Linaro-android May 2014

linaro-android@lists.linaro.org

16 participants
20 discussions

by Mohammad Merajul Islam Molla

Hello Amit/Robert/Others, I was looking into the code for heat_cpu.c in pm-qa. I think there are some issues with the current implementation. Please find details below - 1. #define NR_THREAD 3 pid_t thr_id[NR_THREAD]; I think the array thr_id should be dynamically allocated as number of threads depend on number of cpus. However, we can eliminate the need for this array as mentioned below at no 7. 2. There is no error checking for sysconf (...) call (line 83). If this call fails … [View More]num_cpus will be -1 which is a problem. 3. at line 122, pthread_t *p_thread_ptr = (pthread_t *) malloc( num_cpus * sizeof(pthread_attr_t)); we are allocating space for pthread_t. So sizeof(ptread_t) should be used instead of sizeof(pthread_attr_t). 4. at line 144, SCHED_NORMAL is used. On my Ubuntu 13.04, <sched.h> does not define SCHED_NORMAL. So it gives compile error if ANDROID is defined. heat_cpu.c:145:44: error: ‘SCHED_NORMAL’ undeclared (first use in this function) We can either use SCHED_OTHER (which is equivalent to SCHED_NORMAL) or include <linux/sched.h> 5. at line 168, error message is misleading. This error is related to failure from pthread_create, not during setting affinity. 6. at line 76, the call to gettid results in linking error - heat_cpu.c: In function ‘do_loop’: heat_cpu.c:77:2: warning: implicit declaration of function ‘gettid’ [-Wimplicit-function-declaration] /tmp/ccAhcUGE.o: In function `do_loop': heat_cpu.c:(.text+0x4e): undefined reference to `gettid' There is no libc wrapper for gettid. So I think it should be called as below - #include <sys/syscall.h> thr_id[thread] = syscall(SYS_gettid); With this changes the compilation warning and link error goes away and it works. However, this call can be eliminated as suggested below at no 7. 7. There is a potential race condition between line 171 & 182, despite the synchronization mechanism (mutex & condition variable) used. There is no guaranty that do_loop(...) will execute first and thr_id[thread] = gettid(); will execute before setting affinity as noted from below debug output - setting affinity for : 0 setting affinity for : 0 setting affinity for : 0 setting affinity for : 0 The code is setting affinity for tid 0, which it should not, despite the synchronization code. However, we can eliminate need for thr_id and the associated synchronization if we use pthread_setaffinity_np(...) . Would you care to test the alternative implementation using the patch below (also attached)? -------------------------------------------------------------------- diff --git a/utils/heat_cpu.c b/utils/heat_cpu.c index 90d8a97..236cb1d 100644 --- a/utils/heat_cpu.c +++ b/utils/heat_cpu.c @@ -47,23 +47,12 @@ #include <string.h> #include <math.h> -#define NR_THREAD 3 - int cont = 1; -int cpu_index; long long a = 1, c = 1; float f = M_PI; -pid_t thr_id[NR_THREAD]; int moderate_inst; - -#ifdef ANDROID -int conditionMet; -pthread_cond_t cond = PTHREAD_COND_INITIALIZER; -pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER; -#endif - void *do_loop(void *data) { const long long b = 3; @@ -71,14 +60,6 @@ void *do_loop(void *data) a = 1; c = 1; -#ifdef ANDROID - long int thread = (long int) data; - thr_id[thread] = gettid(); - pthread_mutex_lock(&mutex); - while (!conditionMet) - pthread_cond_wait(&cond, &mutex); - pthread_mutex_unlock(&mutex); -#endif while (cont) { if (!moderate_inst) { @@ -96,9 +77,15 @@ void *do_loop(void *data) int main(int arg_count, char *argv[]) { int ret, i; - int num_cpus = sysconf(_SC_NPROCESSORS_ONLN); + int num_cpus; cpu_set_t cpuset; + num_cpus = sysconf(_SC_NPROCESSORS_ONLN); + if (num_cpus < 0) { + printf("ERROR: sysconf error\n"); + return -1; + } + printf("Num CPUs: %d\n", num_cpus); if (arg_count > 1) { if (!strcmp("moderate", argv[1])) { @@ -120,7 +107,7 @@ int main(int arg_count, char *argv[]) } pthread_t *p_thread_ptr = (pthread_t *) malloc( - num_cpus * sizeof(pthread_attr_t)); + num_cpus * sizeof(pthread_t)); if (!p_thread_ptr) { printf("ERROR: out of memory\n"); @@ -141,7 +128,7 @@ int main(int arg_count, char *argv[]) #endif #else - ret = pthread_attr_setschedpolicy(&p[i], SCHED_NORMAL); + ret = pthread_attr_setschedpolicy(&p[i], SCHED_OTHER); #endif /* for each new object */ CPU_SET(i, &cpuset); @@ -156,7 +143,7 @@ int main(int arg_count, char *argv[]) return ret; } #endif - CPU_CLR(i, &cpuset); + CPU_CLR(i, &cpuset); } @@ -165,29 +152,23 @@ int main(int arg_count, char *argv[]) ret = pthread_create(&p_thread_ptr[i], &p[i], do_loop, (void *)i); if (ret < 0) - printf("Error setting affinity for cpu%d\n", i); + printf("Error creating thread for cpu%d\n", i); #ifdef ANDROID CPU_ZERO(&cpuset); CPU_SET(i, &cpuset); - ret = sched_setaffinity(thr_id[i], sizeof(cpuset), &cpuset); + ret = pthread_setaffinity_np(p_thread_ptr[i], sizeof(cpuset), &cpuset); if (ret) { - printf("Error setting affinity on pthread th_id\n"); + printf("Error setting affinity on pthread\n"); printf("Error: %s\n", strerror(ret)); return ret; } #endif } -#ifdef ANDROID - pthread_mutex_lock(&mutex); - conditionMet = 1; - pthread_cond_broadcast(&cond); - pthread_mutex_unlock(&mutex); -#endif - while (1) sleep(1); + return 0; } -- Thanks, -Meraj [View Less]

