 Check, check. Yeah, raising it. Close mine. I need to clean my nose before I actually can start. Because like every year, I got a cold. I guess that's winter or something. So 10, 29, and 40 seconds. So I'm starting right on time. Welcome, everybody. I'm Thorsten Lindhuis, who am I in case you don't know me. I'm from Germany. I'm actually working for a publishing company. It's called Highsea. We are doing still a paper magazine. It's called CT. And I'm having a column there that's called Colonel Locke. And it regularly talks about what's happening in the Linux kernel. The text does also go online. Obviously, they are German until maybe some of you still remember until a few years ago. They got published on DH in English. But that closed down in 2013. I was actually also a longtime Fedora contributor. Or to be more precise, I actually contributed to Fedora a lot 10 years ago. These days, I'm doing not that much anymore. Actually, the most important thing, I guess, or what most people know me these days for, is a package prep repository where I'm doing kernel vanilla builds for Fedora. So you can grab RPMs there with the latest mainline kernel or the latest stable kernel without additional patches. And you can try to check if a new kernel makes something better or if some problem you see is actually from the patches that are part of the Fedora kernel. What I'm not is a Linux kernel developer. But thanks to my work is actually I'm having a really good overview of what's happening. Actually, sometimes even better than a lot of kernel developers because I look everywhere a little bit and don't have to focus on one area. What I actually did for the Linux kernel is working as a regression tracker a little while ago to check to actually compile a list of what regressions or bugs there were to help getting them fixed in time before a new release. What also helps to get an overview of what's happening is LWN.NET because it's a great resource and they're writing really, really good articles about what's happening in the kernel. So what does this talk about? Actually, there are five parts. There's a short status overview. Then we get a bit into recent developments and we're looking at the last kernel, the next kernel and then we go over to important changes, looking back a bit to the last 12 or 18 months what important things changed there. Then I'm trying to actually look ahead a bit and then I'm taking a different view at the kernel like what's happening in the development is everything working fine in the last part in the meta view. By the way, the slides will continue to change this frequently. So better write questions down. If you have any, then we can go to them at the end of the session if there's time. I have a lot of slides, but I'll try to have some time for questions. So the first part, current status, what's it? Everything works as usual. Current kernel version is Linux 4.9, was released in mid December and actually got submitted as regular update to Fedora 25 and 24. So if you're using those versions, you're running a quite up-to-date kernel with the latest features, some of which I'll mention in a few minutes. One of those things this kernel has is brought support for 500 more devices or device classes, 150 to 200 of them are ACPI, USB or PCIe devices. So this kernel, as every new kernel, really made the hardware support better and make sure it's Linux and Fedora runs well on the latest hardware, especially I have quite new laptop. That's really important often to have latest Linux kernel because all the drivers are in there and these new drivers make those new laptops often run way better. What else is happening currently in the kernel? 4.9 made the system and the Linux kernel more secure. That's something that's trend recently releases that there are lots of things happening that lots of small improvements to make sure the system is more reliable and that attackers cannot get into your system so easily. There's really, really a lot happening there to make everything more secure. We get two details in that in a minute. So, current status 4.9 as I said was released in mid-December, that means we are now heading for 4.10. That's developed right now and actually likely to come out on February the 20th. Maybe one week less, maybe one week more. Remains to be seen, that often depends on if there are bugs and sometimes it depends on if Linux is traveling or things like that because every new version is released around about every 10 weeks. As I said, sometimes on nine, sometimes 11 and that's actually quite predictable. You can actually bet on it if you want. Sometimes you lose if Linux is on holiday for two weeks or something, then sometimes it might be 12 weeks or if there's a big bug somewhere in the kernel found, then it can take a little bit longer to get a new kernel release finished but normally it's really nine or 10 weeks most of the time. All those new versions, as I said, there are new drivers in it that makes the hardware support better. That means changes and those actually are quite a bunch of changes in those nine or 10 weeks that are 13,000 commits around about every new version from around about 1,500 developers and those add about 700,000 lines of code every new version. Also a bit gets obviously thrown out or moved around but in the end it means the kernel grows by nearly 300,000 lines of code every release and you can see it here, the kernel gets bigger and bigger that's from the Git area, two, six, 13 on the left. Back then we were about seven and a half million lines of code around about and now we are about getting to 22.5 million lines of code including documentation, comments and blank lines so you can really see the kernel gets bigger and bigger. Actually all those development is happening quite fast. Take for example the development between 4.8 and the first pre-release from 4.9 that happened, there were 13 days and there were a bit more than 15,000 changes with close to 700,000 insertion of new lines and 430,000 deletions. That means every 35 seconds there was one change. That's