Re: Alignment traps on snmpd 5.7.1 build for ARM9

[Tyler O. <tyl...@gm...>]

This patch has been submitted as ID 3536420.

On Mon, Jun 18, 2012 at 7:30 PM, Tyler Olmstead
<tyl...@gm...> wrote:
> Here's a patch that fixes the problem. I have only tested it on my
> target (arm926ejs), so tests on other architectures would be
> worthwhile. Hope this helps, and let me know if I can do anything
> else.
>
> -- Tyler
>
> diff --git a/agent/mibgroup/ip-mib/data_access/defaultrouter_linux.c
> b/agent/mibgroup/ip-mib/data_access/defaultrouter_linux.c
> index 702128b..4cb5dd9 100644
> --- a/agent/mibgroup/ip-mib/data_access/defaultrouter_linux.c
> +++ b/agent/mibgroup/ip-mib/data_access/defaultrouter_linux.c
> @@ -11,7 +11,7 @@
>
>  #include "ip-mib/ipDefaultRouterTable/ipDefaultRouterTable.h"
>
> -#include <asm/types.h>
> +#include <stdint.h>
>  #ifdef HAVE_LINUX_RTNETLINK_H
>  #include <linux/netlink.h>
>  #include <linux/rtnetlink.h>
> @@ -93,9 +93,18 @@ _load(netsnmp_container *container)
>     netsnmp_defaultrouter_entry *entry;
>     int nlsk;
>     struct sockaddr_nl addr;
> -    unsigned char rcvbuf[RCVBUF_SIZE];
>     int rcvbuf_size = RCVBUF_SIZE;
> -    unsigned char sndbuf[SNDBUF_SIZE];
> +
> +    /* Buffers must be memory aligned. Message structure internal alignment is
> +    * maintained by the netlink API. */
> +    unsigned char rcvbufmem[RCVBUF_SIZE + sizeof(intmax_t)];
> +    unsigned char *rcvbuf = rcvbufmem +
> +        sizeof(intmax_t) - (((intptr_t)rcvbufmem) % sizeof(intmax_t));
> +
> +    unsigned char sndbufmem[SNDBUF_SIZE + sizeof(intmax_t)];
> +    unsigned char *sndbuf = sndbufmem +
> +        sizeof(intmax_t) - (((intptr_t)sndbufmem) % sizeof(intmax_t));
> +
>     struct nlmsghdr *hdr;
>     struct rtmsg *rthdr;
>     int count;
>
>
> On Mon, Jun 11, 2012 at 8:46 PM, Magnus Fromreide <ma...@ly...> wrote:
>> On Mon, 2012-06-11 at 16:32 -0700, Tyler Olmstead wrote:
>>> Hi all,
>>>
>>> I have cross-compiled net-snmp 5.7.1 for a custom port of Linux 2.6.37
>>> on an armv5tejl core (AM1808). I have encountered some alignment traps
>>> in syslog after MIB-walking.
>>>
>>> According to objdump, most of the traps seem to be triggered by an
>>> unaligned access to stack memory in defaultrouter_linux.c:static int
>>> _load(netsnmp_container *container). Has anyone else encountered this?
>>> Is this a known issue or should I submit a patch?
>>
>> I have not heard of this before so a patch or enough information to let
>> us debug this would be highly appreciated.
>>
>> /MF
>>