May 2017 - kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

How to setup network connection for initramfs running in QEMU?
by Douglas Su May 26, 2017

May 26, 2017

Thanks to this mail list, I have already successfully set up a simple initramfs and booted it in QEMU with busybox tools. However, there is no network connection for the initramfs, that makes me hard to exchange files between qemu and the host. Command `ip link show` tells there exists only one 'lo' interface, and I can't ping 127.0.0.1 successfully. So how to boot initramfs in qemu with full network support? Best regards.

2 1

what is the current/ongoing state of userspace access to GPIO?
by Robert P. J. Day May 25, 2017

May 25, 2017

ashamed to admit, i haven't been keeping up with this, so AIUI, the GPIO sysfs interface is deprecated, replaced with character device files (/dev/gpiochip*) to allow access to GPIO. first, i find it curious that this is a move away from sysfs access to using ioctl(), since i had always thought ioctls were massively discouraged in new code. is there a document that goes through the rationale for this change? and i notice under Documentation/ABI/ that sysfs-gpio is listed under "obsolete", while gpio-cdev is listed under "testing". this seems inconsistent -- how can one be obsolete while the other be categorized as testing? that just seems strange. in any event, is the /dev/gpiochip* interface the recommended interface now? thanks. rday -- ======================================================================== Robert P. J. Day Ottawa, Ontario, CANADA http://crashcourse.ca Twitter: http://twitter.com/rpjday LinkedIn: http://ca.linkedin.com/in/rpjday ========================================================================

4 7

Multipath support?
by Vasu M May 25, 2017

May 25, 2017

I would like to know from which version of Linux, ECMP is supported? I am trying to add a route with two next hops. Only one path gets added. And for the other path, the route says "RTNETLINK answers: File exists". I am not able to understand what's going on. The route prefix is not pointing to ecmp because metric is not given for static routes? How can I test multipath for a prefix that goes over 2 links? I tried "echo 7 > /proc/sys/kernel/printk" and dmesg but I don't see anything related to route add events. Just wondering how I should approach this problem. I am running the following versions on two devices running ONL. *Linux localhost 3.16.43-OpenNetworkLinux #1 SMP Thu May 11 18:37:48 UTC 2017 x86_64 GNU/Linux* I tried the following :- root@localhost:~# route -n Kernel IP routing table Destination Gateway Genmask Flags Metric Ref Use Iface 0.0.0.0 192.168.100.1 0.0.0.0 UG 0 0 0 ma1 10.10.10.0 0.0.0.0 255.255.255.0 U 0 0 0 fpPort2 20.20.20.0 0.0.0.0 255.255.255.0 U 0 0 0 fpPort13 192.168.100.0 0.0.0.0 255.255.255.0 U 0 0 0 ma1 root@localhost:~# arp -a ? *(20.20.20.2) at 00:a0:c9:00:00:01 [ether] on fpPort13 * ? (192.168.100.100) at f0:7f:06:1d:56:ff [ether] on ma1 ? (192.168.100.1) at 18:b1:69:33:88:a4 [ether] on ma1 ? (192.168.100.52) at cc:37:ab:7c:ca:16 [ether] on ma1 ? (192.168.100.3) at c8:1f:66:f6:ea:07 [ether] on ma1 ? *(10.10.10.2) at 00:a0:c9:00:00:01 [ether] on fpPort2* ip route add 100.100.100.0/24 via 10.10.10.2 root@localhost:~# route -n Kernel IP routing table Destination Gateway Genmask Flags Metric Ref Use Iface 0.0.0.0 192.168.100.1 0.0.0.0 UG 0 0 0 ma1 10.10.10.0 0.0.0.0 255.255.255.0 U 0 0 0 fpPort2 20.20.20.0 0.0.0.0 255.255.255.0 U 0 0 0 fpPort13 100.100.100.0 10.10.10.2 255.255.255.0 UG 0 0 0 fpPort2 192.168.100.0 0.0.0.0 255.255.255.0 U 0 0 0 ma1 root@localhost:~# ip route add 100.100.100.0/24 via 20.20.20.2 *RTNETLINK answers: File exists*

1 1

Lock contention determination
by Yiqun Chen May 23, 2017

May 23, 2017

Hi all If I have a lockstat report showing that some lock has more waiting time than holding time, can I assert that this lock is (one of) the performance bottlenecks? Here performance bottleneck means we could optimize lock usage, e.g. reduce the size of critical section etc. If not, why? Thanks beforehand -- Best regards Chen, Yiqun

