Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 11 Nov 2009 14:37:51 -0800
From:      Pyun YongHyeon <pyunyh@gmail.com>
To:        freebsd-current@freebsd.org
Subject:   Call for bge(4) testers
Message-ID:  <20091111223751.GE15449@michelle.cdnetworks.com>

next in thread | raw e-mail | index | archive | help

Hi,

I had been working on fixing bus_dma(9) bugs and adding TSO
capability to bge(4). Now TSO is supported for BCM5755 or newer
controllers. Actually some pre-BCM5755 controllers also support
TSO with the help of special firmware but the license issue and
lower performance of firmware based TSO as well as TSO bug I
intentionally excluded TSO support for pre-BCM5755 controllers.
You can get the patch form the following URL. The diff was
generated against latest HEAD.

http://people.freebsd.org/~yongari/bge/bge.tso.1111.diff
or
http://people.freebsd.org/~yongari/bge/if_bge.c
http://people.freebsd.org/~yongari/bge/if_bgereg.h

The diff also has a lot of functional changes which may affect
other bge(4) controllers. Specifically the diff
- Added workarounds for known DMA bug. This workaround shall make
  bge(4) use bounce buffers so this fix requires all previous
  changes made in HEAD. This may slow down overall performance if
  bounce buffers are used but I think it's better than leaving the
  bug.
- Tuned interrupt coalescing parameters.
- Use taskqeue to get better performance on PCIe controllers.
- TSO support for BCM5755 or newer controllers.
- Tx path cleaned up.
- Interrupt handler now clears status word as it does in polling
  case.

I'm interesting in performance changes before/after the diff and
any regressions from the diff. Please test I'll commit the diff
next week unless I get regressions.

Thanks.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20091111223751.GE15449>