Saturday 31 December 2016

Kernel printk statements

The kernel contains tens of thousands of statements that may print various errors, warnings and debug/information messages to the kernel log.  Unsurprisingly, as the kernel grows in size, so does the quantity of these messages.  I've been scraping the kernel source for various kernel printk style statements and macros and scanning these for various typos and spelling mistakes and to make this easier I hacked up kernelscan (a quick and dirty parser) that helps me find literal strings from the kernel for spell checking.

Using kernelscan, I've gathered some statistics for the number of kernel print statements for various kernel releases:


As one can see, we have over 200,000 messages in the 4.9 kernel(!).  Given the kernel growth, we can see this seems to roughly correlate with the kernel source size:



So how many lines of code in the kernel do we have per kernel printk messages over time?


..showing that the trend is to have more lines of code per frequent printk statements over time.  I didn't differentiate between different types of printk message, so it is hard to see any deeper trends on what kinds of messages are being logged more or less frequently over each release, for example,  perhaps there are less debug messages landing in the kernel nowadays.

I find it quite amazing that the kernel contains quite so many printk messages; it would be useful to see just how many of these are actually in a production kernel. I suspect quite large number are for driver debugging and may be conditionally omitted at build time.

Saturday 17 December 2016

A seasonal obfuscated C program for 2016

Another year passes and once more I have another seasonal obfuscated C program.  I was caught short on free time this year to heavily obfuscate the code which is a shame. However, this year I worked a bit harder at animating the output, so hopefully that will make up for lack of obfuscation.

The source is available on github to eyeball.  I've had criticism on previous years that it is hard to figure out the structure of my obfuscated code, so this year I made sure that the if statements were easier to see and hence understand the flow of the code.

This year I've snapped up all my seasonal obfuscated C programs and put them into the snap store as the christmas-obfuscated-c snap.

Below is a video of the program running; it is all ASCII art and one can re-size the window while it is running.


Unlike previous years, I have the pre-obfuscated version of the code available in the git repository at commit c98376187908b2cf8c4d007445b023db67c68691 so hopefully you can see the original hacky C source.

Have a great Christmas and a most excellent New Year. 

Saturday 26 November 2016

stress-ng 0.07.07 released

stress-ng is a tool that I have been developing on-and-off for a few years. It is designed to stress kernels to force out bugs, stress CPU and memory and also contains some performance benchmarking metrics too.

stress-ng is now entering the maturity part of the development phase, however, there is always scope to add new stressors and generally improve the tool.   I've just released version 0.07.07 for the Ubuntu Zesty 17.04 release and it contains a few additional features:
  • SIGUSR2 sent to stress-ng will dump out the current system load and memory statistics
  • Sched policy stress tests for different scheduler configurations
  • Add a missing --sockfd-port option
And various bug fixes:
  • Fixed up some minor memory leaks
  • Missing counter stats on bind-mount, fp-error, personality and resources stressors
  • Fix the --fiemap-bytes option
  • Fix up build warnings with various compilers and static analyzers
The major change to stress-ng over the past month was an internal re-working of system call and GNU features to abstract these into a shim layer to reduce the number build conditional #ifdef paths around code. This simplifies portability, so the code now builds more easily across a range of systems and with various versions of gcc and clang and fixes some issues on older kernels too.   This makes the code also faster to statically analyze with cppcheck.

For more details, visit the stress-ng project page or the quick help guide.

Sunday 28 August 2016

Fixing an overheating Lenovo X230 laptop

Over the past month I've been hitting excessive thermal heating on my laptop and kidle_inject has been kicking in to try and stop the CPU from overheating (melting!).  A quick double-check with older kernels showed me that this issue was not thermal/performance regression caused by software - instead it was time to clean my laptop and renew the thermal paste.

After some quick research, I found that Artic MX-4 Thermal Compound provided an excellent thermal conductivity rating of 8.5W/mK so I ordered a 4g sample as well as a can of pressurized gas cleaner to clean out dust.

