The tpm(4) driver has been added by Sascha Wildner, ported from FreeBSD.  What’s it do?

From the man page: “The tpm driver provides support for various trusted platform modules (TPM) that can store cryptographic keys.” Crypto keys stored in hardware, where they are in theory unmangleable, instead of on the disk. At least, that’s my impression after 30 seconds of research.

Sepherosa Ziehau has posted some numbers showing improvements in ip forwarding rates.  He’s done this before, except this time it’s with bnx(4), probably because of his recent commits.

If you have a sili(4) device, Francois Tigeot needs you to run a particular patch and tell him what happens.  He’s testing a larger I/O request size, and wants to see how it will work out “in the field”.

John Marino brought up a point every operating system project will have to think about: when does support for i386 (i.e. 32-bit x86 processors) stop?  Follow the thread for details.  There’s no final answer, yet.

Matthew Dillon wrote a note about SSDs, HDDs, and swapcache that may be useful for anyone building a system soon.   Conversations about SSDs, swapcache, and so on have happened before.

The upcoming DragonFly 3.4 release will not include the USB4BSD port from Markus Pfeiffer; he’s hoping for it to become default in the next release after 3.4.

You can still try it, as it’s present in DragonFly but not on by default.  Help with driver porting is always welcome, of course.

MSI (Message Signaled Interrupts) has been enabled by default on the re(4), msk(4), and et(4) networking chipsets, by Sepherosa Ziehau.

It seems Sepherosa Ziehau won’t rest until he’s reached peak performance for every network card in DragonFly; he’s added multiple ring/MSI-X support for Broadcom 5709/5716 chipsets in DragonFly.  In more concrete terms, this means better speeds when transmitting and receiving multiple streams of data.

(at least, I think so.)

Following this recent thread, it looks like the best answer for software RAID is: buy hardware.  I’d be interested to hear what people have experience with in the realm of cheap but OK RAID hardware.

Sepherosa Ziehau has posted more statistics on his ifnet/ifaddr per-CPU stats work.  It’s doing so well that he’s very close to reaching the maximum physical capacity of the 4x gigabit ethernet hardware he’s using.

Sepherosa Ziehau has merged the hardware abstraction layer (HAL) for em(4) and igb(4), along with updating em(4)/emx(4) to version 7.3.4 and igb(4) to version 2.3.7.

Sepherosa Ziehau has posted a detailed message showing the speeds he gets with multiple transmission queues, using igb(4).  The short version:

Quick summary, the multiple TX queue support gives me:
+200Kpps for 2 bidirectional normal IP forwarding (now 4.40Mpps)
+160Kpps for 2 bidirectional fast IP forwarding (now 5.23Mpps)

Dave Hayes asked for some “best practices” ideas for setting up a HAMMER (1) system.  I replied, and the conversation turned to RAID, as these often do.  If you’re thinking of purchasing disk hardware in the near future, this will be useful to you.

Adam Sakareassen submitted a patch for AVX support for 64-bit DragonFly, and Alex Hornung has committed it.  If you’re like me and have only the vaguest idea what AVX is, it’s a set of processor instructions added by Intel to Sandy Bridge and later CPUs.

Sepherosa Ziehau has been making a lot of commits to increase packet-per-second rates without increasing CPU usage.  He’s published a sort of progress report/benchmark to show current performance levels.  It sounds like he’s expecting even better performance in the future, though I’m not sure how much more he could push out of it, since the bulk performance appears to be close to the rated capacity of the copper…

The stl(4), bt(4), aic(4), and cy(4) drivers are now PCI-only, which means no COMPAT_OLDISA kernel option, and a time to upgrade your hardware if you’re actually using these devices.  Does anyone even still have ISA slots?

Sepherosa Ziehau is switching a number of network cards over to use ifpoll, which means they will have capabilities similar to MSI-X, even if the network card doesn’t support it.  My suspicion is that it will make these cards perform better in busy situation where they would otherwise get bogged down… but that’s based on hunch rather than empirical testing.  As Sepherosa Ziehau pointed out, it certainly can’t hurt.

Smartmontools will catch impending disk failures about 2/3 of the time, so it’s useful to run it and interpret the results.  The results can be somewhat complex, though.  However, it can be useful to look at other people talking about the output and glean knowledge from the context.

