Re: [PATCH v4 05/12] riscv: ftrace: prepare ftrace for atomic code patching

[Robbin Ehn <rehn@rivosinc.com>]

Hi, thanks,

On Tue, 2025-04-08 at 02:08 +0800, Andy Chiu wrote:
> From: Andy Chiu <andy.chiu@sifive.com>
> 
> We use an AUIPC+JALR pair to jump into a ftrace trampoline. Since
> instruction fetch can break down to 4 byte at a time, it is impossible
> to update two instructions without a race. In order to mitigate it, we
> initialize the patchable entry to AUIPC + NOP4. Then, the run-time code
> patching can change NOP4 to JALR to eable/disable ftrcae from a
> function. This limits the reach of each ftrace entry to +-2KB displacing
> from ftrace_caller.
> 
> Starting from the trampoline, we add a level of indirection for it to
> reach ftrace caller target. Now, it loads the target address from a
> memory location, then perform the jump. This enable the kernel to update
> the target atomically.
> 
> The new don't-stop-the-world text patching on change only one RISC-V
> instruction:
> 
>   |  -8: &ftrace_ops of the associated tracer function.
>   | <ftrace enable>:
>   |   0: auipc  t0, hi(ftrace_caller)
>   |   4: jalr   t0, lo(ftrace_caller)
>   |
>   |  -8: &ftrace_nop_ops
>   | <ftrace disable>:
>   |   0: auipc  t0, hi(ftrace_caller)
>   |   4: nop
> 
> This means that f+0x0 is fixed, and should not be claimed by ftrace,
> e.g. kprobe should be able to put a probe in f+0x0. Thus, we adjust the
> offset and MCOUNT_INSN_SIZE accordingly.
> 
> Co-developed-by: Björn Töpel <bjorn@rivosinc.com>
> Signed-off-by: Björn Töpel <bjorn@rivosinc.com>
> Signed-off-by: Andy Chiu <andy.chiu@sifive.com>
> ---
> Changelog v4:
>  - Include Björn's fix for kprobe
>  - Refactor code for better reading (Robbin, Björn)
>  - Remove make_call_ra and friedns (Björn)
>  - Update comments to match reality (Björn)
>  - Drop code defined by !WITH_ARG
>  - Add a synchronization point when updating ftrace_call_dest (Björn)
> ---
>  arch/riscv/include/asm/ftrace.h |  49 ++++++------
>  arch/riscv/kernel/ftrace.c      | 130 ++++++++++++++++----------------
>  arch/riscv/kernel/mcount-dyn.S  |   9 +--
>  3 files changed, 92 insertions(+), 96 deletions(-)
> 
> diff --git a/arch/riscv/include/asm/ftrace.h b/arch/riscv/include/asm/ftrace.h
> index d8b2138bd9c6..6a5c0a7fb826 100644
> --- a/arch/riscv/include/asm/ftrace.h
> +++ b/arch/riscv/include/asm/ftrace.h
> @@ -20,10 +20,9 @@ extern void *return_address(unsigned int level);
>  #define ftrace_return_address(n) return_address(n)
>  
>  void _mcount(void);
> -static inline unsigned long ftrace_call_adjust(unsigned long addr)
> -{
> -	return addr;
> -}
> +unsigned long ftrace_call_adjust(unsigned long addr);
> +unsigned long arch_ftrace_get_symaddr(unsigned long fentry_ip);
> +#define ftrace_get_symaddr(fentry_ip) arch_ftrace_get_symaddr(fentry_ip)
>  
>  /*
>   * Let's do like x86/arm64 and ignore the compat syscalls.
> @@ -57,12 +56,21 @@ struct dyn_arch_ftrace {
>   * 2) jalr: setting low-12 offset to ra, jump to ra, and set ra to
>   *          return address (original pc + 4)
>   *
> + * The first 2 instructions for each tracable function is compiled to 2 nop