really quick, that's really happening quite a lot of things in the kernel to make sure everything works better and the kernel gets improved and a lot of people say hey that's crazy, how is that working and actually that's not a problem at all because those changes are prepared before a new release and then are simply smashed together. Obviously they are in the last minute sometimes things that need to get adjusted to each other but really the speed and all those growth is not a problem, nothing to worry about. In case you're interested in things like that every year or year and a half the Linux Foundation together with a few people releases a small report, the Linux kernel development, how fast is it going, who's doing it and things like that. They have many more details on things like that. If you want to know them, grab it, download, you have to reduce it for it but it's really insightful how kernel development works. Here for example is an analyst from this document that shows how many people are contributing to new kernel versions and how that changed over time. If you're interested in this, there's also LWNnet articles on every new kernel release which actually looks which people did changes and how many were there and actually there are also employer statistics in case anyone wonders how big or not big redhead is. Until a few years ago it used to be on the top in both columns by change number of changes. These days redhead slowly gets down. Here for example in the left column it's on the second place and on the right column it's at the fourth. It's not that redhead isn't doing so much anymore, it's more that other companies are doing more for example Samsung for embedded things or media and Intel to make sure their hardware works well. So that's how it comes that redhead got a little bit down. That was part one already. I need to drink often today, otherwise my voice might simply break and I can't say anything anymore. Next up, recent developments, part two. So as I said, Linux 4.9 is recent version. Three of the most important features of this version brought, XFS, reflink and deduplication, BPR improvements for tracing and bottleneck bandwidth and RTT. If you wonder what this is, I'm explaining those a bit more in detail. The XFS enhancements, the reflink support for example allows light ways copies using copy and write. That's for example if you copy a file within a file system, you simply say do it with CPU reflink and then a new metadata entry is generated but the payload from the files isn't actually duplicated and that makes it possible to copy really big files within seconds or within parts of a second. And if one of those files is changed, then thanks to copy and write, the other isn't changed, so there are really independent files from each other and that's for example can be of interest for to duplicate a VM because you can copy a VM image over only takes a second and start using it and you don't have to wait like 10 minutes or something like that to get the copy first. Deduplication is another feature that's allowed by this copy and writes support. It's like the copying, it's kind of reversed like the copying, you can scan your file system manually with a tool and check if there are some errors where the same data is stored and then the metadata can be adjusted to make sure that they both point to just one of those data blocks and the others gets deleted so you can save space like that. It's also interesting, especially for big setups, if you have lots of VMs with similar data, you can reduce your storage load and improve things because less data has to be cached. Downside of this, both these features are not yet usable. They need XFS Prox version that's still under development. Remains to be seen how long that takes until this is getting finished. And another detail around that, there are more XFS improvements to be planned. Here's actually a list like there's scrubbing support and sector data loss reporting and other things. And if some of you are interested in file system and look a little bit closer at this list, you'll see XFS is simply gaining a few features, but of S and ZFS are famous for. So for a long time looked like it, that but of S and ZFS are the features or especially but of S in the Linux space is something that will be used as next generation file system, but it looks like XFS might be changed and so far that it's really in a better position to compete with but of S. And maybe in the end it will be XFS that we're using as our next gen file system remains to be seen. These big changes actually mainly developed by Oracle developer, Red Hat developer there was also involved, but it's interesting that Oracle is driving mainly these XFS enhancements at the moment. I need to speed up a little bit. What else? I said something about BPF improvements for tracing. Tracing is like looking at the program or that's running or the kernel while the program is running and check what the kernel is doing to see where the performance gap is or where it would be important to optimize the program to speed things up. And what's now possible with the Berkeley package filter that's a BPF acronym I just used is time sampling and summarizing stack traces. That's actually a pretty complicated topic and I won't go into that too much into detail because otherwise we just need to focus on that. The thing is it makes tracing way easier because in the past you had to do the tracing, analyzing and then put it up to some output that's easily consumable by users and these days you can do it with Linux 4.9 and easier thanks to the new features with a way more simpler script and put it for example to a flame graph and a flame graph is then something like this where you see okay here this is how the program work for example and how long each function was used and that's for example then sometimes easier to see where the time went and actually as I said to optimize the program. In case you wonder yes that's finally Linux gets to a point where we have features like Dtrace that is something Solaris was famous for