The best way to fix this appears to be to bump the counters up to u32 or
s32.
Just an open question: Do you think that we could get better performance if
we would go with u32 instead of reducing the buffer to 16K?
Jerome
----- Original Message -----
From: "Michael S. Tsirkin" <[EMAIL PROTECTED]>
To: "Roland Dreier" <[EMAIL PROTECTED]>; "Tom Duffy"
<[EMAIL PROTECTED]>; "Jerome Pioux" <[EMAIL PROTECTED]>
Cc: <[email protected]>
Sent: Sunday, September 25, 2005 3:45 AM
Subject: Re: FW: SDP problems with 64K page size
Quoting r. Roland Dreier <[EMAIL PROTECTED]>:
Subject: FW: SDP problems with 64K page size
Hi, Jerome asked me to forward this on, since for some reason his
email didn't appear when he sent it.
In any case there seem to be some PAGE_SIZE dependencies in SDP.
Libor provided a patch that fixed this up a while ago, but I don't
know if this is the right way to handle this.
- R.
Roland, thanks very much for forwarding this, and for providing
a patch to Jerome.
The problem is with recv_size/send_size counters in SDP, which are u16,
so that assigning a value of 64K overflows them.
The best way to fix this appears to be to bump the counters up to u32 or
s32.
The patch that Roland posted fixes this by reducing the buffer
size to 16K, so that should work, too.
Roland, I might check in the patch that you posted to work around
this problem for 64K page users, until I have a final fix ready.
Is that OK with everyone?
Thanks,
--
MST
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
