Re: [PATCH] Reclaim _REENT_MP_P5S in _reclaim_reent

On 16/11/2023 8:15 pm, Corinna Vinschen wrote:
> On Nov 16 12:12, [email protected] wrote:
>> From: Chris Johns <[email protected]>
>>
>> The change fixes a memory leak in threads that clean up using
>> _reclaim_reent.
>>
>> RTEMS: Closes #4967
>> ---
>>  newlib/libc/reent/reent.c | 11 +++++++++++
>>  1 file changed, 11 insertions(+)
>>
>> diff --git a/newlib/libc/reent/reent.c b/newlib/libc/reent/reent.c
>> index db80ca06e..6bb271b9a 100644
>> --- a/newlib/libc/reent/reent.c
>> +++ b/newlib/libc/reent/reent.c
>> @@ -59,6 +59,17 @@ _reclaim_reent (struct _reent *ptr)
>>  	}
>>        if (_REENT_MP_RESULT(ptr))
>>  	_free_r (ptr, _REENT_MP_RESULT(ptr));
>> +      if (_REENT_MP_P5S(ptr))
>> +        {
>> +         struct _Bigint *thisone, *nextone;
>> +         nextone = _REENT_MP_P5S(ptr);
>> +         while (nextone)
>> +           {
>> +             thisone = nextone;
>> +             nextone = nextone->_next;
>> +             _free_r (ptr, thisone);
>> +           }
>> +       }
> 
> The p5s data is allocated using Balloc, so the pointers are in freelist
> and should already be free'd at this point, starting at line 42.

Yes however P5S blocks allocated by Balloc via i2b are not passed to Bfree.

> Please explain what happens and why free'ing the freelist is not
> sufficient in this case, preferredly as part of the commit message.

Yes good idea. How about:

The _REENT_MP_P5S blocks are allocated using Balloc via i2b and linked in the
pow5mult call. As a result these blocks are not on the freelist managed by the
Bfree call. This change fixes a memory leak in threads that clean up using
_reclaim_reent.

Chris