You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@httpd.apache.org by Rainer Jung <ra...@kippdata.de> on 2011/01/17 18:51:46 UTC
Slotmem patch for alignment
Hi,
I wrote a small patch fixing an alignment issue for mod_slotmem:
http://people.apache.org/~rjung/patches/mod_slotmem_shm_alignment.patch
The problem is: mod_slotmem uses a header for meta data before the
actual data slots. The addresses used for the data slots are offset by
the size of the meta data. Currently this size is divisible by 4 but not
by 8 and thus the data put into the slots is only aligned at 4 byte
boudaries. If there is data inside the slots needing 8 byte alignment,
then the compiler can not arrange for that.
I fixed it by increasing the offset to the APR default alignment. Please
check whether you think there's a problem with it. It tried it in
combination with a separate proxy patch I'll post in a minute and both
together fix the crashes observed on Solaris Sparc.
One additional remark: I didn't actually understand the original code lines:
apr_size_t basesize = (item_size * item_num);
apr_size_t size = sizeof(sharedslotdesc_t) +
(item_num * sizeof(char)) + basesize;
What's the reason for the "(item_num * sizeof(char))" part? I can't
actually see, where this part of the memory is being used.
Regards,
Rainer
Re: Slotmem patch for alignment
Posted by Jim Jagielski <ji...@jaguNET.com>.
On Jan 17, 2011, at 12:51 PM, Rainer Jung wrote:
> Hi,
>
> I wrote a small patch fixing an alignment issue for mod_slotmem:
>
> http://people.apache.org/~rjung/patches/mod_slotmem_shm_alignment.patch
>
> The problem is: mod_slotmem uses a header for meta data before the actual data slots. The addresses used for the data slots are offset by the size of the meta data. Currently this size is divisible by 4 but not by 8 and thus the data put into the slots is only aligned at 4 byte boudaries. If there is data inside the slots needing 8 byte alignment, then the compiler can not arrange for that.
>
> I fixed it by increasing the offset to the APR default alignment. Please check whether you think there's a problem with it. It tried it in combination with a separate proxy patch I'll post in a minute and both together fix the crashes observed on Solaris Sparc.
>
Sounds good.
> One additional remark: I didn't actually understand the original code lines:
>
> apr_size_t basesize = (item_size * item_num);
> apr_size_t size = sizeof(sharedslotdesc_t) +
> (item_num * sizeof(char)) + basesize;
>
> What's the reason for the "(item_num * sizeof(char))" part? I can't actually see, where this part of the memory is being used.
>
It's the inuse table
> Regards,
>
> Rainer
>