- kernelnewbies@kernelnewbies.org - Kernelnewbies Mailing Lists

[HELP] Documentation on CPU:/sys/devices/system/cpu/cpuX/cache/indexX/shared_cpu_map
by Ren Zhen April 25, 2013

April 25, 2013

Hi all: Can anybody help me to understand the usage of 'shared_cpu_map' in /sys/devices/system/cpu/cpuX/cache/indexX/. when I execute the cmd--'#cat shared_cpu_map', it retures '05'. And my computer use Ubuntu12.04,Intel core i3 CPU. I have read one email in LKML,it says: "The patch also adds a bunch of interfaces under /sys/devices/system/cpu/cpuX/cache, showing various information about the caches. Most useful field being shared_cpu_map, which says what caches are shared among which logical cpus. " But I still cannot catch the meaning of 'which says what caches are shared among which logical cpus'. Thanks. -- Sincerely, Ren Zhen

2 1

［OT] How to use relative path in gdb macro
by hz hanks April 24, 2013

April 24, 2013

Hi, All I'm using gdb macro (.gdb files) in gdb to achieve complex process. And my .gdb files will use command "source" to include the python files to do the actual work. The problem is that the .gdb files only know the relative path of .py files. Does anyone know how to write that kind of source command? Thanks! Best, Hanks

1 0

Re: Dynamic Sysfs Attribute Files
by Matt Davis April 24, 2013

April 24, 2013

> Date: Wed, 24 Apr 2013 15:39:42 +0200 > From: Tobias Boege <tobias(a)gambas-buch.de> > > On Wed, 24 Apr 2013, Matt Davis wrote: >> Hello, >> I want my kernel module to dynamically add sysfs files for its >> corresponding device. Without going into detail, the attribute file >> will be named something like: /sys/class/myclass/mydev0/fileX where >> 'X' is an integer ranging from 1-256. I did not see anything in the >> api to allow this. I suppose I could just dynamically allocate an >> attribute, give it the unique name by tacking on my own integer, and >> pass that off to device_create_file(). However, I have not seen >> anywhere in the codebase where another module does this; I assume it >> is not allowed (or there is a better solution). >> >> The other solution (instead of sysfs) would be to add an ioctl, but >> this means the userland application must know the specific 'file' id >> to query. In the sysfs version above, the userland application can >> just grab the directory listing of /sys/class/myclaass/mydev0/ and see >> how many 'file' instances exist. >> >> -Matt > > Hmm, the following led some places on my system: > > $ shopt -s extglob > $ find /sys/!(module|kernel) -name "*[0-9]" > ... > /sys/firmware/acpi/interrupts/gpe00 > /sys/firmware/acpi/interrupts/gpe01 > /sys/firmware/acpi/interrupts/gpe02 > /sys/firmware/acpi/interrupts/gpe03 > /sys/firmware/acpi/interrupts/gpe04 > /sys/firmware/acpi/interrupts/gpe05 > ... > > >From a quick grep, the source _may_ be somewhere around > > drivers/acpi/sysfs.c: > 658 sprintf(buffer, "gpe%02X", i); > > Hope this helps you further. Tobi, That is exactly what I am looking for; that and the fact no one explicitly flamed me for my idea. -Matt

1 0

anyone else getting "sizeof(struct x86cpu_device_id)" errors compiling kernel?
by Robert P. J. Day April 24, 2013

April 24, 2013

for the first time in a while, i'm compiling the current git pull of the kernel source on my ubuntu system and, for a number of modules, i'm getting: Building modules, stage 2. MODPOST 1943 modules FATAL: arch/x86/kvm/kvm-amd: sizeof(struct x86cpu_device_id)=12 is not a modulo of the size of section __mod_x86cpu_device_table=32. Fix definition of struct x86cpu_device_id in mod_devicetable.h make[2]: *** [__modpost] Error 1 make[1]: *** [modules] Error 2 make: *** [sub-make] Error 2 i already deselected a couple of crypto modules that produced that but it's clear there are a number of modules that are going to cause this grief. i've never seen this before and i'm about to dig into it, but anyone else tripped over this and know the workaround? this is on a 64-bit ubuntu system, with: $ gcc --version gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3 ... it looks like something related to packed structures but that's just a guess right now. rday -- ======================================================================== Robert P. J. Day Ottawa, Ontario, CANADA http://crashcourse.ca Twitter: http://twitter.com/rpjday LinkedIn: http://ca.linkedin.com/in/rpjday ========================================================================

