1\documentclass{article} 2 3% 4% Copyright (C) 2005, 2006 Alan D. Brunelle <Alan.Brunelle@hp.com> 5% 6% This program is free software; you can redistribute it and/or modify 7% it under the terms of the GNU General Public License as published by 8% the Free Software Foundation; either version 2 of the License, or 9% (at your option) any later version. 10% 11% This program is distributed in the hope that it will be useful, 12% but WITHOUT ANY WARRANTY; without even the implied warranty of 13% MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14% GNU General Public License for more details. 15% 16% You should have received a copy of the GNU General Public License 17% along with this program; if not, write to the Free Software 18% Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 19% 20 21\title{blktrace User Guide} 22\author{blktrace: Jens Axboe (jens.axboe@oracle.com)\\ 23 User Guide: Alan D. Brunelle (Alan.Brunelle@hp.com)} 24\date{27 May 2008} 25 26\begin{document} 27\maketitle 28%--------------------- 29\section{\label{sec:intro}Introduction} 30 31blktrace is a block layer IO tracing mechanism which provides detailed 32information about request queue operations up to user space. There are 33three major components that are provided: 34 35\begin{description} 36 \item[Kernel patch] A patch to the Linux kernel which includes the 37 kernel event logging interfaces, and patches to areas within the block 38 layer to emit event traces. If you run a 2.6.17-rc1 or newer kernel, 39 you don't need to patch blktrace support as it is already included. 40 41 \item[blktrace] A utility which transfers event traces from the kernel 42 into either long-term on-disk storage, or provides direct formatted 43 output (via blkparse). 44 45 \item[blkparse] A utility which formats events stored in files, or when 46 run in \emph{live} mode directly outputs data collected by blktrace. 47\end{description} 48 49\subsection{blktrace Download Area} 50 51The blktrace and blkparse utilities and associated kernel patch are provided 52as part of the following git repository: 53 54git://git.kernel.org/pub/scm/linux/kernel/git/axboe/blktrace.git bt 55 56%-------------------------- 57\newpage\section{\label{sec:quick-start}Quick Start Guide} 58 59The following sections outline some quick steps towards utilizing 60blktrace. Some of the specific instructions below may need to be tailored 61to your environment. 62 63\subsection{\label{sec:get-blktrace}Retrieving blktrace} 64 65As noted above, the kernel patch along with the blktrace and blkparse utilities are stored in a git repository. One simple way to get going would be: 66 67\begin{verbatim} 68% git clone git://git.kernel.org/pub/scm/linux/kernel/git/axboe/blktrace.git bt 69% cd bt 70% git checkout 71\end{verbatim} 72 73\subsection{\label{sec:patching}Patching and configuring the Linux kernel} 74 75A patch for a \emph{specific Linux kernel} is provided in bt/kernel (where 76\emph{bt} is the name of the directory from the above git sequence). The 77detailed actual patching instructions for a Linux kernel is outside the 78scope of this document, but the following may be used as a sample template. 79Note that you may skip this step, if you kernel is at least 2.6.17-rc1. 80 81As an example, bt/kernel contains blk-trace-2.6.14-rc1-git-G2, download 82linux-2.6.13.tar.bz2 and patch-2.6.14-rc1.bz2 83 84\begin{verbatim} 85% tar xjf linux-2.6.13.tar.bz2 86% mv linux-2.6.13 linux-2.6.14-rc1 87% cd linux-2.6.14-rc1 88% bunzip2 -c ../patch-2.6.14-rc1.bz2 | patch -p1 89\end{verbatim} 90 91At this point you may (optionally) remove linux-2.6.13.tar.bz2 and 92patch-2.6.14-rc1.bz2. 93 94At this point you should configure the Linux kernel for your specific 95system -- again, outside the scope of this document -- and then enable 96\emph{Support for tracing block io actions.} To do this, run 97 98\begin{verbatim} 99% make menuconfig or make xconfig, or edit .config, or ... 100\end{verbatim} 101 102and navigate through \emph{Device Drivers} and \emph{Block devices} 103and then down to \emph{Support for tracing block io actions} and hit Y. 104 105Install the new kernel (and modules\ldots) and reboot. 106 107\subsection{\label{sec:mount}Mounting the debugfs file system} 108 109blktrace utilizes files under the debug file system, and thus must have 110the mount point set up -- mounted on the directory /sys/kernel/debug. 