ehmry/genode
ehmry
/
genode
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

ahci & exynos5: avoid large DMA buf in AHCI bench 

Normally this bench has read all data to one large buffer and
than written it back to the drive but for SATA 3 (6 Gbps) benchmarks
we would need a buffer of approximately 1.2 GB to do it this way
and reach 2 seconds bench time. Thus we use a buffer of SATA request size
and override it with every request.
This commit is contained in:
Martin Stein 2013-06-25 18:18:41 +02:00 committed by Stefan Kalkowski
parent 0ad284febe
commit 3fb7e74287
1 changed files with 12 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
os/src/drivers/ahci/exynos5/bench/main.cc
View File

@ -45,7 +45,11 @@ static void run_benchmark(Block::Driver &driver,
Operation &operation)
{
using namespace Genode;
if (request_size > buffer_size) {
PERR("undersized buffer %u, need %u", buffer_size, buffer_size);
while (1) ;
}
size_t const block_count = request_size / driver.block_size();
/*
* The goal is to get a test that took 2 s < time < 2.3 s,
* thus we start with count = 32 and then adjust the count
@ -55,26 +59,18 @@ static void run_benchmark(Block::Driver &driver,
unsigned ms = 0;
while (1)
{
/* calculate test parameters */
if (bytes > buffer_size) {
PERR("undersized buffer %u, need %u", buffer_size, bytes);
while (1) ;
}
size_t num_requests = bytes / request_size;
/* do measurement */
size_t const time_before_ms = timer.elapsed_ms();
for (size_t i = 0; i < num_requests; i++)
unsigned const time_before_ms = timer.elapsed_ms();
for (unsigned i = 0; i < num_requests; i++)
{
size_t const block_count = request_size / driver.block_size();
addr_t const block_number = i*block_count;
addr_t const block_number = i * block_count;
operation(driver, block_number, block_count,
buffer_phys + i*request_size,
buffer_virt + i*request_size);
buffer_phys, buffer_virt);
}
/* read results */
size_t const time_after_ms = timer.elapsed_ms();
unsigned const time_after_ms = timer.elapsed_ms();
ms = time_after_ms - time_before_ms;
/*
@ -121,7 +117,7 @@ int main(int argc, char **argv)
1048576, 262144, 16384, 8192, 4096, 2048, 1024, 512, 0 };
/* total size of communication buffer */
size_t const buffer_size = 600*1024*1024;
size_t const buffer_size = 1024*1024;
/* allocate read/write buffer */
static Attached_ram_dataspace buffer(env()->ram_session(), buffer_size,
@ -158,9 +154,7 @@ int main(int argc, char **argv)
/*
* Benchmark writing to SATA device
*
* We write back the content of the buffer, which we just filled during the
* read benchmark. If both read and write succeed, the SATA device
* will retain its original content.
* Attention: Original data will be overridden on target drive
*/
printf("\n");