2 0/6] arm64: Enable UPROBES with GCS

[Jeremy Linton]

Currently uprobes and the Arm Guarded Control Stack (GCS) feature are
exclusive of each other. This restriction needs to be lifted in order
to utilize GCS for generic Linux distro images where the expectation
is that core debugging features like uprobes work.

This series adds some user accessors to read/push/pop the userspace
shadow stack. It then utilizes those functions in the uprobe paths
as needed to synchronize GCS with the changes in control flow at
probe locations. 

Along the way we fix a bug in the core gcs task handling and export
some uprobe quality of life functionality for use in arch specific code.

The KCONFIG restriction is then dropped.

v1->v2:
	Drop uprobe_warn() patch
	Fix copy_thread_gcs() bug created by fixing task_gcs_el0_enabled()
	Comments, now describe issues with reading userspace GCS pages
	Rebased to 6.15

Jeremy Linton (6):
  arm64/gcs: task_gcs_el0_enable() should use passed task
  arm64: probes: Break ret out from bl/blr
  arm64: uaccess: Add additional userspace GCS accessors
  arm64: probes: Add GCS support to bl/blr/ret
  arm64: uprobes: Add GCS support to uretprobes
  arm64: Kconfig: Remove GCS restrictions on UPROBES

 arch/arm64/Kconfig                       |  1 -
 arch/arm64/include/asm/gcs.h             |  2 +-
 arch/arm64/include/asm/uaccess.h         | 42 ++++++++++++++++++++++++
 arch/arm64/kernel/probes/decode-insn.c   |  7 ++--
 arch/arm64/kernel/probes/simulate-insn.c | 38 ++++++++++++++++++---
 arch/arm64/kernel/probes/simulate-insn.h |  3 +-
 arch/arm64/kernel/probes/uprobes.c       | 30 +++++++++++++++++
 arch/arm64/kernel/process.c              |  6 ++--
 8 files changed, 115 insertions(+), 14 deletions(-)

-- 
2.49.0

[PATCH v2 1/6] arm64/gcs: task_gcs_el0_enable() should use passed task

[Jeremy Linton]

Mark Rutland noticed that the task parameter is ignored and
'current' is being used instead. Since this is usually
what its passed, it hasn't yet been causing problems but likely
will as the code gets more testing.

But, once this is fixed, it creates a new bug in copy_thread_gcs()
since the gcs_el_mode isn't yet set for the task before its being
checked. Move gcs_alloc_thread_stack() after the new task's
gcs_el0_mode initialization to avoid this.

Fixes: fc84bc5378a8 ("arm64/gcs: Context switch GCS state for EL0")
Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/include/asm/gcs.h | 2 +-
 arch/arm64/kernel/process.c  | 6 +++---
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/include/asm/gcs.h b/arch/arm64/include/asm/gcs.h
index f50660603ecf..5bc432234d3a 100644
--- a/arch/arm64/include/asm/gcs.h
+++ b/arch/arm64/include/asm/gcs.h
@@ -58,7 +58,7 @@ static inline u64 gcsss2(void)
 
 static inline bool task_gcs_el0_enabled(struct task_struct *task)
 {
-	return current->thread.gcs_el0_mode & PR_SHADOW_STACK_ENABLE;
+	return task->thread.gcs_el0_mode & PR_SHADOW_STACK_ENABLE;
 }
 
 void gcs_set_el0_mode(struct task_struct *task);
diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
index 42faebb7b712..68b7021cda77 100644
--- a/arch/arm64/kernel/process.c
+++ b/arch/arm64/kernel/process.c
@@ -305,13 +305,13 @@ static int copy_thread_gcs(struct task_struct *p,
 	p->thread.gcs_base = 0;
 	p->thread.gcs_size = 0;
 
+	p->thread.gcs_el0_mode = current->thread.gcs_el0_mode;
+	p->thread.gcs_el0_locked = current->thread.gcs_el0_locked;
+
 	gcs = gcs_alloc_thread_stack(p, args);
 	if (IS_ERR_VALUE(gcs))
 		return PTR_ERR((void *)gcs);
 
-	p->thread.gcs_el0_mode = current->thread.gcs_el0_mode;
-	p->thread.gcs_el0_locked = current->thread.gcs_el0_locked;
-
 	return 0;
 }
 
-- 
2.49.0

Re: [PATCH v2 1/6] arm64/gcs: task_gcs_el0_enable() should use passed task

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 332 bytes --]

On Mon, Apr 07, 2025 at 11:19:46AM -0500, Jeremy Linton wrote:
> Mark Rutland noticed that the task parameter is ignored and
> 'current' is being used instead. Since this is usually
> what its passed, it hasn't yet been causing problems but likely
> will as the code gets more testing.

