public inbox for gdb-patches@sourceware.org
 help / color / mirror / Atom feed
From: Luis Machado <luis.machado@arm.com>
To: Yvan Roux <yvan.roux@foss.st.com>, gdb-patches@sourceware.org
Cc: Torbjorn SVENSSON <torbjorn.svensson@foss.st.com>
Subject: Re: [PATCH] gdb/arm: Unwind Non-Secure callbacks from Secure
Date: Mon, 27 Jun 2022 09:38:34 +0100	[thread overview]
Message-ID: <a21c1796-a571-3db6-9abc-16843689ca96@arm.com> (raw)
In-Reply-To: <20220624160525.GA30742@gnbcxd0114.gnb.st.com>

On 6/24/22 17:05, Yvan Roux wrote:
> Hi,
> 
> Thanks for the review Luis, here is a new version of the patch.
> 
> On Tue, Jun 21, 2022 at 12:28:07PM +0100, Luis Machado wrote:
>> Hi,
>>
>> On 6/17/22 13:50, Yvan Roux wrote:
>>> Hi,
>>>
>>> Without this changeset, the unwinding doesn't take into account
>>> Non-Secure to Secure stack unwinding enablement status and
>>> doesn't choose on the proper SP to do the unwinding.
>>
>> choose on -> choose?
>>
>>>
>>> This patch only unwind the stack when Non-Secure to Secure
>>
>> unwind -> unwinds
>>> unwinding is enabled, previous SP is set w/r to the current mode
>>> (Handler -> msp_s, Thread -> psp_s) and then the Secure stack is
>>> unwound. Ensure thumb bit is set in PSR when needed. Also, drop
>>> thumb bit from PC if set.
>>
>> Also, two spaces after `.` according to the GNU Coding Standards.
>>
>>>
>>> Signed-off-by: Torbjörn SVENSSON <torbjorn.svensson@foss.st.com>
>>> Signed-off-by: Yvan ROUX <yvan.roux@foss.st.com>
>>> ---
>>>    gdb/arm-tdep.c | 121 +++++++++++++++++++++++++++++++++++++------------
>>>    1 file changed, 91 insertions(+), 30 deletions(-)
>>>
>>> diff --git a/gdb/arm-tdep.c b/gdb/arm-tdep.c
>>> index 0c907482036..8a84754cfa6 100644
>>> --- a/gdb/arm-tdep.c
>>> +++ b/gdb/arm-tdep.c
>>> @@ -309,6 +309,21 @@ struct arm_prologue_cache
>>>      arm_prologue_cache() = default;
>>>    };
>>> +
>>> +/* Reconstruct T bit in program status register from LR value.  */
>>> +
>>> +static inline ULONGEST
>>> +reconstruct_t_bit(struct gdbarch *gdbarch, CORE_ADDR lr, ULONGEST psr)
>>> +{
>>> +  ULONGEST t_bit = arm_psr_thumb_bit (gdbarch);
>>> +  if (IS_THUMB_ADDR (lr))
>>> +    psr |= t_bit;
>>> +  else
>>> +    psr &= ~t_bit;
>>> +
>>> +  return psr;
>>> +}
>>> +
>>>    /* Initialize stack pointers, and flag the active one.  */
>>>    static inline void
>>> @@ -2342,15 +2357,10 @@ arm_prologue_prev_register (struct frame_info *this_frame,
>>>         but the processor status is likely valid.  */
>>>      if (prev_regnum == ARM_PS_REGNUM)
>>>        {
>>> -      CORE_ADDR lr, cpsr;
>>> -      ULONGEST t_bit = arm_psr_thumb_bit (gdbarch);
>>> +      ULONGEST cpsr = get_frame_register_unsigned (this_frame, prev_regnum);
>>> +      CORE_ADDR lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
>>> -      cpsr = get_frame_register_unsigned (this_frame, prev_regnum);
>>> -      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
>>> -      if (IS_THUMB_ADDR (lr))
>>> -	cpsr |= t_bit;
>>> -      else
>>> -	cpsr &= ~t_bit;
>>> +      cpsr = reconstruct_t_bit (gdbarch, lr, cpsr);
>>>          return frame_unwind_got_constant (this_frame, prev_regnum, cpsr);
>>>        }
>>> @@ -3363,24 +3373,46 @@ arm_m_exception_cache (struct frame_info *this_frame)
>>>          return cache;
>>>        }
>>> -  fnc_return = ((lr & 0xfffffffe) == 0xfefffffe);
>>> +  fnc_return = (((lr >> 24) & 0xff) == 0xfe);
>>
>> Is the above throwing away the comparison with the lower byte 0xfe?
> 
> Yes, it only checks the FNC_RETURN prefix for coherency with how EXC_RETURN is
> handled and how it is also done in arm_m_addr_is_magic function (line 777), but
> according to the reference manual bits [23:1] are all 1, so we can keep the
> current comparison if you prefer.
>   

Thanks for the explanation. No need to keep it if it doesn't make sense.

