Debian Codegraph @

Documentation/driver-api/crypto/iaa/iaa-crypto.rst

raw
1 .. SPDX-License-Identifier: GPL-2.0
2
3 =========================================
4 IAA Compression Accelerator Crypto Driver
5 =========================================
6
7 Tom Zanussi <[email protected]>
8
9 The IAA crypto driver supports compression/decompression compatible
10 with the DEFLATE compression standard described in RFC 1951, which is
11 the compression/decompression algorithm exported by this module.
12
13 The IAA hardware spec can be found here:
14
15 https://cdrdv2.intel.com/v1/dl/getContent/721858
16
17 The iaa_crypto driver is designed to work as a layer underneath
18 higher-level compression devices such as zswap.
19
20 Users can select IAA compress/decompress acceleration by specifying
21 one of the supported IAA compression algorithms in whatever facility
22 allows compression algorithms to be selected.
23
24 For example, a zswap device can select the IAA 'fixed' mode
25 represented by selecting the 'deflate-iaa' crypto compression
26 algorithm::
27
28 # echo deflate-iaa > /sys/module/zswap/parameters/compressor
29
30 This will tell zswap to use the IAA 'fixed' compression mode for all
31 compresses and decompresses.
32
33 Currently, there is only one compression modes available, 'fixed'
34 mode.
35
36 The 'fixed' compression mode implements the compression scheme
37 specified by RFC 1951 and is given the crypto algorithm name
38 'deflate-iaa'. (Because the IAA hardware has a 4k history-window
39 limitation, only buffers <= 4k, or that have been compressed using a
40 <= 4k history window, are technically compliant with the deflate spec,
41 which allows for a window of up to 32k. Because of this limitation,
42 the IAA fixed mode deflate algorithm is given its own algorithm name
43 rather than simply 'deflate').
44
45
46 Config options and other setup
47 ==============================
48
49 The IAA crypto driver is available via menuconfig using the following
50 path::
51
52 Cryptographic API -> Hardware crypto devices -> Support for Intel(R) IAA Compression Accelerator
53
54 In the configuration file the option called CONFIG_CRYPTO_DEV_IAA_CRYPTO.
55
56 The IAA crypto driver also supports statistics, which are available
57 via menuconfig using the following path::
58
59 Cryptographic API -> Hardware crypto devices -> Support for Intel(R) IAA Compression -> Enable Intel(R) IAA Compression Accelerator Statistics
60
61 In the configuration file the option called CONFIG_CRYPTO_DEV_IAA_CRYPTO_STATS.
62
63 The following config options should also be enabled::
64
65 CONFIG_IRQ_REMAP=y
66 CONFIG_INTEL_IOMMU=y
67 CONFIG_INTEL_IOMMU_SVM=y
68 CONFIG_PCI_ATS=y
69 CONFIG_PCI_PRI=y
70 CONFIG_PCI_PASID=y
71 CONFIG_INTEL_IDXD=m
72 CONFIG_INTEL_IDXD_SVM=y
73
74 IAA is one of the first Intel accelerator IPs that can work in
75 conjunction with the Intel IOMMU. There are multiple modes that exist
76 for testing. Based on IOMMU configuration, there are 3 modes::
77
78 - Scalable
79 - Legacy
80 - No IOMMU
81
82
83 Scalable mode
84 -------------
85
86 Scalable mode supports Shared Virtual Memory (SVM or SVA). It is
87 entered when using the kernel boot commandline::
88
89 intel_iommu=on,sm_on
90
91 with VT-d turned on in BIOS.
92
93 With scalable mode, both shared and dedicated workqueues are available
94 for use.
95
96 For scalable mode, the following BIOS settings should be enabled::
97
98 Socket Configuration > IIO Configuration > Intel VT for Directed I/O (VT-d) > Intel VT for Directed I/O
99
100 Socket Configuration > IIO Configuration > PCIe ENQCMD > ENQCMDS
101
102
103 Legacy mode
104 -----------
105
106 Legacy mode is entered when using the kernel boot commandline::
107
108 intel_iommu=off
109
110 or VT-d is not turned on in BIOS.
111
112 If you have booted into Linux and not sure if VT-d is on, do a "dmesg
113 | grep -i dmar". If you don't see a number of DMAR devices enumerated,
114 most likely VT-d is not on.
115
116 With legacy mode, only dedicated workqueues are available for use.
117
118
119 No IOMMU mode
120 -------------
121
122 No IOMMU mode is entered when using the kernel boot commandline::
123
124 iommu=off.