1 0

[HELP]CPU in SYSFS:/sys/devices/system/cpu/cpu0/microcode
by Ren Zhen April 24, 2013

April 24, 2013

Hi all: Can anyone help me to understand the usage of 'microcode' in /sys/devices/system/cpu/cpu0/. ------------------------------------------------------------------------------------- mysystem: /sys/devices/system/cpu/cpu0/ #ls processor_flags reload version ------------------------------------------------------------------------------------- What are the usages of 'processor_flags', 'reload' and 'version'? I have searched in the Documentation and LKML,but failed to find helpful information. Can Anyone who know those give me some more information about this? Any help will be appreciated! -- Sincerely, Ren Zhen

1 0

Dynamic Sysfs Attribute Files
by Matt Davis April 24, 2013

April 24, 2013

Hello, I want my kernel module to dynamically add sysfs files for its corresponding device. Without going into detail, the attribute file will be named something like: /sys/class/myclass/mydev0/fileX where 'X' is an integer ranging from 1-256. I did not see anything in the api to allow this. I suppose I could just dynamically allocate an attribute, give it the unique name by tacking on my own integer, and pass that off to device_create_file(). However, I have not seen anywhere in the codebase where another module does this; I assume it is not allowed (or there is a better solution). The other solution (instead of sysfs) would be to add an ioctl, but this means the userland application must know the specific 'file' id to query. In the sysfs version above, the userland application can just grab the directory listing of /sys/class/myclaass/mydev0/ and see how many 'file' instances exist. -Matt

3 2

access to hardware registers using KGDB on ARM
by Андрей Скворцов April 24, 2013

April 24, 2013

Hello everyone, I have arm SoC AM3517 (Cortex-A8) running linux 2.6.32. I am trying to access hardware registers using kgdb. I can connect to the board via RS232, can read variables in kernel space, control processes using breakpoints, but I can't read hardware register. For example, I am trying to read UART3 registers. Information about UART3 is saved in structure: static struct plat_serial8250_port serial_platform_data2[] = { { .mapbase = OMAP_UART3_BASE, .irq = 74, .iotype = UPIO_MEM, .regshift = 2, ... } } mapbase is a physical address. According to the Technical Reference Manual is should be 0x49020000. I checked this field using kgdb: (gdb) p/x serial_platform_data2[0].mapbase $34 = 0x49020000 This structure has the field membase. As I understand, it contains virtual address of the mapbase physical address. This field is initialized on UART init via function ioremap(). I checked this field as well: (gdb) p/x serial_platform_data2[0].membase $35 = 0xfb020000 Now I am trying to read register MCR_REG, that has offset 0x10 from the base address. (gdb) x serial_platform_data2[0].membase+0x10 0xfb020010: Cannot access memory at address 0xfb020010 (gdb) x serial_platform_data2[0].mapbase+0x10 0x49020010: Cannot access memory at address 0x49020010 As I understand, the membase address should work. Right? Why has access failed? Is it kgdb limitation? -- Best regards, Andrey Skvortsov

1 0

[HELP] cpuidle in SYSFS: /sys/devices/system/cpu/cpu0/cpuidle/state0/desc
by Ren Zhen April 24, 2013

April 24, 2013

Hi all: In /Documentation/cpuidle/sysfs.txt, It says: "desc : Small description about the idle state(string)". when I read desc: ------------------------------------------------------------------------ mysystem:/sys/devices/system/cpu/cpu0/cpuidle/state0/# cat desc CPUIDLE CORE POLL IDLE ------------------------------------------------------------------------ Here is my question:What the meaning of ‘CPUIDLE CORE POLL IDLE'? Any help will be appreciated! -- Sincerely, Ren Zhen

