- l4-hackers - example.com Site

clntsrv example capabilities
by Wesley Miaw 17 Sep '11

17 Sep '11

Hi, I'm still trying to fully understand capabilities, and I guess maybe what I need to do is create a new capability of some sort, but I thought asking might help me understand better. Basically, I am trying to build an example like the clntsrv example, and the server.cc attempts to register itself. If that fails, it prints out an error message saying "Readonly namespace?". I guess this is because the namespace is still 'rom'. I tried understanding the contents of the various .cfg files but I'm afraid I'm still confused. Thanks, -- Wesley Miaw wesley(a)wesman.net

2 1

pthread issue on ARM
by Ivan Loskutov 16 Sep '11

16 Sep '11

Hello, Hackers! I trying to porting a library from Linux to L4Re and I got a issue with libpthread on ARM device (beagleboard clone). If I create a detached thread use pthread_detach() or pthread attributes then error is occured. Error does not occur on x86, tested in qemu. I wrote a simple programm for test. Is problem in L4Re? Or should I make any steps before use pthread? Thank you! Code: #include <stdio.h> #include <string.h> #include <pthread.h> void *thread(void* arg) { int res; int num = (int)arg; printf("Start thread %d\n", num); res = pthread_detach(pthread_self()); printf("Detach thread %d (%d)\n", num, res); usleep(5000*num); printf("End thread %d\n", num); return 0; } int main(int argc, char* argv[]) { pthread_t t1, t2; printf("Test pthread\n"); while(1) { printf("new loop...\n"); pthread_create(&t1, NULL, thread, (void*)1); pthread_create(&t2, NULL, thread, (void*)2); usleep(100000); } return 0; } Log: Starting kernel ... L4 Bootstrapper Build: #101 Срд Сен 14 16:32:30 MSD 2011, 4.4.1 Scanning up to 128 MB RAM Memory size is 128MB (80000000 - 88000000) RAM: 0000000080000000 - 0000000087ffffff: 131072kB Total RAM: 128MB mod04: 810c5000-810e7ec0: pthread_test mod03: 810aa000-810c439c: l4re mod02: 8106f000-810a9b50: moe mod01: 81065000-8106e70c: sigma0 mod00: 81015000-81064f74: fiasco Moving 5 modules to 81100000 with offset eb000 moving module 05 { 810c5000-810e7ec0 } -> { 811b0000-811d2ec0 } moving module 04 { 810aa000-810c439c } -> { 81195000-811af39c } moving module 03 { 8106f000-810a9b50 } -> { 8115a000-81194b50 } moving module 02 { 81065000-8106e70c } -> { 81150000-8115970c } moving module 01 { 81015000-81064f74 } -> { 81100000-8114ff74 } Scanning fiasco Scanning sigma0 Scanning moe --init=rom/pthread_test Relocated mbi to [0x8100f000-0x8100f0e4] Loading fiasco Loading sigma0 Loading moe find kernel info page... found kernel info page at 0x80002000 Regions of list regions [ 80001000, 800019ff] { a00} Kern fiasco [ 80002000, 80061fff] { 60000} Kern fiasco [ 80090000, 800969bf] { 69c0} Sigma0 sigma0 [ 80098000, 8009e17b] { 617c} Sigma0 sigma0 [ 80140000, 8017618b] { 3618c} Root moe [ 80178000, 8018ed0f] { 16d10} Root moe [ 81000000, 810143eb] { 143ec} Boot bootstrap [ 8100f000, 8100f1e1] { 1e2} Root Multiboot info [ 81195000, 811d2ebf] { 3dec0} Root Modules Memory API Version: (87) experimental Sigma0 config ip:800900e0 sp:81013704 Roottask config ip:801401c0 sp:00000000 Starting kernel fiasco at 80001000 Hello from Startup::stage2 Initialize page table Vmem_alloc::init() Cache config: ON ID_PFR0: 00001031 ID_PFR1: 00000011 ID_DFR0: 00010400 ID_AFR0: 00000000 ID_MMFR0: 31100003 ID_MMFR1: 20000000 ID_MMFR2: 01202000 ID_MMFR3: 00000011 FPU0: Arch: VFPv3(3), Part: VFPv3(30), r: 1, v: c, i: 41, t: hard, p: dbl/sngl Welcome to Fiasco.OC (arm)! L4/Fiasco.OC arm microkernel (C) 1998-2011 TU Dresden Rev: rUNKNOWN compiled with gcc 4.4.1 for Beagleboard [] Build: #1 Чтв Сен 8 11:49:39 MSD 2011 Calibrating timer loop... done. SIGMA0: Hello! KIP @ 80002000 allocated 4KB for maintenance structures SIGMA0: Dump of all resource maps RAM:------------------------ [0:80000000;80000fff] [0:80062000;8008ffff] [0:80097000;80097fff] [0:8009f000;8013ffff] [4:80140000;80176fff] [0:80177000;80177fff] [4:80178000;8018efff] [0:8018f000;8100efff] [4:8100f000;8100ffff] [0:81010000;81194fff] [4:81195000;811d2fff] [0:811d3000;86ffffff] IOMEM:---------------------- [0:0;7fffffff] [0:88000000;ffffffff] MOE: Hello world MOE: found 113680 KByte free memory MOE: found RAM from 80000000 to 87000000 MOE: allocated 112 KByte for the page array @0x80062000 MOE: virtual user address space [0-bfffffff] MOE: rom name space cap -> [C:501000] BOOTFS: [81195000-811af39c] [C:503000] l4re BOOTFS: [811b0000-811d2ec0] [C:504000] pthread_test MOE: cmdline: moe --init=rom/pthread_test MOE: Starting: rom/pthread_test MOE: loading 'rom/pthread_test' Test pthread new loop... Start thread 1 Detach thread 1 (0) Start thread 2 Detach thread 2 (0) End thread 1 L4Re[rm]: unhandled read page fault @300d3c pc=1003b54 L4Re: unhandled exception: pc=0x1003b54 End thread 2 new loop... modules.list: modaddr 0x01100000 entry pthread_test roottask moe --init=rom/pthread_test module l4re module pthread_test

