[PATCH] target/hppa: lock both words of function descriptor

Mikulas Patocka posted 1 patch 8 months, 1 week ago
Failed in applying to current master (apply log)
linux-user/hppa/signal.c |    5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
[PATCH] target/hppa: lock both words of function descriptor
Posted by Mikulas Patocka 8 months, 1 week ago
The code in setup_rt_frame reads two words at haddr, but locks only one.
This patch fixes it to lock both.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>

---
 linux-user/hppa/signal.c |    5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Index: qemu/linux-user/hppa/signal.c
===================================================================
--- qemu.orig/linux-user/hppa/signal.c
+++ qemu/linux-user/hppa/signal.c
@@ -149,12 +149,11 @@ void setup_rt_frame(int sig, struct targ
         target_ulong *fdesc, dest;
 
         haddr &= -4;
-        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
+        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 * sizeof(target_ulong), 1)))
             goto give_sigsegv;
-        }
         __get_user(dest, fdesc);
         __get_user(env->gr[19], fdesc + 1);
-        unlock_user_struct(fdesc, haddr, 1);
+        unlock_user(fdesc, haddr, 0);
         haddr = dest;
     }
     env->iaoq_f = haddr;
Re: [PATCH] target/hppa: lock both words of function descriptor
Posted by Helge Deller 8 months, 1 week ago
On 9/16/23 15:52, Mikulas Patocka wrote:
> The code in setup_rt_frame reads two words at haddr, but locks only one.
> This patch fixes it to lock both.
>
> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
>
> ---
>   linux-user/hppa/signal.c |    5 ++---
>   1 file changed, 2 insertions(+), 3 deletions(-)
>
> Index: qemu/linux-user/hppa/signal.c
> ===================================================================
> --- qemu.orig/linux-user/hppa/signal.c
> +++ qemu/linux-user/hppa/signal.c
> @@ -149,12 +149,11 @@ void setup_rt_frame(int sig, struct targ
>           target_ulong *fdesc, dest;
>
>           haddr &= -4;
> -        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
> +        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 * sizeof(target_ulong), 1)))
>               goto give_sigsegv;
> -        }

Patch is Ok, but I think the qemu coding style is to keep the { } braces, even
if they are unnecessary (as in this case).

Acked-by: Helge Deller <deller@gmx.de>

>           __get_user(dest, fdesc);
>           __get_user(env->gr[19], fdesc + 1);
> -        unlock_user_struct(fdesc, haddr, 1);
> +        unlock_user(fdesc, haddr, 0);
>           haddr = dest;
>       }
>       env->iaoq_f = haddr;
>
Re: [PATCH] target/hppa: lock both words of function descriptor
Posted by Richard Henderson 8 months, 1 week ago
On 9/16/23 09:18, Helge Deller wrote:
> On 9/16/23 15:52, Mikulas Patocka wrote:
>> The code in setup_rt_frame reads two words at haddr, but locks only one.
>> This patch fixes it to lock both.
>>
>> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
>>
>> ---
>>   linux-user/hppa/signal.c |    5 ++---
>>   1 file changed, 2 insertions(+), 3 deletions(-)
>>
>> Index: qemu/linux-user/hppa/signal.c
>> ===================================================================
>> --- qemu.orig/linux-user/hppa/signal.c
>> +++ qemu/linux-user/hppa/signal.c
>> @@ -149,12 +149,11 @@ void setup_rt_frame(int sig, struct targ
>>           target_ulong *fdesc, dest;
>>
>>           haddr &= -4;
>> -        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
>> +        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 * sizeof(target_ulong), 1)))
>>               goto give_sigsegv;
>> -        }
> 
> Patch is Ok, but I think the qemu coding style is to keep the { } braces, even
> if they are unnecessary (as in this case).

