def test_trim_basic(nvme0: d.Controller, nvme0n1: d.Namespace, verify):
GB = 1024 * 1024 * 1024
all_zero_databuf = d.Buffer(512)
trimbuf = d.Buffer(4096)
q = d.Qpair(nvme0, 32)
# DUT info
logging.info("model number: %s" % nvme0.id_data(63, 24, str))
logging.info("firmware revision: %s" % nvme0.id_data(71, 64, str))
# write
logging.info("write data in 10G ~ 20G")
io_size = 128 * 1024 // 512
start_lba = 10 * GB // 512
lba_count = 10 * GB // 512
nvme0n1.ioworker(io_size=io_size,
lba_align=io_size,
lba_random=False,
read_percentage=0,
lba_start=start_lba,
io_count=lba_count // io_size,
qdepth=128).start().close()
# verify data after write, data should be modified
with pytest.warns(UserWarning, match="ERROR status: 02/85"):
nvme0n1.compare(q, all_zero_databuf, start_lba).waitdone()
# get the empty trim time
trimbuf.set_dsm_range(0, 0, 0)
trim_cmd = nvme0n1.dsm(q, trimbuf,
1).waitdone() # first call is longer, due to cache?
start_time = time.time()
trim_cmd = nvme0n1.dsm(q, trimbuf, 1).waitdone()
empty_trim_time = time.time() - start_time
# the trim time on device-side only
logging.info("trim the 10G data from LBA 0x%lx" % start_lba)
trimbuf.set_dsm_range(0, start_lba, lba_count)
start_time = time.time()
trim_cmd = nvme0n1.dsm(q, trimbuf, 1).waitdone()
trim_time = time.time() - start_time - empty_trim_time
logging.info("trim bandwidth: %0.2fGB/s" % (10 / trim_time))
# verify after trim
nvme0n1.compare(q, all_zero_databuf, start_lba).waitdone()
def test_dsm_trim(nvme0: d.Controller, nvme0n1: d.Namespace, qpair: d.Qpair):
trimbuf = d.Buffer(4096)
# DUT info
logging.info("model number: %s" % nvme0.id_data(63, 24, str))
logging.info("firmware revision: %s" % nvme0.id_data(71, 64, str))
# single range
start_lba = 0
lba_count = 8 * 1024
trimbuf.set_dsm_range(0, start_lba, lba_count)
nvme0n1.dsm(qpair, trimbuf, 1, attribute=0x4).waitdone()
# multiple range
lba_count = lba_count // 256
for i in range(256):
trimbuf.set_dsm_range(i, start_lba + i * lba_count, lba_count)
nvme0n1.dsm(qpair, trimbuf, 256).waitdone()