2 4

L4Re/L4Android specify RAM size on Beagleboard
by giowck 15 Sep '11

15 Sep '11

Hi, when running L4Linux/L4Android on the Beagleboard (rev. C4), only 128MB of 256MB RAM are detected by the bootstrap process. Here [1] you write that the older Beagleboard has only 128MB, so this is default setting. As described in that thread, I tried to create the Makeconf.boot file, and it works so far (image is build with 256MB RAM), but still only 128MB are actualy detected (limited by scan): L4 Bootstrapper Build: #30 Di 6. Sep 14:38:47 CEST 2011, 4.5.1 Scanning up to 256 MB RAM Memory size is 128MB (Limited by Scan) (80000000 - 88000000) RAM: 0000000080000000 - 0000000087ffffff: 131072kB Total RAM: 128MB Please see [2] for the full bootstrap log. I already tried to adjust manually the support.h and startup.cc file in the src/l4/pkg/bootstrap/server/src directory, resulting exactly the same issue. By the way, I think this is causing L4Android to fail during boot, since memory is not correctly allocated: android | l4x_setup_memory: Can't get main memory of 256MB! MOE: mem_alloc: quota: limit=-1 Byte, used=11572080 Byte MOE: mem_alloc: global: avail=57008128 Byte android | Terminating L4Linux. mag | GC: step this=0x108d220 _life = 0x13228 mag | GC: object=0x13228 Thank you for your help so far! Regards Giorgio [1] http://os.inf.tu-dresden.de/pipermail/l4-hackers/2011/004773.html [2] http://pastebin.com/nHQYGz6s