that makes analyzing performance optimization way more easier and in case you want to know more about this there is Brandon Gregg who writes really good blog articles and develops a set of tools to use these features in the recent kernels to do actually performance monitoring or tracing and there you get way more details and actually he gives talks as well so really if you are interested in this go to Google Dtrace and Google for Dtrace for Linux 2016 then you get to the blog post I just showed you and there's really some interesting insights how to do tracing these days and there are also great talks from Brandon on YouTube where he gives a great overview of how to actually do that. What else was in 4.9? There was a bottleneck bandwagon RTT that's actually a new TCP congressional control algorithm. The kernel already has I think 15 or 20 of those. So yes even one more but it works a bit differently like the other's congressional control in case you don't know what it is actually makes sure the bandwagon is used as good as possible to make sure your data is getting from the server to you as quickly as possible and to reduce congression where they are fighting I can I go to the wire or do you have to? As I said it works differently than most others it was developed by Google mainly. Two of the main features are it improves resilience against random loss if they are on the wire for some reason sometimes packages don't make it through then the speed will won't decrease that much and it avoids some problems that happen due to some bloated buffers that are oftenly common on the last mile like from your house to the next station where the internet gets together by your provider. It can make things really a lot better I guess we actually might use this in Android because it was developed by Google I guess they want to use it there but they haven't said anything. Remains to be seen if it gains traction and how much it will be used. What else? I can talk about 4.9 for hours. There were some for example some security improvements like virtually met kernel stacks as the second one here that makes sure that stack overflows in the kernel are detected and cannot not be used by attackers or the last two features here are something in that are really important for SE Linux and on using for containers because overlay FSS are actually used there quite a lot and the ACL support as a test control list for views is also something for Gluster and SEV because they use views to get access to it. You can use views to get access to those things you start on Gluster and SEV volumes and that makes sure only the right people get proper access there. If you are interested in more details you can read them here in the German version if you want to. I know some people that know my work actually run my German articles through Google Translate and they say it works quite well sometimes but if you want a straight English go to lwn.net and they have an area where they cover their features from 4.9. So that was 4.9, 4.10 is the next one as I said it's due to in three weeks. I also looked again for three important features. One is GPU passthrough and the other are storage related that's right back startling right back cash for rate five. I'll explain those more in detail now in GPU passthrough I guess most of you will know it's simply putting your GPU from the host to the VM or maybe even a container to actually run games if you're on a Windows VM that's running on Linux. That's actually quite hard to do until now but there are really some improvements to make things work better. And this GPU passthrough is done with two different techniques. That's also the first one is mediated device MNF and the other is Intel GBT and KVM GT. MNF is something like where your host kernel can grab parts of a hardware or complete PCI chip and put it securely to a VM. The important part is it can do that with parts of the chip for example with the GPU so your graphics core graphics processor in your device. There you take just for the Intel case just parts of the Intel GPU and put them through the VM to make sure the VM can't use this MNF to get control over the host. And to use MNF you need support for the driver and that's where KVM GT comes in that's actually support in the driver to hand parts of the integrated graphics from your inter processor to a VM that actually still doesn't work that well yet. Pretty rough and work in progress. There's a block entry from Gert Hoffmann that explains the current state. He actually showed a window yet where he got kernel output but it was a bit messy in some places so it will take a little while before you can start using it but maybe in the end it will be possible to run Windows games in a VM that's running on a Linux host because AMD and Nvidia plan something similar they plan to use this as well. So maybe for the gamers that want to use Windows it's you don't have to do a boot anymore you can maybe just open your Linux KVM VM and simply use your graphics card there. What else? Right back throttling that's a bit complicated topic the short version is make background right back less annoying and now you might ask what is background right back, what do I care? I guess a lot of you experience this you try to DD and ease the image to a slow USB key and for example in Fedora image to install Fedora and while this DD was running and transferring the image to the USB key your system became really really slow. That was because the Linux kernel had was writing this data out to the USB key with as you can say maybe with a higher priority and reads from the disk to for example the Firefox user there does when you use it were actually putting a hold and that's why your system suddenly felt so slow and that's something the kernel developers finally took a closer look at and hopefully fix this they actually try to fix this a few times already maybe they did this for real it's also important for your databases or something on your servers and because something like that obviously can lead to long latencies and it's what you don't want