Reviewed-by: Mark Brown <broonie@kernel.org>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[PATCH v2 2/6] arm64: probes: Break ret out from bl/blr

[Jeremy Linton]

Prepare for GCS by breaking RET out into its own function, where
it makes more sense to encapsulate the new behavior independent
from the branch instructions.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/kernel/probes/decode-insn.c   |  7 ++++---
 arch/arm64/kernel/probes/simulate-insn.c | 10 +++++++++-
 arch/arm64/kernel/probes/simulate-insn.h |  3 ++-
 3 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/arch/arm64/kernel/probes/decode-insn.c b/arch/arm64/kernel/probes/decode-insn.c
index 6438bf62e753..4137cc5ef031 100644
--- a/arch/arm64/kernel/probes/decode-insn.c
+++ b/arch/arm64/kernel/probes/decode-insn.c
@@ -108,9 +108,10 @@ arm_probe_decode_insn(u32 insn, struct arch_probe_insn *api)
 	    aarch64_insn_is_bl(insn)) {
 		api->handler = simulate_b_bl;
 	} else if (aarch64_insn_is_br(insn) ||
-	    aarch64_insn_is_blr(insn) ||
-	    aarch64_insn_is_ret(insn)) {
-		api->handler = simulate_br_blr_ret;
+		aarch64_insn_is_blr(insn)) {
+		api->handler = simulate_br_blr;
+	} else if (aarch64_insn_is_ret(insn)) {
+		api->handler = simulate_ret;
 	} else {
 		/*
 		 * Instruction cannot be stepped out-of-line and we don't
diff --git a/arch/arm64/kernel/probes/simulate-insn.c b/arch/arm64/kernel/probes/simulate-insn.c
index 4c6d2d712fbd..09a0b36122d0 100644
--- a/arch/arm64/kernel/probes/simulate-insn.c
+++ b/arch/arm64/kernel/probes/simulate-insn.c
@@ -126,7 +126,7 @@ simulate_b_cond(u32 opcode, long addr, struct pt_regs *regs)
 }
 
 void __kprobes
-simulate_br_blr_ret(u32 opcode, long addr, struct pt_regs *regs)
+simulate_br_blr(u32 opcode, long addr, struct pt_regs *regs)
 {
 	int xn = (opcode >> 5) & 0x1f;
 
@@ -138,6 +138,14 @@ simulate_br_blr_ret(u32 opcode, long addr, struct pt_regs *regs)
 		set_x_reg(regs, 30, addr + 4);
 }
 
+void __kprobes
+simulate_ret(u32 opcode, long addr, struct pt_regs *regs)
+{
+	int xn = (opcode >> 5) & 0x1f;
+
+	instruction_pointer_set(regs, get_x_reg(regs, xn));
+}
+
 void __kprobes
 simulate_cbz_cbnz(u32 opcode, long addr, struct pt_regs *regs)
 {
diff --git a/arch/arm64/kernel/probes/simulate-insn.h b/arch/arm64/kernel/probes/simulate-insn.h
index efb2803ec943..9e772a292d56 100644
--- a/arch/arm64/kernel/probes/simulate-insn.h
+++ b/arch/arm64/kernel/probes/simulate-insn.h
@@ -11,7 +11,8 @@
 void simulate_adr_adrp(u32 opcode, long addr, struct pt_regs *regs);
 void simulate_b_bl(u32 opcode, long addr, struct pt_regs *regs);
 void simulate_b_cond(u32 opcode, long addr, struct pt_regs *regs);
-void simulate_br_blr_ret(u32 opcode, long addr, struct pt_regs *regs);
+void simulate_br_blr(u32 opcode, long addr, struct pt_regs *regs);
+void simulate_ret(u32 opcode, long addr, struct pt_regs *regs);
 void simulate_cbz_cbnz(u32 opcode, long addr, struct pt_regs *regs);
 void simulate_tbz_tbnz(u32 opcode, long addr, struct pt_regs *regs);
 void simulate_ldr_literal(u32 opcode, long addr, struct pt_regs *regs);
-- 
2.49.0

[PATCH v2 3/6] arm64: uaccess: Add additional userspace GCS accessors

[Jeremy Linton]

Uprobes need more advanced read, push, and pop userspace GCS
functionality. Implement those features using the existing gcsstr()
and copy_from_user().

