For Thursday, June 4th 2009, I'm Jon Masters with a summary of today's LKML traf fic.
In today's issue: The Linux Driver Project, Remapping NULL pointers, MCE ring buffer, paravirt operations overhead, Super-H, System 390, Console screen blanking, and kernels listed on kernel.org.
The Linux Driver Project Status Report. Greg Kroah-Hartman posted an update concerning the Linux Driver Project, an effort intended to make life easier for those who want to work with the Linux community in supporting hardware. Greg re-affirms his ascertion that "almost all hardware is currently already supported on Linux, with almost all major vendors shipping Linux drivers". He also draws attention to several areas he had said he would work on last year. This included writing new drivers (many were written), educating people (Greg mentions how training is something that is still very much in high demand), and finally working more in the open (the staging tree). Over the next year, Greg intends to work with more companies on Linux drivers, maintain the staging tree as he currently does, and improve education and training.
Remapping NULL pointers. In yesterday's podcast, I mentioned ongoing discussion of potential NULL or zero page remapping, and in particular how there was some concern that allowing user tasks to map the zero page could expose the kernel to unnecessary risk in the case that there were later a compromise allowing that pointer to be passed into the kernel. Larry Highsmith had proposed changing the use of NULL pointers on error within the kernel to an alternative unmappable address instead. Discussion on that original topic swung off on a tangent - as these things are want to do - into the general issue of allowing tasks to map the NULL page. And mmap_min_addr in particular, whose availability currently depends upon CONFIG_SECURITY being set. But there is no real reason for that requirement any more, so James Morris has taken a fix from Christoph Lameter into the security-testing-2.6 tree, which will make that dependency go away in future kernels.
MCE ring buffer. Huang Ying posted various fixes to the MCE log ring buffer support in the kernel. Chief amongst these was the introduction of per-cpu ring buffers with a lock-less path for writers. The similarities in use between this implementation and Steven Rostedt's generic ring_buffer lead one to wonder whether the latter should be investigated for safe use during MCE.
Paravirt Operations Overhead. Rusty Russell and Linus Torvalds continued their dialog on the performance impact of enabling various kernel options, especially in microbenchmarks. Rusty had been attempting to demonstrate (to Ingo Molnar, as it happens) that the oft-quoted 1% overhead of enabling paravirt ops support was insignificant in comparison with many other options one might enable, take a performance hit in so doing, and not make use of. Linus suggested that Rusty should turn off some of the HIGHMEM config options on his x86 hardware since these were unfairly affecting the benchmarks. Rusty followed up with a simple utility "so everyone can play along at home" that reproduced the figures he was seeing - up to a 48% overhead in his latest tests on a different, laptop, system (Rusty is apparently current traveling, and therefore does not have access to the system he saw 69% overheads on).
Super-H. Paul Mundt posted a summary of "what's in sh-2.6.git for 2.6.31?". Lots of goodies here include support for generic time, new CPU and board support, earlytimer support, power management improvements, sparseirq support, and lots more.
System s390. Somewhat scarily, it looks like powermanagement and hibernate support are coming to a System 390 near you. Martin Schwidefsky posteded a 38-part patch series adding power management and hibernate support that can be used in the usual fashion (by writing to /sys/power/state, for example). This will be of particular interest to anyone who happens to have a mainframe laptop lying around, in addition to the other virtual machine uses.
Console screen blanking. Daniel Mack requested a kernel boot parameter be added through which one could override the blanking timeout on the console, since he has some development systems that might take 10 minutes to boot into the initscripts under certain circumstances, and no way to easily set a console ioctl in time to get at the useful output. Andrew Morton's reply was to feel sympathy, and ask for a more detailed description in a resend. The boot option is named consoleblank=seconds.
Kernels listed on kernel.org. Jean Delvare requested that the kernel.org front page be updated to remove references to the -mm kernel and replace it with something like linux-next.
In today's announcements: Git version 18.104.22.168. Junio C Hamano announced version 22.214.171.124 of the Git SCM (Software Configuration Management) tool originally created by Linus Torvalds and used by the vast majority of Linux kernel developers (with the notable exception of Andrew Morton) to maintain their kernel trees. Since the previous release, a number of bugs have been fixed, including allowing a "git add ." in an empty directory to silently succeed, and fixes for "git add -p" and "git blame path".
The latest kernel release is: 2.6.30-rc8, which was released by Linus on Tuesday afternoon. A handful of fixes have appeared in Linus' tree since.
Andrew Morton uploaded another mm-of-the-moment for 2009-06-03-16-33.
Stephen Rothwell posted a linux-next tree for June 4th. Following a request, I shall no longer break this out in great detail (please let me know which you prefer), but notably, the powerpc tree continues to fail in an allyesconfig, and 8 other tree conflicts remain.
Finally today, your author has ordered a Samson C01U USB microphone for this podcast, which is apparently already supported by ALSA. Hopefully, some time next week we'll all be happier with the sound quality than I'm getting from the existing lower-end equipment I had here. USB seemed like an easier option than also buying a 48V DC Phantom power supply, especially when travelling.
That's a summary of today's LKML traffic. For further information visit kernel.o rg. I'm Jon Masters.