10 years, 10 months

Trinity fuzzer

by $rik＠nth

Hi All, Does Trinity fuzzer is available on Linaro kernel? Can anyone please let me know if it is part of some other git project or where i can get it for Android. -- Thanks & Regards, M.Srikanth Kumar.

10 years, 10 months

Re: [PATCH PMQA] cpuidle_killer: allocate per cpu process array dynamically

by Mohammad Merajul Islam Molla

Hello Sanjay, You are welcome. Earlier I submitted some patches in powedebug and powertop tools. Please feel free to merge if you find any useful. http://lists.linaro.org/pipermail/linaro-dev/2014-April/017117.html http://lists.linaro.org/pipermail/linaro-dev/2014-May/017141.html http://lists.linaro.org/pipermail/linaro-dev/2014-May/017144.html http://lists.linaro.org/pipermail/linaro-dev/2014-May/017146.html -- Thanks, -Meraj Mohammad Merajul Islam Molla (Meraj) Mobile Lab 1, Mobile … [View More]Development Team Samsung R&D Institute Bangladesh (SRBD) Phone: +880-1754380207 Skype: mmm2177 On Mon, May 12, 2014 at 4:27 PM, Sanjay Singh Rawat <sanjay.rawat(a)linaro.org > wrote: > On Wednesday 07 May 2014 03:34 PM, Mohammad Merajul Islam Molla wrote: > >> Hello Sanjay, >> >> One suggestion - its probably better to precede the output lines with >> cpu no. As currently output from child processes are intermixed as below >> and difficult to co-relate. >> >> jiffies are : 10000 usecs >> found 4 cpu(s) >> duration: 120 secs, #sleep: 1200, delay: 100000 us >> duration: 120 secs, #sleep: 1200, delay: 100000 us >> duration: 120 secs, #sleep: 1200, delay: 100000 us >> duration: 120 secs, #sleep: 1200, delay: 100000 us >> counter value 4072718528 >> test duration: 120.057526 secs >> deviation 0.000479 >> counter value 3914063589 >> test duration: 120.057335 secs >> counter value 4015937716 >> test duration: 120.057430 secs >> deviation 0.000479 >> deviation 0.000478 >> counter value 411037603 >> test duration: 120.062716 secs >> deviation 0.000523 >> > > thanks Meraj, done > >> >> >> >> -- >> Thanks, >> -Meraj >> >> >> On Fri, May 2, 2014 at 11:23 PM, Sanjay Singh Rawat >> <sanjay.rawat(a)linaro.org <mailto:sanjay.rawat@linaro.org>> wrote: >> >> On Wednesday 30 April 2014 02:04 PM, Mohammad Merajul Islam Molla >> wrote: >> >> Hello Sanjay, >> As far I know, if option argument is 0, the parent will wait for >> >> >> Looks like waiting is done if childs exit. I checked for the >> offlined CPU case, if there are no child processes, waitpid returns >> -1. Setting errno as "no child processes" >> >> specified child pid to terminate, its not for immediate return as >> in >> case of WNOHANG. This is probably the intended use of the code >> (author >> will be able to confirm). Changing 0 to WNOHANG macro will >> change the >> meaning of the code. >> Also, from header file - >> #define WNOHANG 0x00000001 >> >> >> sorry i referred wrong document >> >> WNOHANG is defined as 1, not zero. >> Which makes me think of two cases below - >> 1. If the real intended purpose is to not to wait infinitely, >> replacing >> >> [...] >> >> >> >> >> >> > > -- > sanjay > [View Less]