because your database doesn't reply to connections in time. So that's hopefully going to get better there. What else for the sys admins here maybe of interest might be the writeback cache for write rate five that's actually where you have a rate five array and put an additional disk to that rate five array duplicate SSD to improve the data until now that was actually this is called a lock device and support for this was added in 4.4 until now that was actually only used to make your system more secure to make sure all the data made it to disk even if the system crashes at some time that's avoided for example the right hole but now it's actually can be used as cache and that makes sure that more write cycle accumulated and then written in one go and that reduces the time to read, modify write cycles and in the end it means better performance for rate five and as I said you don't get it for free you need a caching device attached to it. What else for the 10 just like for nine brings a lot of features for example the Neville driver for NVIDIA cards open source driver got a little bit better for example it's supposed to get 4k displays running probably now there's turbo boost max 3.0 support to make sure your Intel CPU gets as quick as possible to the quickest frequency level and actually a new technique to talk to your CPU how it's using the caches the first and second level and third level caches actually that can be important in high level setups to make sure the performance is better but as I said that's pretty specific if you want to read more about it LWN has articles and I'm writing on it on my German article as well but I haven't written about that yet because 4.0, 4.10 is not out yet. Yeah, that was part two. I guess I need something for my throat, wait a moment. Some magic. So important changes last 12 to 18 months as I said earlier already self-protection kernel hardering to improve the security because kernel developers and outsiders notice more security bugs just happen and some analysts showed it takes often about five years until they are found so the kernel developers came to the conclusion they need to protect the kernel a bit better against common flaws. Truly, sorry. And that's why they are integrating these hardering techniques, security techniques just like cars are prepared for potential crash release days that's pretty normal and it wasn't like that in the 50s or 60s. This work is mainly driven by the kernel self-protection project that was actually founded like one and a half year ago I think and they improved and have merged really a big, big bunch of security and hardening techniques in the past year. Here's a list, I won't go into detail. As you can see, there are six or seven things and that was just what was in 4.8 to make the security better. So really that's happening quite a lot. There were similar new features in all recent kernels and the 4.10 as well and yeah, there's more to come. And if you're wondering, hey, is the security making my system slower and to be honest, yes, that sometimes can happen that the system gets a bit slower due to the security measurements but normally the kernel developers make sure the performance impact is not too big. It's a bit like with cars and those are, get a bit heavier due to protection techniques as well but we accept that today, that's pretty normal and because if there's a crash we, our lives are more important than a little bit speed or something on the road. If you want to get details about this, there are slides from the security summits, there's here for example on Keith Cooke's something that the state of the kernel self-protection project, a talk that was given last fall. If you want overview why this protection is needed and what kernel developers are doing, go and watch that video, it's like 45 minutes well spent. What else, the graphics drivers in the kernel improved quite a lot. Actually in case you don't know, the most important part of your graphics drivers stack in your Linux system actually is a kernel driver and the kernel driver is really important to get the best out of here, out of your system and the user land drivers that are working together with the kernel drivers also improved quite a lot. That's worth it's own talk. In case you wonder, you get only get a quick mention here now and as I said some of the things I mentioned are in the user land parts as well. And we have three major vendors for graphics processors on the desktop, that's for example Intel and the most important thing that's happening there right now is we finally got to open GL 4.5 support and measure 13.0 that was like three months ago. So most of the up to date games should simply work these days. And one more word on Intel graphics, they have a good reputation in Linux but sometimes they are not that good as their reputation Skylake supports. For example, the Core e6000 series as the support in the beginning wasn't that good but it works quite well most of the time. Where really a lot happened in the past two or three years is AMD Radeon drivers, the open source drivers improved really, really much in that timeframe. And you have to really say thanks AMD for that. Most things simply just work these days and it includes really acceleration, high DPI things and other things and the 3D performance is also quite good. It's not perfect but it actually got closer to AMD proprietary drivers and actually sometimes it's better. And Mesa 17.0 which is due to in the next few weeks will actually bring official open GL 4.5 support and that's important to get the latest games running. So it will make things a lot better for drivers. There for AMD the reputation for the drivers not that good, they are actually better but not everything is perfect. For example, the latest cards you can't do audio via DisplayPort or HDMI and things like that. There's, as I said, also a bit complicated topic and worth its own talk. Next up, Nvidia is the third major vendor. As you might know, the free and open source drivers are quite slow and lacking features but they are improved