dmitriy.kazimirov/inxi
1
0
Fork 0
mirror of https://github.com/smxi/inxi.git synced 2025-01-19 08:57:57 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

New version, new tarball. Another stab at finally fixing the cpu / core count failures on fringe

Browse source 
cases. This required fixing some core logic assumptions that are not currently correct, particularly
on two cases, some xeon cpus fail to show core id for each core, showing 0 for all of them, second,
vm cpus do not show physical ids at all for at least intel, nor do they show core id. 

While we can't get HT totally reliable, particularly for vm xeon, since inxi has no way to know in
that case if a core is attached to a physical core or a virtual one, all of them being virtual in that
case, but still inxi is now reporting the correct number of cores, or threads in vm xeons, and is not
showing multicore cpus as single core, which was the main issue.

This required redoing the counter logic for the cpu/core/physical arrays, now they are set independently,
and can handle any of the others not being set, without creating an error or failure condition.

Also added in last check for a certain intel case where core id is 0 but > 1 physical cores exist, that
now also shows the correct cpu / core count.

While this is tested on many data sets of proc cpuinfo, it's still possible there is a fringe case I have
not seen that will trigger yet another unexpected behavior.
This commit is contained in:
inxi-svn 2013-06-20 00:29:19 +00:00
parent a1822fd07f
commit 854f348969
2 changed files with 83 additions and 52 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