111To do this one may do either of the following: 112 113\begin{enumerate} 114 \item Manually mount after each boot: 115\begin{verbatim} 116% mount -t debugfs debugfs /sys/kernel/debug 117\end{verbatim} 118 119 \item Add an entry into /etc/fstab, and have it done automatically at 120 each boot\footnote{Note: after adding the entry to /etc/fstab, you 121 could then mount the directory this time only by doing: \% mount debug}: 122\begin{verbatim} 123debug /sys/kernel/debug debugfs default 0 0 124\end{verbatim} 125\end{enumerate} 126 127\subsection{\label{sec:build}Build the tools} 128 129To build and install the tools, execute the following sequence (as root): 130 131\begin{verbatim} 132% cd bt 133% make && make install 134\end{verbatim} 135 136\subsection{\label{sec:live-blktrace}blktrace -- live} 137 138Now to simply watch what is going on for a specific disk (to stop the 139trace, hit control-C): 140 141\begin{verbatim} 142% blktrace -d /dev/sda -o - | blkparse -i - 143 8,0 3 1 0.000000000 697 G W 223490 + 8 [kjournald] 144 8,0 3 2 0.000001829 697 P R [kjournald] 145 8,0 3 3 0.000002197 697 Q W 223490 + 8 [kjournald] 146 8,0 3 4 0.000005533 697 M W 223498 + 8 [kjournald] 147 8,0 3 5 0.000008607 697 M W 223506 + 8 [kjournald] 148 8,0 3 6 0.000011569 697 M W 223514 + 8 [kjournald] 149 8,0 3 7 0.000014407 697 M W 223522 + 8 [kjournald] 150 8,0 3 8 0.000017367 697 M W 223530 + 8 [kjournald] 151 8,0 3 9 0.000020161 697 M W 223538 + 8 [kjournald] 152 8,0 3 10 0.000024062 697 D W 223490 + 56 [kjournald] 153 8,0 1 11 0.009507758 0 C W 223490 + 56 [0] 154 8,0 1 12 0.009538995 697 G W 223546 + 8 [kjournald] 155 8,0 1 13 0.009540033 697 P R [kjournald] 156 8,0 1 14 0.009540313 697 Q W 223546 + 8 [kjournald] 157 8,0 1 15 0.009542980 697 D W 223546 + 8 [kjournald] 158 8,0 1 16 0.013542170 0 C W 223546 + 8 [0] 159... 160^C 161... 162CPU1 (8,0): 163 Reads Queued: 0, 0KiB Writes Queued: 7, 128KiB 164 Read Dispatches: 0, 0KiB Write Dispatches: 7, 128KiB 165 Reads Completed: 0, 0KiB Writes Completed: 11, 168KiB 166 Read Merges: 0 Write Merges: 25 167 IO unplugs: 0 Timer unplugs: 0 168... 169CPU3 (8,0): 170 Reads Queued: 0, 0KiB Writes Queued: 1, 28KiB 171 Read Dispatches: 0, 0KiB Write Dispatches: 1, 28KiB 172 Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB 173 Read Merges: 0 Write Merges: 6 174 IO unplugs: 0 Timer unplugs: 0 175 176Total (8,0): 177 Reads Queued: 0, 0KiB Writes Queued: 11, 168KiB 178 Read Dispatches: 0, 0KiB Write Dispatches: 11, 168KiB 179 Reads Completed: 0, 0KiB Writes Completed: 11, 168KiB 180 Read Merges: 0 Write Merges: 31 181 IO unplugs: 0 Timer unplugs: 3 182 183Events (8,0): 89 entries, 0 skips 184\end{verbatim} 185 186A \emph{btrace} script is included in the distribution to ease live 187tracing of devices. The above could also be accomplished by issuing: 188 189\begin{verbatim} 190% btrace /dev/sda 191\end{verbatim} 192 193By default, \emph{btrace} runs the trace in quiet mode so it will not 194include statistics when you break the run. Add the \emph{-S} option to 195get that dumped as well. 196 197\subsection{\label{sec:pc-blktrace}blktrace -- SCSI commands} 198 199The previous section showed typical file system io actions, but blktrace 200can also show SCSI commands going in and out of the queue as submitted 201by applications using the SCSI Generic (\emph{sg}) interface. 202 203\begin{verbatim} 204% btrace /dev/cdrom 205[...] 206 3,0 0 25 0.004884107 13528 G R 0 + 0 [inquiry] 207 3,0 0 26 0.004890361 13528 I R 56 (12 00 00 00 38 ..) [inquiry] 208 3,0 0 27 0.004891223 13528 P R [inquiry] 209 3,0 0 28 0.004893250 13528 D R 56 (12 00 00 00 38 ..) [inquiry] 210 3,0 0 29 0.005344910 0 C R (12 00 00 00 38 ..) [0] 211\end{verbatim} 212 213Here we see a program issuing an INQUIRY command to the CDROM device. 214The program requested a read of 56 bytes of data, the CDB is included 215in parenthesis after the data length. The completion event shows shows 216that the command completed successfully. Tracing SCSI commands can be 217very useful for debugging problems with programs talking directly to the 218device. An example of that would be \emph{cdrecord} burning. 219 220\subsection{\label{sec:blktrace-post}blktrace -- post-processing} 221 222Another way to run blktrace is to have blktrace save data away for later 223formatting by blkparse. This would be useful if you want to get 224measurements while running specific loads. 