From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from sfi-mx-3.v28.ch3.sourceforge.com ([172.29.28.123] helo=mx.sourceforge.net) by h25xhf1.ch3.sourceforge.com with esmtp (Exim 4.69) (envelope-from ) id 1My07g-0002d2-As for ltp-list@lists.sourceforge.net; Wed, 14 Oct 2009 09:26:20 +0000 Received: from e28smtp02.in.ibm.com ([59.145.155.2]) by 3b2kzd1.ch3.sourceforge.com with esmtps (TLSv1:AES256-SHA:256) (Exim 4.69) id 1My07c-0005ag-RE for ltp-list@lists.sourceforge.net; Wed, 14 Oct 2009 09:26:20 +0000 Received: from d28relay05.in.ibm.com (d28relay05.in.ibm.com [9.184.220.62]) by e28smtp02.in.ibm.com (8.14.3/8.13.1) with ESMTP id n9E9Ptgx015504 for ; Wed, 14 Oct 2009 14:55:55 +0530 Received: from d28av01.in.ibm.com (d28av01.in.ibm.com [9.184.220.63]) by d28relay05.in.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id n9E9Ptg92580720 for ; Wed, 14 Oct 2009 14:55:55 +0530 Received: from d28av01.in.ibm.com (loopback [127.0.0.1]) by d28av01.in.ibm.com (8.14.3/8.13.1/NCO v10.0 AVout) with ESMTP id n9E9PsKu001608 for ; Wed, 14 Oct 2009 14:55:55 +0530 Date: Wed, 14 Oct 2009 14:55:54 +0530 From: Gautham R Shenoy Message-ID: <20091014092554.GA27706@in.ibm.com> References: <20091013074337.22414.20968.sendpatchset@localhost.localdomain> <20091013074346.22414.9377.sendpatchset@localhost.localdomain> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <20091013074346.22414.9377.sendpatchset@localhost.localdomain> Subject: Re: [LTP] [Patch 2/6] Developed new functions and fixed issues causing ilb test failure Reply-To: ego@in.ibm.com List-Id: Linux Test Project General Discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Errors-To: ltp-list-bounces@lists.sourceforge.net To: Poornima Nayak Cc: ltp-list@lists.sourceforge.net, svaidy@linux.vnet.ibm.com, arun@linux.vnet.ibm.com On Tue, Oct 13, 2009 at 01:13:46PM +0530, Poornima Nayak wrote: > +def is_multi_core(): > + ''' Return true if system has sockets has multiple cores > + ''' > + > + try: > + file_cpuinfo = open("/proc/cpuinfo", 'r') > + for line in file_cpuinfo: > + if line.startswith('siblings'): > + siblings = line.split(":") > + if line.startswith('cpu cores'): > + cpu_cores = line.split(":") > + break I assume your code works only on x86 machine for now. The corresponding kernel code found from arch/x86/kernel/cpu/proc.c is as follows ======================================================================== /* * Get CPU information for use by the procfs. */ static void show_cpuinfo_core(struct seq_file *m, struct cpuinfo_x86 *c, unsigned int cpu) { #ifdef CONFIG_SMP if (c->x86_max_cores * smp_num_siblings > 1) { seq_printf(m, "physical id\t: %d\n", c->phys_proc_id); seq_printf(m, "siblings\t: %d\n", cpumask_weight(cpu_core_mask(cpu))); seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id); seq_printf(m, "cpu cores\t: %d\n", c->booted_cores); seq_printf(m, "apicid\t\t: %d\n", c->apicid); seq_printf(m, "initial apicid\t: %d\n", c->initial_apicid); } #endif } ============================================================================ IIUC, 'siblings' tells us the number of hardware threads available within a socket and 'cpu cores' tells us the number of cores available within a socket. So, if you are checking if the socket has multiple cores or not, 'cpu cores > 1' would suffice, no ? Why do we need the following code below ? > + > + if int( siblings[1] ) == int( cpu_cores[1] ): ^^^^^^^^^^^^^^^^^^^^^ This only means that each core in the socket has only one processing unit or thread. > + if int( cpu_cores[1] ) > 1: > + multi_core = 1 > + else: > + multi_core = 0 > + else: > + num_of_cpus = int(siblings[1]) / int(cpu_cores[1]) ^^^^^^^^^^ This tells us the number of hardware threads per core. > + if num_of_cpus > 1: > + multi_core = 1 > + else: > + multi_core = 0 > + file_cpuinfo.close() > + return multi_core > + except Exception: > + print "Failed to check if system is multi core system" > + sys.exit(1) > + > def