1 0

Fwd: Building kernel modules with debuginfo and printing line numbers in kernel oops message / coredump
by Sankar P April 24, 2013

April 24, 2013

On Mon, Apr 22, 2013 at 6:00 AM, arshad hussain <arshad.super(a)gmail.com> wrote: > > > > On Mon, Apr 22, 2013 at 5:58 AM, arshad hussain <arshad.super(a)gmail.com> > wrote: >> >> >> >> >> On Fri, Apr 19, 2013 at 11:55 PM, Sankar P <sankar.curiosity(a)gmail.com> >> wrote: >>> >>> Hi, >>> >>> I have written a kernel module which I build using: >>> >>> Makefile (edited a bit) >>> ========== >>> obj-m := simple.o >>> simple-objs := s.o >>> >>> make -C /lib/modules/$(shell uname -r)/build M=$(PWD) modules >>> >>> ======= >>> >>> Everything builds fine. But my kernel module caused an oops and the >>> message was something like: >>> >>> myfunctionname +0x2507 +5679 >>> >>> or some hexadecimal gibberish. I know that I can use some external >>> tool to find out what these addresses mean. But, is there a way I can >>> get the kernel oops message to print these information along with line >>> number ? >>> >>> Is there a way to build the kernel module with debug symbols ? In user >>> space, I usually do "-g" or "-ggdb" and it will give me line numbers >>> and filenames neatly while dumping stacktrace. Is there a similar >>> facility available for out of tree kernel modules too ? >>> >>> Even if I give : >>> >>> $ CFLAGS="-ggdb" make >>> >>> The output file size (.ko) is not changed. >>> >>> Can someone help me with building kernel modules with debug info and >>> get the filename and line numbers printed in the oops messages / >>> coredump ? >>> >>> Thanks. >>> >>> >>> -- >>> Sankar P >>> http://psankar.blogspot.com >>> >>> _______________________________________________ >>> Kernelnewbies mailing list >>> Kernelnewbies(a)kernelnewbies.org >>> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies >> >> >> "myfunctionname +0x2507 +5679" >> >> It's not hex gibberish. What it saying is that, at the time of crash the >> EIP was pointing >> withing function "myfunctionname" and offset 0x2507 bytes. Use objdump >> binary to >> more detail of your simple.ko >> Thanks. But is there not a way to get the line numbers and filenames printed automatically instead of the memory addresses ? For example, in gdb, in userspace, if there is a crash, we can get the line numbers printed when there is a crash. Can the kernel crash/oops handler similarly not print the filenames and linenumbers automatically ? Is there a configuration or hack for getting it that way ? -- Sankar P http://psankar.blogspot.com -- Sankar P http://psankar.blogspot.com

2 1

Re: [rrd-users] Can small writes generate a lot of faults?
by Daniel Hilst Selli April 23, 2013

April 23, 2013

Em 20/04/2013 09:09, Tobi Oetiker escreveu: > hi daniel > > On 19.04.2013, at 16:42, Daniel Hilst Selli <danielhilst(a)gmail.com> wrote: > >> AFAIK, major page faults are generated when data that is not yet present >> on RAM is loaded from disk, but in this case data is being write do >> disk, I can't >> see how writes can generate faults, but still, it seems that is >> happening, ... !? >> > rrdtool changes just few bytes in the rrf file with each update > > the OS though can only write a complete block of data to disk. > > so when you write a byte to a file, the OS has to first fetch the block, modify the byte and write the block back > > that is why rrdtool benefits greatly from more RAM as it can hold mor blocks in RAM and does not have to read them first. > > cheers tobi > Thanks Tobi, I had no idea about this read/update/write process, Now I know how to improve the performance of my collectd+rrdtool stuff, Cheers

1 0