crypto: xor - defer and optimize boot time benchmark
v2 cover crypto: xor - defer and optimize boot time benchmark
Doug reports [0] that the XOR boot time benchmark takes more time than necessary, and runs at a time when there is little room for other boot time tasks to run concurrently.
Let's fix this by #1 deferring the benchmark, and #2 uses a faster implementation.
Changes since v2:
- incorporate Doug's review feedback re coarse clocks and the use of pr_info
- add Doug's ack to #1
Ard Biesheuvel (2):
crypto: xor - defer load time benchmark to a later time
crypto: xor - use ktime for template benchmarking
crypto/xor.c | 67 +++++++++++++------- 1 file changed, 44 insertions(+), 23 deletions(-)
Signature · 1 line
2.17.1
PATCH v2 1/2 crypto: xor - defer load time benchmark to a later time
Currently, the XOR module performs its boot time benchmark at core initcall time when it is built-in, to ensure that the RAID code can make use of it when it is built-in as well.
Let's defer this to a later stage during the boot, to avoid impacting the overall boot time of the system. Instead, just pick an arbitrary implementation from the list, and use that as the preliminary default.
--- crypto/xor.c | 29 +++++++++++++++++++- 1 file changed, 28 insertions(+), 1 deletion(-)
crypto/xor.c +28 -1
@@ -54,6 +54,28 @@ EXPORT_SYMBOL(xor_blocks);/* Set of all registered templates. */ static struct xor_block_template *__initdata template_list; +#ifndef MODULE +static void __init do_xor_register (struct xor_block_template *tmpl) +{ + tmpl->next = template_list; + template_list = tmpl; +} + +static int __init register_xor_blocks (void ) +{ + active_template =XOR_SELECT_TEMPLATE (NULL ); + +if (!active_template) { +#define xor_speed do_xor_register +// register all the templates and pick the first as the default +XOR_TRY_TEMPLATES ; +#undef xor_speed + active_template = template_list; + } +return 0 ; +} +#endif +#define BENCH_SIZE (PAGE_SIZE) static void __init
@@ -129,6 +151,7 @@ calibrate_xor_blocks(void)#define xor_speed (templ) do_xor_speed((templ), b1, b2)printk (KERN_INFO "xor: measuring software checksum speed\n" ); + template_list =NULL ;XOR_TRY_TEMPLATES ; fastest = template_list;for (f = fastest; f; f = f->next)
@@ -150,6 +173,10 @@ static __exit void xor_exit(void) { }MODULE_LICENSE ("GPL" ); +#ifndef MODULE /* when built-in xor.o must initialize before drivers/md/md.o */ -core_initcall (calibrate_xor_blocks); +core_initcall (register_xor_blocks); +#endif + +module_init (calibrate_xor_blocks);module_exit (xor_exit);
Signature · 1 line
2.17.1
PATCH v2 2/2 crypto: xor - use ktime for template benchmarking
In reply to Ard BiesheuvelCurrently, we use the jiffies counter as a time source, by staring at it until a HZ period elapses, and then staring at it again and perform as many XOR operations as we can at the same time until another HZ period elapses, so that we can calculate the throughput. This takes longer than necessary, and depends on HZ, which is undesirable, since HZ is system dependent.
Let's use the ktime interface instead, and use it to time a fixed number of XOR operations, which can be done much faster, and makes the time spent depend on the performance level of the system itself, which is much more reasonable. To ensure that we have the resolution we need even on systems with 32 kHz time sources, while not spending too much time in the benchmark on a slow CPU, let's switch to 3 attempts of 800 repetitions each: that way, we will only misidentify algorithms that perform within 10% of each other as the fastest if they are faster than 10 GB/s to begin with, which is not expected to occur on systems with such coarse clocks.