I just wanted to make sure we're not dropping something that may be used by other targets.

>>>      if (tdep->have_sec_ext && fnc_return)
>>>        {
>>> -      int actual_sp;
>>> +      if (!arm_unwind_secure_frames)
>>> +	{
>>> +	  warning (_("Non-secure to secure stack unwinding disabled."));
>>> -      arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_msp_ns_regnum);
>>> -      arm_cache_set_active_sp_value (cache, tdep, sp);
>>> -      if (lr & 1)
>>> -	actual_sp = tdep->m_profile_msp_s_regnum;
>>> +	  /* Terminate any further stack unwinding by referring to self.  */
>>> +	  arm_cache_set_active_sp_value (cache, tdep, sp);
>>> +	  return cache;
>>> +	}
>>> +
>>> +      xpsr = get_frame_register_unsigned (this_frame, ARM_PS_REGNUM);
>>> +      if ((xpsr & 0xff) != 0)
>>> +	/* Handler mode */
>>
>> What is the handler mode? Can we expand on that to make it clear?
> 
> This is the mode in which exceptions are handled and Threadmode is the normal
> or main mode where a program runs in, the comment are more explicits in this
> new version of the patch.
> 

Thanks.

>>> +	arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_msp_s_regnum);
>>>          else
>>> -	actual_sp = tdep->m_profile_msp_ns_regnum;
>>> +	/* Thread mode */
>>
>> Similarly, what is the thread mode?
>>
>>> +	arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_psp_s_regnum);
>>> +
>>> +      unwound_sp = arm_cache_get_prev_sp_value (cache, tdep);
>>> -      arm_cache_switch_prev_sp (cache, tdep, actual_sp);
>>> -      sp = get_frame_register_unsigned (this_frame, actual_sp);
>>> +      /* Stack layout for a function call from Secure to Non-Secure state
>>> +	 (ARMv8-M section B3.16):
>>> -      cache->saved_regs[ARM_LR_REGNUM].set_addr (sp);
>>> +	    SP Offset
>>> -      arm_cache_set_active_sp_value (cache, tdep, sp + 8);
>>> +		    +-------------------+
>>> +	     0x08   |                   |
>>> +		    +-------------------+   <-- Original SP
>>> +	     0x04   |    Partial xPSR   |
>>> +		    +-------------------+
>>> +	     0x00   |   Return Address  |
>>> +		    +===================+   <-- New SP  */
>>> +
>>> +      cache->saved_regs[ARM_PC_REGNUM].set_addr (unwound_sp + 0x00);
>>> +      cache->saved_regs[ARM_LR_REGNUM].set_addr (unwound_sp + 0x00);
>>> +      cache->saved_regs[ARM_PS_REGNUM].set_addr (unwound_sp + 0x04);
>>> +
>>> +      arm_cache_set_active_sp_value (cache, tdep, unwound_sp + 0x08);
>>>          return cache;
>>>        }
>>> @@ -3441,11 +3473,6 @@ arm_m_exception_cache (struct frame_info *this_frame)
>>>    	    arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_msp_regnum);
>>>    	}
>>>        }
>>> -  else
>>> -    {
>>> -      /* Main stack used, use MSP as SP.  */
>>> -      arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_msp_regnum);
>>> -    }
>>>      /* Fetch the SP to use for this frame.  */
>>>      unwound_sp = arm_cache_get_prev_sp_value (cache, tdep);
>>> @@ -3641,6 +3668,20 @@ arm_m_exception_prev_register (struct frame_info *this_frame,
>>>        return frame_unwind_got_constant (this_frame, prev_regnum,
>>>    				      arm_cache_get_prev_sp_value (cache, tdep));
>>> +  /* If we are asked to unwind the PC, strip the saved T bit.  */
>>> +  if (prev_regnum == ARM_PC_REGNUM)
>>> +    {
>>> +      CORE_ADDR pc;
>>> +      struct value *value;
>>> +
>>> +      value = trad_frame_get_prev_register (this_frame, cache->saved_regs,