2 3

Re: Sigma0 raised exception on boot
by Wesley Miaw 14 Sep '11

14 Sep '11

On Wed, 2011-08-03 at 20:51 +0000, Adam Lackorzynski wrote: > On Wed Aug 03, 2011 at 12:58:08 +0100, Anna Fischer wrote: > > I'm running Fiasco on the PandaBoard. On platform boot I'm getting an > > error that Sigma0 raised an exception and then Fiasco panics and the > > boot process stops. > > > > Any idea what the problem could be? I can't find any more information to > > debug this properly. I have compile Fiasco.OC and L4Re for the > > PandaBoard platform. The board has 1GB memory. > > > > KERNEL: Warning: Sigma0 raised an exception --> HALT > > Panic: ... > > Return reboots, "k" enters L4 kernel debugger... > > In src/kern/thread-ipc.cpp, before the message "Sigma0 raised an > exception", please add an 'ts->dump();' and retry. This should show more > info on what's going on. This is something that should definitely not > happen. I'm wondering if a solution was ever found to this problem? I am hitting the same error on my Versatile Express rev. D board. My dump output looks similar to Anna's, but with a different RAM base address: EXCEPTION: swi pfa=6009e00c, error=00200000 R[0]: 6009e00c 60096510 0000000f 60096710 6009e018 60002000 0000000f 00000001 R[8]: 60096510 00000000 00000000 00000000 00000000 61013704 27102178 ffffff04 Here's Anna's dump for comparison: EXCEPTION: swi pfa=8009e00c, error=00200000 R[0]: 8009e00c 80096b50 0000000f 80094dd4 8009e018 80002000 0000000f 00000001 R[8]: 80096b50 00000000 00000000 00000000 80096d48 81013044 45559145 ffffff04 This is happening as a result of the call to sigma0_thread->activate() inside kernel_thread-std.cpp. I haven't been able yet to track it down farther. Thanks, -- Wesley Miaw wesley(a)wesman.net

2 1

Re: build error: No rule to make target crtendS.o
by Wesley Miaw 13 Sep '11

13 Sep '11

On Mon, September 12, 2011 09:45, Adam Lackorzynski wrote: > Hi, > On Mon Sep 12, 2011 at 08:38:56 -0700, Wesley Miaw wrote: >> Do you have any clues as to why I only get garbage printed out when I try to run one of the bootstrap programs? > Hmm, the baud-rate in bootstrap is set to 115200, so that's what you're using? IIRC the boot monitor is using 38400. Thanks, Adam. That was exactly the problem for why I was getting garbage. I didn't think to guess the bootstrap might be changing the console baud rate. I've tried changing the code to use 38400 instead of 115200 (because I can't find a way to change the board's startup to use 115200, only post-startup, which makes things inconvenient) but all of my attempts to change the values of 115200 to 38400 in the code don't seem to have any effect. I had to switch the RAM_BASE to 0x60000000 but after doing that, it looks like things should be working. I also changed the Fiasco globalconfig settings to change it to VEXPRESS and use 0x60000000 as the RAM location. I noticed there are two builds of Fiasco being created also. One of which is multi-core and has a TrustZone option. The other does not: arm-mp-rv-pb11mp and arm-up-rv-9. I'm not sure which one is getting used right now. Anyway, doing this got everything to load until the point where it says there is an invalid System ID. I'm guessing I still need to tweak the kernel configuration but at least I've made some good progress today! Starting kernel fiasco at 60001000 Realview System ID: Rev=3 HBI=190 Build=f Arch=5 FPGA=05 Invalid System ID for this kernel config Expected (3190f505 & ffffff00) == 1190f500 Stopping. Thanks for your help. :) -- Wesley Miaw wesley(a)wesman.net