225 226To do this, one would specify the device (or devices) to be watched. Then 227go run you test cases. Stop the trace, and at your leisure utilize 228blkparse to see the results. 229 230In this example, devices /dev/sdaa, /dev/sdc and /dev/sdo are used in an 231LVM volume called adb3/vol. 232 233\begin{verbatim} 234% blktrace /dev/sdaa /dev/sdc /dev/sdo & 235[1] 9713 236% 237% mkfs -t ext3 /dev/adb3/vol 238mke2fs 1.35 (28-Feb-2004) 239Filesystem label= 240OS type: Linux 241Block size=4096 (log=2) 242Fragment size=4096 (log=2) 24316793600 inodes, 33555456 blocks 2441677772 blocks (5.00%) reserved for the super user 245First data block=0 246Maximum filesystem blocks=4294967296 2471025 block groups 24832768 blocks per group, 32768 fragments per group 24916384 inodes per group 250Superblock backups stored on blocks: 251 32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 252 4096000, 7962624, 11239424, 20480000, 23887872 253 254Writing inode tables: done 255Creating journal (8192 blocks): done 256Writing superblocks and filesystem accounting information: done 257 258This filesystem will be automatically checked every 27 mounts or 259180 days, whichever comes first. Use tune2fs -c or -i to override. 260% 261% kill -15 9713 262\end{verbatim} 263 264Then you could process the events later: 265 266\begin{verbatim} 267% 268% blkparse sdaa sdc sdo > events 269% less events 270 8,32 1 1 0.000000000 9728 G R 384 + 32 [mkfs.ext3] 271 8,32 1 2 0.000001959 9728 P R [mkfs.ext3] 272 8,32 1 3 0.000002446 9728 Q R 384 + 32 [mkfs.ext3] 273 8,32 1 4 0.000005110 9728 D R 384 + 32 [mkfs.ext3] 274 8,32 3 5 0.000200570 0 C R 384 + 32 [0] 275 8,224 3 1 0.021658989 9728 G R 384 + 32 [mkfs.ext3] 276... 277 65,160 3 163392 41.117070504 0 C W 87469088 + 1376 [0] 278 8,32 3 163374 41.122683668 0 C W 88168160 + 1376 [0] 279 65,160 3 163393 41.129952433 0 C W 87905984 + 1376 [0] 280 65,160 3 163394 41.130049431 0 D W 89129344 + 1376 [swapper] 281 65,160 3 163395 41.130067135 0 D W 89216704 + 1376 [swapper] 282 65,160 3 163396 41.130083785 0 D W 89304096 + 1376 [swapper] 283 65,160 3 163397 41.130099455 0 D W 89391488 + 1376 [swapper] 284 65,160 3 163398 41.130114732 0 D W 89478848 + 1376 [swapper] 285 65,160 3 163399 41.130128885 0 D W 89481536 + 64 [swapper] 286 8,32 3 163375 41.134758196 0 C W 86333152 + 1376 [0] 287 65,160 3 163400 41.142229726 0 C W 89129344 + 1376 [0] 288 65,160 3 163401 41.144952314 0 C W 89481536 + 64 [0] 289 8,32 3 163376 41.147441930 0 C W 88342912 + 1376 [0] 290 65,160 3 163402 41.155869604 0 C W 89478848 + 1376 [0] 291 8,32 3 163377 41.159466082 0 C W 86245760 + 1376 [0] 292 65,160 3 163403 41.166944976 0 C W 89216704 + 1376 [0] 293 65,160 3 163404 41.178968252 0 C W 89304096 + 1376 [0] 294 65,160 3 163405 41.191860173 0 C W 89391488 + 1376 [0] 295... 296Events (sdo): 0 entries, 0 skips 297 298CPU0 (65,160): 299 Reads Queued: 0, 0KiB Writes Queued: 9, 5,520KiB 300 Read Dispatches: 0, 0KiB Write Dispatches: 0, 0KiB 301 Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB 302 Read Merges: 0 Write Merges: 336 303 IO unplugs: 0 Timer unplugs: 0 304CPU1 (65,160): 305 Reads Queued: 2,411, 38,576KiB Writes Queued: 769, 425,408KiB 306 Read Dispatches: 2,407, 38,512KiB Write Dispatches: 118, 61,680KiB 307 Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB 308 Read Merges: 0 Write Merges: 25,819 309 IO unplugs: 0 Timer unplugs: 4 310CPU2 (65,160): 311 Reads Queued: 2, 32KiB Writes Queued: 18, 10,528KiB 312 Read Dispatches: 2, 32KiB Write Dispatches: 3, 1,344KiB 313 Reads Completed: 0, 0KiB Writes Completed: 0, 0KiB 314 Read Merges: 0 Write Merges: 640 315 IO unplugs: 0 Timer unplugs: 0 316CPU3 (65,160): 317 Reads Queued: 20,572, 329,152KiB Writes Queued: 594, 279,712KiB 318 Read Dispatches: 20,576, 329,216KiB Write Dispatches: 1,474, 740,720KiB 319 Reads Completed: 22,985, 367,760KiB Writes Completed: 1,390, 721,168KiB 320 Read Merges: 0 Write Merges: 16,888 321 IO unplugs: 0 Timer unplugs: 0 322 323Total (65,160): 324 Reads Queued: 22,985, 367,760KiB Writes Queued: 1,390, 721,168KiB 325 Read Dispatches: 22,985, 367,760KiB Write Dispatches: 1,595, 803,744KiB 326 Reads Completed: 22,985, 367,760KiB Writes Completed: 1,390, 721,168KiB 327 Read Merges: 0 Write Merges: 43,683 328 IO unplugs: 0 Timer unplugs: 4 329... 