On ThunderX2, I get the following results:
Before:
[72625.956765] xor: measuring software checksum speed [72625.993104] 8regs : 10169.000 MB/sec [72626.033099] 32regs : 12050.000 MB/sec [72626.073095] arm64_neon: 11100.000 MB/sec [72626.073097] xor: using function: 32regs (12050.000 MB/sec)
After:
[72599.650216] xor: measuring software checksum speed [72599.651188] 8regs : 10491 MB/sec [72599.652006] 32regs : 12345 MB/sec [72599.652871] arm64_neon : 11402 MB/sec [72599.652873] xor: using function: 32regs (12345 MB/sec)
--- crypto/xor.c | 38 +++++++++----------- 1 file changed, 16 insertions(+), 22 deletions(-)
crypto/xor.c +16 -22
@@ -76,49 +76,43 @@ static int __init register_xor_blocks(void) }#endif -#define BENCH_SIZE (PAGE_SIZE) +#define BENCH_SIZE 4096 +#define REPS 800Ustatic void __init do_xor_speed (struct xor_block_template *tmpl,void *b1,void *b2) {int speed; -unsigned long now, j; -int i, count, max; +int i, j, count; +ktime_t min, start, diff; tmpl->next = template_list; template_list = tmpl;preempt_disable (); -/* -* Count the number of XORs done during a whole jiffy, and use -* this to calculate the speed of checksumming. We use a 2-page -* allocation to have guaranteed color L1-cache layout. -*/ - max =0 ; -for (i =0 ; i <5 ; i++) { - j = jiffies; - count =0 ; -while ((now = jiffies) == j) -cpu_relax (); -while (time_before (jiffies, now +1 )) { + min = (ktime_t )S64_MAX ; +for (i =0 ; i <3 ; i++) { + start =ktime_get (); +for (j =0 ; j <REPS ; j++) {mb ();/* prevent loop optimzation */ tmpl->do_2 (BENCH_SIZE , b1, b2);mb (); count++;mb (); } -if (count > max) - max = count; + diff =ktime_sub (ktime_get (), start); +if (diff < min) + min = diff; }preempt_enable (); - speed = max * (HZ *BENCH_SIZE /1024 ); +// bytes/ns == GB/s, multiply by 1000 to get MB/s [not MiB/s] + speed = (1000 *REPS *BENCH_SIZE ) / (unsigned int )ktime_to_ns (min); tmpl->speed = speed; -printk (KERN_INFO " %-10s: %5d.%03d MB/sec\n" , tmpl->name, - speed /1000 , speed %1000 ); +pr_info (" %-16s: %5d MB/sec\n" , tmpl->name, speed); } static int __init
@@ -158,8 +152,8 @@ calibrate_xor_blocks(void)if (f->speed > fastest->speed) fastest = f; -printk (KERN_INFO "xor: using function: %s (%d.%03d MB/sec)\n" , - fastest->name, fastest->speed /1000 , fastest->speed %1000 ); +pr_info ("xor: using function: %s (%d MB/sec)\n" , + fastest->name, fastest->speed);#undef xor_speed
Signature · 1 line
2.17.1
Re: [PATCH v2 2/2] crypto: xor - use ktime for template benchmarking
Hi,
Quoted 29 lines · Ard Biesheuvel
Currently, we use the jiffies counter as a time source, by staring at it until a HZ period elapses, and then staring at it again and perform as many XOR operations as we can at the same time until another HZ period elapses, so that we can calculate the throughput. This takes longer than necessary, and depends on HZ, which is undesirable, since HZ is system dependent.
Let's use the ktime interface instead, and use it to time a fixed number of XOR operations, which can be done much faster, and makes the time spent depend on the performance level of the system itself, which is much more reasonable. To ensure that we have the resolution we need even on systems with 32 kHz time sources, while not spending too much time in the benchmark on a slow CPU, let's switch to 3 attempts of 800 repetitions each: that way, we will only misidentify algorithms that perform within 10% of each other as the fastest if they are faster than 10 GB/s to begin with, which is not expected to occur on systems with such coarse clocks.
On ThunderX2, I get the following results:
Before:
[72625.956765] xor: measuring software checksum speed [72625.993104] 8regs : 10169.000 MB/sec [72626.033099] 32regs : 12050.000 MB/sec [72626.073095] arm64_neon: 11100.000 MB/sec [72626.073097] xor: using function: 32regs (12050.000 MB/sec)
Quoted 7 lines
After:
[72599.650216] xor: measuring software checksum speed [72599.651188] 8regs : 10491 MB/sec [72599.652006] 32regs : 12345 MB/sec [72599.652871] arm64_neon : 11402 MB/sec [72599.652873] xor: using function: 32regs (12345 MB/sec)
What are the chances of 12345 coming up? ;-)
Quoted 6 lines
Signed-off-by Ard Biesheuvel <[email protected]>--- crypto/xor.c | 38 +++++++++----------- 1 file changed, 16 insertions(+), 22 deletions(-)
This looks good to me. Thanks for taking this on!
Herbert Xu
maintainerQuoted 16 lines · Ard Biesheuvel
Doug reports [0] that the XOR boot time benchmark takes more time than necessary, and runs at a time when there is little room for other boot time tasks to run concurrently.
Let's fix this by #1 deferring the benchmark, and #2 uses a faster implementation.
Changes since v2:
- incorporate Doug's review feedback re coarse clocks and the use ofpr_info
- add Doug's ack to #1Cc Douglas Anderson <[email protected]>Cc David Laight <[email protected]>
Quoted 6 lines
Ard Biesheuvel (2):
crypto: xor - defer load time benchmark to a later time
crypto: xor - use ktime for template benchmarkingcrypto/xor.c | 67 +++++++++++++------- 1 file changed, 44 insertions(+), 23 deletions(-)
All applied. Thanks.
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt