dmitriy.kazimirov/inxi
1
0
Fork 0
mirror of https://github.com/smxi/inxi.git synced 2024-11-16 16:21:39 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Huge refactor of CPU physical/core/cache logic. That was some very old logic

Browse source 
with a lot of hacks and patches, but it had never been actually rewritten to
take advantage of Perl's far more powerful and robust data structures and logic.
This caused a continuous stream of error cases in subtle ways, or not so subtle,
and fixes were just changing how the errors manifested.

Tnanks very much to linuxquestions.org Slackware forum people for massive help,
and also to linux.org forum members for ongoing help and data and debugging.

Note Changes 5, change of default widths in display to 80 columns, and out (aka,
console, or ssh into remote system), 100. You can still use other widths if you
like something wider using the configuration options shown. Also upped max
columns wrapping line starters to own rows to 110 columns from 90, again the
idea being to make output more readable to other users when posted in public.

I've been thinking of this change for a long time, but was hoping -y would
register with users, but it hasn't gained enough traction, so the result is way
too many super hard to read issue reports, forum posts, linux kernel issues,
etc, it's honestly gotten sort of embarrassing because they make it look like
inxi has bad output. Sidescrolling code blocks in forums in particular are
absurdly hard to read and scan rapidly for data.

Going along with the width and indentation changes, for most main row types, if
they wrap to a second row, they are further indented 2 column2, to make it
easier to see what they belong to. The two levels of indentation contain more
useful visual cues as to what belongs to what.

There was a temptation to release this as either 3.4.00 or 4.0.00 but in the
end, I decided to follow the numbering rules, and to just roll it to 3.3.10
since there aren't really any primary new features even though CPU was basically
rewritten in large part, and big parts of inxi were also changed, upgraded, and
enhanced. But no truly new features, just some display control items like -Y,
--indent, --indents.

I hope this refactor meets its primary goals, and that the new defaults for
display help resolve public posting issues which have grown increasingly
annoying for anyone trying to read those pasted in too long outputs.

--------------------------------------------------------------------------------
KNOWN ISSUES:

1. Android seems to have CPU cache data, but does not show any 'size' item. But
it does have the other data for each cache type internally, which is odd.

2. In some instances, the parent key:value pairs with '' as value, those are
parents of children key:value pairs, are left hanging at end of line, with the
children on the following line. This can look awkward, but in other cases,
actually looks very good, it depends if it's at the start or end of the line.

I won't say this is not correctable, but it would be very difficult, and outside
the scope of this release, but that is something that I may look at for a future
release now that the output generator logic was reworked slightly for Change 5b.

It's tricky though, because in cases where it's the first item on the line, you
want that behavior, but when it's last, you don't. But this may be worth
revisiting in the future.

3. In some cases, -Y + -y1 may lead to the start of the block scrolling off the
top of the visible screen. This isn't really correctable, so if that's an issue
for you, just don't use -y1 with -Y and all the output will wrap nicely.

4. There is an unaccountable ~10-20ms delay reading cpufreq/scaling_cur_freq,
per thread/core, which really adds up on high thread count CPUs. There is a
workaround in inxi to use cpufreq_cur_freq if it is readable, ie, if you are
root or use sudo, but to fallback to scaling_cur_freq if can't read cpuinfo_...

This is a drag, and really looks like a kernel bug, or a frequency driver bug.

--------------------------------------------------------------------------------
BUGS:

1. 3.3.09 and 3.3.10 CPU bug fixes:
* Failed to filter out certain virtual machine CPU core speeds, and showed more
speeds than the instance actually has. Noticed this with KVM running on Xeon
CPUs.
* For many cases, L2 cache, particularly for Intel, was completely wrong, it was
showing L3 caches, or L3+L2. Failed to handle cases where L2 cache belongs to
more than 1 core, except for using a crude hack for AMD Bulldozer microarch.
Older Intel Core 4 core CPUs would sometimes be 1 L2 per die, and the 4 cores
were actually 2 core duo cpu dies, with one L2 cache per die.
* Shows wrong core count for complex core complexes like those found in Intel
Alder Lake, now shows correct count of actual cores, regardless of the MT or ST
state of each core.
* Showed invalid L3 cache values in some legacy cpus that had no L3 cache, that
is due to a bug in the dmidecode data itself. Solution is to never use dmidecode
cache data if any other valid L1, L2, or L3 cache data found for Linux, and to
only use dmidecode data for bsds if no L1, or L2, or L3 data found. Or if forced
with --dmidecode.

2. An unfortunately long standing bug found and fixed, thanks slackware users!
cp_cpu_arch was, and has been for a while, failing to convert hex stepping to
decimal, or test if the string it gets is even a possible hex value, this
resulted in all Intel CPUs with stepping > 9 failing to ID correctly for cpu
arch.

3. In a related bug, hex to decimal tool used to create --admin hex/decimal
output for family/model/stepping was also not testing if the string was an
actual valid hex number. Case in particular, power pc with revision field
contained a long string, which was of course not a valid hex number, and that
tripped a Perl error when it was asked to convert a non hex string to decimal.

4. Long standing bug found while doing Change 5: inxi actually never applied
separate in/out of dispay to widths because using a legacy boolean that was not
updated, so it was always using out of display widths.

--------------------------------------------------------------------------------
FIXES:

1. Incorrectly calling PowerPC 'revision' 'stepping' for -Ca, that is now stored
as $cpu{'revision'} to avoid mixing up the logics there. For PowerPC shows as
rev: [string].

2. Microarch:
* AMD family 15, model 2 as bulldozer, actually piledriver.
* AMD family 17, model 18, was supposed to be zen/zen+, since I can't tell those
apart, seen stepping 1 is zen+, but had incorrect match.
* Intel family 6, model 25, stepping 2 as nehalem, should have been westmere.
* Changed Penryn to Core Penryn, intel family 6, model 17
* misc other micro arch fine tunings.

3. Code fix 8, switched to global %risc for arm, mips, ppc, riscv, sparc. This
corrects many sloppy handlers, and makes all risc processing the same, and calls
device tree readers for all risc systems, not just arm or arm and mips.

4. In cases where bogomips were 0 due to false values in risc results, show N/A.

5. Removed all attempts to guess at what /proc/cpuinfo cache size: refers to,
it can literally be anything, a per core L1, a per core or cpu L2, or an L3.
So applying any math to it is just a random guess at that point. If any L1,2,3
cache data is found, don't use the cache: value at all, but that will only be
present if no /sys data was found anyway, and if cpuinfo had no specific cache
type fields, only generic cache.

6. Added failsafe tests for stepping and model id before doing conversion to
hex. Make sure integer!

7. Added L1 D cache, was only using I cache for BSDs. Output will show total for
L1 A + L1 D. No idea why I didn't use L1 D, makes little sense, but that's how
it goes.

8. Made bogomips tests more granular, now only rejects low sub 50 bogomips
values if %risc cpu type. Legacy ancient cpus like 486 could and did have
bogomip counts below 50.
https://tldp.org/HOWTO/BogoMips/bogo-list.html

9. See Enhancement 12 as well. If OpenBSD, which has no per core data or
physical cpu data, is running on MT capable cpu, but for security OpenBSD has
disabled MT, will now force MT to be not shown via the hw.smt value. This
removes a small glitch that would have bothered OpenBSD users who know that
OpenBSD has disabled MT for security purposes.

10. Changed BSD hack to use L2 cache totals to deduce > 1 physical cpus, that
was flat out dumb, since we can just use dmidecode type 4 to iterate physical
cpu counts and skip the pointless logic. Thus, if dmidecode, and if > 1
dmidecode type 4 found, and if physical cpu counts equal 1, then replace the
found counts with the dmidecode physical cpu counts.

11. Corrected bad assumption that threads would always be 2 per core for MT
tests. Still no way to reliably determine threads per core for non x86 cpus like
powerpc however, but those are very fringe and should rarely be an issue since
that data is only missing on very old linux now I think.

12. Fixed 'parameters:' going to its own line with -Sa, that wasn't supposed to.
-S is two lines, the kernel / host stuff, and the desktop/console/distro stuff.

13. Fixed case when key: value first word plus other parts of line longer than
max width, failed to wrap as expected.

14. Added start/end ' and " start / end \s to main filters.

--------------------------------------------------------------------------------
ENHANCEMENTS:

1. CPU: most Linux will now show L1 and L3 cache with -Cx without needing
sudo/root, and it will be more accurate than ever before.

2. CPU: shows per CPU L1/L2/L3 totals, and shows actual full system physical
processor count * L1/L2/L3 total in parentheses, like: L2: 1.5 MiB (3 MiB).

3. CPU: A long standing annoyance, previously for main CPU 'Speed:' item, showed
the fastest core speed found, now shows avg: [speed] and with -Cx, shows the
'high:' as well if > 1 cores, and if 1 or more cores have a higher speed than
the other(s).

4. CPU: Handles advanced cases of new architectures, like Alder Lake with
Performance and Efficiency cores, future Zen, and existing ARM CPUs with 2 or
more different core sets, with different max/min frequencies. Previously a hack
was used to handle only ARM CPUs with this type of architecture. Will show
correct CPU core counts, which previous inxi versions would fail to do for Alder
Lake type scenarios of 8 single threaded CPUs and 4-8 multithreaded )MT)
perforance cores.

This should also in theory show different the different min/max speeds if they
were detected. Those did not seem to be set correctly in Alder Lake sample data
I saw however, P and E cores were set to the same min/max speeds.

5. Added CPU types MST (Multi+Single Thread), AMP (Asymmetric Multi Processing),
and AMCP (Asymmetrical Multi Core Processor). This will be applied to any CPU
that has this type of complex topology that has been dynamically detected, like
Alder Lake or different core count or min/max speed RISC CPUs.

6. CPU: shows with -Ca for cases where different L1/L2/L3 caches found per
physical CPU, as with Alder Lake, but also many other variants that were poorly
or not at all handled before, how many of each cache type (L1 Data, instruction)
were found, otherwise will show how many of each cache were found.

7. CPU: shows with -Ca in Topology: report, for cases like Alder Lake with
different core types in one physical CPU (type: MST AMCP), the number that are
single threaded (st) and number that are multi-threaded (mt).

8. Basic support for rsyc-v systems, going along with code fix 8, fix 3, now
it's easy to add this type of support.

9. Added shortcut options for --filter-label (--zl), --filter-uuid (--zu), and
andded new filter option, --filter-vulnerabilities (--zv). The latter is added
by request, a decent idea to have option to not show cpu vulnerabilities.

10. Going with fix 7, switched to a sort of pseudo L1 d/i with desc report for
any BSD with L1 I/D cache found, or elbrus cache0 (icache) / cache1 (d cache).
Elbrus should hopefully be handled by the /sys tool. Guesses on the L1 are ok,
since those are almost always per core, so it's fine. Didn't expect to enhance
any BSD cpu data this time around, but there you go!! If they have the data,
then it will be used. Not going to go overboard though in that, quite useless
overall since usually can't see how many CPUs are present, at least not usually.

11. For -Ca, full CPU topology report if any complex topogy is detected,
otherwise shows the same basic Info: 2x 6-core or Info: dual core as before, no
point in wasting a line for something with no more data than the short string.
Complex types include MT CPUs since they will have different thread counts etc,
and will have 2 or more threads per core, which will also be listed.

12. If smt status is defined (0/1), shows smt: enabled|disabled in Topology
section, can be useful for systems with disabled MT, but supporting it. If no
topology data found (OpenBSD for example), for -C shows 'smt: disabled' after
'type:' section, and enabled if -Cxxx (since MT really already tells you that).

13. For -Ca Speed: report, added 'governor:' item, if found. Can show 1 or more
active governors.

14. Output height (in lines) control: -Y [-2|-1|0|1-xxx]]. This lets you break
up any of the output into whatever number of lines you want. Also useful out of
DISPLAY for reading -h options menu items etc.

It came tp my attention that the long standing shift+pgup/pgdown (aka
'softscrollback) behavior had stopped working, and in fact has been removed from
the current Linux kernel, at least until it is rewritten to be more clean and
understandable. Read more about it in these kernel post/commit messages:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=50145474f6ef4a9c19205b173da6264a644c7489
https://lwn.net/ml/linux-kernel/CAHk-=whe4ZdTdCebneWqC4gSQZwsVJ5-Emg0BucOGCwPhOAJpw@mail.gmail.com/

Options for -Y are:
* -Y 0 or -Y: Set maximum block height to terminal line height.
* -Y [1-xxx]: Set maximum block line height to given integer.
* -Y -1: Print out one primary data block item at a time, with -F for example.
* -Y -2: Restore default unlimited height if LINES_MAX configuration item used.

15. And finally, more disk vendors/vendor ids. As usual. As expected.

--------------------------------------------------------------------------------
CHANGES:

1. If /sys or /proc/cpuinfo speed data available:
* For -b CPU item:
   speed: [speed MHz] min/max: [min]/[max] MHz
  becomes:
   speed (MHz): avg: [speed] min/max: [min]/[max]
* For -C, Speed item
   Speed: [speed MHz] min/max: [min]/[max] MHz Cores (MHz): ...
  becomes:
   Speed (MHz): avg: [speed] min/max: [min]/[max] cores: ...
* For short form, shows speed/min/max but uses average speed if available.

For -b and -C, only shows one MHz in Speed line starter, which slightly shortens
the line even with the added 'avg:' item since 3 MHz are replaced with 1.

2. Going with change 1, now the 'avg:' item shows not the fastest cpu speed
found, which was the case before, but shows an average of all cpu speeds found.
Showing the fastest made some sense back in the days of single core, or even
dual core CPUs, but makes little sense today with many core/threaded cpus.

With -x, it will show the high: [speed] item as well, after 'avg:'.

3. By suggestion, wrapped first Type item in Vulnerabilities to its own line,
that's a verbose --admin option after all, no need to save lines!

4. Going along with Fix 5, give up on trying to pretend we can guess at L2
cache, now if only 'cache' data was available from cpuinfo, will just say:
cache: [cache size]
 note: check

and call it a day.

5a. Change default width to 80 columns, in and out of display. Too many users
are posting horribly wrapping inxi output in forums, issue trackers, etc, and it
frankly makes inxi look really bad, creates awful side scrolling code boxes,
etc. So now default widths in and out of console are 80 (since often data is
generated in SSH or out of X/Wayland) for issues.

This essentially makes -y 80/-y the default width. This is what I've been using
for a few years now, and after seeing far too many side scrolling or badly
wrapping inxi outputs online, I think it's probably time to just force 80 column
widths as default and call it a day.

You can change these new defaults using configuration options (these are the
previous options, though due to a bug, COLS_MAX_CONSOLE was never being used):

 COLS_MAX_CONSOLE=115 # in display, terminal client max width
 COLS_MAX_IRC=100
 COLS_MAX_NO_DISPLAY=130 # not in display, no X/Wayland running

5b. Changed output wrapped indent to 1 column from 2, and make second and
greater rows of a line indent +1 to make it more clear that it is a child row of
its parent row. Note that because no arg short form, -S, and -I are special
types of rows, this behavior is not used, they just print out as usual. This 1
column indent also applies to -y1, making for a little more data per line but
more readable and easy to follow.