The X230 has an excellent hardware maintenance manual, and following the instructions I stripped the laptop right down so I could pop the heat pipe contacts off the CPU and GPU.  I carefully cleaned off the old dry and cracked thermal paste and applied about 0.2g of MX-4 thermal compound to the CPU and GPU and re-seated the heat pipe.  With the pressurized gas I cleaned out the fan and airways to maximize airflow over the heatpipe.   The entire procedure took about an hour to complete and for once I didn't have any screws left over after re-assembly!

I normally take photos of the position of components during the strip down of a laptop for reference in case I cannot figure out exactly how parts are meant to fix on the re-assembly phase.  In this case, the X230 maintenance manual is sufficiently detailed so I didn't take any photos this time.

I'm glad to report that my X230 is now no-longer overheating. Heat is being effectively pumped away from the CPU and GPU and one can feel the additional heat being pushed out of the laptop.  Once again I can fully max out the CPU and GPU without passive thermal cooling mechanisms being kicked into action, so I've now got 100% of my CPU performance back again; as good as new!

Now and again I see laptop overheating bugs being filed in LaunchPad.  While some are legitimate issues with broken software, I do wonder if the majority of issues with the older laptops is simply due to accumulation of dust and/or old and damaged thermal paste.

Saturday 30 July 2016

Scanning the Linux kernel for error messages

The Linux kernel contains lots of error/warning/information messages; over 130,000 in the current 4.7 kernel.  One of the tests in the Firmware Test Suite (FWTS) is to find BIOS/ACPI/UEFI related kernel error messages in the kernel log and try to provide some helpful advice on each error message since some can be very cryptic to the untrained eye.

The FWTS kernel error log database is currently approaching 800 entries and I have been slowly working through another 800 or so more relevant and recently added messages.  Needless to say, this is taking a while to complete.  The hardest part was finding relevant error messages in the kernel as they appear in different forms (e.g. printk(), dev_err(), ACPI_ERROR() etc).

In order to scrape the Linux kernel source for relevant error messages I hacked up the kernelscan parser to find error messages and dump these to stdout.  kernelscan can scan 43,000 source files (17,900,000 lines of source) in under 10 seconds on my Lenovo X230 laptop, so it is relatively fast.

I also have been using kernelscan to find spelling mistakes in kernel messages and I've been punting trivial fixes upstream to fix these.  These mistakes are small and petty, but I find it a little irksome when I see the kernel emit a message that contains a typo or spelling mistake - it just looks a bit unprofessional.

I've created a kernelscan snap (which was really easy and fast to do using scancraft), so it is now available Ubuntu.  The source code is also available from the kernel team git web at http://kernel.ubuntu.com/git/cking/kernelscan.git/

The code is designed to only parse kernel source, and it is a very rough and ready parser designed for speed;  fundamentally, it is a big quick hack.  When I get a few spare minutes I will try and see if there is any correlation between the number of error messages with the size of the kernel over the various releases.

Wednesday 29 June 2016

What's new in stress-ng 0.06.07?

Since my last blog post about stress-ng, I've pushed out several more small releases that incorporate new features and (as ever) a bunch more bug fixes.  I've been eyeballing gcov kernel coverage stats to find more regions in the kernel where stress-ng needs to exercise.   Also, testing on a range of hardware (arm64, s390x, etc) and a range of kernels has eeked out some bugs and helped me to improve stress-ng.  So what's new?