125
126 With no IOMMU mode, only dedicated workqueues are available for use.
127
128
129 Usage
130 =====
131
132 accel-config
133 ------------
134
135 When loaded, the iaa_crypto driver automatically creates a default
136 configuration and enables it, and assigns default driver attributes.
137 If a different configuration or set of driver attributes is required,
138 the user must first disable the IAA devices and workqueues, reset the
139 configuration, and then re-register the deflate-iaa algorithm with the
140 crypto subsystem by removing and reinserting the iaa_crypto module.
141
142 The :ref:`iaa_disable_script` in the 'Use Cases'
143 section below can be used to disable the default configuration.
144
145 See :ref:`iaa_default_config` below for details of the default
146 configuration.
147
148 More likely than not, however, and because of the complexity and
149 configurability of the accelerator devices, the user will want to
150 configure the device and manually enable the desired devices and
151 workqueues.
152
153 The userspace tool to help doing that is called accel-config. Using
154 accel-config to configure device or loading a previously saved config
155 is highly recommended. The device can be controlled via sysfs
156 directly but comes with the warning that you should do this ONLY if
157 you know exactly what you are doing. The following sections will not
158 cover the sysfs interface but assumes you will be using accel-config.
159
160 The :ref:`iaa_sysfs_config` section in the appendix below can be
161 consulted for the sysfs interface details if interested.
162
163 The accel-config tool along with instructions for building it can be
164 found here:
165
166 https://github.com/intel/idxd-config/#readme
167
168 Typical usage
169 -------------
170
171 In order for the iaa_crypto module to actually do any
172 compression/decompression work on behalf of a facility, one or more
173 IAA workqueues need to be bound to the iaa_crypto driver.
174
175 For instance, here's an example of configuring an IAA workqueue and
176 binding it to the iaa_crypto driver (note that device names are
177 specified as 'iax' rather than 'iaa' - this is because upstream still
178 has the old 'iax' device naming in place) ::
179
180 # configure wq1.0
181
182 accel-config config-wq --group-id=0 --mode=dedicated --type=kernel --priority=10 --name="iaa_crypto" --driver-name="crypto" iax1/wq1.0
183
184 accel-config config-engine iax1/engine1.0 --group-id=0
185
186 # enable IAA device iax1
187
188 accel-config enable-device iax1
189
190 # enable wq1.0 on IAX device iax1
191
192 accel-config enable-wq iax1/wq1.0
193
194 Whenever a new workqueue is bound to or unbound from the iaa_crypto
195 driver, the available workqueues are 'rebalanced' such that work
196 submitted from a particular CPU is given to the most appropriate
197 workqueue available. Current best practice is to configure and bind
198 at least one workqueue for each IAA device, but as long as there is at
199 least one workqueue configured and bound to any IAA device in the
200 system, the iaa_crypto driver will work, albeit most likely not as
201 efficiently.
202
203 The IAA crypto algorigthms is operational and compression and
204 decompression operations are fully enabled following the successful
205 binding of the first IAA workqueue to the iaa_crypto driver.
206
207 Similarly, the IAA crypto algorithm is not operational and compression
208 and decompression operations are disabled following the unbinding of
209 the last IAA worqueue to the iaa_crypto driver.
210
211 As a result, the IAA crypto algorithms and thus the IAA hardware are
212 only available when one or more workques are bound to the iaa_crypto
213 driver.
214
215 When there are no IAA workqueues bound to the driver, the IAA crypto
216 algorithms can be unregistered by removing the module.
217
218
219 Driver attributes
220 -----------------
221
222 There are a couple user-configurable driver attributes that can be
223 used to configure various modes of operation. They're listed below,
224 along with their default values. To set any of these attributes, echo
225 the appropriate values to the attribute file located under
226 /sys/bus/dsa/drivers/crypto/
227
228 The attribute settings at the time the IAA algorithms are registered
229 are captured in each algorithm's crypto_ctx and used for all compresses
230 and decompresses when using that algorithm.
231
232 The available attributes are:
233
234 - verify_compress
235
236 Toggle compression verification. If set, each compress will be
237 internally decompressed and the contents verified, returning error
238 codes if unsuccessful. This can be toggled with 0/1::
239
240 echo 0 > /sys/bus/dsa/drivers/crypto/verify_compress
241
242 The default setting is '1' - verify all compresses.
243
244 - sync_mode
245
246 Select mode to be used to wait for completion of each compresses
247 and decompress operation.
