[PATCH v4 0/7] crypto: SHA1 multibuffer implementation

[Tim Chen]

Herbert,

I've updated my implementation from v3 to flush the jobs early when cpu goes to idle.
The flush routine was moved out of the notifier path to the crypto
thread.  To check that there's no other jobs running, I've added the
nr_running_cpu function to obtain the information.  I've also cleaned
up the shash_ahash_mcryptd_digest to mcryptd.c as you have suggested.
Please note also that a separate bug fix to the crypto scatter gather list walk
for the null string needs to be incorporated, which I encountered during
my testing: http://marc.info/?l=linux-crypto-vger&m=140503429412699&w=2

In this patch series, we introduce the multi-buffer crypto algorithm on
x86_64 and apply it to SHA1 hash computation.  The multi-buffer technique
takes advantage of the 8 data lanes in the AVX2 registers and allows
computation to be performed on data from multiple jobs in parallel.
This allows us to parallelize computations when data inter-dependency in
a single crypto job prevents us to fully parallelize our computations.
The algorithm can be extended to other hashing and encryption schemes
in the future.

On multi-buffer SHA1 computation with AVX2, we see throughput increase
up to 2.2x over the existing x86_64 single buffer AVX2 algorithm.

The multi-buffer crypto algorithm is described in the following paper:
Processing Multiple Buffers in Parallel to Increase Performance on
Intel Architecture Processors
http://www.intel.com/content/www/us/en/communications/communications-ia-multi-buffer-paper.html

The outline of the algorithm is sketched below:
Any driver requesting the crypto service will place an async crypto
request on the workqueue.  The multi-buffer crypto daemon will pull
request from work queue and put each request in an empty data lane for
multi-buffer crypto computation.  When all the empty lanes are filled,
computation will commence on the jobs in parallel and the job with the
shortest remaining buffer will get completed and be returned.  To prevent
prolonged stall when there is no new jobs arriving, we will flush a
crypto job if it has not been completed after a maximum allowable delay,
or when cpu becomes idle and cpu cycles become available.

The multi-buffer algorithm necessitates mapping multiple scatter gather
buffers to linear addresses simultaneously. The crypto daemon may need
to sleep and yield the cpu to work on something else from time to time.
We made a change to not use kmap_atomic to do scatter-gather buffer
mapping and take advantage of the fact that we can directly translate
address the buffer's address to its linear address with x86_64.
To accommodate the fragmented nature of scatter-gather, we will keep
submitting the next scatter-buffer fragment for a job for multi-buffer
computation until a job is completed and no more buffer fragments remain.
At that time we will pull a new job to fill the now empty data slot.
We call a get_completed_job function to check whether there are other
jobs that have been completed when we job when we have no new job arrival
to prevent extraneous delay in returning any completed jobs.

The multi-buffer algorithm should be used for cases where crypto jobs
submissions are at a reasonable high rate.  For low crypto job submission
rate, this algorithm will not be beneficial. The reason is at low rate,
we do not fill out the data lanes before flushing the jobs instead of
processing them with all the data lanes full.  We will miss the benefit
of parallel computation, and adding delay to the processing of the crypto
job at the same time.  Some tuning of the maximum latency parameter may
be needed to get the best performance.

Note that the tcrypt SHA1 speed test, we wait for a previous job to
be completed before submitting a new job.  Hence this is not a valid
test for multi-buffer algorithm as it requires multiple outstanding jobs
submitted to fill the all data lanes to be effective (i.e. 8 outstanding
jobs for the AVX2 case).

Feedbacks and testings will be most welcomed.

Tim Chen

Change log:
v4
1. Move the early flush of jobs when cpu becomes idle to crypto thread.
2. Move shash_ahash_mcryptd_digest to mcryptd.c 

v3
1. Add notifier to multi-buffer algorithm to flush job when the cpu
goes to idle to take advantage of available cpu cycles. 
2. Clean up of error messages.
http://marc.info/?l=linux-crypto-vger&m=140252063401632&w=2

v2
1. Change the sha1 crypto walk to use the new crypto_ahash_walk
interface for proper kmap.
2. Drop the hack that map buffer in crypto_hash_walk without kmap_atomic
as the new crypto_ahash_walk interface is merged.
3. Reorganize some
of the mcryptd hash interface code from ahash.c to mcryptd.c
http://marc.info/?l=linux-crypto-vger&m=140088627927559&w=2

v1
refer to: http://www.spinics.net/lists/linux-crypto/msg10993.html


Tim Chen (7):
  crypto: SHA1 multibuffer crypto hash infrastructure
  crypto: SHA1 multibuffer algorithm data structures
  crypto: SHA1 multibuffer submit and flush routines for AVX2
  crypto: SHA1 multibuffer crypto computation (x8 AVX2)
  crypto: SHA1 multibuffer scheduler
  sched: add function nr_running_cpu to expose number of tasks running
    on     cpu
  crypto: SHA1 multibuffer - flush the jobs early if cpu becomes idle

 arch/x86/crypto/Makefile                         |    2 +
 arch/x86/crypto/sha-mb/Makefile                  |   11 +
 arch/x86/crypto/sha-mb/sha1_mb.c                 | 1004 ++++++++++++++++++++++
 arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S  |  287 +++++++
 arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S  |  327 +++++++
 arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c   |   64 ++
 arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S |  228 +++++
 arch/x86/crypto/sha-mb/sha1_x8_avx2.S            |  472 ++++++++++
 arch/x86/crypto/sha-mb/sha_mb_ctx.h              |  136 +++
 arch/x86/crypto/sha-mb/sha_mb_mgr.h              |  110 +++
 crypto/Kconfig                                   |   30 +
 crypto/Makefile                                  |    1 +
 crypto/mcryptd.c                                 |  622 ++++++++++++++
 include/crypto/internal/hash.h                   |    9 +
 include/crypto/mcryptd.h                         |  110 +++
 include/linux/sched.h                            |    1 +
 kernel/sched/core.c                              |    8 +
 17 files changed, 3422 insertions(+)
 create mode 100644 arch/x86/crypto/sha-mb/Makefile
 create mode 100644 arch/x86/crypto/sha-mb/sha1_mb.c
 create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_datastruct.S
 create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_flush_avx2.S
 create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_init_avx2.c
 create mode 100644 arch/x86/crypto/sha-mb/sha1_mb_mgr_submit_avx2.S
 create mode 100644 arch/x86/crypto/sha-mb/sha1_x8_avx2.S
 create mode 100644 arch/x86/crypto/sha-mb/sha_mb_ctx.h
 create mode 100644 arch/x86/crypto/sha-mb/sha_mb_mgr.h
 create mode 100644 crypto/mcryptd.c
 create mode 100644 include/crypto/mcryptd.h

-- 
1.7.11.7


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/