[Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[chengang]

From: Chen Gang <chengang@emindsoft.com.cn>

Original implementation uses do_rt_sigreturn directly in host space,
when a guest program is in unwind procedure in guest space, it will get
an incorrect restore address, then causes unwind failure.

Also cleanup the original incorrect indentation.

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
---
 linux-user/signal.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/linux-user/signal.c b/linux-user/signal.c
index 919aa83..0e3b1c6 100644
--- a/linux-user/signal.c
+++ b/linux-user/signal.c
@@ -5566,8 +5566,13 @@ struct target_rt_sigframe {
     unsigned char save_area[16]; /* caller save area */
     struct target_siginfo info;
     struct target_ucontext uc;
+    abi_ulong retcode[2];
 };
 
+#define INSN_MOVELI_R10_139  0x00045fe551483000ULL /* { moveli r10, 139 } */
+#define INSN_SWINT1          0x286b180051485000ULL /* { swint1 } */
+
+
 static void setup_sigcontext(struct target_sigcontext *sc,
                              CPUArchState *env, int signo)
 {
@@ -5643,9 +5648,12 @@ static void setup_rt_frame(int sig, struct target_sigaction *ka,
     __put_user(target_sigaltstack_used.ss_size, &frame->uc.tuc_stack.ss_size);
     setup_sigcontext(&frame->uc.tuc_mcontext, env, info->si_signo);
 
-    restorer = (unsigned long) do_rt_sigreturn;
     if (ka->sa_flags & TARGET_SA_RESTORER) {
-            restorer = (unsigned long) ka->sa_restorer;
+        restorer = (unsigned long) ka->sa_restorer;
+    } else {
+        __put_user(INSN_MOVELI_R10_139, &frame->retcode[0]);
+        __put_user(INSN_SWINT1, &frame->retcode[1]);
+        restorer = (unsigned long)frame->retcode;
     }
     env->pc = (unsigned long) ka->_sa_handler;
     env->regs[TILEGX_R_SP] = (unsigned long) frame;
-- 
1.9.3

Re: [Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[Chen Gang]

Hello All:

Please help check this patch when you have time.

After this patch, we can let gcc testsuite cleanup-10 run successfully.

Next, I shall continue to implement floating point instructions: remove
(u)int64_to_float64 from fdouble implementation.


Thanks.

On 3/15/16 05:51, chengang@emindsoft.com.cn wrote:
> From: Chen Gang <chengang@emindsoft.com.cn>
> 
> Original implementation uses do_rt_sigreturn directly in host space,
> when a guest program is in unwind procedure in guest space, it will get
> an incorrect restore address, then causes unwind failure.
> 
> Also cleanup the original incorrect indentation.
> 
> Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
> ---
>  linux-user/signal.c | 12 ++++++++++--
>  1 file changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/linux-user/signal.c b/linux-user/signal.c
> index 919aa83..0e3b1c6 100644
> --- a/linux-user/signal.c
> +++ b/linux-user/signal.c
> @@ -5566,8 +5566,13 @@ struct target_rt_sigframe {
>      unsigned char save_area[16]; /* caller save area */
>      struct target_siginfo info;
>      struct target_ucontext uc;
> +    abi_ulong retcode[2];
>  };
>  
> +#define INSN_MOVELI_R10_139  0x00045fe551483000ULL /* { moveli r10, 139 } */
> +#define INSN_SWINT1          0x286b180051485000ULL /* { swint1 } */
> +
> +
>  static void setup_sigcontext(struct target_sigcontext *sc,
>                               CPUArchState *env, int signo)
>  {
> @@ -5643,9 +5648,12 @@ static void setup_rt_frame(int sig, struct target_sigaction *ka,
>      __put_user(target_sigaltstack_used.ss_size, &frame->uc.tuc_stack.ss_size);
>      setup_sigcontext(&frame->uc.tuc_mcontext, env, info->si_signo);
>  
> -    restorer = (unsigned long) do_rt_sigreturn;
>      if (ka->sa_flags & TARGET_SA_RESTORER) {
> -            restorer = (unsigned long) ka->sa_restorer;
> +        restorer = (unsigned long) ka->sa_restorer;
> +    } else {
> +        __put_user(INSN_MOVELI_R10_139, &frame->retcode[0]);
> +        __put_user(INSN_SWINT1, &frame->retcode[1]);
> +        restorer = (unsigned long)frame->retcode;
>      }
>      env->pc = (unsigned long) ka->_sa_handler;
>      env->regs[TILEGX_R_SP] = (unsigned long) frame;
> 