New stressors:
  • ioprio  - exercises ioprio_get(2) and ioprio_set(2) (I/O scheduling classes and priorities)
  • opcode - generates random object code and executes this, generating and catching illegal instructions, bus errors,  segmentation  faults,  traps and floating  point errors.
  • stackmmap - allocates a 2MB stack that is memory mapped onto a temporary file. A recursive function works down the stack and flushes dirty stack pages back to the memory mapped file using msync(2) until the end of the stack is reached (stack overflow). This exercises dirty page and stack exception handling.
  • madvise - applies random madvise(2) advise settings on pages of a 4MB file backed shared memory mapping.
  • pty - exercise pseudo terminal operations.
  • chown - trivial chown(2) file ownership exerciser.
  • seal - fcntl(2) file SEALing exerciser.
  • locka - POSIX advisory locking exerciser.
  • lockofd - fcntl(2) F_OFD_SETLK/GETLK open file description lock exerciser.
Improved stressors:
  • msg: add in IPC_INFO, MSG_INFO, MSG_STAT msgctl calls
  • vecmath: add more ops to make vecmath more demanding
  • socket: add --sock-type socket type option, e.g. stream or seqpacket
  • shm and shm-sysv: add msync'ing on the shm regions
  • memfd: add hole punching
  • mremap: add MAP_FIXED remappings
  • shm: sync, expand, shrink shm regions
  • dup: use dup2(2)
  • seek: add SEEK_CUR, SEEK_END seek options
  • utime: exercise UTIME_NOW and UTIME_OMIT settings
  • userfaultfd: add zero page handling
  • cache:  use cacheflush() on systems that provide this syscall
  • key:  add request_key system call
  • nice: add some randomness to the delay to unsync nicenesses changes
If any new features land in Linux 4.8 I may add stressors for them, but for now I suspect that's about it for the big changes for stress-ng for the Ubuntu Yakkey 16.10 release.

Wednesday 4 May 2016

Some extra features landing in stress-ng 0.06.00

Recently I've been adding a few more features into stress-ng to get improve kernel code coverage.   I'm currently using a kernel built with gcov enabled and using the most excellent lcov tool to collate the coverage data and produce some rather useful coverage charts.

With a gcov enabled kernel, gathering coverage stats is a trivial process with lcov:

 sudo apt-get install lcov  
 sudo lcov --zerocounters  
 stress-ng --seq 0 -t 60  
 sudo lcov -c -o kernel.info  
 sudo genhtml -o html kernel.info  

..and the html output appears in the html directory.

In the latest 0.06.00 release of stress-ng, the following new features have been introduced:

  • af-alg stressor, added skciphers and rngs
  • new Translation Lookaside Buffer (TLB) shootdown stressor
  • new /dev/full stressor
  • hdd stressor now works through all the different hdd options if --maximize is used
  • wider procfs stressing
  • added more keyctl commands to the key stressor
  • new msync stressor, exercise msync of mmap'd memory back to file and from file back to memory.
  • Real Time Clock (RTC) stressor (via /dev/rtc and /proc/driver/rtc)
  • taskset option, allowing one to run stressors on specific CPUs (affinity setting)
  • inotify stressor now also exercises the FIONREAD ioctl()
  • and some bug fixes found when testing stress-ng on various architectures.
The --taskset option allows one to keep stress-ng stressors bound to specific CPUs, for example, to run 5 CPU stressors tied to CPUs 1, 3, 5, 6 and 7:

 stress-ng --taskset 1,3,5-7 --cpu 5  

..thanks to Jim Rowan (Intel) for the CPU affinity ideas.

stress-ng 0.06.00 will be landing in Ubunty Yakkety soon, and also in my power utilities PPA ppa:colin-king/white

Thursday 21 April 2016

Ensure your stack is aligned correctly on aarch64 clone()

I got bitten this week with the clone() system call returning -EINVAL on aarch64 on code that worked fine on x86.  After re-reading the manual several times and looking at my code, I resorted to shoving in debug into the kernel to track down where the -EINVAL was occurring.

The answer to my issue is in arch/arm64/kernel/process.c, copy_thread():

         if (stack_start) {  
             if (is_compat_thread(task_thread_info(p)))  
                 childregs->compat_sp = stack_start;  
             /* 16-byte aligned stack mandatory on AArch64 */  
             else if (stack_start & 15)  
                 return -EINVAL;  
             else  
                 childregs->sp = stack_start;  
         }  

Ahah! The stack being passed into clone() has to be 16 byte aligned.  With this simple fix to my code, clone() worked.   Pity this was not in the documentation.

Friday 11 March 2016

A frequently used incorrect realloc() idiom

While running static analysis on a lot of C source code, I keep on finding a common incorrect programming idiom used with realloc() allocation failures where a NULL is returned and the code returns with some kind of exit failure status, something like the following:

 ptr = realloc(ptr, new_size);  
 if (!ptr)   
     return -ENOMEM;  /* Failed, no memory! */  

However, when realloc() fails it returns NULL and the original object remains unchanged and thus it is not freed.  So the above code leaks the memory pointed to by ptr if realloc() returns NULL.

