For Friday, May 15th, 2009 I'm Jon Masters with a summary of today's LKML traffi c.
In today's issue: Generic DMA, KVM, DRBD, and a new kernel RC release.
Generic DMA. Fujita Tomonori posted version 2 of a patchset aimed at unifying in-kernel handling of DMA operations, through the introduction of a generic struct dma_map_ops. Initially, the patch series had targetted x86 and IA64, but Fujita also posted some patches implementing the new dma_map_ops for the sparc architecture. In the process, he was able to remove both of the (non-unified at the architecture level) dma-mapping_32.h and dma-mapping_64.h headers provided by the sparc port itself.
KVM. Nicholas A. Bellinger posted the results of various testing of 10GB Ethernet using PCIe passthrough with Linux/iSCSI and large block sizes under the KVM hypervisor. These benchmarks show the performance of both of the popular iSCSI initiators with various other settings and performance ranging up to 880MB/s for READ tests. Also on the subject of KVM, Gregory Haskins posted version 2 of an iosignalfd patch series. With the patch series, the KVM kernel can be configured to emit a signal whenever the guest writes to arbitrarily defined regions of memory. Using this patch series, one can trap on certain IO and perform a more efficient operation in the host than a pure IO emulation. For example, one could trigger a host-based DMA operation.
DRBD. Philipp Reisner reposted a 16 part DRBD patch series. For those just tuning in, DRBD is a shared-nothing, synchronously replicated block device designed to serve as a building block for high availability clusters. It has triggered a number of discussions about implementations of HA clustering storage over the past week, much of it positive (some of it typically tangential in nature). The latest patch series includes a number of cleanups, and the DRBD folks are seeking further reviews in lieu of a full merge. For more information, listeners can also visit http://www.drbd.org/.
In today's announcements: No major announcements today. For those experimenting with the latest RT trees, it's worth noting that a quiet rt13 release was made that fixes a build failure and adds some miscellanenous ARM fixes to the tree also.
The latest kernel release is 2.6.30-rc6, which was released by Linus late on Friday evening. This is won't be the final 2.6.30, however, since a number of last minute fixes have already been published - including a number from Benjamin Herrenschmidt (benh), several ext4 patches from Ted T'so, a number of driver updates to libata from Jeff Garzik, and a fix to the PCI MSI usage issues experienced by the Sun Neptune (NIU) driver . Ingo Molnar followed up to a question from Jeremy Fitzhardinge concerning when his Xen patches will be pulled, indicating that the patches are fine but that Ingo needs a bit longer to perform necessary code review before pulling into the x86 tree.
Stephen Rothwell posted a linux-next for May 15th. Since Thursday, linux-next gained a new tree 'asm-generic' with the aforementioned generic assembly fixes, which was temporarily dropped because it doesn't build currently. The fixes tree gained three net fixes, and the kvm tree lost its conflict. The addition of the new 'asm-generic ' tree brings the total to 137 subtrees.
And finally. An apology that today's podcast was a little later than planned. Updates typically go out around midnight during the week and in the early morning hours at weekends, though this is subject to variation. The next weekend summary podcast will be released on Monday morning.
That's a summary of today's LKML traffic. For further information visit kernel.o rg. I'm Jon Masters.