-- 
Chen Gang (陈刚)

Managing Natural Environments is the Duty of Human Beings.

Re: [Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[Laurent Vivier]

Le 27/03/2016 11:44, Chen Gang a écrit :
> Hello All:
> 
> Please help check this patch when you have time.
> 
> After this patch, we can let gcc testsuite cleanup-10 run successfully.
> 
> Next, I shall continue to implement floating point instructions: remove
> (u)int64_to_float64 from fdouble implementation.
> 
> 
> Thanks.
> 
> On 3/15/16 05:51, chengang@emindsoft.com.cn wrote:
>> From: Chen Gang <chengang@emindsoft.com.cn>
>>
>> Original implementation uses do_rt_sigreturn directly in host space,
>> when a guest program is in unwind procedure in guest space, it will get
>> an incorrect restore address, then causes unwind failure.
>>
>> Also cleanup the original incorrect indentation.
>>
>> Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
>> ---
>>  linux-user/signal.c | 12 ++++++++++--
>>  1 file changed, 10 insertions(+), 2 deletions(-)
>>
>> diff --git a/linux-user/signal.c b/linux-user/signal.c
>> index 919aa83..0e3b1c6 100644
>> --- a/linux-user/signal.c
>> +++ b/linux-user/signal.c
>> @@ -5566,8 +5566,13 @@ struct target_rt_sigframe {
>>      unsigned char save_area[16]; /* caller save area */
>>      struct target_siginfo info;
>>      struct target_ucontext uc;
>> +    abi_ulong retcode[2];
>>  };
>>  
>> +#define INSN_MOVELI_R10_139  0x00045fe551483000ULL /* { moveli r10, 139 } */
>> +#define INSN_SWINT1          0x286b180051485000ULL /* { swint1 } */
>> +
>> +
>>  static void setup_sigcontext(struct target_sigcontext *sc,
>>                               CPUArchState *env, int signo)
>>  {
>> @@ -5643,9 +5648,12 @@ static void setup_rt_frame(int sig, struct target_sigaction *ka,
>>      __put_user(target_sigaltstack_used.ss_size, &frame->uc.tuc_stack.ss_size);
>>      setup_sigcontext(&frame->uc.tuc_mcontext, env, info->si_signo);
>>  
>> -    restorer = (unsigned long) do_rt_sigreturn;
>>      if (ka->sa_flags & TARGET_SA_RESTORER) {
>> -            restorer = (unsigned long) ka->sa_restorer;
>> +        restorer = (unsigned long) ka->sa_restorer;
>> +    } else {
>> +        __put_user(INSN_MOVELI_R10_139, &frame->retcode[0]);
>> +        __put_user(INSN_SWINT1, &frame->retcode[1]);
>> +        restorer = (unsigned long)frame->retcode;

The address of retcode in host and guest can differ.
You need something like:

	restorer = (unsigned long)(frame_addr + offsetof(struct
target_rt_sigframe, retcode));

I've experienced this on sh4 (see commit 2a0fa68)

>>      }
>>      env->pc = (unsigned long) ka->_sa_handler;
>>      env->regs[TILEGX_R_SP] = (unsigned long) frame;
>>
> 

Laurent

Re: [Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[Chen Gang]