>>> +					    prev_regnum);
>>> +
>>> +      pc = value_as_address (value);
>>> +      return frame_unwind_got_constant (this_frame, prev_regnum,
>>> +				        UNMAKE_THUMB_ADDR (pc));
>>> +    }
>>> +
>>>      /* The value might be one of the alternative SP, if so, use the
>>>         value already constructed.  */
>>>      if (arm_cache_is_sp_register (cache, tdep, prev_regnum))
>>> @@ -3649,6 +3690,29 @@ arm_m_exception_prev_register (struct frame_info *this_frame,
>>>          return frame_unwind_got_constant (this_frame, prev_regnum, sp_value);
>>>        }
>>> +  /* If we are asked to unwind the xPSR, set T bit if PC is in thumb mode.
>>> +   * LR register is unreliable as it contains FNC_RETURN or EXC_RETURN pattern.
>>> +   */
>>
>> I think this style of comments doesn't match GDB's coding standards.
>>
>>> +  if (prev_regnum == ARM_PS_REGNUM)
>>> +    {
>>> +      CORE_ADDR pc;
>>> +      ULONGEST xpsr;
>>
>>
>> You could declare both pc and xpsr at their assignment locations.
>>
>>> +      struct value *value;
>>
>> Same thing as above. Declare it in its assignment.
>>
>>> +      struct gdbarch *gdbarch = get_frame_arch (this_frame);
>>> +
>>> +      value = trad_frame_get_prev_register (this_frame, cache->saved_regs,
>>> +					    ARM_PC_REGNUM);
>>
>> here...
>>
>>> +      pc = value_as_address (value);
>>
>> here...
>>> +
>>> +      value = trad_frame_get_prev_register (this_frame, cache->saved_regs,
>>> +					    ARM_PS_REGNUM);
>>> +      xpsr = value_as_long (value);
>>
>> and here.
>>> +
>>> +      /* Reconstruct the T bit; see arm_prologue_prev_register for details.  */
>>
>> /* Reconstruct the T bit.  See arm_prologue_prev_register for details.  */
>>
>>> +      xpsr = reconstruct_t_bit (gdbarch, pc, xpsr);
>>> +      return frame_unwind_got_constant (this_frame, ARM_PS_REGNUM, xpsr);
>>> +    }
>>> +
>>>      return trad_frame_get_prev_register (this_frame, cache->saved_regs,
>>>    				       prev_regnum);
>>>    }
>>> @@ -3711,8 +3775,8 @@ arm_dwarf2_prev_register (struct frame_info *this_frame, void **this_cache,
>>>    {
>>>      struct gdbarch * gdbarch = get_frame_arch (this_frame);
>>>      arm_gdbarch_tdep *tdep = (arm_gdbarch_tdep *) gdbarch_tdep (gdbarch);
>>> -  CORE_ADDR lr, cpsr;
>>> -  ULONGEST t_bit = arm_psr_thumb_bit (gdbarch);
>>> +  CORE_ADDR lr;
>>> +  ULONGEST cpsr;
>>>      switch (regnum)
>>>        {
>>> @@ -3741,10 +3805,7 @@ arm_dwarf2_prev_register (struct frame_info *this_frame, void **this_cache,
>>>          /* Reconstruct the T bit; see arm_prologue_prev_register for details.  */
>>>          cpsr = get_frame_register_unsigned (this_frame, regnum);
>>>          lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
>>> -      if (IS_THUMB_ADDR (lr))
>>> -	cpsr |= t_bit;
>>> -      else
>>> -	cpsr &= ~t_bit;
>>> +      cpsr = reconstruct_t_bit (gdbarch, lr, cpsr);
>>>          return frame_unwind_got_constant (this_frame, regnum, cpsr);
>>>        default
> 
> Updated version:
> 
> gdb/arm: Unwind Non-Secure callbacks from Secure
> 
> Without this changeset, the unwinding doesn't take into account
> Non-Secure to Secure stack unwinding enablement status and
> doesn't choose the proper SP to do the unwinding.
> 
> This patch only unwinds the stack when Non-Secure to Secure
> unwinding is enabled, previous SP is set w/r to the current mode