Its important to note that GCS pages can be read by normal
instructions, but the hardware validates that pages used by GCS
specific operations, have a GCS privilege set. We aren't validating this
in load_user_gcs because it requires stabilizing the VMA over the read
which may fault.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/include/asm/uaccess.h | 42 ++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)

diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h
index 5b91803201ef..34a8b2cc8935 100644
--- a/arch/arm64/include/asm/uaccess.h
+++ b/arch/arm64/include/asm/uaccess.h
@@ -20,6 +20,7 @@
 
 #include <asm/asm-extable.h>
 #include <asm/cpufeature.h>
+#include <asm/gcs.h>
 #include <asm/mmu.h>
 #include <asm/mte.h>
 #include <asm/ptrace.h>
@@ -539,6 +540,47 @@ static inline void put_user_gcs(unsigned long val, unsigned long __user *addr,
 	uaccess_ttbr0_disable();
 }
 
+static __always_inline unsigned long __must_check
+copy_from_user(void *to, const void __user *from, unsigned long n);
+
+/*
+ * Unlike put_user_gcs() above, the use of copy_from_user() may provide
+ * an opening for non GCS pages to be used to source data. Therefore this
+ * should only be used in contexts where that is acceptable.
+ */
+static inline u64 load_user_gcs(unsigned long __user *addr, int *err)
+{
+	unsigned long ret;
+	u64 load = 0;
+
+	gcsb_dsync();
+	ret = copy_from_user(&load, addr, sizeof(load));
+	if (ret != 0)
+		*err = ret;
+	return load;
+}
+
+static inline void push_user_gcs(unsigned long val, int *err)
+{
+	u64 gcspr = read_sysreg_s(SYS_GCSPR_EL0);
+
+	gcspr -= sizeof(u64);
+	put_user_gcs(val, (unsigned long __user *)gcspr, err);
+	if (!*err)
+		write_sysreg_s(gcspr, SYS_GCSPR_EL0);
+}
+
+static inline u64 pop_user_gcs(int *err)
+{
+	u64 gcspr = read_sysreg_s(SYS_GCSPR_EL0);
+	u64 read_val;
+
+	read_val = load_user_gcs((unsigned long __user *)gcspr, err);
+	if (!*err)
+		write_sysreg_s(gcspr + sizeof(u64), SYS_GCSPR_EL0);
+
+	return read_val;
+}
 
 #endif /* CONFIG_ARM64_GCS */
 
-- 
2.49.0

[PATCH v2 4/6] arm64: probes: Add GCS support to bl/blr/ret

[Jeremy Linton]

The arm64 probe simulation doesn't currently have logic in place
to deal with GCS and this results in core dumps if probes are inserted
at control flow locations. Fix-up bl, blr and ret to manipulate the
shadow stack as needed.

While we manipulate and validate the shadow stack correctly, the
hardware provides additional security by only allowing GCS operations
against pages which are marked to support GCS. For writing there is
gcssttr() which enforces this, but there isn't an equivalent for
reading. This means that uprobe users should be aware that probing on
control flow instructions which require reading the shadow stack (ex:
ret) offers lower security guarantees than what is achieved without
the uprobe active.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/kernel/probes/simulate-insn.c | 28 ++++++++++++++++++++----
 1 file changed, 24 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/kernel/probes/simulate-insn.c b/arch/arm64/kernel/probes/simulate-insn.c
index 09a0b36122d0..1fc9bb69b1eb 100644
--- a/arch/arm64/kernel/probes/simulate-insn.c
+++ b/arch/arm64/kernel/probes/simulate-insn.c
@@ -13,6 +13,7 @@
 #include <asm/traps.h>
 
 #include "simulate-insn.h"
+#include "asm/gcs.h"
 
 #define bbl_displacement(insn)		\
 	sign_extend32(((insn) & 0x3ffffff) << 2, 27)
@@ -49,6 +50,18 @@ static inline u32 get_w_reg(struct pt_regs *regs, int reg)
 	return lower_32_bits(pt_regs_read_reg(regs, reg));
 }
 
+static inline void update_lr(struct pt_regs *regs, long addr)
+{
+	int err = 0;
+
+	if (user_mode(regs) && task_gcs_el0_enabled(current)) {
+		push_user_gcs(addr + 4,	 &err);
+		if (err)
+			force_sig(SIGSEGV);
+	}
+	procedure_link_pointer_set(regs, addr + 4);
+}
+
 static bool __kprobes check_cbz(u32 opcode, struct pt_regs *regs)
 {
 	int xn = opcode & 0x1f;
@@ -107,9 +120,8 @@ simulate_b_bl(u32 opcode, long addr, struct pt_regs *regs)
 {
 	int disp = bbl_displacement(opcode);
 
-	/* Link register is x30 */
 	if (opcode & (1 << 31))
-		set_x_reg(regs, 30, addr + 4);
+		update_lr(regs, addr);
 
 	instruction_pointer_set(regs, addr + disp);
 }
@@ -133,17 +145,25 @@ simulate_br_blr(u32 opcode, long addr, struct pt_regs *regs)
 	/* update pc first in case we're doing a "blr lr" */
 	instruction_pointer_set(regs, get_x_reg(regs, xn));
 
-	/* Link register is x30 */
 	if (((opcode >> 21) & 0x3) == 1)
-		set_x_reg(regs, 30, addr + 4);
+		update_lr(regs, addr);
 }
 
 void __kprobes
 simulate_ret(u32 opcode, long addr, struct pt_regs *regs)
 {
+	u64 ret_addr;
+	int err = 0;
 	int xn = (opcode >> 5) & 0x1f;
 
 	instruction_pointer_set(regs, get_x_reg(regs, xn));
+
+	if (user_mode(regs) && task_gcs_el0_enabled(current)) {
+		ret_addr = pop_user_gcs(&err);
+		if (err || ret_addr != procedure_link_pointer(regs))
+			force_sig(SIGSEGV);
+	}
+
 }
 
 void __kprobes
-- 
2.49.0

Re: [PATCH v2 4/6] arm64: probes: Add GCS support to bl/blr/ret

[Jeremy Linton]

Hi,

On 4/7/25 11:19 AM, Jeremy Linton wrote:
> The arm64 probe simulation doesn't currently have logic in place
> to deal with GCS and this results in core dumps if probes are inserted
> at control flow locations. Fix-up bl, blr and ret to manipulate the
> shadow stack as needed.
> 
> While we manipulate and validate the shadow stack correctly, the
> hardware provides additional security by only allowing GCS operations
> against pages which are marked to support GCS. For writing there is
> gcssttr() which enforces this, but there isn't an equivalent for
> reading. This means that uprobe users should be aware that probing on
> control flow instructions which require reading the shadow stack (ex:
> ret) offers lower security guarantees than what is achieved without
> the uprobe active.
> 
> Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
> ---
>   arch/arm64/kernel/probes/simulate-insn.c | 28 ++++++++++++++++++++----
>   1 file changed, 24 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm64/kernel/probes/simulate-insn.c b/arch/arm64/kernel/probes/simulate-insn.c
> index 09a0b36122d0..1fc9bb69b1eb 100644
> --- a/arch/arm64/kernel/probes/simulate-insn.c
> +++ b/arch/arm64/kernel/probes/simulate-insn.c
> @@ -13,6 +13,7 @@
>   #include <asm/traps.h>
>   
>   #include "simulate-insn.h"
> +#include "asm/gcs.h"
>   
>   #define bbl_displacement(insn)		\
>   	sign_extend32(((insn) & 0x3ffffff) << 2, 27)
> @@ -49,6 +50,18 @@ static inline u32 get_w_reg(struct pt_regs *regs, int reg)
>   	return lower_32_bits(pt_regs_read_reg(regs, reg));
>   }
>   
> +static inline void update_lr(struct pt_regs *regs, long addr)
> +{
> +	int err = 0;
> +
> +	if (user_mode(regs) && task_gcs_el0_enabled(current)) {
> +		push_user_gcs(addr + 4,	 &err);
> +		if (err)
> +			force_sig(SIGSEGV);
> +	}
> +	procedure_link_pointer_set(regs, addr + 4);
> +}
> +
>   static bool __kprobes check_cbz(u32 opcode, struct pt_regs *regs)
>   {
>   	int xn = opcode & 0x1f;
> @@ -107,9 +120,8 @@ simulate_b_bl(u32 opcode, long addr, struct pt_regs *regs)
>   {
>   	int disp = bbl_displacement(opcode);
>   
> -	/* Link register is x30 */
>   	if (opcode & (1 << 31))
> -		set_x_reg(regs, 30, addr + 4);
> +		update_lr(regs, addr);
>   
>   	instruction_pointer_set(regs, addr + disp);
>   }
> @@ -133,17 +145,25 @@ simulate_br_blr(u32 opcode, long addr, struct pt_regs *regs)
>   	/* update pc first in case we're doing a "blr lr" */
>   	instruction_pointer_set(regs, get_x_reg(regs, xn));
>   
> -	/* Link register is x30 */
>   	if (((opcode >> 21) & 0x3) == 1)
> -		set_x_reg(regs, 30, addr + 4);
> +		update_lr(regs, addr);
>   }
>   
>   void __kprobes
>   simulate_ret(u32 opcode, long addr, struct pt_regs *regs)
>   {
> +	u64 ret_addr;
> +	int err = 0;
>   	int xn = (opcode >> 5) & 0x1f;
>   
>   	instruction_pointer_set(regs, get_x_reg(regs, xn));
> +
> +	if (user_mode(regs) && task_gcs_el0_enabled(current)) {
> +		ret_addr = pop_user_gcs(&err);
> +		if (err || ret_addr != procedure_link_pointer(regs))
> +			force_sig(SIGSEGV);
> +	}
> +
>   }