On 3/29/16 06:17, Laurent Vivier wrote:
>> On 3/15/16 05:51, chengang@emindsoft.com.cn wrote:
>>>
>>> Original implementation uses do_rt_sigreturn directly in host space,
>>> when a guest program is in unwind procedure in guest space, it will get
>>> an incorrect restore address, then causes unwind failure.
>>>
>>> Also cleanup the original incorrect indentation.
>>>
>>> Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
>>> ---
>>>  linux-user/signal.c | 12 ++++++++++--
>>>  1 file changed, 10 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/linux-user/signal.c b/linux-user/signal.c
>>> index 919aa83..0e3b1c6 100644
>>> --- a/linux-user/signal.c
>>> +++ b/linux-user/signal.c
>>> @@ -5566,8 +5566,13 @@ struct target_rt_sigframe {
>>>      unsigned char save_area[16]; /* caller save area */
>>>      struct target_siginfo info;
>>>      struct target_ucontext uc;
>>> +    abi_ulong retcode[2];
>>>  };
>>>  
>>> +#define INSN_MOVELI_R10_139  0x00045fe551483000ULL /* { moveli r10, 139 } */
>>> +#define INSN_SWINT1          0x286b180051485000ULL /* { swint1 } */
>>> +
>>> +
>>>  static void setup_sigcontext(struct target_sigcontext *sc,
>>>                               CPUArchState *env, int signo)
>>>  {
>>> @@ -5643,9 +5648,12 @@ static void setup_rt_frame(int sig, struct target_sigaction *ka,
>>>      __put_user(target_sigaltstack_used.ss_size, &frame->uc.tuc_stack.ss_size);
>>>      setup_sigcontext(&frame->uc.tuc_mcontext, env, info->si_signo);
>>>  
>>> -    restorer = (unsigned long) do_rt_sigreturn;
>>>      if (ka->sa_flags & TARGET_SA_RESTORER) {
>>> -            restorer = (unsigned long) ka->sa_restorer;
>>> +        restorer = (unsigned long) ka->sa_restorer;
>>> +    } else {
>>> +        __put_user(INSN_MOVELI_R10_139, &frame->retcode[0]);
>>> +        __put_user(INSN_SWINT1, &frame->retcode[1]);
>>> +        restorer = (unsigned long)frame->retcode;
> 
> The address of retcode in host and guest can differ.
> You need something like:
> 
> 	restorer = (unsigned long)(frame_addr + offsetof(struct
> target_rt_sigframe, retcode));
> 
> I've experienced this on sh4 (see commit 2a0fa68)
> 

OK, thanks. What you said above sounds reasonable to me. :-)

I shall send patch v2 for it (although tilegx is a pure 64-bit target,
with this patch, I guess, tilegx target should still run correctly under
32-bit host).

By the way, it looks that s390x and microblaze targets also have the
same issue.


Thanks.
-- 
Chen Gang (陈刚)

Managing Natural Environments is the Duty of Human Beings.

Re: [Qemu-devel] [PATCH] linux-user/signal.c: Generate opcode data for restorer in setup_rt_frame

[Chen Gang]

On 3/29/16 06:57, Chen Gang wrote:
> On 3/29/16 06:17, Laurent Vivier wrote:
>>
>> The address of retcode in host and guest can differ.
>> You need something like:
>>
>> 	restorer = (unsigned long)(frame_addr + offsetof(struct
>> target_rt_sigframe, retcode));
>>
>> I've experienced this on sh4 (see commit 2a0fa68)
>>
> 
> OK, thanks. What you said above sounds reasonable to me. :-)
> 
> I shall send patch v2 for it (although tilegx is a pure 64-bit target,
> with this patch, I guess, tilegx target should still run correctly under
> 32-bit host).
> 
> By the way, it looks that s390x and microblaze targets also have the
> same issue.

It looks the alpha target may also need consider about retcode, for "r26
= frame_addr" in setup_rt_frame().

Thanks.
-- 
Chen Gang (陈刚)

Managing Natural Environments is the Duty of Human Beings.