6. If > 1 physical cpu detected, no longer uses single/dual/triple/quad core
strings, rather uses: 2x 2-core. Also uses lower case -core, not -Core.

7. Only show die counts for CPU (on rare occasions > 1 found) with -xx. Not
particularly important bit of data afterall.

8. Make L1, L3 cache data show with -Cx, not -Cxx, now that it's working well.

9. Removed CPU die for -Cxx, that's only going to show with -Ca now.

10. If -Ca, and if certain complexity conditions are met, shows a separate
Topology line rather than the Info: 6-core type item. For -b, short, -Cx, -Cxxx
shows the Info: topology short form.

11. Bogomips always shows before flags data, whether -f or just -Cx trips flag
output.

12. Flags/Features now shows in the same place, under Speeds: always, whether
-Cx shortlist, or -Cf full list. Makes more sense that way, and code is much
cleaner too.

13. Bogomips, being essentially bogus units of speed for cpu, are moved into
Speed: report.

--------------------------------------------------------------------------------
DOCUMENTATION:

1. Updated man/help for new CPU extra data options and output changes.

2. Cleaned up and added sample outputs for man CPU items.

3. Now that doas is getting into Linux distros, removed all mentions of doas as
a BSD option, and made it a general doas/sudo item. Glad to see doas making it
into linux distros, it's a good tool, much easier to configure and use than
sudo. Good job OpenBSD guys. Note that inxi already has had full doas support
for a while now, but this finalizes it, and makes it fully agnostic. Internally
doas is actually preferred over sudo, by the way.

4. Added documention items for INDENT (--indent), INDENT_MIN (--inident-min).

5. Re-ordered help menu and man page, created new Filters and Output Controls
sections to make stuff easier to find. In man page, also added on top a list of
OPTIONS sections to make finding stuff easier.

--------------------------------------------------------------------------------
CODE:

1. Removed legacy /sys cpu functions: cpu_bugs_sys(); get_boost_status();
set_cpu_speeds_sys(). cpu_speeds() is deprecated and now will only be used for
legacy Linux and BSDs if they had any per core speeds found; get_caches() was
only a placeholder for the full featured cpu_sys data source, and was removed;
cpu_speeds() no longer needed, integrated into other logic; cpu_dies_sys()
removed, integrated into other logic.

This logic is now integrated into cpu_data_sys() data generator.

2. Changed the main CpuItems functions to use array/hash references, not passing
full hashes or arrays in most cases now.

3. For machine_data_soc(), switched to CpuItem::cpuinfo_grabber() which then
sets the global @cpuinfo and %cpuinfo_machine items, which will be used again
in Cpu if cpu data is requested. This gets rid of a full parsing of cpuinfo
just to get the machine data section, and also makes it so cpuinfo in cpu does
not need to worry about the machine data block, which is not related to the
processor blocks anyway, that was always a hack done by the kernel guys to toss
that SOC data somewhere as far as I can tell.

4. New tools:
* either_or() - takes a list, and returns the first defined element of list.
* regex_range() - generate ranges from comma, space, or ranges like 2-29, or any
combination of those, like 3,6,12-29

5. Added --force cpuinfo to bypass all /sys based cpu logic, useful for testing
to see what would have happened using old logic.

6. Added --dbg switches 39, 40, 41, for the new cpu sys data features, also made
more consistent --dbg 8 and --dbg 38 switches.

7. Added sys/cpuinfo pair debugger to support debugging complex sys/cpuinfo
issues.

8. Got rid of $b_arm,$b_mips,$b_ppc,$b_sparc, replaced with global %risc, also
added $risc{'riscv'} type. this makes general risc type feature testing a lot
easier since inxi can either test for %risc defined, or for a specific type of
risc cpu. This is much cleaner, and use $risc{'id'} for print purposes, which
got rid of a lot of tests. Also made all risc tests consistent, some were ARM
only, or arm/mips, but were supposed to be for all risc cpus.

9. Set help menu code to roughly 80 columns width assuming 2 space tab
indentation.

10. Changed all xxx_cleaner subs to clean_xxx, all filter subs to filter_xxx,
and row_defaults() to message().

11. Dumped redundant fallback logic in get_kernel_bits, if first getconf method
fails, use $sys_bits, and call it good, it was repeating the 32/64 bit tests
pointlessly.

12. Cleaned up print_data() to allow for more fine tuned indentation for the new
2 indent levels.

13. Made help menu code more or less wrap to 80 columns, or close. Ongoing to
bring to 80 columns where practical, but never at expense of clarity or logic.