This may be a moot point, since the error handling paths normally abort the program because we are out of memory if can't proceed any further.  However, there are occasions in code where ENOMEM may not be fatal, for example the program may reallocate smaller buffers and retry or free up space on the heap and retry.

A more correct programming idiom for realloc() perhaps should be:

 tmp = realloc(ptr, new_size);  
 if (!tmp) {  
     free(ptr);  
     return -ENOMEM;  /* Failed, no memory! */  
 }  
 ptr = tmp;  

..which is not aesthetically pleasing, but does the trick of free'ing memory before we return.

Anyhow, it is something to bear in mind next time one uses realloc().

Monday 22 February 2016

ZFS quick start reference guide

Dustin Kirkland recently announced ZFS being officially supported by Canonical for Ubuntu Xenial 16.04.    We've written a short reference guide to help with getting started, and understanding the ZFS terminology.  This touches the basics on setting up ZFS pools as well as creating and using ZFS file systems. 

If you are new to ZFS, I recommend having a look at the reference guide to get you started.








Monday 15 February 2016

New "top" mode in eventstat

I wrote eventstat a few years ago to track wakeup events that keep a machine from being fully idle.  For Ubuntu Xenial Xerus 16.04 I've added a 'top' like mode (enabled using the -T option).

 
By widening the terminal one can see more of the Task, Init Function and Callback text, which is useful as these details can be rather lengthy.

Anyhow, just a minor feature change, but hopefully a useful one.

Thursday 4 February 2016

Intel Platform Quality of Service and Cache Allocation Technology

One issue when running parallel processes is contention of shared resources such as the Last Level Cache (aka LLC or L3 Cache).  For example, a server may be running a set of Virtual Machines with processes that are memory and cache intensive hence producing a large amount of cache activity. This can impact on the other VMs and is known as the "Noisy Neighbour" problem.

Fortunately the next generation Intel processors allow one to monitor and also fine tune cache allocation using Intel Cache Monitoring Technology (CMT) and Cache Allocation Technology (CAT).

Intel kindly loaned me a 12 thread development machine with CMT and CAT support to experiment with this technology using the Intel pqos tool.   For my experiment, I installed Ubuntu Xenial Server on the machine. I then installed KVM and an VM instance of Ubuntu Xenial Server.   I then loaded the instance using stress-ng running a memory bandwidth stressor:

 stress-ng --stream 1 -v --stream-l3-size 16M  
..which allocates 16MB in 4 buffers and performs various read/compute and writes to these, hence causing a "noisy neighbour".