all the time as well and they are sometimes good enough these days for things as well. Actually, Nvidia since two years now participates a little in the development but it didn't make things much better. A big problem there is that 3D acceleration requires assigned firmware these days for the 900 series. It took more than a year for the 900 series to Nvidia to release this firmware. Support was merged in 4.6 and it's finally usable these days. For the latest series like 1070 and 1080 support was merged in 4.8 but Nvidia still didn't release the firmware. That's really annoying. Yeah. So, what else? ButtaFs. There were lots of fixes and few small improvements in the past 12, 18 months but no big changes recently. So as I said earlier already, ButtaFs was supposed to be the next generation file system for Linux which everybody would be using. That was actually how it was told or how it made the media when ButtaFs was introduced in 2008, 2009. And you might wonder is it finally stable? It's also a complicated question, not easy to answer in case you wonder it's a bit like, can I swim here? The answer depends on the local conditions and your swimming abilities. Yeah, I actually have three pictures. I mean, most of us can swim here and even kids would feel safe to get your kids swimming here. But I mean, you can drown here. That's like every file system can eat your data if you're doing something stupid. And, but for comparison, that's a bit like X4s these days. It simply works. ButtaFs, I put this picture. I mean, it looks a little bit more dangerous but I go swim there. I'd be careful with the stones there but I guess you can swim there if you know what you're doing but you wouldn't do let your kids swim here without overseeing them. The problem with ButtaFs is there are sometimes areas where they're a bit more dangerous because some parts of ButtaFs are still working progress and unstable. And the most important thing is rate five and six apart which is included in ButtaFs directly. It's known to be unfinished and that's not the only feature which is unfinished. Luckily, we now have a table in the ButtaFs wiki that explains which features are considered stable and which are still unfinished and rate five and six is one of them. In the end, The message I want to give you is ButtaFs requires some time to learn due to things like that. And that's only worth it if you benefit from ButtaFs features. So if you're interested to look at it if there's something like the snapshot support like open Susie uses and relies on that's worth going for ButtaFs then do it and expect some time to learn and then it is worth the trouble but not simply use X4 or XFS if you want to. What else? The Raspberry Pi support got better. You can boot the recent kernels directly on the Raspid for example and there are actually graphics drivers for the Raspberry Pi now open source in the kernel and in Mesa and that makes it possible to run Fedora 25 simply on the Raspid now because you don't need anything that Fedora can't ship. So it's way important to make Fedora run way better on the Raspid. What else? There's a lot of other things I could talk about it. I'm running out of time because I thought I had till 20 and let's make five minutes for Q&A, yeah? Okay. Then that was part three. Then let's go to part four that's also a bit quicker that's looking ahead a bit what's coming might be coming in the next few months or what's happening in case you haven't heard about it not KDE bus. It was an IPC inter-process communication thing developed for the kernel. Actually it's the developers abandoned this and they actually started on something else. It's called bus one. Actually LWN that has an article about it and there was some discussions on the kernel summit and it's going forward and looking quite good but it's still working for progress and not yet submitted for official proper review on the mailing list. So it will take some time until it's merged in the kernel. And if you wonder what is different in bus one to KDE bus what's design is a bit more inspired by Android Binder and other IPC techniques and it's a bit more universal than KDE bus. It can be used to transport D bus messages but it can also be used for a whole lot of other things. What else, as a lot could be talked about what the kernel developers are also doing is kind of live patching improvements. You might have heard of it. You can now fix security problems in your kernel while it's running. It only works for 90% of the typical bugs right now. There's some really big changes were prepared and parts of them are already in the kernel to make it 95 or even a bit more percent of the fixes be able to fix them while the system is running. And these improvements are forthcoming but it also remains to be seen if they are getting merged next month or maybe in half a year. The kernel is also made secure for year 2038. Some people might know the year 2000 bug. Something similar will happen in the year 2038 for unit systems, only on 32 bit systems only. But the kernel developers make sure this is addressed and to change a few things because today's hardware like in your car there are ARM chips 32 bits that will be used in 2038 because some cars are used for 20 years or something. Support for new hard disk is made. Some kernel developers are looking at making the kernel a bit smaller to make it able to run better on small systems, really small embedded systems. And there are also some improvements to get more out of your CPUs like for example ARM cores that have eight socks that have eight cores for slower ones and for quicker ones that the kernel is better able to spread the load depending on your current condition to the slow or the quicker cores. But it's also a lot of changes and work and it's processing quite slowly. That was part four. Now we're getting to the meta view and look at the kernel development a little bit itself. In case you're compiling your own kernels and not rely on REL, Fedora or your distribution in general there's a new long-term kernel coming now every January that