106
inxi
View file

@ -1,8 +1,8 @@
#!/usr/bin/env bash #!/usr/bin/env bash
######################################################################## ########################################################################
#### Script Name: inxi #### Script Name: inxi
#### Version: 1.9.9 #### Version: 1.9.10
#### Date: June 16 2013 #### Date: June 19 2013
#### Patch Number: 00 #### Patch Number: 00
######################################################################## ########################################################################
#### SPECIAL THANKS #### SPECIAL THANKS
@ -3592,7 +3592,7 @@ get_cpu_ht_multicore_smp_data()
# in /proc/cpuinfo # in /proc/cpuinfo
local temp_array='' local temp_array=''
# note: known bug with xeon intel, they show core_id/physical_id as 0 for ht 4 core # note: known bug with xeon intel, they show a_core_id/physical_id as 0 for ht 4 core
if [[ $B_CPUINFO_FILE == 'true' ]]; then if [[ $B_CPUINFO_FILE == 'true' ]]; then
A_CPU_TYPE_PCNT_CCNT=( $( A_CPU_TYPE_PCNT_CCNT=( $(
gawk ' gawk '
@ -3601,16 +3601,22 @@ get_cpu_ht_multicore_smp_data()
IGNORECASE = 1 IGNORECASE = 1
num_of_cores = 0 num_of_cores = 0
num_of_processors = 0 num_of_processors = 0
num_of_cpus = 0 num_of_physical_cpus = 0
cpu_core_count = 0 cpu_core_count = 0
siblings = 0 siblings = 0
core_id[0] # these 3 arrays cannot be declared because that sets the first element
processor_id[0] # but leaving this here so that we avoid doing that in the future
cpu_id[0] # a_core_id = ""
type = "-" # a_processor_id = ""
iter = 0 # a_physical_id = ""
cpu_type = "-"
# note: we need separate iterators because some cpuinfo data has only
# processor, no core id or phys id
proc_iter = 0
core_iter = "" # set from actual NF data
phys_iter = "" # set from actual NF data
# needed to handle arm cpu, no processor number cases # needed to handle arm cpu, no processor number cases
count = 0 arm_count = 0
nr = 0 nr = 0
bArm = "false" bArm = "false"
bXeon = "false" bXeon = "false"
@ -3619,7 +3625,7 @@ get_cpu_ht_multicore_smp_data()
/^model name/ && ( $0 ~ /Xeon/ ) { /^model name/ && ( $0 ~ /Xeon/ ) {
bXeon = "true" bXeon = "true"
} }
# only do this once since sibling count does not change # only do this once since sibling count does not change.
/^siblings/ && ( bXeon == "true" ) && ( siblings == 0 ) { /^siblings/ && ( bXeon == "true" ) && ( siblings == 0 ) {
gsub(/[^0-9]/,"",$NF) gsub(/[^0-9]/,"",$NF)
if ( $NF != "" ) { if ( $NF != "" ) {
@ -3632,30 +3638,34 @@ get_cpu_ht_multicore_smp_data()
gsub(/,/, " ", $NF) gsub(/,/, " ", $NF)
gsub(/^ +| +$/, "", $NF) gsub(/^ +| +$/, "", $NF)
if ( $NF ~ "^[0-9]+$" ) { if ( $NF ~ "^[0-9]+$" ) {
processor_id[iter] = $NF a_processor_id[proc_iter] = $NF
} }
else { else {
if ( $NF ~ "^ARM" ) { if ( $NF ~ "^ARM" ) {
bArm = "true" bArm = "true"
} }
count += 1 arm_count += 1
nr = count - 1 nr = arm_count - 1
processor_id[iter] = nr a_processor_id[proc_iter] = nr
} }
proc_iter++
}
} # array of physical cpu ids, note, this will be unset for vm cpus in many cases
# array of physical cpus ids # because they have no physical cpu, so we cannot assume this will be here.
/^physical/ { /^physical/ {
cpu_id[iter] = $NF phys_iter = $NF
a_physical_id[phys_iter] = $NF
} }
# array of core ids # array of core ids, again, here we may have HT, so we need to create an array of the
# actual core ids. As With physical, we cannot assume this will be here in a vm
/^core id/ { /^core id/ {
core_id[iter] = $NF core_iter = $NF
iter++ a_core_id[core_iter] = $NF
} }
# this will be used to fix an intel glitch if needed, cause, intel # this will be used to fix an intel glitch if needed, cause, intel
# sometimes reports core id as the same number for each core, 0 # sometimes reports core id as the same number for each core,
# so if cpu cores shows greater value than number of cores, use this # so if cpu cores shows greater value than number of cores, use this.
/^cpu cores/ { /^cpu cores/ {
cpu_core_count = $NF cpu_core_count = $NF
} }
@ -3664,49 +3674,41 @@ get_cpu_ht_multicore_smp_data()
## only unique numbers required ## only unique numbers required
## this is to get an accurate count ## this is to get an accurate count
## we are only concerned with array length ## we are only concerned with array length
i = 0 i = 0
## count unique processors ## ## count unique processors ##
# note, this fails for intel cpus at times # note, this fails for intel cpus at times
for ( i in processor_id ) { for ( i in a_processor_id ) {
procHolder[processor_id[i]] = 1
}
for ( i in procHolder ) {
num_of_processors++ num_of_processors++
} }
i = 0 i = 0
## count unique physical cpus ## ## count unique physical cpus ##
for ( i in cpu_id ) { for ( i in a_physical_id ) {
cpuHolder[cpu_id[i]] = 1 num_of_physical_cpus++
} }
for ( i in cpuHolder ) {
num_of_cpus++
}
i = 0 i = 0
## count unique cores ## ## count unique cores ##
for ( i in core_id ) { for ( i in a_core_id ) {
coreHolder[core_id[i]] = 1
}
for ( i in coreHolder ) {
num_of_cores++ num_of_cores++
} }
# xeon may show wrong core / physical id count, if it does, fix it # xeon may show wrong core / physical id count, if it does, fix it. A xeon
if ( num_of_cores == 1 && bXeon == "true" && siblings > 1 ) { # may show a repeated core id : 0 which gives a fake num_of_cores=1
if ( bXeon == "true" && num_of_cores == 1 && siblings > 1 ) {
num_of_cores = siblings/2 num_of_cores = siblings/2
} }
# final check, override the num of cores value if it clearly is wrong # final check, override the num of cores value if it clearly is wrong
# and use the raw core count and synthesize the total instead of real count # and use the raw core count and synthesize the total instead of real count
if ( ( num_of_cores == 1 ) && ( cpu_core_count * num_of_cpus > 1 ) ) { if ( ( num_of_cores == 0 ) && ( cpu_core_count * num_of_physical_cpus > 1 ) ) {
num_of_cores = cpu_core_count * num_of_cpus num_of_cores = cpu_core_count * num_of_physical_cpus
} }
# last check, seeing some intel cpus and vms with intel cpus that do not show any # last check, seeing some intel cpus and vms with intel cpus that do not show any
# core id data at all, or siblings. # core id data at all, or siblings.
if ( num_of_cores == 0 && num_of_processors > 0 ) { if ( num_of_cores == 0 && num_of_processors > 0 ) {
num_of_cores = num_of_processors num_of_cores = num_of_processors
} }
# print "NoCpu: " num_of_physical_cpus
# print "NoCores: " num_of_cores
# print "NoProc:" num_of_processors
# print "CpuCoreCount:" cpu_core_count
#################################################################### ####################################################################
# algorithm # algorithm
# if > 1 processor && processor id (physical id) == core id then Hyperthreaded (HT) # if > 1 processor && processor id (physical id) == core id then Hyperthreaded (HT)
@ -3716,25 +3718,25 @@ get_cpu_ht_multicore_smp_data()
if ( num_of_processors > 1 || ( bXeon == "true" && siblings > 0 ) ) { if ( num_of_processors > 1 || ( bXeon == "true" && siblings > 0 ) ) {
# non-multicore HT # non-multicore HT
if ( num_of_processors == (num_of_cores * 2) ) { if ( num_of_processors == (num_of_cores * 2) ) {
type = type "HT-" cpu_type = cpu_type "HT-"
} }
else if ( bXeon == "true" && siblings > 1 ) { else if ( bXeon == "true" && siblings > 1 ) {
type = type "HT-" # num_of_processors num_of_cores num_of_cpus cpu_type = cpu_type "HT-"
} }
# non-HT multi-core or HT multi-core # non-HT multi-core or HT multi-core
if (( num_of_processors == num_of_cores) || ( num_of_cpus < num_of_cores)) { if (( num_of_processors == num_of_cores) || ( num_of_physical_cpus < num_of_cores)) {
type = type "MCP-" cpu_type = cpu_type "MCP-"
} }
# >1 cpu sockets active # >1 cpu sockets active
if ( num_of_cpus > 1 ) { if ( num_of_physical_cpus > 1 ) {
type = type "SMP-" cpu_type = cpu_type "SMP-"
} }
} }
else { else {
type = type "UP-" cpu_type = cpu_type "UP-"
} }
print type " " num_of_cpus " " num_of_cores print cpu_type " " num_of_physical_cpus " " num_of_cores
} }
' $FILE_CPUINFO ) ) ' $FILE_CPUINFO ) )
fi fi