Using pqos,  one can monitor and see the cache/memory activity:
sudo apt-get install intel-cmt-cat
sudo modprobe msr  
sudo pqos -r  
TIME 2016-02-04 10:25:06
    CORE   IPC  MISSES    LLC[KB]  MBL[MB/s]  MBR[MB/s]
       0  0.59 168259k     9144.0    12195.0        0.0
       1  1.33    107k        0.0        3.3        0.0
       2  0.20      2k        0.0        0.0        0.0
       3  0.70    104k        0.0        2.0        0.0
       4  0.86     23k        0.0        0.7        0.0
       5  0.38     42k       24.0        1.5        0.0
       6  0.12      2k        0.0        0.0        0.0
       7  0.24     48k        0.0        3.0        0.0
       8  0.61     26k        0.0        1.6        0.0
       9  0.37     11k      144.0        0.9        0.0
      10  0.48      1k        0.0        0.0        0.0
      11  0.45      2k        0.0        0.0        0.0
Now to run a stress-ng stream stressor on the host and see the performance while the noisy neighbour is also running:
stress-ng --stream 4 --stream-l3-size 2M --perf --metrics-brief -t 60
stress-ng: info:  [2195] dispatching hogs: 4 stream
stress-ng: info:  [2196] stress-ng-stream: stressor loosely based on a variant of the STREAM benchmark code
stress-ng: info:  [2196] stress-ng-stream: do NOT submit any of these results to the STREAM benchmark results
stress-ng: info:  [2196] stress-ng-stream: Using L3 CPU cache size of 2048K
stress-ng: info:  [2196] stress-ng-stream: memory rate: 1842.22 MB/sec, 736.89 Mflop/sec (instance 0)
stress-ng: info:  [2198] stress-ng-stream: memory rate: 1847.88 MB/sec, 739.15 Mflop/sec (instance 2)
stress-ng: info:  [2199] stress-ng-stream: memory rate: 1833.89 MB/sec, 733.56 Mflop/sec (instance 3)
stress-ng: info:  [2197] stress-ng-stream: memory rate: 1847.16 MB/sec, 738.86 Mflop/sec (instance 1)
stress-ng: info:  [2195] successful run completed in 60.01s (1 min, 0.01 secs)
stress-ng: info:  [2195] stressor      bogo ops real time  usr time  sys time   bogo ops/s   bogo ops/s
stress-ng: info:  [2195]                          (secs)    (secs)    (secs)   (real time) (usr+sys time)
stress-ng: info:  [2195] stream           22101     60.01    239.93      0.04       368.31        92.10
stress-ng: info:  [2195] stream:
stress-ng: info:  [2195]            547,520,600,744 CPU Cycles                     9.12 B/sec
stress-ng: info:  [2195]             69,959,954,760 Instructions                   1.17 B/sec (0.128 instr. per cycle)
stress-ng: info:  [2195]             11,066,905,620 Cache References               0.18 B/sec
stress-ng: info:  [2195]             11,065,068,064 Cache Misses                   0.18 B/sec (99.98%)
stress-ng: info:  [2195]              8,759,154,716 Branch Instructions            0.15 B/sec
stress-ng: info:  [2195]                  2,205,904 Branch Misses                 36.76 K/sec ( 0.03%)
stress-ng: info:  [2195]             23,856,890,232 Bus Cycles                     0.40 B/sec
stress-ng: info:  [2195]            477,143,689,444 Total Cycles                   7.95 B/sec
stress-ng: info:  [2195]                         36 Page Faults Minor              0.60 sec
stress-ng: info:  [2195]                          0 Page Faults Major              0.00 sec
stress-ng: info:  [2195]                         96 Context Switches               1.60 sec
stress-ng: info:  [2195]                          0 CPU Migrations                 0.00 sec
stress-ng: info:  [2195]                          0 Alignment Faults               0.00 sec
.. so about 1842 MB/sec memory rate and 736 Mflop/sec per CPU across 4 CPUs.  And pqos shows the cache/memory actitivity as:
sudo pqos -r
TIME 2016-02-04 10:35:27
    CORE   IPC  MISSES    LLC[KB]  MBL[MB/s]  MBR[MB/s]
       0  0.14  43060k     1104.0     2487.9        0.0
       1  0.12 3981523k     2616.0     2893.8        0.0
       2  0.26    320k       48.0       18.0        0.0
       3  0.12 3980489k     1800.0     2572.2        0.0
       4  0.12 3979094k     1728.0     2870.3        0.0
       5  0.12 3970996k     2112.0     2734.5        0.0
       6  0.04     20k        0.0        0.3        0.0
       7  0.04     29k        0.0        1.9        0.0
       8  0.09    143k        0.0        5.9        0.0
       9  0.15      0k        0.0        0.0        0.0
      10  0.07      2k        0.0        0.0        0.0
      11  0.13      0k        0.0        0.0        0.0
Using pqos again, we can find out how much LLC cache the processor has:
sudo pqos -v
NOTE:  Mixed use of MSR and kernel interfaces to manage
       CAT or CMT & MBM may lead to unexpected behavior.
INFO: Monitoring capability detected
INFO: CPUID.0x7.0: CAT supported
INFO: CAT details: CDP support=0, CDP on=0, #COS=16, #ways=12, ways contention bit-mask 0xc00
INFO: LLC cache size 9437184 bytes, 12 ways
INFO: LLC cache way size 786432 bytes
INFO: L3CA capability detected
INFO: Detected PID API (perf) support for LLC Occupancy
INFO: Detected PID API (perf) support for Instructions/Cycle
INFO: Detected PID API (perf) support for LLC Misses
ERROR: IPC and/or LLC miss performance counters already in use!
Use -r option to start monitoring anyway.
Monitoring start error on core(s) 5, status 6
So this CPU has 12 cache "ways", each of 786432 bytes (768K).  One or more  "Class of Service" (COS)  types can be defined that can use one or more of these ways.  One uses a bitmap with each bit representing a way to indicate how the ways are to be used by a COS.  For example, to use all the 12 ways on my example machine, the bit map is 0xfff  (111111111111).   A way can be exclusively mapped to a COS or shared, or not used at all.   Note that the ways in the bitmap must be contiguously allocated, so a mask such as 0xf3f (111100111111) is invalid and cannot be used.