248
249 The crypto async interface support implemented by iaa_crypto
250 provides an implementation that satisfies the interface but does
251 so in a synchronous manner - it fills and submits the IDXD
252 descriptor and then loops around waiting for it to complete before
253 returning. This isn't a problem at the moment, since all existing
254 callers (e.g. zswap) wrap any asynchronous callees in a
255 synchronous wrapper anyway.
256
257 The iaa_crypto driver does however provide true asynchronous
258 support for callers that can make use of it. In this mode, it
259 fills and submits the IDXD descriptor, then returns immediately
260 with -EINPROGRESS. The caller can then either poll for completion
261 itself, which requires specific code in the caller which currently
262 nothing in the upstream kernel implements, or go to sleep and wait
263 for an interrupt signaling completion. This latter mode is
264 supported by current users in the kernel such as zswap via
265 synchronous wrappers. Although it is supported this mode is
266 significantly slower than the synchronous mode that does the
267 polling in the iaa_crypto driver previously mentioned.
268
269 This mode can be enabled by writing 'async_irq' to the sync_mode
270 iaa_crypto driver attribute::
271
272 echo async_irq > /sys/bus/dsa/drivers/crypto/sync_mode
273
274 Async mode without interrupts (caller must poll) can be enabled by
275 writing 'async' to it (please see Caveat)::
276
277 echo async > /sys/bus/dsa/drivers/crypto/sync_mode
278
279 The mode that does the polling in the iaa_crypto driver can be
280 enabled by writing 'sync' to it::
281
282 echo sync > /sys/bus/dsa/drivers/crypto/sync_mode
283
284 The default mode is 'sync'.
285
286 Caveat: since the only mechanism that iaa_crypto currently implements
287 for async polling without interrupts is via the 'sync' mode as
288 described earlier, writing 'async' to
289 '/sys/bus/dsa/drivers/crypto/sync_mode' will internally enable the
290 'sync' mode. This is to ensure correct iaa_crypto behavior until true
291 async polling without interrupts is enabled in iaa_crypto.
292
293 .. _iaa_default_config:
294
295 IAA Default Configuration
296 -------------------------
297
298 When the iaa_crypto driver is loaded, each IAA device has a single
299 work queue configured for it, with the following attributes::
300
301 mode "dedicated"
302 threshold 0
303 size Total WQ Size from WQCAP
304 priority 10
305 type IDXD_WQT_KERNEL
306 group 0
307 name "iaa_crypto"
308 driver_name "crypto"
309
310 The devices and workqueues are also enabled and therefore the driver
311 is ready to be used without any additional configuration.
312
313 The default driver attributes in effect when the driver is loaded are::
314
315 sync_mode "sync"
316 verify_compress 1
317
318 In order to change either the device/work queue or driver attributes,
319 the enabled devices and workqueues must first be disabled. In order
320 to have the new configuration applied to the deflate-iaa crypto
321 algorithm, it needs to be re-registered by removing and reinserting
322 the iaa_crypto module. The :ref:`iaa_disable_script` in the 'Use
323 Cases' section below can be used to disable the default configuration.
324
325 Statistics
326 ==========
327
328 If the optional debugfs statistics support is enabled, the IAA crypto
329 driver will generate statistics which can be accessed in debugfs at::
330
331 # ls -al /sys/kernel/debug/iaa-crypto/
332 total 0
333 drwxr-xr-x 2 root root 0 Mar 3 07:55 .
334 drwx------ 53 root root 0 Mar 3 07:55 ..