--------------------------------------------------------------------------------
-- Harald Hope - Tue, 13 Dec 2021 10:25:49 -0800
This commit is contained in:
Harald Hope 2021-12-14 10:35:42 -08:00
parent 2feaf0b853
commit 4c3ab65d46
3 changed files with 4199 additions and 2757 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5685
inxi
View file

File diff suppressed because it is too large Load diff

834
inxi.1
View file

File diff suppressed because it is too large Load diff

433
inxi.changelog
View file

@ -1,3 +1,436 @@
================================================================================
Version: 3.3.10
Patch: 01
Date: 2021-12-13
--------------------------------------------------------------------------------
RELEASE NOTES:
--------------------------------------------------------------------------------
Huge refactor of CPU physical/core/cache logic. That was some very old logic
with a lot of hacks and patches, but it had never been actually rewritten to
take advantage of Perl's far more powerful and robust data structures and logic.
This caused a continuous stream of error cases in subtle ways, or not so subtle,
and fixes were just changing how the errors manifested.
Tnanks very much to linuxquestions.org Slackware forum people for massive help,
and also to linux.org forum members for ongoing help and data and debugging.
Note Changes 5, change of default widths in display to 80 columns, and out (aka,
console, or ssh into remote system), 100. You can still use other widths if you
like something wider using the configuration options shown. Also upped max
columns wrapping line starters to own rows to 110 columns from 90, again the
idea being to make output more readable to other users when posted in public.
I've been thinking of this change for a long time, but was hoping -y would
register with users, but it hasn't gained enough traction, so the result is way
too many super hard to read issue reports, forum posts, linux kernel issues,
etc, it's honestly gotten sort of embarrassing because they make it look like
inxi has bad output. Sidescrolling code blocks in forums in particular are
absurdly hard to read and scan rapidly for data.
Going along with the width and indentation changes, for most main row types, if
they wrap to a second row, they are further indented 2 column2, to make it
easier to see what they belong to. The two levels of indentation contain more
useful visual cues as to what belongs to what.
There was a temptation to release this as either 3.4.00 or 4.0.00 but in the
end, I decided to follow the numbering rules, and to just roll it to 3.3.10
since there aren't really any primary new features even though CPU was basically
rewritten in large part, and big parts of inxi were also changed, upgraded, and
enhanced. But no truly new features, just some display control items like -Y,
--indent, --indents.
I hope this refactor meets its primary goals, and that the new defaults for
display help resolve public posting issues which have grown increasingly
annoying for anyone trying to read those pasted in too long outputs.
--------------------------------------------------------------------------------
KNOWN ISSUES:
1. Android seems to have CPU cache data, but does not show any 'size' item. But
it does have the other data for each cache type internally, which is odd.
2. In some instances, the parent key:value pairs with '' as value, those are
parents of children key:value pairs, are left hanging at end of line, with the
children on the following line. This can look awkward, but in other cases,
actually looks very good, it depends if it's at the start or end of the line.
I won't say this is not correctable, but it would be very difficult, and outside
the scope of this release, but that is something that I may look at for a future
release now that the output generator logic was reworked slightly for Change 5b.
It's tricky though, because in cases where it's the first item on the line, you
want that behavior, but when it's last, you don't. But this may be worth
revisiting in the future.
3. In some cases, -Y + -y1 may lead to the start of the block scrolling off the
top of the visible screen. This isn't really correctable, so if that's an issue
for you, just don't use -y1 with -Y and all the output will wrap nicely.
4. There is an unaccountable ~10-20ms delay reading cpufreq/scaling_cur_freq,
per thread/core, which really adds up on high thread count CPUs. There is a
workaround in inxi to use cpufreq_cur_freq if it is readable, ie, if you are
root or use sudo, but to fallback to scaling_cur_freq if can't read cpuinfo_...
This is a drag, and really looks like a kernel bug, or a frequency driver bug.
--------------------------------------------------------------------------------
BUGS:
1. 3.3.09 and 3.3.10 CPU bug fixes:
* Failed to filter out certain virtual machine CPU core speeds, and showed more
speeds than the instance actually has. Noticed this with KVM running on Xeon
CPUs.
* For many cases, L2 cache, particularly for Intel, was completely wrong, it was
showing L3 caches, or L3+L2. Failed to handle cases where L2 cache belongs to
more than 1 core, except for using a crude hack for AMD Bulldozer microarch.
Older Intel Core 4 core CPUs would sometimes be 1 L2 per die, and the 4 cores
were actually 2 core duo cpu dies, with one L2 cache per die.
* Shows wrong core count for complex core complexes like those found in Intel
Alder Lake, now shows correct count of actual cores, regardless of the MT or ST
state of each core.
* Showed invalid L3 cache values in some legacy cpus that had no L3 cache, that