10 years, 10 months

About USB Mass Storage Support.

by Botao Sun

Hi Vishal, A question about USB mass storage feature on Linaro Android images. I would say there are 2 different ways: 1. Using Android device as a host: In this way, by plugging in a USB mass storage device, a USB stick for example, there should be a pop-up window shows up on screen to say a USB mass storage device has been plugged in, similar as you do it on your PC. The current status is after plugged in the USB stick, some manual procedures need to be done to access the device, as … [View More]

10 years, 10 months

Re: linaro-android Digest, Vol 30, Issue 3

by gmail

Now the "repo sync" is ok. I think the reason is that my Internet connection is not stable. thanks On 05/09/2014 08:00 AM, linaro-android-request(a)lists.linaro.org wrote: > Send linaro-android mailing list submissions to > linaro-android(a)lists.linaro.org > > To subscribe or unsubscribe via the World Wide Web, visit > http://lists.linaro.org/mailman/listinfo/linaro-android > or, via email, send a message with subject or body 'help' to > linaro-android-request(a)… [View More]

10 years, 11 months

Detect UIFreeze and Blank screen

by $rik＠nth

Hi Linaro/Android/Dev's , I am running few Graphics benchmark tests on a Pandaboard. Recently after few changes in Overlay layer i am seeing UI is getting freeze randomly and sometime it is showing black screen even though shell is able to access. If i launch any activity through am, UI is not getting updated and it needs a hard reboot. It is very hard to debug the root cause. How to stop any test/benchmark if the UI freeze or a blank screen. Is there any way to do it from kernel or userspace.… [View More]

10 years, 11 months

repo sync stops at the 99% (410/414)

by gmail

i have tried some times， but all stop at 99% Fetching project platform/external/linaro-android-tools Fetching project platform/hardware/samsung/origen/gralloc_ump Fetching project platform/frameworks/compile/libbcc Fetching projects: 98% (406/414) Fetching project platform/external/chromium_org/third_party/skia/gyp Fetching project platform/hardware/invensense Fetching project platform/external/ipsec-tools Fetching project device/samsung/proprietary-open Fetching projects: 99% (410/414) Fetching project platform/packages/wallpapers/Basic

10 years, 11 months

Linaro Android on cubieboard?

by Vyacheslav Evgrafov

Congrads on partership with Allwinner. Are you guys planning Android release for Cubieboard anytime soon?

10 years, 11 months

how to update the source after the environment MANIFEST_BRANCH have been changed ?

by gmail

I have downloaded the source a month ago , at that time the MANIFEST_BRANCH was "MANIFEST_BRANCH=linaro_android_4.4". Now, i found that the environment is "MANIFEST_BRANCH=linaro_android_4.4.2" how can i update my source code ? i follow this page https://android-build.linaro.org/builds/~linaro-android-member-ti/panda-lin… thang you !