335 -rw-r--r-- 1 root root 0 Mar 3 07:55 global_stats
336 -rw-r--r-- 1 root root 0 Mar 3 07:55 stats_reset
337 -rw-r--r-- 1 root root 0 Mar 3 07:55 wq_stats
338
339 The global_stats file shows a set of global statistics collected since
340 the driver has been loaded or reset::
341
342 # cat global_stats
343 global stats:
344 total_comp_calls: 4300
345 total_decomp_calls: 4164
346 total_sw_decomp_calls: 0
347 total_comp_bytes_out: 5993989
348 total_decomp_bytes_in: 5993989
349 total_completion_einval_errors: 0
350 total_completion_timeout_errors: 0
351 total_completion_comp_buf_overflow_errors: 136
352
353 The wq_stats file shows per-wq stats, a set for each iaa device and wq
354 in addition to some global stats::
355
356 # cat wq_stats
357 iaa device:
358 id: 1
359 n_wqs: 1
360 comp_calls: 0
361 comp_bytes: 0
362 decomp_calls: 0
363 decomp_bytes: 0
364 wqs:
365 name: iaa_crypto
366 comp_calls: 0
367 comp_bytes: 0
368 decomp_calls: 0
369 decomp_bytes: 0
370
371 iaa device:
372 id: 3
373 n_wqs: 1
374 comp_calls: 0
375 comp_bytes: 0
376 decomp_calls: 0
377 decomp_bytes: 0
378 wqs:
379 name: iaa_crypto
380 comp_calls: 0
381 comp_bytes: 0
382 decomp_calls: 0
383 decomp_bytes: 0
384
385 iaa device:
386 id: 5
387 n_wqs: 1
388 comp_calls: 1360
389 comp_bytes: 1999776
390 decomp_calls: 0
391 decomp_bytes: 0
392 wqs:
393 name: iaa_crypto
394 comp_calls: 1360
395 comp_bytes: 1999776
396 decomp_calls: 0
397 decomp_bytes: 0
398
399 iaa device:
400 id: 7
401 n_wqs: 1
402 comp_calls: 2940
403 comp_bytes: 3994213
404 decomp_calls: 4164
405 decomp_bytes: 5993989
406 wqs:
407 name: iaa_crypto
408 comp_calls: 2940
409 comp_bytes: 3994213
410 decomp_calls: 4164
411 decomp_bytes: 5993989
412 ...
413
414 Writing to 'stats_reset' resets all the stats, including the
415 per-device and per-wq stats::
416
417 # echo 1 > stats_reset
418 # cat wq_stats
419 global stats:
420 total_comp_calls: 0
421 total_decomp_calls: 0
422 total_comp_bytes_out: 0
423 total_decomp_bytes_in: 0
424 total_completion_einval_errors: 0
425 total_completion_timeout_errors: 0
426 total_completion_comp_buf_overflow_errors: 0
427 ...
428
429
430 Use cases
431 =========
432
433 Simple zswap test
434 -----------------
435
436 For this example, the kernel should be configured according to the
437 dedicated mode options described above, and zswap should be enabled as
438 well::
439
440 CONFIG_ZSWAP=y
441
442 This is a simple test that uses iaa_compress as the compressor for a
443 swap (zswap) device. It sets up the zswap device and then uses the
444 memory_memadvise program listed below to forcibly swap out and in a
445 specified number of pages, demonstrating both compress and decompress.
446
447 The zswap test expects the work queues for each IAA device on the
448 system to be configured properly as a kernel workqueue with a
449 workqueue driver_name of "crypto".
450
451 The first step is to make sure the iaa_crypto module is loaded::
452
453 modprobe iaa_crypto
454
455 If the IAA devices and workqueues haven't previously been disabled and
456 reconfigured, then the default configuration should be in place and no
457 further IAA configuration is necessary. See :ref:`iaa_default_config`
458 below for details of the default configuration.
459
460 If the default configuration is in place, you should see the iaa
461 devices and wq0s enabled::
462
463 # cat /sys/bus/dsa/devices/iax1/state
464 enabled
465 # cat /sys/bus/dsa/devices/iax1/wq1.0/state
466 enabled
467
468 To demonstrate that the following steps work as expected, these
469 commands can be used to enable debug output::
470
471 # echo -n 'module iaa_crypto +p' > /sys/kernel/debug/dynamic_debug/control
472 # echo -n 'module idxd +p' > /sys/kernel/debug/dynamic_debug/control
473
474 Use the following commands to enable zswap::
475
476 # echo 0 > /sys/module/zswap/parameters/enabled
477 # echo 50 > /sys/module/zswap/parameters/max_pool_percent
478 # echo deflate-iaa > /sys/module/zswap/parameters/compressor
479 # echo 1 > /sys/module/zswap/parameters/enabled
480 # echo 100 > /proc/sys/vm/swappiness
481 # echo never > /sys/kernel/mm/transparent_hugepage/enabled
482 # echo 1 > /proc/sys/vm/overcommit_memory
483
484 Now you can now run the zswap workload you want to measure. For
485 example, using the memory_memadvise code below, the following command
486 will swap in and out 100 pages::
487
488 ./memory_madvise 100
489
490 Allocating 100 pages to swap in/out
491 Swapping out 100 pages
492 Swapping in 100 pages
493 Swapped out and in 100 pages
494
495 You should see something like the following in the dmesg output::
496
497 [ 404.202972] idxd 0000:e7:02.0: iaa_comp_acompress: dma_map_sg, src_addr 223925c000, nr_sgs 1, req->src 00000000ee7cb5e6, req->slen 4096, sg_dma_len(sg) 4096
498 [ 404.202973] idxd 0000:e7:02.0: iaa_comp_acompress: dma_map_sg, dst_addr 21dadf8000, nr_sgs 1, req->dst 000000008d6acea8, req->dlen 4096, sg_dma_len(sg) 8192
499 [ 404.202975] idxd 0000:e7:02.0: iaa_compress: desc->src1_addr 223925c000, desc->src1_size 4096, desc->dst_addr 21dadf8000, desc->max_dst_size 4096, desc->src2_addr 2203543000, desc->src2_size 1568
500 [ 404.202981] idxd 0000:e7:02.0: iaa_compress_verify: (verify) desc->src1_addr 21dadf8000, desc->src1_size 228, desc->dst_addr 223925c000, desc->max_dst_size 4096, desc->src2_addr 0, desc->src2_size 0
501 ...