> + * instructions. Then, the kernel initializes the first instruction to auipc
> at
> + * boot time (<ftrace disable>). The second instruction is patched to jalr to
> + * start the trace.
> + *
> + *<Image>:
> + * 0: nop
> + * 4: nop
> + *
>   *<ftrace enable>:
> - * 0: auipc  t0/ra, 0x?
> - * 4: jalr   t0/ra, ?(t0/ra)
> + * 0: auipc  t0, 0x?
> + * 4: jalr   t0, ?(t0)
>   *
>   *<ftrace disable>:
> - * 0: nop
> + * 0: auipc  t0, 0x?
>   * 4: nop
>   *
>   * Dynamic ftrace generates probes to call sites, so we must deal with
> @@ -75,10 +83,9 @@ struct dyn_arch_ftrace {
>  #define AUIPC_OFFSET_MASK	(0xfffff000)
>  #define AUIPC_PAD		(0x00001000)
>  #define JALR_SHIFT		20
> -#define JALR_RA			(0x000080e7)
> -#define AUIPC_RA		(0x00000097)
>  #define JALR_T0			(0x000282e7)
>  #define AUIPC_T0		(0x00000297)
> +#define JALR_RANGE		(JALR_SIGN_MASK - 1)
>  
>  #define to_jalr_t0(offset)						\
>  	(((offset & JALR_OFFSET_MASK) << JALR_SHIFT) | JALR_T0)
> @@ -96,26 +103,14 @@ do
> {									\
>  	call[1] = to_jalr_t0(offset);					\
>  } while (0)
>  
> -#define to_jalr_ra(offset)						\
> -	(((offset & JALR_OFFSET_MASK) << JALR_SHIFT) | JALR_RA)
> -
> -#define to_auipc_ra(offset)						\
> -	((offset & JALR_SIGN_MASK) ?					\
> -	(((offset & AUIPC_OFFSET_MASK) + AUIPC_PAD) | AUIPC_RA) :	\
> -	((offset & AUIPC_OFFSET_MASK) | AUIPC_RA))
> -
> -#define make_call_ra(caller, callee, call)				\
> -do {									\
> -	unsigned int offset =						\
> -		(unsigned long) (callee) - (unsigned long) (caller);	\
> -	call[0] = to_auipc_ra(offset);					\
> -	call[1] = to_jalr_ra(offset);					\
> -} while (0)
> -
>  /*
> - * Let auipc+jalr be the basic *mcount unit*, so we make it 8 bytes here.
> + * Only the jalr insn in the auipc+jalr is patched, so we make it 4
> + * bytes here.
>   */
> -#define MCOUNT_INSN_SIZE 8
> +#define MCOUNT_INSN_SIZE	4
> +#define MCOUNT_AUIPC_SIZE	4
> +#define MCOUNT_JALR_SIZE	4
> +#define MCOUNT_NOP4_SIZE	4
>  
>  #ifndef __ASSEMBLY__
>  struct dyn_ftrace;
> diff --git a/arch/riscv/kernel/ftrace.c b/arch/riscv/kernel/ftrace.c
> index 1fd10555c580..cf78eef073a0 100644
> --- a/arch/riscv/kernel/ftrace.c
> +++ b/arch/riscv/kernel/ftrace.c
> @@ -8,10 +8,21 @@
>  #include <linux/ftrace.h>
>  #include <linux/uaccess.h>
>  #include <linux/memory.h>
> +#include <linux/irqflags.h>
>  #include <linux/stop_machine.h>
>  #include <asm/cacheflush.h>
>  #include <asm/text-patching.h>
>  
> +unsigned long ftrace_call_adjust(unsigned long addr)
> +{
> +	return addr + MCOUNT_AUIPC_SIZE;
> +}
> +
> +unsigned long arch_ftrace_get_symaddr(unsigned long fentry_ip)
> +{
> +	return fentry_ip - MCOUNT_AUIPC_SIZE;
> +}
> +
>  #ifdef CONFIG_DYNAMIC_FTRACE
>  void ftrace_arch_code_modify_prepare(void) __acquires(&text_mutex)
>  {
> @@ -32,51 +43,32 @@ void ftrace_arch_code_modify_post_process(void)
> __releases(&text_mutex)
>  	mutex_unlock(&text_mutex);
>  }
>  
> -static int ftrace_check_current_call(unsigned long hook_pos,
> -				     unsigned int *expected)
> +static int __ftrace_modify_call(unsigned long source, unsigned long target,
> bool validate)
>  {
> +	unsigned int call[2], offset;