If you are using an Intel 10G Ethernet card with a 82598GB chipset, you’re using ixgbe(4).  You may want to set the net.inet.tcp.sosend_agglim sysctl to a value over 12 in certain circumstances, as described by Francois Tigeot.

Much of this new document has been around in other forms for a while, but now, there’s a brief guide on porting drivers to DragonFly in the source tree.

Because here’s some recommendations on good models, and here’s a way to check SSD health.  Seriously, they’re great.

If you have a LSI RAID card, meaning you are using the mfi(4) driver, Sascha Wildner has added  /proc/devices to linprocfs, so that LSI’s MegaCLI configuration utility will run.

I noticed that this recent commit from Sepherosa Ziehau is a bug fix for jme(4).  The commit thanks a JMicron employee for help.  It’s always appreciated when a vendor is helpful to an open-source project for hardware support.  It’s also something you should consider the next time you are shopping for computer parts.

Sepherosa Ziehau has added MSI support and cleaned up the alc(4) driver.  If you’re using a network card with the Atheros AR8131 or AR8132 chipsets, you should see an improvement.

Pierre Abbat is curious about using Hammer on an SSD.  The discussion that came from that has some useful points, including notes that a straightforward SSD as disk works for most anything with Hammer other than very intensive database use, due to the history retention.  If space is an issue, swapcache on the SSD and attaching a normal HDD is a fine alternative.  A SSD with Hammer can leave some features off, though I’d argue that dedup is totally worth is.  Also, SSD speed is directly correlated with size.

Sepherosa Ziehau’s added TSO support (that’s TCP Segmentation Offloading”, or “Large Segment Offload” going by Wikipedia) within IPv4 on DragonFly, pushing segmentation work from the CPU to the network card.  There’s also some DragonFly-specific improvements.

There’s been a lot of commits from him lately focused around network card improvements; they haven’t been easily summarizable, but it’s worth watching if you are interested in high-bandwidth usage and the hardware to support it.

Not all flavors of Atom CPU support frequency scaling, as Sven Gaerner found out.  This means more heat and more power usage.  There’s further details scattered through the thread, but Sascha Wildner found what seems to be the definitive answer of which variants do and do not.

Thanks to David Christensen of Broadcom, Sepherosa Ziehau was able to add BCM5718 and BCM57785 support in the new bnx(4) driver.

Sepherosa Ziehau has added support for a variety of bge(4) chipsets.

From a thread on users@, I bring you Visible Capacitor Failures.  If the problems pictured are new to you…  trust me, you will see them up close someday.

Someone trying DragonFly couldn’t get it to start, and appeared to have a confused disk.  It looks like the system BIOS were at fault, and Matt Dillon has an explanation of this minefield.  (Including some comments on 4k physical disk sectors.)

Sepherosa Ziehau has added MSI-X support to igb(4), the Intel PRO/1000 gigabit network card.  What does that mean?  The commit message mentions a default transmit rate of 1.48Mpps small packets, which is good?

Francois Tigeot has added the Intel PRO/10GbE driver from FreeBSD, or ixgbe(4).  A couple features are turned off, for now.

If you have a Broadcom BCM570x-series gigabit ethernet adapter, Sepherosa Ziehau’s made a lot of commits for the bge(4) driver recently; they may interest you.  (not sure if he’s even done yet; he tends to commit a lot of work.)

Sepherosa Ziehau has added igb(4) version 2.2.3 direct from Intel, for support of  their 82575 and 82576 Ethernet controller chips.  It now shares a hardware abstraction layer with the em(4) driver, too.

If you are having USB issues on boot with DragonFly, Sepherosa Ziehau’s sysctl suggestions may help you.

Venkatesh Srinivas posted an explanation of the virtio update he’s working on.  I linked to the work before, but not his explanation, which goes into the ‘vm_balloon’ device.

Sascha Wildner’s posted an update to the acpi_asus(4) module, so it’s worth updating if you have an appropriate Asus machine and are running DragonFly-current.

