December 2017 - kernelnewbies@kernelnewbies.org

sched_child_runs_first doesn't work
by Rock Lee Dec. 8, 2017

Dec. 8, 2017

Hi, I ran my test code but always got the father process run first, even after setting sched_child_runs_first. Could anyone give me a hint? Here is my test code. #include <stdio.h> #include <unistd.h> #include <sys/time.h> int main(void) { struct timeval tv; struct timezone tz; int err; pid_t ret = fork(); if (ret == 0) { err = gettimeofday(&tv, &tz); if (!err) printf("child time sec %d, usec %d\n", tv.tv_sec, tv.tv_usec); printf("this child.\n"); } else if (ret > 1) { err = gettimeofday(&tv, &tz); if (!err) printf("father time sec %d, usec %d\n", tv.tv_sec, tv.tv_usec); printf("this father.\n"); } else printf("err!!\n"); return 0; } -- Cheers, Rock

4 6

What happened when the user space application accesss device registers via mapped memory on /dev/mem?
by kipade Dec. 7, 2017

Dec. 7, 2017

Hi all, I want to know the details of device(not physical ram device) registers via mapped memory on /dev/mem. For example, a Soc CPU has some internal device register resources, the driver will map the memory block to a virtual memory to kernel. But, the user application can use /dev/mem to access it(devmem2). I want to know why? and what's happened on the backend? Thanks.

1 0

Re: Any trick to remove a D-state dead process without rebooting?
by Yun-Chih Chen Dec. 6, 2017

Dec. 6, 2017

Hi, Valdis: Thanks for the informative reply! There are actually plenty of processes stuck in D-state, among them are systemd-logind and snmpd. > 1) Check your dmesg/syslog for any WARN, BUG, or OOPS messages with tracebacks, > which probably indicates the kernel bug that caused the problem. http://ix.io/CML > 2) cat /proc/NNN/stack will give you a hint where the syscall is wedged. http://ix.io/CMN It seems that they all have something to do with autofs or filesystem. As shown in this graph: https://i.imgur.com/C1bPHTX.png , the ultimate cause of the massive process massacre is due to memory excessive memory usage of some processes in the noon of Dec 2. However, I did not see OOM in dmesg. Is there a way to adjust system overall OOM aggressiveness so the kernel will kill the offending processes early, before the kernel become mal-functioning? Many thanks, Yun-Chih

3 3

Struggling to get kernel to preserve framebuffer memory when booting
by Dushara Jayasinghe Dec. 5, 2017

Dec. 5, 2017

Hi all, I'm struggling to solve a problem which in theory should be quite straight forward. This email is a bit of a brain dump so apologies for being all over the place. I'm working on an embedded device where u-boot displays a logo which I want the kernel to preserve. However I find the display gets blanked as soon as the kernel starts executing. As I understand it, the first step is to get the kernel to reserve the frame-buffer memory (and hand it to the FB driver later on). So I added the following device tree entry: reserved-memory { #address-cells = <1>; #size-cells = <1>; ranges; display_reserved: fb@0 { no-map; reg = <0x3ffa5000 0x0005b000>; }; }; The memory area reflects the FB memory allocated by u-boot. This is an atmel SAMA5D2 based board with 512M RAM in the 0x20000000-0x40000000 range. However, I find the display is blanked even before the FB driver starts. I have confirmed that the device tree entry is processed by placing printks in drivers/of/of_reserved_mem.c (http://elixir.free-electrons.com/linux/latest/source/drivers/of/of_reserved…) I also added the following line at the same spot to see if that area is at least accessible: *(unsigned int*)0x3ffa5000 = 0xffffffff And I observed the following result: 1. The kernel hangs 2. The display is not yet blanked at this point Writing to the same location from within the u-boot shell places a white pixel on the display. Could anyone suggest what areas I need to look at to solve this problem? Any tips and suggestions will be greatly appreciated. Thank you Dushara