There is a mix of unsigned int and u32.
It doesn't seem to serve any purpose?

>  	unsigned int replaced[2];
> -	unsigned int nops[2] = {RISCV_INSN_NOP4, RISCV_INSN_NOP4};
>  
> -	/* we expect nops at the hook position */
> -	if (!expected)
> -		expected = nops;
> +	offset = target - source;
> +	call[1] = to_jalr_t0(offset);
>  
> -	/*
> -	 * Read the text we want to modify;
> -	 * return must be -EFAULT on read error
> -	 */
> -	if (copy_from_kernel_nofault(replaced, (void *)hook_pos,
> -			MCOUNT_INSN_SIZE))
> -		return -EFAULT;
> -
> -	/*
> -	 * Make sure it is what we expect it to be;
> -	 * return must be -EINVAL on failed comparison
> -	 */
> -	if (memcmp(expected, replaced, sizeof(replaced))) {
> -		pr_err("%p: expected (%08x %08x) but got (%08x %08x)\n",
> -		       (void *)hook_pos, expected[0], expected[1],
> replaced[0],
> -		       replaced[1]);
> -		return -EINVAL;
> +	if (validate) {
> +		call[0] = to_auipc_t0(offset);
> +		/*
> +		 * Read the text we want to modify;
> +		 * return must be -EFAULT on read error
> +		 */
> +		if (copy_from_kernel_nofault(replaced, (void *)source, 2 *
> MCOUNT_INSN_SIZE))
> +			return -EFAULT;
> +
> +		if (replaced[0] != call[0]) {
> +			pr_err("%p: expected (%08x) but got (%08x)\n",
> +			       (void *)source, call[0], replaced[0]);
> +			return -EINVAL;
> +		}
>  	}
>  
> -	return 0;
> -}
> -
> -static int __ftrace_modify_call(unsigned long hook_pos, unsigned long target,
> -				bool enable, bool ra)
> -{
> -	unsigned int call[2];
> -	unsigned int nops[2] = {RISCV_INSN_NOP4, RISCV_INSN_NOP4};
> -
> -	if (ra)
> -		make_call_ra(hook_pos, target, call);
> -	else
> -		make_call_t0(hook_pos, target, call);
> -
> -	/* Replace the auipc-jalr pair at once. Return -EPERM on write error.
> */
> -	if (patch_insn_write((void *)hook_pos, enable ? call : nops,
> MCOUNT_INSN_SIZE))
> +	/* Replace the jalr at once. Return -EPERM on write error. */
> +	if (patch_insn_write((void *)(source + MCOUNT_AUIPC_SIZE), call + 1,
> MCOUNT_JALR_SIZE))
>  		return -EPERM;
>  
>  	return 0;
> @@ -84,22 +76,21 @@ static int __ftrace_modify_call(unsigned long hook_pos,
> unsigned long target,
>  
>  int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
>  {
> -	unsigned int call[2];
> +	unsigned long distance, orig_addr, pc = rec->ip - MCOUNT_AUIPC_SIZE;
>  
> -	make_call_t0(rec->ip, addr, call);
> -
> -	if (patch_insn_write((void *)rec->ip, call, MCOUNT_INSN_SIZE))
> -		return -EPERM;
> +	orig_addr = (unsigned long)&ftrace_caller;
> +	distance = addr > orig_addr ? addr - orig_addr : orig_addr - addr;
> +	if (distance > JALR_RANGE)
> +		return -EINVAL;
>  
> -	return 0;
> +	return __ftrace_modify_call(pc, addr, false);
>  }
>  
> -int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec,
> -		    unsigned long addr)
> +int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec, unsigned long
> addr)
>  {
> -	unsigned int nops[2] = {RISCV_INSN_NOP4, RISCV_INSN_NOP4};
> +	u32 nop4 = RISCV_INSN_NOP4;
>  
> -	if (patch_insn_write((void *)rec->ip, nops, MCOUNT_INSN_SIZE))
> +	if (patch_insn_write((void *)rec->ip, &nop4, MCOUNT_NOP4_SIZE))
>  		return -EPERM;
>  
>  	return 0;
> @@ -114,21 +105,38 @@ int ftrace_make_nop(struct module *mod, struct
> dyn_ftrace *rec,
>   */
>  int ftrace_init_nop(struct module *mod, struct dyn_ftrace *rec)
>  {
> -	int out;
> +	unsigned long pc = rec->ip - MCOUNT_AUIPC_SIZE;
> +	unsigned int nops[2], offset;
> +	int ret;
> +
> +	offset = (unsigned long) &ftrace_caller - pc;
> +	nops[0] = to_auipc_t0(offset);
> +	nops[1] = RISCV_INSN_NOP4;
>  
>  	mutex_lock(&text_mutex);
> -	out = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
> +	ret = patch_insn_write((void *)pc, nops, 2 * MCOUNT_INSN_SIZE);
>  	mutex_unlock(&text_mutex);
>  
> -	return out;
> +	return ret;
>  }
>  
> +ftrace_func_t ftrace_call_dest = ftrace_stub;
>  int ftrace_update_ftrace_func(ftrace_func_t func)
>  {
> -	int ret = __ftrace_modify_call((unsigned long)&ftrace_call,
> -				       (unsigned long)func, true, true);
> -
> -	return ret;
> +	WRITE_ONCE(ftrace_call_dest, func);
> +	/*
> +	 * The data fence ensure that the update to ftrace_call_dest happens
> +	 * before the write to function_trace_op later in the generic ftrace.
> +	 * If the sequence is not enforced, then an old ftrace_call_dest may
> +	 * race loading a new function_trace_op set in ftrace_modify_all_code
> +	 *
> +	 * If we are in stop_machine, then we don't need to call remote fence
> +	 * as there is no concurrent read-side of ftrace_call_dest.
> +	 */
> +	smp_wmb();
> +	if (!irqs_disabled())
> +		smp_call_function(ftrace_sync_ipi, NULL, 1);
> +	return 0;