is due to a bug in the dmidecode data itself. Solution is to never use dmidecode
cache data if any other valid L1, L2, or L3 cache data found for Linux, and to
only use dmidecode data for bsds if no L1, or L2, or L3 data found. Or if forced
with --dmidecode.
2. An unfortunately long standing bug found and fixed, thanks slackware users!
cp_cpu_arch was, and has been for a while, failing to convert hex stepping to
decimal, or test if the string it gets is even a possible hex value, this
resulted in all Intel CPUs with stepping > 9 failing to ID correctly for cpu
arch.
3. In a related bug, hex to decimal tool used to create --admin hex/decimal
output for family/model/stepping was also not testing if the string was an
actual valid hex number. Case in particular, power pc with revision field
contained a long string, which was of course not a valid hex number, and that
tripped a Perl error when it was asked to convert a non hex string to decimal.
4. Long standing bug found while doing Change 5: inxi actually never applied
separate in/out of dispay to widths because using a legacy boolean that was not
updated, so it was always using out of display widths.
--------------------------------------------------------------------------------
FIXES:
1. Incorrectly calling PowerPC 'revision' 'stepping' for -Ca, that is now stored
as $cpu{'revision'} to avoid mixing up the logics there. For PowerPC shows as
rev: [string].
2. Microarch:
* AMD family 15, model 2 as bulldozer, actually piledriver.
* AMD family 17, model 18, was supposed to be zen/zen+, since I can't tell those
apart, seen stepping 1 is zen+, but had incorrect match.
* Intel family 6, model 25, stepping 2 as nehalem, should have been westmere.
* Changed Penryn to Core Penryn, intel family 6, model 17
* misc other micro arch fine tunings.
3. Code fix 8, switched to global %risc for arm, mips, ppc, riscv, sparc. This
corrects many sloppy handlers, and makes all risc processing the same, and calls
device tree readers for all risc systems, not just arm or arm and mips.
4. In cases where bogomips were 0 due to false values in risc results, show N/A.
5. Removed all attempts to guess at what /proc/cpuinfo cache size: refers to,
it can literally be anything, a per core L1, a per core or cpu L2, or an L3.
So applying any math to it is just a random guess at that point. If any L1,2,3
cache data is found, don't use the cache: value at all, but that will only be
present if no /sys data was found anyway, and if cpuinfo had no specific cache
type fields, only generic cache.
6. Added failsafe tests for stepping and model id before doing conversion to
hex. Make sure integer!
7. Added L1 D cache, was only using I cache for BSDs. Output will show total for
L1 A + L1 D. No idea why I didn't use L1 D, makes little sense, but that's how
it goes.
8. Made bogomips tests more granular, now only rejects low sub 50 bogomips
values if %risc cpu type. Legacy ancient cpus like 486 could and did have
bogomip counts below 50.
https://tldp.org/HOWTO/BogoMips/bogo-list.html
9. See Enhancement 12 as well. If OpenBSD, which has no per core data or
physical cpu data, is running on MT capable cpu, but for security OpenBSD has
disabled MT, will now force MT to be not shown via the hw.smt value. This
removes a small glitch that would have bothered OpenBSD users who know that
OpenBSD has disabled MT for security purposes.
10. Changed BSD hack to use L2 cache totals to deduce > 1 physical cpus, that
was flat out dumb, since we can just use dmidecode type 4 to iterate physical
cpu counts and skip the pointless logic. Thus, if dmidecode, and if > 1
dmidecode type 4 found, and if physical cpu counts equal 1, then replace the
found counts with the dmidecode physical cpu counts.
11. Corrected bad assumption that threads would always be 2 per core for MT
tests. Still no way to reliably determine threads per core for non x86 cpus like
powerpc however, but those are very fringe and should rarely be an issue since
that data is only missing on very old linux now I think.
12. Fixed 'parameters:' going to its own line with -Sa, that wasn't supposed to.
-S is two lines, the kernel / host stuff, and the desktop/console/distro stuff.
13. Fixed case when key: value first word plus other parts of line longer than
max width, failed to wrap as expected.
14. Added start/end ' and " start / end \s to main filters.
--------------------------------------------------------------------------------
ENHANCEMENTS:
1. CPU: most Linux will now show L1 and L3 cache with -Cx without needing
sudo/root, and it will be more accurate than ever before.
2. CPU: shows per CPU L1/L2/L3 totals, and shows actual full system physical
processor count * L1/L2/L3 total in parentheses, like: L2: 1.5 MiB (3 MiB).
3. CPU: A long standing annoyance, previously for main CPU 'Speed:' item, showed
the fastest core speed found, now shows avg: [speed] and with -Cx, shows the
'high:' as well if > 1 cores, and if 1 or more cores have a higher speed than
the other(s).
4. CPU: Handles advanced cases of new architectures, like Alder Lake with
Performance and Efficiency cores, future Zen, and existing ARM CPUs with 2 or
more different core sets, with different max/min frequencies. Previously a hack
was used to handle only ARM CPUs with this type of architecture. Will show
correct CPU core counts, which previous inxi versions would fail to do for Alder
Lake type scenarios of 8 single threaded CPUs and 4-8 multithreaded )MT)
perforance cores.
This should also in theory show different the different min/max speeds if they
were detected. Those did not seem to be set correctly in Alder Lake sample data
I saw however, P and E cores were set to the same min/max speeds.