Venkatesh Srinivas has been working on integration of Tim Bisson’s virtio-bhyve drivers into DragonFly.  This would make throughput better in KVM/Qemu.  His bug ticket has some questions that could use answers.

Francois Tigeot has added ichwd(4), a driver for the watchdog function on some Intel ICH motherboard chipsets.  Sascha Wildner has also made the kernel option for it on by default.  (Look for /dev/wdog.)

Update: Francois Tigeot sent a link to an excellent page explaining hardware watchdogs.

Sepherosa Ziehau has updated the em(4) driver from Intel; it only matters if you are using the specific chipsets mentioned in the commit message.

Thanks to Sascha Wildner, the Areca RAID controller driver, arcmsr(4), now supports MSI.   It should only make things better, but if it doesn’t, you can turn it off.

Sascha Wildner has updated mfi(4), the LSI MegaRAID SAS driver , via FreeBSD and LSI.  SAS2208-based controllers are now supported.

Carsten Mattner wrote out his notes on EFI booting on a Mac.  This gets you closer to booting DragonFly on there, but I don’t think it is completely working yet.

Update: Carsten Mattner has a better summation than what I wrote.

If you do, acpi_hp could use some testing.  Sascha Wildner just brought some improvements in for that module.   I’ve seen discounted HP laptops show up in various places, recently.

Notice how the 2.12 release never really happened, and 3.0 came out about 6 months later than usual?  A lot of that delay was caused by a vigorous search for a weird bug.  Multi-threaded buildworlds would crash, seemingly randomly and rarely.  It turns out we have confirmation from AMD that it is, indeed, a CPU hardware bug.

Is it possible to boot with only 48M of RAM in a DragonFly system?  Probably not.  128M would be better.  I usually talk about the lower memory limit for Hammer, since it’s so relatively low for a snapshotting file system, but the converse applies here.  128M is probably the comfortable lower limit, though it’s pretty hard to find a system that would limit you that way without doing it on purpose.  128M sticks of RAM are practically disposable these days, really.

Sascha Wildner is looking for the donation of a Intel Raid Controller RS25DB080.  If you were able to give him access to one, or even purchase it (ow my wallet), that would greatly assist development on DragonFly.

I need to catch up on some older stuff, so here is a longer list of recent updates: libarchive to 3.0.2, xz to 5.0.3, mfi(4) and mfiutil(8) (LSI MegaRAID driver) updated, ATI SB7x0/SB8x0/SB9x0 AHCI devices (on motherboards I assume) updated, and the PHY ID for the Atheros F1 added.  Thanks to everyone who did the work!  I bet I missed something.

That’s Managed System Interrupts, for when your hardware is passing a lot of data and generating a lot of corresponding hardware interrupts.  MSI is what deals with all that traffic.  High-bandwidth (10G) network cards, for instance.  Anyway, Sepherosa Ziehau’s made more commits than what I’m linking to here, for support with various devices.

There’s many other MSIs out there, oddly enough.

Sepherosa Ziehau has added updated the ‘ecc’ device, for Intel E3-1200 series systems.  What’s it do?  It will report on memory errors, and potentially fix them.

You should have ECC memory in your server already.  If not, you oughta.

Update: as Sascha Wildner pointed out, ecc(4) already existed, but didn’t support Intel controllers.  Also, the Xeon X3400 series is supported now too.

I think a lot of people don’t realize rcrun exists, and run files in /etc/rc.d directly.  If you’re one of those people, read the rcrun(8) manpage.  Then, notice that Sascha Wildner has extended rcrun to work with the etc/rc.d in /usr/local and /usr/pkg too.

Matthew Dillon has written a contiguous memory mapper, which is designed to fix problems with video cards and USB drives that need a big chunk of memory to keep.  This can affect booting or later on, when disconnecting/reconnecting a USB drive.  If this still doesn’t fix the problem for you, try adjusting the sysctl ‘vm.dma_reserved’ to something bigger, like 64M.  It defaults to 16M.

(Normal mailarchive isn’t updating because of an ongoing upgrade to crater.dragonflybsd.org – sorry!)

John Marino added tuning support within GCC 4.4 for the Geode CPU.   Waaaay back when, these were x86 -compatible Cyrix chips.  Nowadays I think they are most common in single-board computers.