In my experiment, I want to create 2 COS types, the first COS will have just 1 cache way assigned to it and CPU 0 will be bound to this COS as well as pinning the VM instance to CPU 0  The second COS will have the other 11 cache ways assigned to it, and all the other CPUs can use this COS.

So, create COS #1 with just 1 way of cache, and bind CPU 0 to this COS, and pin the VM to CPU 0:
sudo pqos -e llc:1=0x0001
sudo pqos -a llc:1=0
sudo taskset  -apc 0 $(pidof qemu-system-x86_64)
And create COS #2, with 11 ways of cache and bind CPUs 1-11 to this COS:
sudo pqos -e "llc:2=0x0ffe"
sudo pqos -a "llc:2=1-11"
And let's see the new configuration:
sudo pqos  -s
NOTE:  Mixed use of MSR and kernel interfaces to manage
       CAT or CMT & MBM may lead to unexpected behavior.
L3CA COS definitions for Socket 0:
    L3CA COS0 => MASK 0xfff
    L3CA COS1 => MASK 0x1
    L3CA COS2 => MASK 0xffe
    L3CA COS3 => MASK 0xfff
    L3CA COS4 => MASK 0xfff
    L3CA COS5 => MASK 0xfff
    L3CA COS6 => MASK 0xfff
    L3CA COS7 => MASK 0xfff
    L3CA COS8 => MASK 0xfff
    L3CA COS9 => MASK 0xfff
    L3CA COS10 => MASK 0xfff
    L3CA COS11 => MASK 0xfff
    L3CA COS12 => MASK 0xfff
    L3CA COS13 => MASK 0xfff
    L3CA COS14 => MASK 0xfff
    L3CA COS15 => MASK 0xfff
Core information for socket 0:
    Core 0 => COS1, RMID0
    Core 1 => COS2, RMID0
    Core 2 => COS2, RMID0
    Core 3 => COS2, RMID0
    Core 4 => COS2, RMID0
    Core 5 => COS2, RMID0
    Core 6 => COS2, RMID0
    Core 7 => COS2, RMID0
    Core 8 => COS2, RMID0
    Core 9 => COS2, RMID0
    Core 10 => COS2, RMID0
    Core 11 => COS2, RMID0
..showing Core 0 bound to COS1, and Cores 1-11 bound to COS2, with COS1 with 1 cache way and COS2 with the remaining 11 cache ways.