Here we allow the read of cpu_online_mask being reordered with the store to
ftrace_call_dest.

Can't this happen then?

Hart X                 | Hart new
load cpu_online_mask   |
                       | store cpu_online_mask
                       | tlb shootdown/icache flush
store ftrace_call_dest |
ftrace_sync_ipi        | // not hit
                       | sees old value ?

Thanks, Robbin

>  }
>  
>  struct ftrace_modify_param {
> @@ -172,17 +180,11 @@ void arch_ftrace_update_code(int command)
>  int ftrace_modify_call(struct dyn_ftrace *rec, unsigned long old_addr,
>  		       unsigned long addr)
>  {
> +	unsigned long caller = rec->ip - MCOUNT_AUIPC_SIZE;
>  	unsigned int call[2];
> -	unsigned long caller = rec->ip;
> -	int ret;
>  
>  	make_call_t0(caller, old_addr, call);
> -	ret = ftrace_check_current_call(caller, call);
> -
> -	if (ret)
> -		return ret;
> -
> -	return __ftrace_modify_call(caller, addr, true, false);
> +	return __ftrace_modify_call(caller, addr, true);
>  }
>  #endif
>  
> diff --git a/arch/riscv/kernel/mcount-dyn.S b/arch/riscv/kernel/mcount-dyn.S
> index 3f06b40bb6c8..8aa554d56096 100644
> --- a/arch/riscv/kernel/mcount-dyn.S
> +++ b/arch/riscv/kernel/mcount-dyn.S
> @@ -13,7 +13,6 @@
>  
>  	.text
>  
> -#define FENTRY_RA_OFFSET	8
>  #define ABI_SIZE_ON_STACK	80
>  #define ABI_A0			0
>  #define ABI_A1			8
> @@ -62,8 +61,7 @@
>  * After the stack is established,
>  *
>  * 0(sp) stores the PC of the traced function which can be accessed
> -* by &(fregs)->epc in tracing function. Note that the real
> -* function entry address should be computed with -FENTRY_RA_OFFSET.
> +* by &(fregs)->epc in tracing function.
>  *
>  * 8(sp) stores the function return address (i.e. parent IP) that
>  * can be accessed by &(fregs)->ra in tracing function.
> @@ -140,7 +138,7 @@
>  	.endm
>  
>  	.macro PREPARE_ARGS
> -	addi	a0, t0, -FENTRY_RA_OFFSET
> +	addi	a0, t0, -MCOUNT_JALR_SIZE	// ip (callsite's jalr insn)
>  	la	a1, function_trace_op
>  	REG_L	a2, 0(a1)
>  	mv	a1, ra
> @@ -153,7 +151,8 @@ SYM_FUNC_START(ftrace_caller)
>  	PREPARE_ARGS
>  
>  SYM_INNER_LABEL(ftrace_call, SYM_L_GLOBAL)
> -	call	ftrace_stub
> +	REG_L	ra, ftrace_call_dest
> +	jalr	ra, 0(ra)
>  
>  	RESTORE_ABI_REGS
>  	bnez	t1, .Ldirect