get_hyper_thread_count(): > ''' Return number of threads in CPU. For eg for x3950 this function > would return 2. In future if 4 threads are supported in CPU, this > @@ -153,6 +184,40 @@ def map_cpuid_pkgid(): > sys.exit(1) > > > +def generate_sibling_list(): > + ''' Routine to generate siblings list > + ''' > + try: > + for i in range(0, cpu_count): > + siblings_file = '/sys/devices/system/cpu/cpu%s' % i > + siblings_file += '/topology/thread_siblings_list' > + threads_sibs = open(siblings_file).read().rstrip() > + thread_ids = threads_sibs.split("-") A string '0-3' means that 0,1,2,3 are the siblings and not just 0 and 3. Also, you can have strings like '0,4' which means that only 0 and 4 are siblings. Thus I am not sure if the parsing that you're doing is sufficient to generate the list. > + > + if not thread_ids in siblings_list: > + siblings_list.append(thread_ids) > + except Exception, details: > + print "Exception in generate_siblings_list", details > + sys.exit(1) > + > +def get_siblings(cpu_id): > + ''' Return siblings of cpu_id > + ''' > + try: > + cpus = "" > + for i in range(0, len(siblings_list)): > + for cpu in siblings_list[i]: > + if cpu_id == cpu: > + for j in siblings_list[i]: > + # Exclude cpu_id in the list of siblings > + if j != cpu_id: > + cpus += j > + return cpus > + return cpus > + except Exception, details: > + print "Exception in get_siblings", details > + sys.exit(1) > + > def get_proc_data(stats_list): > ''' Read /proc/stat info and store in dictionary > ''' > @@ -168,18 +233,18 @@ def get_proc_data(stats_list): > sys.exit(1) > > def get_proc_loc_count(loc_stats): > - ''' Read /proc/stat info and store in dictionary > + ''' Read /proc/interrupts info and store in list > ''' > try: > file_procstat = open("/proc/interrupts", 'r') > for line in file_procstat: > - if line.startswith('LOC:'): > + if line.startswith(' LOC:') or line.startswith('LOC:'): > data = line.split() > for i in range(0, cpu_count): > # To skip LOC > loc_stats.append(data[i+1]) > - print data[i+1] > - file_procstat.close() > + file_procstat.close() > + return > except Exception, details: > print "Could not read interrupt statistics", details > sys.exit(1) > @@ -192,6 +257,8 @@ def set_sched_mc_power(sched_mc_level): > os.system('echo %s > \ > /sys/devices/system/cpu/sched_mc_power_savings 2>/dev/null' > % sched_mc_level) > + > + get_proc_data(stats_start) > except OSError, e: > print "Could not set sched_mc_power_savings to", sched_mc_level, e > sys.exit(1) > @@ -203,6 +270,8 @@ def set_sched_smt_power(sched_smt_level) > os.system('echo %s > \ > /sys/devices/system/cpu/sched_smt_power_savings 2>/dev/null' > % sched_smt_level) > + > + get_proc_data(stats_start) > except OSError, e: > print "Could not set sched_smt_power_savings to", sched_smt_level, e > sys.exit(1) > @@ -218,21 +287,36 @@ def set_timer_migration_interface(value) > print "Could not set timer_migration to ", value, e > sys.exit(1) > > -def trigger_ebizzy (sched_smt, stress, duration, background, pinned): > - ''' Triggers ebizzy workload for sched_mc=1 > - testing > +def get_job_count(stress, workload, sched_smt): > + ''' Returns number of jobs/threads to be triggered > ''' > + > try: > if stress == "thread": > threads = get_hyper_thread_count() I am assuming get_hyper_thread_count() returns the number of hyperthreads within a core. So, if stress = "thread", the number of software threads we create is equal to the numbef of h/w threads within a core ? > if stress == "partial": > threads = cpu_count / socket_count If stress = "partial", the number of software threads we create is equal to the number of h/w threads within a socket ? > + if is_hyper_threaded(): > + if workload == "ebizzy" and int(sched_smt) ==0: > + threads = threads / get_hyper_thread_count() Thus if the workload is ebizzy and sched_smt is 0, the number of software threads we create is equal to the number of cores within the socket, right ? > + if workload == "kernbench" and int(sched_smt) < 2: > + threads = threads / get_hyper_thread_count() > if stress == "full": > - threads = cpu_count > + threads = cpu_count > if stress == "single_job": > threads = 1 > duration = 180 > + return threads > + except Exception, details: > + print "get job count failed ", details > + sys.exit(1) > > +def trigger_ebizzy (sched_smt, stress, duration, background, pinned): > + ''' Triggers ebizzy workload for sched_mc=1 > + testing > + ''' > + try: > + threads = get_job_count(stress, "ebizzy", sched_smt) > olddir = os.getcwd() > path = '%s/utils/benchmark' % os.environ['LTPROOT'] > os.chdir(path) > @@ -282,23 +366,14 @@ def trigger_ebizzy (sched_smt, stress, d > print "Ebizzy workload trigger failed ", details > sys.exit(1) > > -def trigger_kernbench (sched_smt, stress, background, pinned): > +def trigger_kernbench (sched_smt, stress, background, pinned, perf_test): > ''' Trigger load on system like kernbench. > Copys existing copy of LTP into as LTP2 and then builds it > with make -j > ''' > olddir = os.getcwd() > try: > - if stress == "thread": > - threads = 2 > - if stress == "partial": > - threads = cpu_count / socket_count > - if is_hyper_threaded() and int(sched_smt) !=2: > - threads = threads / get_hyper_thread_count() > - if stress == "full": > - threads = cpu_count > - if stress == "single_job": > - threads = 1 > + threads = get_job_count(stress, "kernbench", sched_smt) > > dst_path = "/root" > olddir = os.getcwd() > @@ -335,24 +410,35 @@ def trigger_kernbench (sched_smt, stress > get_proc_loc_count(intr_start) > if pinned == "yes": > os.system ( 'taskset -c %s %s/kernbench -o %s -M -H -n 1 \ > - >/dev/null 2>&1' % (cpu_count-1, benchmark_path, threads)) > + >/dev/null 2>&1 &' % (cpu_count-1, benchmark_path, threads)) > + > + # We have to delete import in future > + import time > + time.sleep(240) > + stop_wkld("kernbench") > else: > if background == "yes": > os.system ( '%s/kernbench -o %s -M -H -n 1 >/dev/null 2>&1 &' \ > % (benchmark_path, threads)) > else: > - os.system ( '%s/kernbench -o %s -M -H -n 1 >/dev/null 2>&1' \ > - % (benchmark_path, threads)) > + if perf_test == "yes": > + os.system ( '%s/kernbench -o %s -M -H -n 1 >/dev/null 2>&1' \ > + % (benchmark_path, threads)) > + else: > + os.system ( '%s/kernbench -o %s -M -H -n 1 >/dev/null 2>&1 &' \ > + % (benchmark_path, threads)) > + # We have to delete import in future > + import time > + time.sleep(240) > + stop_wkld("kernbench") > > print "INFO: Workload kernbench triggerd" > os.chdir(olddir) > - #get_proc_data(stats_stop) > - #get_proc_loc_count(intr_stop) > except Exception, details: > print "Workload kernbench trigger failed ", details > sys.exit(1) > > -def trigger_workld(sched_smt, workload, stress, duration, background, pinned): > +def trigger_workld(sched_smt, workload, stress, duration, background, pinned, perf_test): > ''' Triggers workload passed as argument. Number of threads > triggered is based on stress value. > ''' > @@ -360,7 +446,7 @@ def trigger_workld(sched_smt, workload, > if workload == "ebizzy": > trigger_ebizzy (sched_smt, stress, duration, background, pinned) > if workload == "kernbench": > - trigger_kernbench (sched_smt, stress, background, pinned) > + trigger_kernbench (sched_smt, stress, background, pinned, perf_test) > except Exception, details: > print "INFO: Trigger workload failed", details > sys.exit(1) > @@ -434,7 +520,7 @@ def generate_report(): > print >> keyvalfile, "package-%s=%3.4f" % \ > (pkg, (float(total_idle)*100/total)) > except Exception, details: > - print "Generating reportfile failed: ", details > + print "Generating utilization report failed: ", details > sys.exit(1) > > #Add record delimiter '\n' before closing these files > @@ -454,20 +540,18 @@ def generate_loc_intr_report(): > > get_proc_loc_count(intr_stop) > > - print "Before substracting" > - for i in range(0, cpu_count): > - print "CPU",i, intr_start[i], intr_stop[i] > - reportfile = open('/procstat/cpu-loc_interrupts', 'a') > - print >> reportfile, "==============================================" > - print >> reportfile, " Local timer interrupt stats " > - print >> reportfile, "==============================================" > + reportfile = open('/procstat/cpu-loc_interrupts', 'a') > + print >> reportfile, "==============================================" > + print >> reportfile, " Local timer interrupt stats " > + print >> reportfile, "==============================================" > + > for i in range(0, cpu_count): > intr_stop[i] = int(intr_stop[i]) - int(intr_start[i]) > print >> reportfile, "CPU%s: %s" %(i, intr_stop[i]) > print >> reportfile > reportfile.close() > except Exception, details: > - print "Generating reportfile failed: ", details > + print "Generating interrupt report failed: ", details > sys.exit(1) > > def record_loc_intr_count(): > @@ -542,25 +626,24 @@ def validate_cpugrp_map(cpu_group, sched > modi_cpu_grp.remove(core_cpus[i]) > if len(modi_cpu_grp) == 0: > return 0 > - else: > + #This code has to be deleted > + #else: > # If sched_smt == 0 then its oky if threads run > # in different cores of same package > - if sched_smt_level == 1: > - sys.exit(1) > - else: > - if len(cpu_group) == 2 and \ > - len(modi_cpu_grp) < len(cpu_group): > - print "INFO:CPUs utilized not in a core" > - return 1 > - print "INFO: CPUs utilized is not in same package or core" > - return(1) > + #if sched_smt_level > 0 : > + #return 1 > else: > for pkg in sorted(cpu_map.keys()): > pkg_cpus = cpu_map[pkg] > - if pkg_cpus == cpu_group: > - return(0) > - > - return(1) > + if len(cpu_group) == len(pkg_cpus): > + if pkg_cpus == cpu_group: > + return(0) > + else: > + if int(cpus_utilized[0]) in cpu_map[pkg] or int(cpus_utilized[1]) in cpu_map[pkg]: > + return(0) > + > + return(1) > + > except Exception, details: > print "Exception in validate_cpugrp_map: ", details > sys.exit(1) > @@ -605,36 +688,70 @@ def verify_sched_domain_dmesg(sched_mc_l > print "Reading dmesg failed", details > sys.exit(1) > > -def validate_cpu_consolidation(work_ld, sched_mc_level, sched_smt_level): > +def get_cpu_utilization(cpu): > + ''' Return cpu utilization of cpu_id > + ''' > + try: > + for l in sorted(stats_percentage.keys()): What's the key used to index the elements of this hash table ? > + if cpu == stats_percentage[l][0]: > + return stats_percentage[l][1] > + return -1 > + except Exception, details: > + print "Exception in get_cpu_utilization", details > + sys.exit(1) > + > +def validate_cpu_consolidation(stress, work_ld, sched_mc_level, sched_smt_level): > ''' Verify if cpu's on which threads executed belong to same > package > ''' > cpus_utilized = list() > + threads = get_job_count(stress, work_ld, sched_smt_level) > try: > for l in sorted(stats_percentage.keys()): > #modify threshold > + cpu_id = stats_percentage[l][0].split("cpu") > + if cpu_id[1] == '': > + continue > + if int(cpu_id[1]) in cpus_utilized: > + continue > if is_hyper_threaded(): Does is_hyper_threaded() check the status of the system from procfs or sysfs every time it's called ? An easier way could be to obtain that once and save the state in a variable, if it's not done that way already. > - if stats_percentage[l][1] > 25 and work_ld == "kernbench": > - cpu_id = stats_percentage[l][0].split("cpu") > - if cpu_id[1] != '': > + if work_ld == "kernbench" and sched_smt_level < sched_mc_level: > + siblings = get_siblings(cpu_id[1]) > + if siblings != "": > + sib_list = siblings.split() > + utilization = int(stats_percentage[l][1]) > + for i in range(0, len(sib_list)): > + utilization += int(get_cpu_utilization("cpu%s" %sib_list[i])) > + else: > + utilization = stats_percentage[l][1] > + if utilization > 40: > cpus_utilized.append(int(cpu_id[1])) > + if siblings != "": > + for i in range(0, len(sib_list)): > + cpus_utilized.append(int(sib_list[i])) > else: > - if stats_percentage[l][1] > 70: > - cpu_id = stats_percentage[l][0].split("cpu") > - if cpu_id[1] != '': > - cpus_utilized.append(int(cpu_id[1])) > + # This threshold wuld be modified based on results It's easier to have these constants stored in some variables with appropriate names. That would be much easier to edit and update later instead of searching the code every time. > + if stats_percentage[l][1] > 40: > + cpus_utilized.append(int(cpu_id[1])) > else: > - if stats_percentage[l][1] > 70: > - cpu_id = stats_percentage[l][0].split("cpu") > - if cpu_id[1] != '': > + if work_ld == "kernbench" : > + if stats_percentage[l][1] > 50: > cpus_utilized.append(int(cpu_id[1])) > - cpus_utilized.sort() > + else: > + if stats_percentage[l][1] > 70: > + cpus_utilized.append(int(cpu_id[1])) > + cpus_utilized.sort() > print "INFO: CPU's utilized ", cpus_utilized > > + # If length of CPU's utilized is not = number of jobs exit with 1 > + if len(cpus_utilized) < threads: > + return 1 > + > status = validate_cpugrp_map(cpus_utilized, sched_mc_level, \ > sched_smt_level) > if status == 1: > print "INFO: CPUs utilized is not in same package or core" > + > return(status) > except Exception, details: > print "Exception in validate_cpu_consolidation: ", details > @@ -645,7 +762,8 @@ def get_cpuid_max_intr_count(): > try: > highest = 0 > second_highest = 0 > - global cpu1_max_intr, cpu2_max_intr > + cpus_utilized = [] > + > #Skipping CPU0 as it is generally high > for i in range(1, cpu_count): > if int(intr_stop[i]) > int(highest): > @@ -658,15 +776,19 @@ def get_cpuid_max_intr_count(): > if int(intr_stop[i]) > int(second_highest): > second_highest = int(intr_stop[i]) > cpu2_max_intr = i > + cpus_utilized.append(cpu1_max_intr) > + cpus_utilized.append(cpu2_max_intr) > + > for i in range(1, cpu_count): > if i != cpu1_max_intr and i != cpu2_max_intr: > diff = second_highest - intr_stop[i] > ''' Threshold of difference has to be manipulated ''' > if diff < 10000: > print "INFO: Diff in interrupt count is below threshold" > - return 1 > + cpus_utilized = [] > + return cpus_utilized > print "INFO: Interrupt count in other CPU's low as expected" > - return 0 > + return cpus_utilized > except Exception, details: > print "Exception in get_cpuid_max_intr_count: ", details > sys.exit(1) > @@ -675,14 +797,12 @@ def validate_ilb (sched_mc_level, sched_ > ''' Validate if ilb is running in same package where work load is running > ''' > try: > - status = get_cpuid_max_intr_count() > - if status == 1: > + cpus_utilized = get_cpuid_max_intr_count() > + if not cpus_utilized: > return 1 > - for pkg in sorted(cpu_map.keys()): > - if cpu1_max_intr in cpu_map[pkg] and cpu2_max_intr in cpu_map[pkg]: > - return 0 > - print "INFO: CPUs with higher interrupt count is not in same package" > - return 1 > + > + status = validate_cpugrp_map(cpus_utilized, sched_mc_level, sched_smt_level) > + return status > except Exception, details: > print "Exception in validate_ilb: ", details > sys.exit(1) > @@ -706,3 +826,14 @@ def reset_schedsmt(): > except OSError, e: > print "Could not set sched_smt_power_savings to 0", e > sys.exit(1) > + > +def stop_wkld(work_ld): > + ''' Kill workload triggered in background > + ''' > + try: > + os.system('pkill %s 2>/dev/null' %work_ld) > + if work_ld == "kernbench": > + os.system('pkill make 2>/dev/null') > + except OSError, e: > + print "Exception in stop_wkld", e > + sys.exit(1) -- Thanks and Regards gautham ------------------------------------------------------------------------------ Come build with us! The BlackBerry(R) Developer Conference in SF, CA is the only developer event you need to attend this year. Jumpstart your developing skills, take BlackBerry mobile applications to market and stay ahead of the curve. Join us from November 9 - 12, 2009. Register now! http://p.sf.net/sfu/devconference _______________________________________________ Ltp-list mailing list Ltp-list@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ltp-list