2 1

Re: build error: No rule to make target crtendS.o
by Wesley Miaw 12 Sep '11

12 Sep '11

Hi Adam, Do you have any clues as to why I only get garbage printed out when I try to run one of the bootstrap programs? I thought it might be because I need to use the generic-hosted.ld linker script packaged with arm-none-eabi, but since you said you think everything should work with arm-none-linux-gnueabi, I'm questioning my conclusion now. (I haven't gotten around to testing a build with arm-none-eabi and generic-hosted.ld yet.) Thanks, -- Wesley Miaw wesley(a)wesman.net On Sep 12, 2011, at 3:00 AM, l4-hackers-request(a)os.inf.tu-dresden.de wrote: > > On Sun Sep 11, 2011 at 10:48:57 -0700, Wesley Miaw wrote: >> On Sep 11, 2011, at 3:00 AM, l4-hackers-request(a)os.inf.tu-dresden.de wrote: >>> >>> On Sat Sep 10, 2011 at 12:26:05 -0700, Wesley Miaw wrote: >>>> I'm not getting any errors printed out on the console when I run the >>>> ELF or RAW bootstrap binaries. It just prints out a line of garbage >>>> (e.g. ? or other random characters). Then I have to hit the reset >>>> button. This is >>> >>> You were mentioning a 'too many headers error' when compiling/linking >>> with arm-none-linux-gnueabi which you seemed to not get with >>> arm-none-eabi. I would like to understand what makes the difference in >>> your setup and then see further. >> >> Ah, I see. I don't know why I got that error originally when compiling >> L4, but I am no longer getting that error. Maybe I did something wrong >> (or remembering things incorrectly)--I see in the bootstrap linker >> script that it purposefully combines the data and bss sections with a >> comment that sounds like it is explicitly working around the problem >> of a regular linker script creating too many headers. >> >> If I compile some other program I write myself to ELF, like a single >> file Hello World, using arm-none-linux-gnueabi then I get the too many >> headers error when the ARM Boot Manager tries to flash the ELF binary. >> Switching to arm-none-eabi and using the generic-hosted.ld linker >> script fixes this problem. > > That's also the reason why bootstrap is generated into a single header, > because there are loaders that can only do one unfortunately, like the > one on this platform.

2 1

Re: build error: No rule to make target crtendS.o
by Wesley Miaw 11 Sep '11

11 Sep '11

On Sep 11, 2011, at 3:00 AM, l4-hackers-request(a)os.inf.tu-dresden.de wrote: > > On Sat Sep 10, 2011 at 12:26:05 -0700, Wesley Miaw wrote: >> I'm not getting any errors printed out on the console when I run the >> ELF or RAW bootstrap binaries. It just prints out a line of garbage >> (e.g. ? or other random characters). Then I have to hit the reset >> button. This is > > You were mentioning a 'too many headers error' when compiling/linking > with arm-none-linux-gnueabi which you seemed to not get with > arm-none-eabi. I would like to understand what makes the difference in > your setup and then see further. Ah, I see. I don't know why I got that error originally when compiling L4, but I am no longer getting that error. Maybe I did something wrong (or remembering things incorrectly)--I see in the bootstrap linker script that it purposefully combines the data and bss sections with a comment that sounds like it is explicitly working around the problem of a regular linker script creating too many headers. If I compile some other program I write myself to ELF, like a single file Hello World, using arm-none-linux-gnueabi then I get the too many headers error when the ARM Boot Manager tries to flash the ELF binary. Switching to arm-none-eabi and using the generic-hosted.ld linker script fixes this problem. Thanks, -- Wesley Miaw wesley(a)wesman.net