makes it more predictable. Long-term kernel means it's supported for two years, makes it more predictable for hardware renders or distributors to see when will there be a kernel that gets support for not only three months but support for two years or even more. The next one that kernel that gets long-term support is 4.9, which I mentioned earlier already. It was recently announced. What else? I mentioned security a few times already. As I said, there are two or three security issues every week in the kernel, so these days it's pretty much, you have to prepare for this, especially because every year there are really a few critical bugs in the kernel that you can use to take over your system so really be better prepared and don't say, hey, I don't need to update my kernels. That's not an important because sooner or later attackers will get you, maybe not on your desktop system but on your service, it's really important to make sure all the security bugs in the kernel gets fixed. Big distros know that and they take care of it. Some of the hardware vendors, especially in the embedded space, don't understand that that well and that's slowly becoming a real problem with the Internet of Things. There was this distributed denial of service attacks with video recorders or toys that were made together to a big botnet that really got to, was really annoying for some people that were attacked by this botnet and as I said, it's video recorders and some other IoT hardware. Hopefully that gets better. GPL enforcement, that's something the kernel developers talked about in the past half year, a little bit more. It was a long debate. Some kernel developers didn't like it that much. It was triggered by Christoph Helwig getting to court against VMware. Remains to be seen if what's happening there maybe is think it looks a little bit like things continue as they are. GPL is there and people honor it but I wouldn't expect a lot of things going to court. What else the kernel developers noticed that the stable and long-term kernels are not working as good as expected. Like there are sometimes too many regressions and the embedded developers or other companies don't use this kernel because they don't know about them and another example are Android phones running quite old kernels. This is a mess and it has been like that quite a while. There's no magic vent, no magic thing that can see things go away but the kernel developers are talking a bit more to the hardware guys to improve matters a little bit. What else testing and regression tracking is a problem as I said, I did some of that like compiling lists of bugs and things and I think that's important because bug reporting for the kernel is hard. You have to basically know where to go to because in bugzilla.kernel.org doesn't work well and on mailing lists there's sometimes not easy especially if you're not used to mailing lists and this is some, for example, an area where lots of room for improvement is there so if you always wanted to do something with the kernel or get involved a bit there is some area where you can get involved even if you're not a great programmer because there are simply people, there's simply help needed by people that know how to talk to others and help inexperienced users to get their bugs fixed. What else Linux competitors are emerging for IoT? There was Sapphire and Magenta that were a bit smaller than Linux. Maybe Linux is simply too heavy for IoT devices. Maybe the tinefication supports that I mentioned might make things better but maybe the hardware gets quicker so Linux is more suitable for IoT devices and there's always complaints in the media about issues with kernel developments like kernel developers getting too old or the tone is too unfriendly and things like that I'm looking at the kernel for 10 or 15 years now it's really complicated and complex topic and a lot of things are really not that bad as the media makes it sound and a lot of things are way better than five or 10 years ago but as always there are something that could be better so that was part five and I fear I'm getting thrown from stage in two or three minutes but we finally get to the end so one more details on any of this use the web there's really a lot of sources on any thing I mentioned you can find more things or maybe ask me three more things quickly as I said already testing is something I'm interested in and I want you to consider helping testing the kernel especially if you use not that widespread hardware or maybe quite old hardware or use your distribution in an unusual way because it's simply like that if you don't test it nobody might and the problems get only noticed when they are really, really old maybe when the change that introduced the buck was one or two years old already and then it's way harder to track them down and fix them so it's really if you're doing something unusual or have your exotic hardware it's in your interest to test and get involved with testing and to quickly say hey something broke between kernel 4.8 and 4.9 because if you're doing this a year later after the change nobody will be able to help you one more thing tell me what you think about this talk or my work in general has always something to improve or if it's really awful then you have to tell me otherwise I'm doing it again like this you don't have to tell me directly the organizers have a feedback system on the web page where you can simply give stars if you want to be kept up-to-date you can likely not that much of interest LWNS is a better source for you I guess it's English really good topics and if you want to follow me on Google+, there you can get a glimpse of what's happening in the Linux kernel as well because I'm posting there about it now and then that's it for the careers that was slide 256 and do we have time for one or two questions or only on the hallway? yeah? anyone? don't be... try it otherwise grab me on the hallway did I talk too much and nobody has questions or did you all sleep? if there really are no questions then thanks for listening enjoy the conference yeah