502
503 Now that basic functionality has been demonstrated, the defaults can
504 be erased and replaced with a different configuration. To do that,
505 first disable zswap::
506
507 # echo lzo > /sys/module/zswap/parameters/compressor
508 # swapoff -a
509 # echo 0 > /sys/module/zswap/parameters/accept_threshold_percent
510 # echo 0 > /sys/module/zswap/parameters/max_pool_percent
511 # echo 0 > /sys/module/zswap/parameters/enabled
512 # echo 0 > /sys/module/zswap/parameters/enabled
513
514 Then run the :ref:`iaa_disable_script` in the 'Use Cases' section
515 below to disable the default configuration.
516
517 Finally turn swap back on::
518
519 # swapon -a
520
521 Following all that the IAA device(s) can now be re-configured and
522 enabled as desired for further testing. Below is one example.
523
524 The zswap test expects the work queues for each IAA device on the
525 system to be configured properly as a kernel workqueue with a
526 workqueue driver_name of "crypto".
527
528 The below script automatically does that::
529
530 #!/bin/bash
531
532 echo "IAA devices:"
533 lspci -d:0cfe
534 echo "# IAA devices:"
535 lspci -d:0cfe | wc -l
536
537 #
538 # count iaa instances
539 #
540 iaa_dev_id="0cfe"
541 num_iaa=$(lspci -d:${iaa_dev_id} | wc -l)
542 echo "Found ${num_iaa} IAA instances"
543
544 #
545 # disable iaa wqs and devices
546 #
547 echo "Disable IAA"
548
549 for ((i = 1; i < ${num_iaa} * 2; i += 2)); do
550 echo disable wq iax${i}/wq${i}.0
551 accel-config disable-wq iax${i}/wq${i}.0
552 echo disable iaa iax${i}
553 accel-config disable-device iax${i}
554 done
555
556 echo "End Disable IAA"
557
558 echo "Reload iaa_crypto module"
559
560 rmmod iaa_crypto
561 modprobe iaa_crypto
562
563 echo "End Reload iaa_crypto module"
564
565 #
566 # configure iaa wqs and devices
567 #
568 echo "Configure IAA"
569 for ((i = 1; i < ${num_iaa} * 2; i += 2)); do
570 accel-config config-wq --group-id=0 --mode=dedicated --wq-size=128 --priority=10 --type=kernel --name="iaa_crypto" --driver-name="crypto" iax${i}/wq${i}.0
571 accel-config config-engine iax${i}/engine${i}.0 --group-id=0
572 done
573
574 echo "End Configure IAA"
575
576 #
577 # enable iaa wqs and devices
578 #
579 echo "Enable IAA"
580
581 for ((i = 1; i < ${num_iaa} * 2; i += 2)); do
582 echo enable iaa iax${i}
583 accel-config enable-device iax${i}
584 echo enable wq iax${i}/wq${i}.0
585 accel-config enable-wq iax${i}/wq${i}.0
586 done
587
588 echo "End Enable IAA"
589
590 When the workqueues are bound to the iaa_crypto driver, you should
591 see something similar to the following in dmesg output if you've
592 enabled debug output (echo -n 'module iaa_crypto +p' >
593 /sys/kernel/debug/dynamic_debug/control)::
594
595 [ 60.752344] idxd 0000:f6:02.0: add_iaa_wq: added wq 000000004068d14d to iaa 00000000c9585ba2, n_wq 1
596 [ 60.752346] iaa_crypto: rebalance_wq_table: nr_nodes=2, nr_cpus 160, nr_iaa 8, cpus_per_iaa 20
597 [ 60.752347] iaa_crypto: rebalance_wq_table: iaa=0
598 [ 60.752349] idxd 0000:6a:02.0: request_iaa_wq: getting wq from iaa_device 0000000042d7bc52 (0)
599 [ 60.752350] idxd 0000:6a:02.0: request_iaa_wq: returning unused wq 00000000c8bb4452 (0) from iaa device 0000000042d7bc52 (0)
600 [ 60.752352] iaa_crypto: rebalance_wq_table: assigned wq for cpu=0, node=0 = wq 00000000c8bb4452