Now re-run the stream stressor and see if the VM has less impact on the LL3 cache:
stress-ng --stream 4 --stream-l3-size 1M --perf --metrics-brief -t 60
stress-ng: info:  [2232] dispatching hogs: 4 stream
stress-ng: info:  [2233] stress-ng-stream: stressor loosely based on a variant of the STREAM benchmark code
stress-ng: info:  [2233] stress-ng-stream: do NOT submit any of these results to the STREAM benchmark results
stress-ng: info:  [2233] stress-ng-stream: Using L3 CPU cache size of 1024K
stress-ng: info:  [2235] stress-ng-stream: memory rate: 2616.90 MB/sec, 1046.76 Mflop/sec (instance 2)
stress-ng: info:  [2233] stress-ng-stream: memory rate: 2562.97 MB/sec, 1025.19 Mflop/sec (instance 0)
stress-ng: info:  [2234] stress-ng-stream: memory rate: 2541.10 MB/sec, 1016.44 Mflop/sec (instance 1)
stress-ng: info:  [2236] stress-ng-stream: memory rate: 2652.02 MB/sec, 1060.81 Mflop/sec (instance 3)
stress-ng: info:  [2232] successful run completed in 60.00s (1 min, 0.00 secs)
stress-ng: info:  [2232] stressor      bogo ops real time  usr time  sys time   bogo ops/s   bogo ops/s
stress-ng: info:  [2232]                          (secs)    (secs)    (secs)   (real time) (usr+sys time)
stress-ng: info:  [2232] stream           62223     60.00    239.97      0.00      1037.01       259.29
stress-ng: info:  [2232] stream:
stress-ng: info:  [2232]            547,364,185,528 CPU Cycles                     9.12 B/sec
stress-ng: info:  [2232]             97,037,047,444 Instructions                   1.62 B/sec (0.177 instr. per cycle)
stress-ng: info:  [2232]             14,396,274,512 Cache References               0.24 B/sec
stress-ng: info:  [2232]             14,390,808,440 Cache Misses                   0.24 B/sec (99.96%)
stress-ng: info:  [2232]             12,144,372,800 Branch Instructions            0.20 B/sec
stress-ng: info:  [2232]                  1,732,264 Branch Misses                 28.87 K/sec ( 0.01%)
stress-ng: info:  [2232]             23,856,388,872 Bus Cycles                     0.40 B/sec
stress-ng: info:  [2232]            477,136,188,248 Total Cycles                   7.95 B/sec
stress-ng: info:  [2232]                         44 Page Faults Minor              0.73 sec
stress-ng: info:  [2232]                          0 Page Faults Major              0.00 sec
stress-ng: info:  [2232]                         72 Context Switches               1.20 sec
stress-ng: info:  [2232]                          0 CPU Migrations                 0.00 sec
stress-ng: info:  [2232]                          0 Alignment Faults               0.00 sec
Now with the noisy neighbour VM constrained to use just 1 way of LL3 cache, the stream stressor on the host now can achieve about 2592 MB/sec and about 1030 Mflop/sec per CPU across 4 CPUs.

This is a relatively simple example.  With the ability to monitor cache and memory bandwidth activity with one can carefully tune a system to make best use of the limited LL3 cache resource and maximise throughput where needed.

There are many applications where Intel CMT/CAT can be useful, for example fine tuning containers or VM instances, or pinning user space networking buffers to cache ways in DPDK for improved throughput.

Sunday 31 January 2016

Pagemon improvements

Over the past month I've been finding the odd moments [1] to add some small improvements and fix a few bugs to pagemon (a tool to monitor process memory).  The original code went from a sketchy proof of concept prototype to a somewhat more usable tool in a few weeks, so my main concern recently was to clean up the code and make it more efficient.

With the use of tools such as valgrind's cachegrind and perf I was able to work on some of the code hot-spots [2] and reduce it from ~50-60% CPU down to 5-9% CPU utilisation on my laptop, so it's definitely more machine friendly now.  In addition I've added the following small features:
  • Now one can specify the name of a process to monitor as well as the PID.  This also allows one to run pagemon on itself(!), which is a bit meta.
  • Perf events showing Page Faults and Kernel Page Allocates and Frees, toggled on/off with the 'p' key.
  • Improved and snappier clean up and exit when a monitored process exits.
  • Far more efficient page map reading and rendering.
  • Out of Memory (OOM) scores added to VM statistics window.
  • Process activity (busy, sleeping, etc) to VM statistics window.
  • Zoom mode min/max with '[' (min) and ']' (max) keys.
  • Close pop-up windows with key 'c'.
  • Improved handling of rapid map expansion and shrinking.
  • Jump to end of map using 'End' key.
  • Improve the man page.
I've tried to keep the tool small and focused and I don't want feature bloat to make it unwieldy and overly complexed.  "Do one job, and do it well" is the philosophy behind pagemon. At just 1500 lines of C, it is as complex as I want it to be for now.