I was looking at this earlier this week and decided that this routine 
should have a return added following the force_sig(), and moving the 
instruction_pointer_set() below that. That syncs the behavior with what 
would happen without the uprobe active when ret_addr != LR. Then I 
dumped the 'xn' too.

[PATCH v2 5/6] arm64: uprobes: Add GCS support to uretprobes

[Jeremy Linton]

Ret probes work by changing the value in the link register at
the probe location to return to the probe rather than the calling
routine. Thus the GCS needs to be updated with this address as well.

Since its possible to insert probes at locations where the
current value of the LR doesn't match the GCS state this needs
to be detected and handled in order to maintain the existing
no-fault behavior.

Co-developed-by: Steve Capper <steve.capper@arm.com>
Signed-off-by: Steve Capper <steve.capper@arm.com>
(updated to use new gcs accessors, and handle LR/GCS mismatches)
Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/kernel/probes/uprobes.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/arch/arm64/kernel/probes/uprobes.c b/arch/arm64/kernel/probes/uprobes.c
index cb3d05af36e3..5e72409a255a 100644
--- a/arch/arm64/kernel/probes/uprobes.c
+++ b/arch/arm64/kernel/probes/uprobes.c
@@ -159,11 +159,41 @@ arch_uretprobe_hijack_return_addr(unsigned long trampoline_vaddr,
 				  struct pt_regs *regs)
 {
 	unsigned long orig_ret_vaddr;
+	unsigned long gcs_ret_vaddr;
+	int err = 0;
+	u64 gcspr;
 
 	orig_ret_vaddr = procedure_link_pointer(regs);
+
+	if (task_gcs_el0_enabled(current)) {
+		gcspr = read_sysreg_s(SYS_GCSPR_EL0);
+		gcs_ret_vaddr = load_user_gcs((unsigned long __user *)gcspr, &err);
+		if (err) {
+			force_sig(SIGSEGV);
+			goto out;
+		}
+		/*
+		 * If the LR and GCS entry don't match, then some kind of PAC/control
+		 * flow happened. Likely because the user is attempting to retprobe
+		 * on something that isn't a function boundary or inside a leaf
+		 * function. Explicitly abort this retprobe because it will generate
+		 * a GCS exception.
+		 */
+		if (gcs_ret_vaddr != orig_ret_vaddr)	{
+			orig_ret_vaddr = -1;
+			goto out;
+		}
+		put_user_gcs(trampoline_vaddr, (unsigned long __user *) gcspr, &err);
+		if (err) {
+			force_sig(SIGSEGV);
+			goto out;
+		}
+	}
+
 	/* Replace the return addr with trampoline addr */
 	procedure_link_pointer_set(regs, trampoline_vaddr);
 
+out:
 	return orig_ret_vaddr;
 }
 
-- 
2.49.0

[PATCH v2 6/6] arm64: Kconfig: Remove GCS restrictions on UPROBES

[Jeremy Linton]

Now that the uprobe paths have been made GCS compatible
drop the Kconfig restriction.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 arch/arm64/Kconfig | 1 -
 1 file changed, 1 deletion(-)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index a182295e6f08..b962a1321ebf 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -2236,7 +2236,6 @@ config ARM64_GCS
 	default y
 	select ARCH_HAS_USER_SHADOW_STACK
 	select ARCH_USES_HIGH_VMA_FLAGS
-	depends on !UPROBES
 	help
 	  Guarded Control Stack (GCS) provides support for a separate
 	  stack with restricted access which contains only return
-- 
2.49.0