601 [ 60.752354] iaa_crypto: rebalance_wq_table: iaa=0
602 [ 60.752355] idxd 0000:6a:02.0: request_iaa_wq: getting wq from iaa_device 0000000042d7bc52 (0)
603 [ 60.752356] idxd 0000:6a:02.0: request_iaa_wq: returning unused wq 00000000c8bb4452 (0) from iaa device 0000000042d7bc52 (0)
604 [ 60.752358] iaa_crypto: rebalance_wq_table: assigned wq for cpu=1, node=0 = wq 00000000c8bb4452
605 [ 60.752359] iaa_crypto: rebalance_wq_table: iaa=0
606 [ 60.752360] idxd 0000:6a:02.0: request_iaa_wq: getting wq from iaa_device 0000000042d7bc52 (0)
607 [ 60.752361] idxd 0000:6a:02.0: request_iaa_wq: returning unused wq 00000000c8bb4452 (0) from iaa device 0000000042d7bc52 (0)
608 [ 60.752362] iaa_crypto: rebalance_wq_table: assigned wq for cpu=2, node=0 = wq 00000000c8bb4452
609 [ 60.752364] iaa_crypto: rebalance_wq_table: iaa=0
610 .
611 .
612 .
613
614 Once the workqueues and devices have been enabled, the IAA crypto
615 algorithms are enabled and available. When the IAA crypto algorithms
616 have been successfully enabled, you should see the following dmesg
617 output::
618
619 [ 64.893759] iaa_crypto: iaa_crypto_enable: iaa_crypto now ENABLED
620
621 Now run the following zswap-specific setup commands to have zswap use
622 the 'fixed' compression mode::
623
624 echo 0 > /sys/module/zswap/parameters/enabled
625 echo 50 > /sys/module/zswap/parameters/max_pool_percent
626 echo deflate-iaa > /sys/module/zswap/parameters/compressor
627 echo 1 > /sys/module/zswap/parameters/enabled
628
629 echo 100 > /proc/sys/vm/swappiness
630 echo never > /sys/kernel/mm/transparent_hugepage/enabled
631 echo 1 > /proc/sys/vm/overcommit_memory
632
633 Finally, you can now run the zswap workload you want to measure. For
634 example, using the code below, the following command will swap in and
635 out 100 pages::
636
637 ./memory_madvise 100
638
639 Allocating 100 pages to swap in/out
640 Swapping out 100 pages
641 Swapping in 100 pages
642 Swapped out and in 100 pages
643
644 You should see something like the following in the dmesg output if
645 you've enabled debug output (echo -n 'module iaa_crypto +p' >
646 /sys/kernel/debug/dynamic_debug/control)::
647
648 [ 404.202972] idxd 0000:e7:02.0: iaa_comp_acompress: dma_map_sg, src_addr 223925c000, nr_sgs 1, req->src 00000000ee7cb5e6, req->slen 4096, sg_dma_len(sg) 4096
649 [ 404.202973] idxd 0000:e7:02.0: iaa_comp_acompress: dma_map_sg, dst_addr 21dadf8000, nr_sgs 1, req->dst 000000008d6acea8, req->dlen 4096, sg_dma_len(sg) 8192
650 [ 404.202975] idxd 0000:e7:02.0: iaa_compress: desc->src1_addr 223925c000, desc->src1_size 4096, desc->dst_addr 21dadf8000, desc->max_dst_size 4096, desc->src2_addr 2203543000, desc->src2_size 1568
651 [ 404.202981] idxd 0000:e7:02.0: iaa_compress_verify: (verify) desc->src1_addr 21dadf8000, desc->src1_size 228, desc->dst_addr 223925c000, desc->max_dst_size 4096, desc->src2_addr 0, desc->src2_size 0
652 [ 409.203227] idxd 0000:e7:02.0: iaa_comp_adecompress: dma_map_sg, src_addr 21ddd8b100, nr_sgs 1, req->src 0000000084adab64, req->slen 228, sg_dma_len(sg) 228
653 [ 409.203235] idxd 0000:e7:02.0: iaa_comp_adecompress: dma_map_sg, dst_addr 21ee3dc000, nr_sgs 1, req->dst 000000004e2990d0, req->dlen 4096, sg_dma_len(sg) 4096
654 [ 409.203239] idxd 0000:e7:02.0: iaa_decompress: desc->src1_addr 21ddd8b100, desc->src1_size 228, desc->dst_addr 21ee3dc000, desc->max_dst_size 4096, desc->src2_addr 0, desc->src2_size 0
655 [ 409.203254] idxd 0000:e7:02.0: iaa_comp_adecompress: dma_map_sg, src_addr 21ddd8b100, nr_sgs 1, req->src 0000000084adab64, req->slen 228, sg_dma_len(sg) 228
656 [ 409.203256] idxd 0000:e7:02.0: iaa_comp_adecompress: dma_map_sg, dst_addr 21f1551000, nr_sgs 1, req->dst 000000004e2990d0, req->dlen 4096, sg_dma_len(sg) 4096
657 [ 409.203257] idxd 0000:e7:02.0: iaa_decompress: desc->src1_addr 21ddd8b100, desc->src1_size 228, desc->dst_addr 21f1551000, desc->max_dst_size 4096, desc->src2_addr 0, desc->src2_size 0
658
659 In order to unregister the IAA crypto algorithms, and register new
660 ones using different parameters, any users of the current algorithm
661 should be stopped and the IAA workqueues and devices disabled.
662
663 In the case of zswap, remove the IAA crypto algorithm as the
664 compressor and turn off swap (to remove all references to
665 iaa_crypto)::
666
667 echo lzo > /sys/module/zswap/parameters/compressor
668 swapoff -a
669
670 echo 0 > /sys/module/zswap/parameters/accept_threshold_percent
671 echo 0 > /sys/module/zswap/parameters/max_pool_percent
672 echo 0 > /sys/module/zswap/parameters/enabled
673
674 Once zswap is disabled and no longer using iaa_crypto, the IAA wqs and
675 devices can be disabled.
676
677 .. _iaa_disable_script:
678
679 IAA disable script
680 ------------------
681
682 The below script automatically does that::
683
684 #!/bin/bash
685
686 echo "IAA devices:"
687 lspci -d:0cfe
688 echo "# IAA devices:"
689 lspci -d:0cfe | wc -l
690
691 #
692 # count iaa instances
693 #
694 iaa_dev_id="0cfe"
695 num_iaa=$(lspci -d:${iaa_dev_id} | wc -l)
696 echo "Found ${num_iaa} IAA instances"
697
698 #
699 # disable iaa wqs and devices
700 #
701 echo "Disable IAA"
702
703 for ((i = 1; i < ${num_iaa} * 2; i += 2)); do
704 echo disable wq iax${i}/wq${i}.0
705 accel-config disable-wq iax${i}/wq${i}.0
706 echo disable iaa iax${i}
707 accel-config disable-device iax${i}
708 done
709
710 echo "End Disable IAA"
711
712 Finally, at this point the iaa_crypto module can be removed, which
713 will unregister the current IAA crypto algorithms::
714
715 rmmod iaa_crypto
716
717
718 memory_madvise.c (gcc -o memory_memadvise memory_madvise.c)::
719
720 #include <stdio.h>
721 #include <stdlib.h>
722 #include <string.h>
723 #include <unistd.h>
724 #include <sys/mman.h>
725 #include <linux/mman.h>
726
727 #ifndef MADV_PAGEOUT
728 #define MADV_PAGEOUT 21 /* force pages out immediately */
729 #endif
730
731 #define PG_SZ 4096
732
733 int main(int argc, char **argv)
734 {
735 int i, nr_pages = 1;
736 int64_t *dump_ptr;
737 char *addr, *a;
738 int loop = 1;
739
740 if (argc > 1)
741 nr_pages = atoi(argv[1]);
742
743 printf("Allocating %d pages to swap in/out\n", nr_pages);
744
745 /* allocate pages */
746 addr = mmap(NULL, nr_pages * PG_SZ, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0);
747 *addr = 1;
748
749 /* initialize data in page to all '*' chars */
750 memset(addr, '*', nr_pages * PG_SZ);
751
752 printf("Swapping out %d pages\n", nr_pages);
753
754 /* Tell kernel to swap it out */