330\end{verbatim} 331 332%---------------------------- 333\newpage\section{\label{sec:blktrace-ug}blktrace User Guide} 334 335The \emph{blktrace} utility extracts event traces from the kernel (via 336the relaying through the debug file system). Some background details 337concerning the run-time behaviour of blktrace will help to understand some 338of the more arcane command line options: 339 340\begin{itemize} 341 \item blktrace receives data from the kernel in buffers passed up 342 through the debug file system (relay). Each device being traced has 343 a file created in the mounted directory for the debugfs, which defaults 344 to \emph{/sys/kernel/debug} -- this can be overridden with the \emph{-r} 345 command line argument. 346 347 \item blktrace defaults to collecting \emph{all} events that can be 348 traced. To limit the events being captured, you can specify one or 349 more filter masks via the \emph{-a} option. 350 351 Alternatively, one may specify the entire mask utilizing a hexadecimal 352 value that is version-specific. (Requires understanding of the internal 353 representation of the filter mask.) 354 355 \item As noted above, the events are passed up via a series of buffers 356 stored into debugfs files. The size and number of buffers can be 357 specified via the \emph{-b} and \emph{-n} arguments respectively. 358 359 \item blktrace stores the extracted data into files stored in the 360 \emph{local} directory. The format of the file names is (by default) 361 \emph{device}.blktrace.\emph{cpu}, where \emph{device} is the base 362 device name (e.g, if we are tracing /dev/sda, the base device name would 363 be \emph{sda}); and \emph{cpu} identifies a CPU for the event stream. 364 365 The \emph{device} portion of the event file name can be changed via 366 the \emph{-o} option. 367 368 \item blktrace may also be run concurrently with blkparse to produce 369 \emph{live} output -- to do this specify \emph{-o -} for blktrace. 370 371 \item The default behaviour for blktrace is to run forever until explicitly killed by the user (via a control-C, or \emph{kill} utility invocation). There are two ways to modify this: 372 373 \begin{enumerate} 374 \item You may utilize the blktrace utility itself to \emph{kill} 375 a running trace -- via the \emph{-k} option. 376 377 \item You can specify a run-time duration for blktrace via the 378 \emph{-w} option -- then blktrace will run for the specified number 379 of seconds, and then halt. 380 \end{enumerate} 381\end{itemize} 382 383\subsection{\label{sec:blktrace-args}Command line arguments} 384\begin{tabular}{|l|l|l|}\hline 385Short & Long & Description \\ \hline\hline 386-A \emph{hex-mask} & --set-mask=\emph{hex-mask} & Set filter mask to \emph{hex-mask} \\ \hline 387-a \emph{mask} & --act-mask=\emph{mask} & Add \emph{mask} to current filter (see below for masks) \\ \hline 388-b \emph{size} & --buffer-size=\emph{size} & Specifies buffer size for event extraction (scaled by $2^{10}$) \\ \hline 389-d \emph{dev} & --dev=\emph{dev} & Adds \emph{dev} as a device to trace \\ \hline 390-k & --kill & Kill on-going trace \\ \hline 391-n \emph{num-sub} & --num-sub=\emph{num-sub} & Specifies number of buffers to use \\ \hline 392-o \emph{file} & --output=\emph{file} & Prepend \emph{file} to output file name(s) \\ \hline 393-r \emph{rel-path} & --relay=\emph{rel-path} & Specifies debugfs mount point \\ \hline 394-V & --version & Outputs version \\ \hline 395-w \emph{seconds} & --stopwatch=\emph{seconds} & Sets run time to the number of seconds specified \\ \hline 396-I \emph{devs file}& --input-devs=\emph{devs file}& Adds devices found in \emph{devs file} to list of devices to trace. \\ 397 & & (One device per line.) \\ \hline 398\end{tabular} 399 400\subsubsection{\label{sec:filter-mask}Filter Masks} 401The following masks may be passed with the \emph{-a} command line 402option, multiple filters may be combined via multiple \emph{-a} command 403line options.