10 years, 11 months

Problems building Linaro Android on Ubuntu 10.04 for Versatile Express

by Dious, Kevin

To whom it may concern, My name is Kevin Dious and I am writing because I am trying to use the linaro_android_build_cmds.sh script to build Linaro Android for the Versatile Express. Specifically, I am trying to build the following: https://android-build.linaro.org/builds/~linaro-android/vexpress-linaro-14.… Script link: http://snapshots.linaro.org/android/~linaro-android/vexpress-linaro-14.04-r… When I run the script, I get the following error initially (highlighted in yellow): Package … [View More]acpica-tools is not available, but is referred to by another package. This may mean that the package is missing, has been obsoleted, or is only available from another source E: Package acpica-tools has no installation candidate kdious@eco-build1:/local/mnt/workspace/vexpress-linaro-build-309$ I noticed that the acpica-tools package is only available for Ubuntu 14.04 (Trusty Tahr). However, the script mentioned that it can be used for Ubunuu versions as old as 10.04 (see yellow heighted portion below). The acpica-tools package is part of the list of packages that the script tries to download (see below highlighted line below): UBUNTU=`cat /etc/issue.net | cut -d' ' -f2` HOST_ARCH=`uname -m` if [ ${HOST_ARCH} == "x86_64" ] ; then PKGS='gnupg flex bison gperf build-essential zip curl zlib1g-dev libc6-dev lib32ncurses5-dev x11proto-core-dev libx11-dev lib32z1-dev libgl1-mesa-dev g++-multilib mingw32 tofrodos python-markdown libxml2-utils xsltproc uboot-mkimage openjdk-6-jdk openjdk-6-jre vim-common python-parted python-yaml wget uuid-dev acpica-tools' else echo "ERROR: Only 64bit Host(Build) machines are supported at the moment." exit 1 fi if [[ ${UBUNTU} =~ "13." || ${UBUNTU} =~ "12.10" ]]; then #Install basic dev package missing in chrooted environments sudo apt-get install software-properties-common sudo dpkg --add-architecture i386 PKGS+=' libstdc++6:i386 git-core' elif [[ ${UBUNTU} =~ "12.04" || ${UBUNTU} =~ "10.04" ]] ; then #Install basic dev package missing in chrooted environments sudo apt-get install python-software-properties if [[ ${UBUNTU} =~ "12.04" ]]; then PKGS+=' libstdc++6:i386 git-core' else PKGS+=' ia32-libs libssl-dev libcurl4-gnutls-dev libexpat1-dev gettext' fi else echo "ERROR: Only Ubuntu 10.04, 12.* and 13.04 versions are supported." exit 1 fi If I remove the reference to the acpica-tools package (the blue highlighted section above), the script does progress further but the following error is hit (See green highlighted error message below): Your Name [Kevin Dious]: Your Email [kdious(a)eco-build1.qualcomm.com]: kdious(a)quicinc.com Your identity is: Kevin Dious <kdious(a)quicinc.com> is this correct [y/N]? y repo has been initialized in /local/mnt/workspace/vexpress-linaro-build-309/git-1.7.9.5/android cp: missing destination file operand after `.repo/manifest.xml' Try `cp --help' for more information. kdious@eco-build1:/local/mnt/workspace/vexpress-linaro-build-309$ >From looking at the script, it seems as if the following line is failing (See green highlighted line below): # download the android code ./repo init -u ${MANIFEST_REPO} -b ${MANIFEST_BRANCH} -m ${MANIFEST_FILENAME} --repo-url=git://android.git.linaro.org/tools/repo -g common,vexpress if [ ${EXACT} -eq 1 ] ; then rm .repo/manifest.xml cp $MANIFEST .repo/manifest.xml fi So I have a few questions: a) Is the acpica-tools package truly necessary in order to build Linaro Android? If so, how are you able to build it on versions of Ubuntu older than 14.04? b) IS there some workaround that we can use to get around the cp error referenced above? Any help that you can provide would be greatly appreciated. Thank you. Regards, Kevin Dious Qualcomm Innovation Center, Inc. 5755 Morehouse Drive San Diego, CA 92121-1714 (858) 651-2014 (Phone) (858) 845-7478 (Fax) kdious(a)quicinc.com [View Less]

10 years, 11 months

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-android May 2014