5. Added CPU types MST (Multi+Single Thread), AMP (Asymmetric Multi Processing),
and AMCP (Asymmetrical Multi Core Processor). This will be applied to any CPU
that has this type of complex topology that has been dynamically detected, like
Alder Lake or different core count or min/max speed RISC CPUs.
6. CPU: shows with -Ca for cases where different L1/L2/L3 caches found per
physical CPU, as with Alder Lake, but also many other variants that were poorly
or not at all handled before, how many of each cache type (L1 Data, instruction)
were found, otherwise will show how many of each cache were found.
7. CPU: shows with -Ca in Topology: report, for cases like Alder Lake with
different core types in one physical CPU (type: MST AMCP), the number that are
single threaded (st) and number that are multi-threaded (mt).
8. Basic support for rsyc-v systems, going along with code fix 8, fix 3, now
it's easy to add this type of support.
9. Added shortcut options for --filter-label (--zl), --filter-uuid (--zu), and
andded new filter option, --filter-vulnerabilities (--zv). The latter is added
by request, a decent idea to have option to not show cpu vulnerabilities.
10. Going with fix 7, switched to a sort of pseudo L1 d/i with desc report for
any BSD with L1 I/D cache found, or elbrus cache0 (icache) / cache1 (d cache).
Elbrus should hopefully be handled by the /sys tool. Guesses on the L1 are ok,
since those are almost always per core, so it's fine. Didn't expect to enhance
any BSD cpu data this time around, but there you go!! If they have the data,
then it will be used. Not going to go overboard though in that, quite useless
overall since usually can't see how many CPUs are present, at least not usually.
11. For -Ca, full CPU topology report if any complex topogy is detected,
otherwise shows the same basic Info: 2x 6-core or Info: dual core as before, no
point in wasting a line for something with no more data than the short string.
Complex types include MT CPUs since they will have different thread counts etc,
and will have 2 or more threads per core, which will also be listed.
12. If smt status is defined (0/1), shows smt: enabled|disabled in Topology
section, can be useful for systems with disabled MT, but supporting it. If no
topology data found (OpenBSD for example), for -C shows 'smt: disabled' after
'type:' section, and enabled if -Cxxx (since MT really already tells you that).
13. For -Ca Speed: report, added 'governor:' item, if found. Can show 1 or more
active governors.
14. Output height (in lines) control: -Y [-2|-1|0|1-xxx]]. This lets you break
up any of the output into whatever number of lines you want. Also useful out of
DISPLAY for reading -h options menu items etc.
It came tp my attention that the long standing shift+pgup/pgdown (aka
'softscrollback) behavior had stopped working, and in fact has been removed from
the current Linux kernel, at least until it is rewritten to be more clean and
understandable. Read more about it in these kernel post/commit messages:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=50145474f6ef4a9c19205b173da6264a644c7489
https://lwn.net/ml/linux-kernel/CAHk-=whe4ZdTdCebneWqC4gSQZwsVJ5-Emg0BucOGCwPhOAJpw@mail.gmail.com/
Options for -Y are:
* -Y 0 or -Y: Set maximum block height to terminal line height.
* -Y [1-xxx]: Set maximum block line height to given integer.
* -Y -1: Print out one primary data block item at a time, with -F for example.
* -Y -2: Restore default unlimited height if LINES_MAX configuration item used.
15. And finally, more disk vendors/vendor ids. As usual. As expected.
--------------------------------------------------------------------------------
CHANGES:
1. If /sys or /proc/cpuinfo speed data available:
* For -b CPU item:
speed: [speed MHz] min/max: [min]/[max] MHz
becomes:
speed (MHz): avg: [speed] min/max: [min]/[max]
* For -C, Speed item
Speed: [speed MHz] min/max: [min]/[max] MHz Cores (MHz): ...
becomes:
Speed (MHz): avg: [speed] min/max: [min]/[max] cores: ...
* For short form, shows speed/min/max but uses average speed if available.
For -b and -C, only shows one MHz in Speed line starter, which slightly shortens
the line even with the added 'avg:' item since 3 MHz are replaced with 1.
2. Going with change 1, now the 'avg:' item shows not the fastest cpu speed
found, which was the case before, but shows an average of all cpu speeds found.
Showing the fastest made some sense back in the days of single core, or even
dual core CPUs, but makes little sense today with many core/threaded cpus.
With -x, it will show the high: [speed] item as well, after 'avg:'.
3. By suggestion, wrapped first Type item in Vulnerabilities to its own line,
that's a verbose --admin option after all, no need to save lines!
4. Going along with Fix 5, give up on trying to pretend we can guess at L2
cache, now if only 'cache' data was available from cpuinfo, will just say:
cache: [cache size]
note: check
and call it a day.
5a. Change default width to 80 columns, in and out of display. Too many users
are posting horribly wrapping inxi output in forums, issue trackers, etc, and it
frankly makes inxi look really bad, creates awful side scrolling code boxes,
etc. So now default widths in and out of console are 80 (since often data is
generated in SSH or out of X/Wayland) for issues.
This essentially makes -y 80/-y the default width. This is what I've been using
for a few years now, and after seeing far too many side scrolling or badly
wrapping inxi outputs online, I think it's probably time to just force 80 column
widths as default and call it a day.
You can change these new defaults using configuration options (these are the