\smallskip 404 405\begin{tabular}{|l|l|}\hline 406barrier & \emph{barrier} attribute \\ \hline 407complete & \emph{completed} by driver \\ \hline 408fs & \emph{FS} requests \\ \hline 409issue & \emph{issued} to driver \\ \hline 410pc & \emph{packet command} events \\ \hline 411queue & \emph{queue} operations \\ \hline 412read & \emph{read} traces \\ \hline 413requeue & \emph{requeue} operations \\ \hline 414sync & \emph{synchronous} attribute \\ \hline 415write & \emph{write} traces \\ \hline 416notify & \emph{notify} trace messages \\ \hline 417\end{tabular} 418 419\subsubsection{\label{sec:request-types}Request types} 420blktrace disguingishes between two types of block layer requests, 421file system and scsi commands. The former are dubbed \emph{fs} 422requests, the latter \emph{pc} requests. File system requests are 423normal read/write operations, ie any type of read or write from a 424specific disk location at a given size. These requests typically 425originate from a user process, but they may also be initiated by 426the vm flushing dirty data to disk or the file system syncing 427a super or journal block to disk. \emph{pc} requests are SCSI 428commands. blktrace sends the command data block as a payload 429so that blkparse can decode it. 430 431%---------------------------- 432\newpage\section{\label{sec:blkparse-ug}blkparse User Guide} 433 434The \emph{blkparse} utility will attempt to combine streams of events 435for various devices on various CPUs, and produce a formatted output of 436the event information. As with blktrace, some details concerning blkparse 437will help in understanding the command line options presented below. 438 439\begin{itemize} 440 \item By default, blkparse expects to run in a post-processing mode 441 -- one where the trace events have been saved by a previous run 442 of blktrace, and blkparse is combining event streams and dumping 443 formatted data. 444 445 blkparse \emph{may} be run in a \emph{live} manner concurrently with 446 blktrace by specifying \emph{-i -} to blkparse, and combining it with 447 the live option for blktrace. An example would be: 448 449 \begin{verbatim} 450 % blktrace -d /dev/sda -o - | blkparse -i - 451 \end{verbatim} 452 453 \item You can set how many blkparse batches event reads via the 454 \emph{-b} option, the default is to handle events in batches of 512. 455 456 \item If you have saved event traces in blktrace with different output 457 names (via the \emph{-o} option to blktrace), you must specify the 458 same \emph{input} name via the \emph{-i} option. 459 460 \item The format of the output data can be controlled via the \emph{-f} 461 or \emph{-F} options -- see section~\ref{sec:blkparse-format} for details. 462 463 By default, blkparse sends formatted data to standard output. This may 464 be changed via the \emph{-o} option, or text output can be disabled 465 via the\emph{-O} option. A merged binary stream can be produced using 466 the \emph{-d} option. 467 468\end{itemize} 469 470\newpage\subsection{\label{sec:blkparse-args}Command line arguments} 471\begin{tabular}{|l|l|l|}\hline 472Short & Long & Description \\ \hline\hline 473-b \emph{batch} & --batch={batch} & Standard input read batching \\ \hline 474 475-i \emph{file} & --input=\emph{file} & Specifies base name for input files -- default is \emph{device}.blktrace.\emph{cpu}. \\ 476 & & As noted above, specifying \emph{-i -} runs in \emph{live} mode with blktrace \\ 477 & & (reading data from standard in). \\ \hline 478 479-F \emph{typ,fmt} & --format=\emph{typ,fmt} & Sets output format \\ 480-f \emph{fmt} & --format-spec=\emph{fmt} & (See section~\ref{sec:blkparse-format} for details.) \\ 481 & & \\ 482 & & The -f form specifies a format for all events \\ 483 & & \\ 484 & & The -F form allows one to specify a format for a specific \\ 485 & & event type. The single-character \emph{typ} field is one of the \\ 486 & & action specifiers in section~\ref{sec:act-table} \\ \hline 487 488 489-m & --missing & Print missing entries\\ \hline 490 491-h & --hash-by-name & Hash processes by name, not by PID\\ \hline 492 493-o \emph{file} & --output=\emph{file} & Output file \\ \hline 494-O & --no-text-output & Do \emph{not} produce text output, used for binary (-d) only \\ \hline 495 496-d \emph{file} & --dump-binary=\emph{file} & Binary output file \\ \hline 497 498-q & --quiet & Quite mode \\ \hline 499 500-s & --per-program-stats & Displays data sorted by program \\ \hline 501 502-t & --track-ios & Display time deltas per IO \\ \hline 503 504-w \emph{span} & --stopwatch=\emph{span} & Display traces for the \emph{span} specified -- where span can be: \\ 505 & & \emph{end-time} -- Display traces from time 0 through \emph{end-time} (in ns) \\ 506 & & or \\ 507 & & \emph{start:end-time} -- Display traces from time \emph{start} \\ 508 & & through {end-time} (in ns). \\ \hline 509 510-M & --no-msgs & Do not add messages to binary output file \\\hline 511-v & --verbose & More verbose marginal on marginal errors \\ \hline 512-V & --version & Display version \\ \hline 513 514\end{tabular} 515 516\newpage 517\subsection{\label{sec:blkparse-actions}Trace actions} 518 519\begin{description} 520 \item[C -- complete] A previously issued request has been completed. 