Coding style also separates the assignment from the if condition.


r~

> 
> Acked-by: Helge Deller <deller@gmx.de>
> 
>>           __get_user(dest, fdesc);
>>           __get_user(env->gr[19], fdesc + 1);
>> -        unlock_user_struct(fdesc, haddr, 1);
>> +        unlock_user(fdesc, haddr, 0);
>>           haddr = dest;
>>       }
>>       env->iaoq_f = haddr;
>>
> 


Re: [PATCH] target/hppa: lock both words of function descriptor
Posted by Helge Deller 8 months, 1 week ago
On 9/16/23 19:17, Richard Henderson wrote:
> On 9/16/23 09:18, Helge Deller wrote:
>> On 9/16/23 15:52, Mikulas Patocka wrote:
>>> The code in setup_rt_frame reads two words at haddr, but locks only one.
>>> This patch fixes it to lock both.
>>>
>>> Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
>>>
>>> ---
>>>   linux-user/hppa/signal.c |    5 ++---
>>>   1 file changed, 2 insertions(+), 3 deletions(-)
>>>
>>> Index: qemu/linux-user/hppa/signal.c
>>> ===================================================================
>>> --- qemu.orig/linux-user/hppa/signal.c
>>> +++ qemu/linux-user/hppa/signal.c
>>> @@ -149,12 +149,11 @@ void setup_rt_frame(int sig, struct targ
>>>           target_ulong *fdesc, dest;
>>>
>>>           haddr &= -4;
>>> -        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
>>> +        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 * sizeof(target_ulong), 1)))
>>>               goto give_sigsegv;
>>> -        }
>>
>> Patch is Ok, but I think the qemu coding style is to keep the { } braces, even
>> if they are unnecessary (as in this case).
>
> Coding style also separates the assignment from the if condition.

I'll fix it up.

Thanks!

Helge


>
> r~
>
>>
>> Acked-by: Helge Deller <deller@gmx.de>
>>
>>>           __get_user(dest, fdesc);
>>>           __get_user(env->gr[19], fdesc + 1);
>>> -        unlock_user_struct(fdesc, haddr, 1);
>>> +        unlock_user(fdesc, haddr, 0);
>>>           haddr = dest;
>>>       }
>>>       env->iaoq_f = haddr;
>>>
>>
>
>
[PATCH v2] qemu-hppa: lock both words of function descriptor
Posted by Mikulas Patocka 8 months, 1 week ago


On Sat, 16 Sep 2023, Helge Deller wrote:

> On 9/16/23 15:52, Mikulas Patocka wrote:
> > The code in setup_rt_frame reads two words at haddr, but locks only one.
> > This patch fixes it to lock both.
> >
> > Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
> >
> > ---
> >   linux-user/hppa/signal.c |    5 ++---
> >   1 file changed, 2 insertions(+), 3 deletions(-)
> >
> > Index: qemu/linux-user/hppa/signal.c
> > ===================================================================
> > --- qemu.orig/linux-user/hppa/signal.c
> > +++ qemu/linux-user/hppa/signal.c
> > @@ -149,12 +149,11 @@ void setup_rt_frame(int sig, struct targ
> >           target_ulong *fdesc, dest;
> >
> >           haddr &= -4;
> > -        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
> > +        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 *
> > sizeof(target_ulong), 1)))
> >               goto give_sigsegv;
> > -        }
> 
> Patch is Ok, but I think the qemu coding style is to keep the { } braces, even
> if they are unnecessary (as in this case).
> 
> Acked-by: Helge Deller <deller@gmx.de>

OK, here I resend it:


From: Mikulas Patocka <mpatocka@redhat.com>

The code in setup_rt_frame reads two words at haddr, but locks only one.
This patch fixes it to lock both.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Acked-by: Helge Deller <deller@gmx.de>

---
 linux-user/hppa/signal.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: qemu/linux-user/hppa/signal.c
===================================================================
--- qemu.orig/linux-user/hppa/signal.c
+++ qemu/linux-user/hppa/signal.c
@@ -149,12 +149,12 @@ void setup_rt_frame(int sig, struct targ
         target_ulong *fdesc, dest;
 
         haddr &= -4;
-        if (!lock_user_struct(VERIFY_READ, fdesc, haddr, 1)) {
+        if (!(fdesc = lock_user(VERIFY_READ, haddr, 2 * sizeof(target_ulong), 1))) {
             goto give_sigsegv;
         }
         __get_user(dest, fdesc);
         __get_user(env->gr[19], fdesc + 1);
-        unlock_user_struct(fdesc, haddr, 1);
+        unlock_user(fdesc, haddr, 0);
         haddr = dest;
     }
     env->iaoq_f = haddr;