29
inxi.changelog
View file

@ -1,3 +1,32 @@
=====================================================================================
Version: 1.9.10
Patch Version: 00
Script Date: June 19 2013
-----------------------------------
Changes:
-----------------------------------
New version, new tarball. Another stab at finally fixing the cpu / core count failures on fringe
cases. This required fixing some core logic assumptions that are not currently correct, particularly
on two cases, some xeon cpus fail to show core id for each core, showing 0 for all of them, second,
vm cpus do not show physical ids at all for at least intel, nor do they show core id.
While we can't get HT totally reliable, particularly for vm xeon, since inxi has no way to know in
that case if a core is attached to a physical core or a virtual one, all of them being virtual in that
case, but still inxi is now reporting the correct number of cores, or threads in vm xeons, and is not
showing multicore cpus as single core, which was the main issue.
This required redoing the counter logic for the cpu/core/physical arrays, now they are set independently,
and can handle any of the others not being set, without creating an error or failure condition.
Also added in last check for a certain intel case where core id is 0 but > 1 physical cores exist, that
now also shows the correct cpu / core count.
While this is tested on many data sets of proc cpuinfo, it's still possible there is a fringe case I have
not seen that will trigger yet another unexpected behavior.
-----------------------------------
-- Harald Hope - Wed, 19 Jun 2013 17:22:42 -0700
===================================================================================== =====================================================================================
Version: 1.9.9 Version: 1.9.9
Patch Version: 00 Patch Version: 00