Oct 15 2020
Sep 23 2020
revised wait_for_completion_timeout() call to avoid missing complete message
add saved_done to avoid missing wake-up.
Sep 20 2020
Aug 31 2020
Jul 30 2020
Looks good to me. Thanks for fixing this!
Jul 28 2020
Is this only for test purpose? When will this be defined? Thanks.
May 22 2020
May 20 2020
May 11 2020
Do you have performance test results for already existing linux implementation?
From a quick look it seems to me there will be bottleneck regarding locking that seems can be reduced using CK and epoch. But this task can be done in future, if you plan support this code.
Apr 15 2020
Out for broader review
Mar 13 2020
Jul 9 2019
Nov 1 2018
Oct 31 2018
Oct 29 2018
Oct 22 2018
Nov 22 2015
Nov 13 2015
This has been tested by several customers who were experiencing packet loss on their 10.2 gateway VMs. The patch fixed their problems. I just have one minor comment to add a little detailed comments in the code.
Jul 30 2015
Jul 28 2015
Jul 27 2015
Jul 22 2015
Jul 15 2015
Jul 7 2015
Jul 1 2015
Bjoern helped fixed the problem in the NO_INET and NO_INET6 builds. Also MFC his fix r284889 to 10 stable.
Also MFC r284889 Fix compilation without INET6 and without INET and INET6
after offload support was introduced in r284746.
Jun 26 2015
Are you aware of https://lists.freebsd.org/pipermail/svn-src-head/2015-June/073483.html? This needs to be solved before any MFC can be done.
Jun 24 2015
Thanks Roger for the review!
Jun 17 2015
- More review comments from royger
Sorry it took a while to get this done. I was out of office due to family issues.
May 28 2015
- Incorporate review comments from royger
May 25 2015
Thanks for the review. I have addressed most of them. Only have a question regarding to creating a new tag for M_DEVBUF.
May 22 2015
May 19 2015
- Fix white space issue left from previous change
- Incorporate internal review comments
May 18 2015
- Incorporate review comments from rogyer.
Thanks Roger! I will re-do the merge and upload a new review.
I also see such mergeinfo in the svn diff output. I am not sure why it happens and how it could effect the commit. What I did was:
May 12 2015
May 11 2015
May 4 2015
Code committed to head as r282212. Thanks all for the review!
Apr 29 2015
Apr 28 2015
OK, but getting rid of this custom bounce buffer should be a high priority.
Apr 27 2015
The Microsoft Ignite event is coming up next week. The team here would like to see the code in head before this event so the management can make some announcement.
Apr 23 2015
To be clear, if PG_KEEP_OFFSET isn't sufficient, I'd be tempted to go with the private allocator for now. If we teach bus_dma about this type of restriction, I think I'd prefer it to be a new flag. I'm less sure about extending the filter callback. I'd actually like to remove the filter callback longer term. Also, changing it would be a fairly invasive API/ABI change that would be difficult to MFC.
I've crafted a patch to add information about first/last segments to bus_dma_filter, it is only implemented for x86 right now, so it breaks other arches:
Can you test if this makes it possible to use bus_dma with your use-case?
Apr 21 2015
Regarding to the comment from John:
Apr 20 2015
Apr 17 2015
Apr 16 2015
Apr 14 2015
- Change the comment in storvsc driver to better state the reason
Apr 13 2015
- Add comment in storvsc driver for the potential BUS_DMA fix and using it for bounce buffer.
Apr 1 2015
Mar 31 2015
Mar 23 2015
- Incoporate review comments from royger.
Mar 9 2015
Thanks John for the review. I just updated the code with your review comments.
- Incorporate review comments from jhb.
Mar 4 2015
Mar 3 2015
Feb 28 2015
Thanks. I have committed the change.