previous options, though due to a bug, COLS_MAX_CONSOLE was never being used):
COLS_MAX_CONSOLE=115 # in display, terminal client max width
COLS_MAX_IRC=100
COLS_MAX_NO_DISPLAY=130 # not in display, no X/Wayland running
5b. Changed output wrapped indent to 1 column from 2, and make second and
greater rows of a line indent +1 to make it more clear that it is a child row of
its parent row. Note that because no arg short form, -S, and -I are special
types of rows, this behavior is not used, they just print out as usual. This 1
column indent also applies to -y1, making for a little more data per line but
more readable and easy to follow.
6. If > 1 physical cpu detected, no longer uses single/dual/triple/quad core
strings, rather uses: 2x 2-core. Also uses lower case -core, not -Core.
7. Only show die counts for CPU (on rare occasions > 1 found) with -xx. Not
particularly important bit of data afterall.
8. Make L1, L3 cache data show with -Cx, not -Cxx, now that it's working well.
9. Removed CPU die for -Cxx, that's only going to show with -Ca now.
10. If -Ca, and if certain complexity conditions are met, shows a separate
Topology line rather than the Info: 6-core type item. For -b, short, -Cx, -Cxxx
shows the Info: topology short form.
11. Bogomips always shows before flags data, whether -f or just -Cx trips flag
output.
12. Flags/Features now shows in the same place, under Speeds: always, whether
-Cx shortlist, or -Cf full list. Makes more sense that way, and code is much
cleaner too.
13. Bogomips, being essentially bogus units of speed for cpu, are moved into
Speed: report.
--------------------------------------------------------------------------------
DOCUMENTATION:
1. Updated man/help for new CPU extra data options and output changes.
2. Cleaned up and added sample outputs for man CPU items.
3. Now that doas is getting into Linux distros, removed all mentions of doas as
a BSD option, and made it a general doas/sudo item. Glad to see doas making it
into linux distros, it's a good tool, much easier to configure and use than
sudo. Good job OpenBSD guys. Note that inxi already has had full doas support
for a while now, but this finalizes it, and makes it fully agnostic. Internally
doas is actually preferred over sudo, by the way.
4. Added documention items for INDENT (--indent), INDENT_MIN (--inident-min).
5. Re-ordered help menu and man page, created new Filters and Output Controls
sections to make stuff easier to find. In man page, also added on top a list of
OPTIONS sections to make finding stuff easier.
--------------------------------------------------------------------------------
CODE:
1. Removed legacy /sys cpu functions: cpu_bugs_sys(); get_boost_status();
set_cpu_speeds_sys(). cpu_speeds() is deprecated and now will only be used for
legacy Linux and BSDs if they had any per core speeds found; get_caches() was
only a placeholder for the full featured cpu_sys data source, and was removed;
cpu_speeds() no longer needed, integrated into other logic; cpu_dies_sys()
removed, integrated into other logic.
This logic is now integrated into cpu_data_sys() data generator.
2. Changed the main CpuItems functions to use array/hash references, not passing
full hashes or arrays in most cases now.
3. For machine_data_soc(), switched to CpuItem::cpuinfo_grabber() which then
sets the global @cpuinfo and %cpuinfo_machine items, which will be used again
in Cpu if cpu data is requested. This gets rid of a full parsing of cpuinfo
just to get the machine data section, and also makes it so cpuinfo in cpu does
not need to worry about the machine data block, which is not related to the
processor blocks anyway, that was always a hack done by the kernel guys to toss
that SOC data somewhere as far as I can tell.
4. New tools:
* either_or() - takes a list, and returns the first defined element of list.
* regex_range() - generate ranges from comma, space, or ranges like 2-29, or any
combination of those, like 3,6,12-29
5. Added --force cpuinfo to bypass all /sys based cpu logic, useful for testing
to see what would have happened using old logic.
6. Added --dbg switches 39, 40, 41, for the new cpu sys data features, also made
more consistent --dbg 8 and --dbg 38 switches.
7. Added sys/cpuinfo pair debugger to support debugging complex sys/cpuinfo
issues.
8. Got rid of $b_arm,$b_mips,$b_ppc,$b_sparc, replaced with global %risc, also
added $risc{'riscv'} type. this makes general risc type feature testing a lot
easier since inxi can either test for %risc defined, or for a specific type of
risc cpu. This is much cleaner, and use $risc{'id'} for print purposes, which
got rid of a lot of tests. Also made all risc tests consistent, some were ARM
only, or arm/mips, but were supposed to be for all risc cpus.
9. Set help menu code to roughly 80 columns width assuming 2 space tab
indentation.
10. Changed all xxx_cleaner subs to clean_xxx, all filter subs to filter_xxx,
and row_defaults() to message().
11. Dumped redundant fallback logic in get_kernel_bits, if first getconf method
fails, use $sys_bits, and call it good, it was repeating the 32/64 bit tests
pointlessly.
12. Cleaned up print_data() to allow for more fine tuned indentation for the new
2 indent levels.
13. Made help menu code more or less wrap to 80 columns, or close. Ongoing to
bring to 80 columns where practical, but never at expense of clarity or logic.
--------------------------------------------------------------------------------
-- Harald Hope - Tue, 13 Dec 2021 10:25:49 -0800
================================================================================
Version: 3.3.09
Patch: 01