2 1

Re: build error: No rule to make target crtendS.o
by Wesley Miaw 11 Sep '11

11 Sep '11

I'm not getting any errors printed out on the console when I run the ELF or RAW bootstrap binaries. It just prints out a line of garbage (e.g. ? or other random characters). Then I have to hit the reset button. This is I tried modifying the load and run memory addresses but that didn't make any difference. There weren't any build errors as far as I could see. Is there some other error you would like me to try capturing? Thanks, -- Wesley Miaw wesley(a)wesman.net On Sep 10, 2011, at 3:00 AM, l4-hackers-request(a)os.inf.tu-dresden.de wrote: > Hi, > > On Fri Sep 09, 2011 at 13:09:09 -0700, Wesley Miaw wrote: >>>> % arm-none-eabi-gcc --version >>>> arm-none-eabi-gcc (Linaro GCC 4.5-2011.02-0) 4.5.2 >>>> make[8]: *** No rule to make target >> `/opt/armgcc/lib/gcc/arm-none-eabi/4.5.2//crtendS.o', needed by >> `libl4sys.so'. Stop. >>> >>> Please use the arm-none-linux-gnueabi version of gcc. Looks like we >>> might not need to link crtendS.o at least with 4.5 but I'm not sure >>> currently for other versions. >> >> I'm having trouble understanding how the bootstrap images are supposed to >> be run on an ARM board if they are compiled with arm-none-linux-gnueabi >> instead of arm-none-eabi. Reading the Sourcery G++ Lite documentation, and >> talking to ARM support, bare metal should be compiled using arm-none-eabi. >> >> I am able to compile a hello world application for execution off the >> Versatile Express Boot Monitor using arm-none-eabi but not when using >> arm-none-linux-gnueabi (too many headers error). I saw in the bootstrap >> linker script that the data and bss sections are combined to fix this >> problem. But when I try to run one of the E=hello bootstrap images (ELF or >> RAW) I get garbage printed out. With arm-none-eabi I have to specify a >> 'hosted' linker script to get it to output text. >> >> Fiasco.OC's Features web page states it supports the Versatile Express >> board. I'm unable to find any documentation on it though. >> >> At this point, would I need to modify the build to use arm-none-eabi and >> the Makeconf.local to use the correct arm-none-eabi linker script and >> build option (e.g. -mcpu=cortex-a9)? > > I would like to see the errors you are getting with a non-modified > setup. For basic things like bootstrap it does not matter whether it is > the arm-none-eabi or arm-none-linux-gnueabi as it does not make use of > any of those specifics. I for example use the arm-none-linux-gnueabi > variant just fine. Also, this does not have to do anything with the > specific board such as Versatile Express or others, or with the specific > CPU variant the compiler is targeting for. This is 'just' a compile/link > issue.

2 1

Re: build error: No rule to make target crtendS.o
by Wesley Miaw 09 Sep '11

09 Sep '11

Hi Adam, >> % arm-none-eabi-gcc --version >> arm-none-eabi-gcc (Linaro GCC 4.5-2011.02-0) 4.5.2 >> make[8]: *** No rule to make target `/opt/armgcc/lib/gcc/arm-none-eabi/4.5.2//crtendS.o', needed by `libl4sys.so'. Stop. > > Please use the arm-none-linux-gnueabi version of gcc. Looks like we > might not need to link crtendS.o at least with 4.5 but I'm not sure > currently for other versions. I'm having trouble understanding how the bootstrap images are supposed to be run on an ARM board if they are compiled with arm-none-linux-gnueabi instead of arm-none-eabi. Reading the Sourcery G++ Lite documentation, and talking to ARM support, bare metal should be compiled using arm-none-eabi. I am able to compile a hello world application for execution off the Versatile Express Boot Monitor using arm-none-eabi but not when using arm-none-linux-gnueabi (too many headers error). I saw in the bootstrap linker script that the data and bss sections are combined to fix this problem. But when I try to run one of the E=hello bootstrap images (ELF or RAW) I get garbage printed out. With arm-none-eabi I have to specify a 'hosted' linker script to get it to output text. Fiasco.OC's Features web page states it supports the Versatile Express board. I'm unable to find any documentation on it though. At this point, would I need to modify the build to use arm-none-eabi and the Makeconf.local to use the correct arm-none-eabi linker script and build option (e.g. -mcpu=cortex-a9)? Thanks, -- Wesley Miaw wesley(a)wesman.net

2 1

loading boot task
by Chen Tian 09 Sep '11

09 Sep '11

Hi, When Fiasco.OC loads the boot task, how are physical pages allocated for different elf sections? I am experiencing a page fault when accessing the second page that contains .bss section data. Is this normal? It will be great if someone can tell me where the boot task loading is implemented in the kernel. Thanks a lot. Chen

2 2