1 0

Boot issues with custom 64-bit kernel v3.19
by Aditya Basu May 22, 2017

May 22, 2017

I am trying to compile and boot Linux Kernel v3.19 (64-bit version) using - make defconfig make kvmconfig make -j8 The compilation works out without any errors but it doesn't boot inside qemu or my host machine. I use the following command line parameters to linux: "debug earlyprintk loglevel=7" # The boot hangs at the point - #################################### Booting from ROM... early console in setup code early console in decompress_kernel Decompressing Linux... Parking ELF... done. Booting the kernel . #################################### # Things I have tried so far: - Interestingly if I uncheck the 64-bit build in config (from make menuconfig), the resulting 32-bit kernel boots correctly. - Have tried running qemu with different cpus (host) + with & without kvm. sudo qemu-system-x86_64 -m 4G -append 'debug earlyprintk loglevel=7' -kernel arch/x86/boot/bzImage -cpu host -enable-kvm I know the kernel will panic because of no root filesystem, but it happens later in the boot process. I am basing this from booting the 32 bit version of the kernel which prints out a lot of stuff and then panics due to no vfs_root. - The 64-bit versions of kernel v4.2 and above don't have this issue - meaning they boot correctly. Any pointers to what the mistake is? *Linux Kernel Repo used* git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git *My Config* https://dl.dropboxusercontent.com/u/9020146/errors/linux_2017_05_05/config-… * QEMU version* QEMU emulator version 2.5.0 (Debian 1:2.5+dfsg-5ubuntu10.11) -- Regards, Aditya Basu

3 4

Mechnism for sharing data between kernel and user space
by jitendra kumar khasdev May 22, 2017

May 22, 2017

Hi All, I am exploring the mechanism for sharing data between kernel and user space. There are following mechanism I have found 1. copy_to_user/copy_from_user 2. Netlink sock 3. /proc, /debugfs etc. Other than this, Is there any efficient way of which I could read/write linked-list that I have created from my kernel module. --- Jitendra

3 2

Some orientation on topics
by Thomas Korimort May 21, 2017

May 21, 2017

Hello! I am new to kernel development, but i feel somehow committed to officially contribute to open source and Linux. My personal computer runs Win 10, however i am also operating 3 RaspBerry Pi ( 2 x Model 3 and 1 x Model 2) on Raspbian. One is Raspi 3 is my Mediacenter connected to my TV, another is my private server and another is my old Raspi 2, which i have connected to my modem. In the endeavour to build up my private cloud, Apache server and so on, i got into troubles first with incompatible WLAN stick for Raspi 2 with very bad data transmission rates unsuitable even for home network purposes. After moving to Raspi 3 and in one new flat, that problem seems to be improved, and another problem came about: making my intended servers like DNS, Apache, owncloud available to the outside world. My ISP even provided for free a dynamically changing IP open to the internet. So i tried with a free dynamic domain name which was updated by my ISP-provided router (3HuiGate, Huawei LTE-Router with proprietary firmware) automatically. Another issue i encountered was the flash codec not being available on the ARM platform, on which the Raspi3 is based. There are open-source projects like lightspark or gnash, that dealt already with the flash codec in open source context, but they seem not so far developed (gnash abandoned some years ago, lightspark not up-to-date with implementation). Although i operate my phone and tablet on Android comfortably, and my day-to-day business on my PC on Win 10, i strongly want to switch completely to Linux as during my university times i was very much a fan of Linux. My personal idea (or vision) about Linux kernels future is: a universal executable format, more far-reaching than just ELF format. Even the byte-code stored in platform specific form in the text section of ELF binary should be code on a universal platform, that can be easily and fastly compiled and translated on-the-run to the respective platform on which the operating system is actually running, such that >>everything<< on the Linux platforms design is platform independent but only with exception of the very very elementary basic layers. I.e. i am suggesting to design kind of virtual machine specification in an open-source way like for example ARM architecture specification or the Java Virtual machine and make that the standard for a virtual machine capable of most-efficient and perfect code execution according to cutting edge requirements on computing. Another topic i observe is the scheduling policies of the drivers and buffering and paging mechanisms and maybe it is possible to develop a global buffering, synchronizing and caching strategy, that is base on online combinatorial optimization techniques that can yield more perfect scheduling and still achieve greater transparency in the layer of storage media, thereby developing a greater abstraction and higher consistency of computation like for example when the file systems technology progressed from ext2 to ext3 and ext4 or ReiserFS as journaling file system with high failure resistance. What are the perfect policies for dealing with hardware that can be disconnected, reconnected and spoiled anytime by another operating system? Let us say you take a FAT formatted harddisk and plug it in Linux, unplug it and modify it in Windows. What happens then with the Inode superstructure of Linux VFS? Isn't it spoiled and dirty? In such a setting where hardware integrity cannot be guarantueed completely for example by hardware assisted encryption of hardware and strong tying to one operating faculty only, how should an operating system (or specifically Linux) deal with such an issue (on a general scale)? By reading a book on the kernel technology of kernel 2.6 (Wolfgang Maurer: Linux-Kernelarchitektur 2.6) i found that Linux has already good policies implemented since many years, but maybe it is still possible to widen the view also to the legal environment: If you buy one harddisk and format it with ext4, the risk is different than if you share one NTFS formatted harddisk between your Windows and Linux systems. Can such problems be tackled on an operating systems level CONSCIOUSLY without introducing "Linux-certified hardware" that could again spoil away the original idea of Linux? I am a mathematician, computer scientist (Facebook: Thomas Korimort, LinkedIn: Thomas Korimort, ResearchGate: Thomas Korimort) who has also some competence in bible, law and medicine. Thus, personally i would rank myself in the Linux universe among the following: lawgiver, pioneer, executor, judge, lawyer, computer scientist, mathematician, programmer, tester, designer, leader, manager, priest, bishop, magician,... i hope that this list does not offend anyone. I just want to offer my skills to this great project of GNU, Linux, Debian, Open-Source, visionary, prophet,.... in a way that is compatible with my personal style and habit. I am looking forward to your estimated comments and informations. Many greetings, Thomas Korimort.