1 0

dts: Exposing local-mac-address to dm9000 from external eeprom
by Mathieu Malaterre Dec. 5, 2017

Dec. 5, 2017

Hi there, I am trying to setup local-mac-address reading for the dm9000 ethernet on a MIPS Creator CI20. The efuse block contains the actual mac address, so I exposed it using nvmem framework: efuse: efuse@134100d0 { compatible = "ingenic,jz4780-efuse"; reg = <0x134100d0 0xff>; clocks = <&cgu JZ4780_CLK_AHB2>; clock-names = "bus_clk"; #address-cells = <1>; #size-cells = <1>; eth_mac: eth_mac@12 { /* six byte/48bit MAC address stored as 8-bit integers */ reg = <0x12 0x6>; }; }; while dm9000 is defined by (boring stuff trimmed): dm9000@6 { compatible = "davicom,dm9000"; davicom,no-eeprom; /* local-mac-address = [00 00 00 00 00 00]; */ }; dm9000 driver handles platform data (dm9000_parse_dt -> of_get_mac_address) just fine. On this SOC there is no eeprom attached to dm9000, so I could potentially hack my way using C code and pretend there is an attached eeprom, using the dm9000 pdata API: $ cat include/linux/dm9000.h [...] struct dm9000_plat_data { /* allow replacement IO routines */ void (*inblk)(void __iomem *reg, void *data, int len); void (*outblk)(void __iomem *reg, void *data, int len); void (*dumpblk)(void __iomem *reg, int len); }; but it feels like there should be a simpler way using DTS declaration only, right ? Maybe I am missing the whole point and local-mac-address should only be setup by U-Boot in this case. Thanks for comments, -- Mathieu

1 0

Any trick to remove a D-state dead process without rebooting?
by Yun-Chih Chen Dec. 5, 2017

Dec. 5, 2017

Hi, fellows: Does anyone know of any hacks that remove a D-state dead process whose parent is pid=1 (actually systemd), without rebooting? Someone posed a trick on the net, but not sure how he did it: https://serverfault.com/questions/76263/how-to-kill-a-defunct-process-with-… Because the dead process is wrapped as a Systemd unit, I've also asked at Systemd's mailing list without avail: https://lists.freedesktop.org/archives/systemd-devel/2017-December/039941.h… Appreciate any hint or even dirty hacks! Thanks in advance! Yunchih

2 1

SD card wont work with dw_mmc driver in linux. Error -84 EILSEQ
by John Smith Dec. 2, 2017

Dec. 2, 2017

2 2

Public Kernel Development
by Chris Obbard Dec. 2, 2017

Dec. 2, 2017

Hi list, Currently I am writing drivers for ARM boards (think along the lines of Raspberry Pi HATs, it doesn't belong in mainline; as well as other various hacks), which just consist of series of patches which I apply by checking out Mainline and then applying the patches by hand. Naturally, this is a bit of a hack... I want to have this as one repo on Github (or the like) with a _lot_ of branches with my changes in. This is what Git is made for, but what happens when Linus releases a new version? How do I make my changes apply to the new tag? Also, what happens if I want to pull from someone's -next repo, or pull in other patches? Maybe there is a tutorial somewhere that can help me. I am used to using Git for small projects :-). Cheers!

4 5

Debugging a kernel freeze
by Victor Ascroft Dec. 1, 2017

Dec. 1, 2017

Hello, I have a iMX6 running a 4.9 kernel with a custom kernel driver communicating with a FPGA over PCIe. The driver is not built in to the kernel but loaded as a module after complete boot up. During the running of the system, after a few hours the kernel completely freezes. No kernel panics or stack traces, nothing. I have access to the serial console. In such a scenario what are the ways to debug and try locating the source of the problem? I am not looking for a solution for my problem but things or approaches one can go about trying while trying to fix such a scenario? Thank you for any pointers in advance. Regards.

3 2