Version 0.01.08 should be hitting the Ubuntu 16.04 Xenial Xerus archive in the next 24 hours or so.  I have also the lastest version in my PPA (ppa:colin-king/pagemon) built for Trusty, Vivid, Wily and Xenial.


Pagemon is useful for spotting unexpected memory activity and it is just interesting watching the behaviour memory hungry processes such as web-browsers and Virtual Machines.

Notes:
[1] Mainly very late at night when I can't sleep (but that's another story...).  The git log says it all.
[2] Reading in /proc/$PID/maps and efficiently reading per page data from /proc/$PID/pagemap

Thursday 28 January 2016

Forcing out bugs with stress-ng

stress-ng logo
Over the past few months I've been adding several new stress tests and a lot more stressor options to stress-ng for Ubuntu 16.04 Xenial Xerus.  I try to track new system calls and features landing in the kernel and where appropriate add a stress test to try and force out bugs.

Stress-ng has found various kernel bugs, such as CVE-2015-1333 and LP:#1526811 as well as bugs in user space (for example, daemons crashing) when memory pressure is very high.  Simple abusive tricks, such as aggressively trying to allocate every free page in memory are useful in finding drivers that don't necessary check for memory allocation failures.  For example, today I was caught out when a USB ethernet dongle driver didn't check for a null pointer due to an allocation failure and stress-ng ended up triggering a kernel oops (fortunately, this bug was fixed in a recent kernel).

The underlying philosophy for stress-ng is "use and abuse standard Linux interfaces and see how far we can push them to destruction".  I'm pretty sure there are plenty of creative folk out there who can dream up dastardly ways to make stress-ng even more stressy, so contributions are always warmly accepted!  I have a mirrorred copy of the git repository on github to make it easy for developers to get their hands on the code.

We've been using stress-ng on ARM based SoC kernels to force out bugs and this has been useful in finding areas where non-swap based systems break. You really don't want your kernel oopsing or processes segfaulting when a IoT device has run low on memory.

My original intent for stress-ng was just to make a system run hot and force thermal overruns. However, I soon discovered it is useful to force kernel bugs out by attempting to (pathologically) thrash most of the system calls.  I've also added perf stats to stress-ng to track performance of standard stress scenarios over kernel versions to get an early warning of any potential performance regressions.  So stress-ng is a bit of a mixed bag of stress tests and performance measuring goodness.

When I get some free time I hope to run stress-ng against a GCOV instrumented kernel at see how much test coverage I get on a kernel. I suspect there are a lot of core kernel functionality still not being touched by stress-ng.

I've also tried to make stress-ng portable, so it can build fine on GNU/Hurd and Debian kFreeBSD (with Linux specific tests not built-in of course). It also contains some architecture specific features, such as handling the data and instruction cache as well as the x86 rdrand instruction and cache line locking. If there are any ARM specific features than can be stressed I'd like to know and perhaps implement stressors for them.

Anyhow, I believe stress-ng is almost feature complete for Ubuntu Xenial, however, I expect it to grow in features over time since there is always new functionality landing in the Linux kernel that needs to be thrashed tested.

Friday 8 January 2016

FIXME and TODO comments in the Linux kernel source

While looking at some code in the Linux Kernel this morning I spotted a few FIXME comments and that got me wondering just how many there are in the source code.  After a quick grep I found nearly 4200 in v4.4.0-rc8 and that got me thinking about other similar comment tags such as TODO that are in the source and how this has been changing over time.


So the trends are certainly upwards, but then again, so is the size of the kernel source:

Note: Data gathered using sloccount on the lines of C in the kernel source.

Using the sloccount data I then calculated the number of FIXME and TODOs per 1000 lines of code to see what the underlying trend is:

So FIXMEs are actually dropping in relative terms to the size of the kernel where as TODOs are increasing.

Of course, these statistics are bogus because it is dependent on kernel developers adding and removing FIXMEs and TODOs in a consistent manner, however, it is interesting to see how many comments exist and hence how much work has been tagged in comments as work to be done later. I wonder how this compares to other large open source projects.