1 0

Re: Question regarding RT patches for ARM
by Fawad Lateef May 18, 2017

May 18, 2017

Hi Sebastian, On 17 May 2017 at 18:02, Sebastian Andrzej Siewior <bigeasy(a)linutronix.de> wrote: > On 2017-05-17 07:45:39 [-0700], Eric Nelson wrote: >> Hi Fawad, > Hi Fawad, > >> On 05/17/2017 06:40 AM, Fawad Lateef wrote: >> > > > I'm not sure about the unhandled page fault, but the stalls may be >> > > > because of the SDMA driver. >> > > > >> > > > See this patch in the Freescale/NXP community for kernel 4.1: >> > > > https://github.com/Freescale/meta-freescale/blob/master/recipes-kernel/linu… > > That is something I see for the first time. The description says that it > leads to this stall but it fails to explain _how_ it got there. > >> > > Thanks for the patch link. I will give it a try. >> > >> > Talked to hardware manufacturer and they applied the patch. >> > >> > Still that SDMA is only used for NOR flash and only u-boot and its >> > environment are stored in nor flash. So not sure if that is going to >> > help in any case. >> > >> >> SDMA is used for much more than NOR flash (SSI, USB, and sometimes >> UARTs). >> >> > So I would really like few more suggestions OR confirmations about ARM >> > (cortex-A9 based imx6) (preemept)-RT support/patches are stable and >> > can be used for final products. >> > >> >> I don't have any experience with the RT patches on 4.9 or later >> kernels, but do know that they're being used on i.MX6 with the >> vendor kernels (3.10.x, 4.1.x). > > I just looked into a few boot logs and it seems the Phyflex board (imx6 > quad) bootet v4.9.27-rt18, run cyclictest including some tests and > nothing produced this "unhandled page fault" you mention here. This > includes the previous RT version in the v4.9 line. > Thanks, seems like RT patches are quite stable for ARM then. > Of this "page fault" happens on boot, it has to do something without > your .config + hw combination. If it happens at rune-time it is probably > triggered by a driver or an application is triggering this. > Yes, this happens at run-time and not always. Sometimes hardware just hangs/stalls. As now RT is widely used on ARM so to me seems like it something related to hardware design/setting OR application. Another question: Is RT user-space application can make kernel/system crash? As to me kernel should kill the app but let itself running _but_ not sure in the case of RT patched kernel. >> Regards, >> >> >> Eric > > Sebastian Thanks, Fawad Lateef

2 1

device tree nodes and platform drivers
by Max Ruttenberg May 18, 2017

May 18, 2017

Hi all, I'm trying to write a platform driver for a device node that I have defined inside of a device tree. The node looks like this: my-node0 { compatible = "my,id"; dcok-gpios = <&gpio1 6 0>; }; In my kernel module I have something like this: const static struct of_device_id my_ids [] = { { .compatible = "my,id" }, { /* sentinal */ } }; struct platform_driver my_drv = { .driver = { .name = "my-node", .owner = THIS_MODULE, .of_match_table = my_ids, }; .probe = my_probe_func, .remove = my_remove_func, }; static int __init my_init(void) { /* some extra stuff */ return platform_driver_register(&my_drv); } module_init(my_init); And yet... the driver for my probe function is never called! I've looked in /proc for device tree node, and I can see it there. Furthermore, just to be sure, I call of_find_compatbile_node(NULL,NULL,"my,id") and it returns my device node. I'm tempted to just extract what I need from it (i.e. those gpio descriptors) in my init function and skip the platform driver stuff. Can anyone help me understand what I might be missing? Thank you, Max

2 4

Re: I want write more important things
by Roman Storozhenko May 18, 2017

May 18, 2017

> Hello, > I am interested in wrinting code in Linux kernel and I started write kernel modules. I have some basics about kernel development- I red Robert's Love book. My the biggest project with kernel code is simple module, which create char device which can execute commands as root. Actually I'm reading Linux device drivers 3rd edition. And I haven't any idea what I could doing. I have many ebooks about kernel, but I prefer learn by writing some code. Sorry for my banal question (propably you haven't many time), but when someone answer me, I will be very very thankful. I would suggest you to: 1) Complete eudyptula challenge. It contains 20 kernel-related tasks. I have completed 8th task and awaiting to response and I want to say that this is a very good challenge for newbies. 2) Pick one of the areas in which you want to participate in. It might be virtualization, networking, storage, drivers, etc.. Any of that. Go to the appropriate kernel bug tracker, pick a bug and then try to reproduce it, fix it and send it as a patch. I tried many different strategies in last 8 months, but it seems that this one is the most obvious and efficient. After you complete it you will know enough to continue without any advices )

2 1