755 madvise(addr, nr_pages * PG_SZ, MADV_PAGEOUT);
756
757 while (loop > 0) {
758 /* Wait for swap out to finish */
759 sleep(5);
760
761 a = addr;
762
763 printf("Swapping in %d pages\n", nr_pages);
764
765 /* Access the page ... this will swap it back in again */
766 for (i = 0; i < nr_pages; i++) {
767 if (a[0] != '*') {
768 printf("Bad data from decompress!!!!!\n");
769
770 dump_ptr = (int64_t *)a;
771 for (int j = 0; j < 100; j++) {
772 printf(" page %d data: %#llx\n", i, *dump_ptr);
773 dump_ptr++;
774 }
775 }
776
777 a += PG_SZ;
778 }
779
780 loop --;
781 }
782
783 printf("Swapped out and in %d pages\n", nr_pages);
784
785 Appendix
786 ========
787
788 .. _iaa_sysfs_config:
789
790 IAA sysfs config interface
791 --------------------------
792
793 Below is a description of the IAA sysfs interface, which as mentioned
794 in the main document, should only be used if you know exactly what you
795 are doing. Even then, there's no compelling reason to use it directly
796 since accel-config can do everything the sysfs interface can and in
797 fact accel-config is based on it under the covers.
798
799 The 'IAA config path' is /sys/bus/dsa/devices and contains
800 subdirectories representing each IAA device, workqueue, engine, and
801 group. Note that in the sysfs interface, the IAA devices are actually
802 named using iax e.g. iax1, iax3, etc. (Note that IAA devices are the
803 odd-numbered devices; the even-numbered devices are DSA devices and
804 can be ignored for IAA).
805
806 The 'IAA device bind path' is /sys/bus/dsa/drivers/idxd/bind and is
807 the file that is written to enable an IAA device.
808
809 The 'IAA workqueue bind path' is /sys/bus/dsa/drivers/crypto/bind and
810 is the file that is written to enable an IAA workqueue.
811
812 Similarly /sys/bus/dsa/drivers/idxd/unbind and
813 /sys/bus/dsa/drivers/crypto/unbind are used to disable IAA devices and
814 workqueues.
815
816 The basic sequence of commands needed to set up the IAA devices and
817 workqueues is:
818
819 For each device::
820 1) Disable any workqueues enabled on the device. For example to
821 disable workques 0 and 1 on IAA device 3::
822
823 # echo wq3.0 > /sys/bus/dsa/drivers/crypto/unbind
824 # echo wq3.1 > /sys/bus/dsa/drivers/crypto/unbind
825
826 2) Disable the device. For example to disable IAA device 3::
827
828 # echo iax3 > /sys/bus/dsa/drivers/idxd/unbind
829
830 3) configure the desired workqueues. For example, to configure
831 workqueue 3 on IAA device 3::
832
833 # echo dedicated > /sys/bus/dsa/devices/iax3/wq3.3/mode
834 # echo 128 > /sys/bus/dsa/devices/iax3/wq3.3/size
835 # echo 0 > /sys/bus/dsa/devices/iax3/wq3.3/group_id
836 # echo 10 > /sys/bus/dsa/devices/iax3/wq3.3/priority
837 # echo "kernel" > /sys/bus/dsa/devices/iax3/wq3.3/type
838 # echo "iaa_crypto" > /sys/bus/dsa/devices/iax3/wq3.3/name
839 # echo "crypto" > /sys/bus/dsa/devices/iax3/wq3.3/driver_name
840
841 4) Enable the device. For example to enable IAA device 3::
842
843 # echo iax3 > /sys/bus/dsa/drivers/idxd/bind
844
845 5) Enable the desired workqueues on the device. For example to
846 enable workques 0 and 1 on IAA device 3::
847
848 # echo wq3.0 > /sys/bus/dsa/drivers/crypto/bind
849 # echo wq3.1 > /sys/bus/dsa/drivers/crypto/bind