521 The output will detail the sector and size of that request, as well 522 as the success or failure of it. 523 524 \item[D -- issued] A request that previously resided on the block layer 525 queue or in the io scheduler has been sent to the driver. 526 527 \item[I -- inserted] A request is being sent to the io scheduler for 528 addition to the internal queue and later service by the driver. The 529 request is fully formed at this time. 530 531 \item[Q -- queued] This notes intent to queue io at the given location. 532 No real requests exists yet. 533 534 \item[B -- bounced] The data pages attached to this \emph{bio} are 535 not reachable by the hardware and must be bounced to a lower memory 536 location. This causes a big slowdown in io performance, since the data 537 must be copied to/from kernel buffers. Usually this can be fixed with 538 using better hardware - either a better io controller, or a platform 539 with an IOMMU. 540 541 \item[m -- message] Text message generated via kernel call to 542 \texttt{blk\_add\_trace\_msg}. 543 544 \item[M -- back merge] A previously inserted request exists that ends 545 on the boundary of where this io begins, so the io scheduler can merge 546 them together. 547 548 \item[F -- front merge] Same as the back merge, except this io ends 549 where a previously inserted requests starts. 550 551 \item[G -- get request] To send any type of request to a block device, 552 a \emph{struct request} container must be allocated first. 553 554 \item[S -- sleep] No available request structures were available, so 555 the issuer has to wait for one to be freed. 556 557 \item[P -- plug] When io is queued to a previously empty block device 558 queue, Linux will plug the queue in anticipation of future ios being 559 added before this data is needed. 560 561 \item[U -- unplug] Some request data already queued in the device, 562 start sending requests to the driver. This may happen automatically 563 if a timeout period has passed (see next entry) or if a number of 564 requests have been added to the queue. 565 566 \item[T -- unplug due to timer] If nobody requests the io that was queued 567 after plugging the queue, Linux will automatically unplug it after a 568 defined period has passed. 569 570 \item[X -- split] On raid or device mapper setups, an incoming io may 571 straddle a device or internal zone and needs to be chopped up into 572 smaller pieces for service. This may indicate a performance problem due 573 to a bad setup of that raid/dm device, but may also just be part of 574 normal boundary conditions. dm is notably bad at this and will clone 575 lots of io. 576 577 \item[A -- remap] For stacked devices, incoming io is remapped to device 578 below it in the io stack. The remap action details what exactly is 579 being remapped to what. 580 581\end{description} 582 583\subsection{\label{sec:blkparse-format}Output Description and Formatting} 584 585The output from blkparse can be tailored for specific use - in particular, 586to ease parsing of output, and/or limit output fields to those the user 587wants to see. The data for fields which can be output include: 588 589\smallskip 590\begin{tabular}{|l|l|}\hline 591Field & Description \\ 592Specifier & \\ \hline\hline 593\emph{a} & Action, a (small) string (1 or 2 characters) -- see table below for more details \\ \hline 594\emph{c} & CPU id \\ \hline 595\emph{C} & Command \\ \hline 596\emph{d} & RWBS field, a (small) string (1-3 characters) -- see section below for more details \\ \hline 597\emph{D} & 7-character string containing the major and minor numbers of 598the event's device \\ 599 & (separated by a comma). \\ \hline 600\emph{e} & Error value \\ \hline 601\emph{m} & Minor number of event's device. \\ \hline 602\emph{M} & Major number of event's device. \\ \hline 603\emph{n} & Number of blocks \\ \hline 604\emph{N} & Number of bytes \\ \hline 605\emph{p} & Process ID \\ \hline 606\emph{P} & Display packet data -- series of hexadecimal values\\ \hline 607\emph{s} & Sequence numbers \\ \hline 608\emph{S} & Sector number \\ \hline 609\emph{t} & Time stamp (nanoseconds) \\ \hline 610\emph{T} & Time stamp (seconds) \\ \hline 611\emph{u} & Elapsed value in microseconds (\emph{-t} command line option) \\ \hline 612\emph{U} & Payload unsigned integer \\ \hline 613\end{tabular} 614 615Note that the user can optionally specify field display width, and 616optionally a left-aligned specifier. These precede field specifiers, 617with a '\%' character, followed by the optional left-alignment specifer 618(-) followed by the width (a decimal number) and then the field. 619 620Thus, to specify the command in a 12-character field that is left aligned: 621 622\begin{verbatim} 623-f "%-12C" 624\end{verbatim} 625 626\newpage 627\subsubsection{\label{sec:act-table}Action Table} 628The following table shows the various actions which may be output. 629 630\begin{tabular}{|l|l|}\hline 631Act & Description \\ \hline\hline 632A & IO was remapped to a different device \\ \hline 633B & IO bounced \\ \hline 634C & IO completion \\ \hline 635D & IO issued to driver \\ \hline 636F & IO front merged with request on queue \\ \hline 637G & Get request \\ \hline 638I & IO inserted onto request queue \\ \hline 639M & IO back merged with request on queue \\ \hline 640P & Plug request \\ \hline 641Q & IO handled by request queue code \\ \hline 642S & Sleep request \\ \hline 643T & Unplug due to timeout \\ \hline 644U & Unplug request \\ \hline 645X & Split \\ \hline 646\end{tabular} 647 648\subsubsection{\label{sec:act-table}RWBS Description} 649This is a small string containing at least one character ('R' for read, 650'W' for write, or 'D' for block discard operation), and optionally either 651a 'B' (for barrier operations) or 'S' (for synchronous operations). 652 653\subsubsection{\label{sec:default-output}Default output} 654 655The standard \emph{header} (or initial fields displayed) include: 656 657\begin{verbatim} 658"%D %2c %8s %5T.%9t %5p %2a %3d " 659\end{verbatim} 660 661Breaking this down: 662 663\begin{description} 664 \item[\%D] Displays the event's device major/minor as: \%3d,\%-3d. 665 \item[\%2c] CPU ID (2-character field). 666 \item[\%8s] Sequence number 667 \item[\%5T.\%9t] 5-charcter field for the seconds portion of the 668 time stamp and a 9-character field for the nanoseconds in the time stamp. 669 \item[\%5p] 5-character field for the process ID. 670 \item[\%2a] 2-character field for one of the actions. 671 \item[\%3d] 3-character field for the RWBS data. 672\end{description} 673 674Seeing this in action: 675 676\begin{verbatim} 677 8,0 3 1 0.000000000 697 G W 223490 + 8 [kjournald] 678\end{verbatim} 679 680The header is the data in this line up to the 223490 (starting block). 681 682The default output for all event types includes this header. 683 684\paragraph{Default output per action} 685 686\begin{description} 687 \item[C -- complete] If a payload is present, this is presented between 688 parenthesis following the header, followed by the error value. 689 690 If no payload is present, the sector and number of blocks are presented 691 (with an intervening plus (+) character). If the \emph{-t} option 692 was specified, then the elapsed time is presented. In either case, 693 it is followed by the error value for the completion. 694 695 \item[D -- issued] 696 \item[I -- inserted] 697 \item[Q -- queued] 698 \item[B -- bounced] If a payload is present, the number of payload bytes 699 is output, followed by the payload in hexadecimal between parenthesis. 700 701 If no payload is present, the sector and number of blocks are presented 702 (with an intervening plus (+) character). If the \emph{-t} option was 703 specified, then the elapsed time is presented (in parenthesis). In 704 either case, it is followed by the command associated with the event 705 (surrounded by square brackets). 706 707 \item[M -- back merge] 708 \item[F -- front merge] 709 \item[G -- get request] 710 \item[S -- sleep] The starting sector and number of blocks is output 711 (with an intervening plus (+) character), followed by the command 712 associated with the event (surrounded by square brackets). 713 714 \item[P -- plug] The command associated with the event (surrounded by 715 square brackets) is output. 716 717 \item[U -- unplug] 718 \item[T -- unplug due to timer] The command associated with the event 719 (surrounded by square brackets) is output, followed by the number of 720 requests outstanding. 721 722 \item[X -- split] The original starting sector followed by the new 723 sector (separated by a slash (/) is output, followed by the command 724 associated with the event (surrounded by square brackets). 725 726 \item[A -- remap] Sector and length is output, along with the original 727 device and sector offset. 728 729 \item[m -- message] The supplied message is appended to the end of 730 the standard header. 731 732\end{description} 733 734%------------------------------ 735\newpage 736\newpage\section*{\label{sec:blktrace-kg}Appendix: blktrace Kernel Guide} 737 738The blktrace facility provides an efficient event transfer mechanism which 739supplies block IO layer state transition data via the relay 740filesystem. This section provides some details as to the interfaces 741blktrace utilizes in the kernel to effect this. It is good background data 742to help understand some of the outputs and command-line options above. 743 744\subsection{blktrace.h Definitions} 745Files which include $<linux/blktrace.h>$ are supplied with the following 746definitions: 747 748\subsubsection{Trace Action Specifiers} 749\begin{tabular}{|l|l|}\hline 750 BLK\_TA\_QUEUE & (RQ) Command queued to request\_queue. \\ 751 & (BIO) Command queued by elevator. \\ \hline 752 BLK\_TA\_BACKMERGE & Back merging elevator operation \\ \hline 753 BLK\_TA\_FRONTMERGE & Front merging elevator operation \\ \hline 754 BLK\_TA\_GETRQ & Free request retrieved. \\ \hline 755 BLK\_TA\_SLEEPRQ & No requests available, device unplugged. \\ \hline 756 BLK\_TA\_REQUEUE & Request requeued. \\ \hline 757 BLK\_TA\_ISSUE & Command set to driver for request\_queue. \\ \hline 758 BLK\_TA\_COMPLETE & Command completed by driver. \\ \hline 759 BLK\_TA\_PLUG & Device is plugged \\ \hline 760 BLK\_TA\_UNPLUG\_IO & Unplug device as IO is made available. \\ \hline 761 BLK\_TA\_UNPLUG\_TIMER & Unplug device after timer expired. \\ \hline 762 BLK\_TA\_INSERT & Insert request into queue. \\ \hline 763 BLK\_TA\_SPLIT & BIO split into 2 or more requests. \\ \hline 764 BLK\_TA\_BOUNCE & BIO was bounced \\ \hline 765 BLK\_TA\_REMAP & BIO was remapped \\ \hline 766\end{tabular} 767 768%.......................................... 769\subsection{blktrace.h Routines} 770Files which include $<linux/blktrace.h>$ are supplied with the following 771kernel routine invocable interfaces: 772 773\begin{description} 774 \item[blk\_add\_trace\_rq(struct request\_queue *q, struct request\_queue 775 *rq, u32 what)] 776 Adds a trace event describing the state change of the passed in 777 request\_queue. The \emph{what} parameter describes the change in 778 the request\_queue state, and is one of the request queue action 779 specifiers -- BLK\_TA\_QUEUE, BLK\_TA\_REQUEUE, BLK\_TA\_ISSUE, 780 or BLK\_TA\_COMPLETE. 781 782 \item[blk\_add\_trace\_bio(struct request\_queue *q, struct bio *bio, 783 u32 what)] 784 Adds a trace event for the BIO passed in. The \emph{what} parameter 785 describes the action being performed on the BIO, and is one of 786 BLK\_TA\_BACKMERGE, BLK\_TA\_FRONTMERGE, or BLK\_TA\_QUEUE. 787 788 \item[blk\_add\_trace\_generic(struct request\_queue *q, struct bio *bio, 789 int rw, u32 what)] 790 Adds a \emph{generic} trace event -- not one of the request queue 791 or BIO traces. The \emph{what} parameter describes the action being 792 performed on the BIO (if bio is non-NULL), and is one of 793 BLK\_TA\_PLUG, BLK\_TA\_GETRQ or BLK\_TA\_SLEEPRQ. 794 795 \item[blk\_add\_trace\_pdu\_int(struct request\_queue *q, u32 what, 796 u32 pdu)] 797 Adds a trace with some payload data -- in this case, an unsigned 798 32-bit entity (the \emph{pdu} parameter). The \emph{what} parameter 799 describes the nature of the payload, and is one of 800 BLK\_TA\_UNPLUG\_IO or BLK\_TA\_UNPLUG\_TIMER. 801 802 \item[blk\_add\_trace\_remap(struct request\_queue *q, struct bio *bio, 803 dev\_t dev, sector\_t sector)] 804 Adds a trace with a remap event. \emph{dev} and \emph{sector} denote 805 the original device this \emph{bio} was mapped from. 806 807 \item[blk\_add\_trace\_msg(struct request\_queue *q, char *fmt, ...)] 808 Adds a formatted message to the output stream. The total message 809 size can not exceed BLK\_TN\_MSG\_MSG characters (currently 810 1024). Standard format conversions are supported (as supplied 811 by \texttt{vscnprintf}. 812 813\end{description} 814\end{document} 815