> (Handler -> msp_s, Thread -> psp_s) and then the Secure stack is
> unwound.  Ensure thumb bit is set in PSR when needed. Also, drop
> thumb bit from PC if set.
> 
> Signed-off-by: Torbjörn SVENSSON <torbjorn.svensson@foss.st.com>
> Signed-off-by: Yvan ROUX <yvan.roux@foss.st.com>
> ---
>   gdb/arm-tdep.c | 117 ++++++++++++++++++++++++++++++++++++-------------
>   1 file changed, 87 insertions(+), 30 deletions(-)
> 
> diff --git a/gdb/arm-tdep.c b/gdb/arm-tdep.c
> index 96d70d40b28..8ae0c9fecd7 100644
> --- a/gdb/arm-tdep.c
> +++ b/gdb/arm-tdep.c
> @@ -309,6 +309,21 @@ struct arm_prologue_cache
>     arm_prologue_cache() = default;
>   };
>   
> +
> +/* Reconstruct T bit in program status register from LR value.  */
> +
> +static inline ULONGEST
> +reconstruct_t_bit(struct gdbarch *gdbarch, CORE_ADDR lr, ULONGEST psr)
> +{
> +  ULONGEST t_bit = arm_psr_thumb_bit (gdbarch);
> +  if (IS_THUMB_ADDR (lr))
> +    psr |= t_bit;
> +  else
> +    psr &= ~t_bit;
> +
> +  return psr;
> +}
> +
>   /* Initialize stack pointers, and flag the active one.  */
>   
>   static inline void
> @@ -2348,15 +2363,10 @@ arm_prologue_prev_register (struct frame_info *this_frame,
>        but the processor status is likely valid.  */
>     if (prev_regnum == ARM_PS_REGNUM)
>       {
> -      CORE_ADDR lr, cpsr;
> -      ULONGEST t_bit = arm_psr_thumb_bit (gdbarch);
> +      ULONGEST cpsr = get_frame_register_unsigned (this_frame, prev_regnum);
> +      CORE_ADDR lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
>   
> -      cpsr = get_frame_register_unsigned (this_frame, prev_regnum);
> -      lr = frame_unwind_register_unsigned (this_frame, ARM_LR_REGNUM);
> -      if (IS_THUMB_ADDR (lr))
> -	cpsr |= t_bit;
> -      else
> -	cpsr &= ~t_bit;
> +      cpsr = reconstruct_t_bit (gdbarch, lr, cpsr);
>         return frame_unwind_got_constant (this_frame, prev_regnum, cpsr);
>       }
>   
> @@ -3369,24 +3379,46 @@ arm_m_exception_cache (struct frame_info *this_frame)
>         return cache;
>       }
>   
> -  fnc_return = ((lr & 0xfffffffe) == 0xfefffffe);
> +  fnc_return = (((lr >> 24) & 0xff) == 0xfe);
>     if (tdep->have_sec_ext && fnc_return)
>       {
> -      int actual_sp;
> +      if (!arm_unwind_secure_frames)
> +	{
> +	  warning (_("Non-secure to secure stack unwinding disabled."));
>   
> -      arm_cache_switch_prev_sp (cache, tdep, tdep->m_profile_msp_ns_regnum);
> -      arm_cache_set_active_sp_value (cache, tdep, sp);
> -      if (lr & 1)
> -	actual_sp = tdep->m_profile_msp_s_regnum;
> +	  /* Terminate any further stack unwinding by referring to self.  */
> +	  arm_cache_set_active_sp_value (cache, tdep, sp);
> +	  return cache;
> +	}
> +
> +      xpsr = get_frame_register_unsigned (this_frame, ARM_PS_REGNUM);
> +      if ((xpsr & 0xff) != 0)
> +	/* Handler mode: This is mode that exceptions are handled in.  */


This is mode -> This is the mode.

Otherwise LGTM.

Thanks!

      reply	other threads:[~2022-06-27  8:38 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-06-17 12:50 Yvan Roux
2022-06-21 11:28 ` Luis Machado
2022-06-24 16:05   ` Yvan Roux
2022-06-27  8:38     ` Luis Machado [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=a21c1796-a571-3db6-9abc-16843689ca96@arm.com \
    --to=luis.machado@arm.com \
    --cc=gdb-patches@sourceware.org \
    --cc=torbjorn.svensson@foss.st.com \
    --cc=yvan.roux@foss.st.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).