Date: Tue, 23 Nov 2004 04:09:20 +0100 (CET) From: Sten Spans <sten@blinkenlights.nl> To: John Baldwin <jhb@FreeBSD.org> Cc: freebsd-alpha@FreeBSD.org Subject: Re: alpha and em mtu Message-ID: <Pine.SOC.4.61.0411230406010.10997@tea.blinkenlights.nl> In-Reply-To: <Pine.SOC.4.61.0411230259030.10997@tea.blinkenlights.nl> References: <Pine.SOC.4.61.0411142153430.26307@tea.blinkenlights.nl> <Pine.SOC.4.61.0411222147180.10997@tea.blinkenlights.nl> <Pine.SOC.4.61.0411230026430.10997@tea.blinkenlights.nl> <Pine.SOC.4.61.0411230259030.10997@tea.blinkenlights.nl>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 23 Nov 2004, Sten Spans wrote:
>> doesn't seem to print anything, but ...
>>
>> if_em.c
>> 2442
>> 2443 /*if (ifp->if_mtu <= ETHERMTU) { */
>> 2444 m_adj(mp, ETHER_ALIGN);
>> 2445 /*} */
>> 2446
>>
>> does seem to fix the crash, also trashes the performance,
>> but that's another matter. It looks like mbuf alignment is
>> needed, if_bge seems to provide reasonable examples.
>
> And looking at netbsd/openbsd clarifies the whole issue,
>
> #ifdef __STRICT_ALIGNMENT
> /*
> * The ethernet payload is not 32-bit aligned when
> * Jumbo packets are enabled, so on architectures
> with
> * strict alignment we need to shift the entire
> packet
> * ETHER_ALIGN bytes. Ugh.
> */
>
>
> This diff probably should be merged.
> http://www.openbsd.org/cgi-bin/cvsweb/src/sys/dev/pci/if_em.c.diff?r1=1.22&r2=1.23
>
> Although I don't know wether freebsd has the STRICT_ALIGNMENT define.
This is an initial patch based on the openbsd code,
which solves the if_em issue, and seems to give ok performance
( note to self: turn off debugging when testing network performance ).
Comments welcome.
--- if_em.c.orig Tue Nov 23 03:03:23 2004
+++ if_em.c Tue Nov 23 03:20:15 2004
@@ -2784,8 +2784,57 @@
/* Assign correct length to the current fragment */
mp->m_len = len;
+#ifndef __i386__
+ /*
+ * The ethernet payload is not 32-bit aligned when
+ * Jumbo packets are enabled, so on architectures with
+ * strict alignment we need to shift the entire packet
+ * ETHER_ALIGN bytes. Ugh.
+ */
+ if (ifp->if_mtu > ETHERMTU) {
+ unsigned char tmp_align_buf[ETHER_ALIGN];
+ int tmp_align_buf_len = 0;
+
+ if (prev_len_adj > adapter->align_buf_len)
+ prev_len_adj -= adapter->align_buf_len;
+ else
+ prev_len_adj = 0;
+
+ if (mp->m_len > MCLBYTES - ETHER_ALIGN) {
+ bcopy(mp->m_data +
+ (MCLBYTES - ETHER_ALIGN),
+ &tmp_align_buf,
+ ETHER_ALIGN);
+ tmp_align_buf_len = mp->m_len -
+ (MCLBYTES - ETHER_ALIGN);
+ mp->m_len -= ETHER_ALIGN;
+ }
+
+ if (mp->m_len) {
+ bcopy(mp->m_data,
+ mp->m_data + ETHER_ALIGN,
+ mp->m_len);
+ if (!adapter->align_buf_len)
+ mp->m_data += ETHER_ALIGN;
+ }
+
+ if (adapter->align_buf_len) {
+ mp->m_len += adapter->align_buf_len;
+ bcopy(&adapter->align_buf,
+ mp->m_data,
+ adapter->align_buf_len);
+ }
+
+ if (tmp_align_buf_len)
+ bcopy(&tmp_align_buf,
+ &adapter->align_buf,
+ tmp_align_buf_len);
+ adapter->align_buf_len = tmp_align_buf_len;
+ }
+#endif /* __386__ */
+
if (adapter->fmp == NULL) {
- mp->m_pkthdr.len = len;
+ mp->m_pkthdr.len = mp->m_len;
adapter->fmp = mp; /* Store the first mbuf */
adapter->lmp = mp;
} else {
@@ -2801,7 +2850,7 @@
}
adapter->lmp->m_next = mp;
adapter->lmp = adapter->lmp->m_next;
- adapter->fmp->m_pkthdr.len += len;
+ adapter->fmp->m_pkthdr.len += mp->m_len;
}
if (eop) {
--- if_em.h.orig Wed Nov 10 03:18:52 2004
+++ if_em.h Tue Nov 23 03:33:01 2004
@@ -347,6 +347,12 @@
u_int8_t unit;
struct mtx mtx;
+#ifndef __i386__
+ /* Used for carrying forward alignment adjustments */
+ unsigned char align_buf[ETHER_ALIGN]; /* tail of unaligned packet */
+ u_int8_t align_buf_len; /* bytes in tail */
+#endif /* __i386__ */
+
/* Info about the board itself */
u_int32_t part_num;
u_int8_t link_active;
--
Sten Spans
"There is a crack in everything, that's how the light gets in."
Leonard Cohen - Anthem
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SOC.4.61.0411230406010.10997>