blob: 01ff28b58828ab297260d960bdae109323f54afa [file] [log] [blame]
#!/usr/bin/env python3
#
# Copyright © 2020 Google LLC
# SPDX-License-Identifier: MIT
import argparse
import re
import sys
from serial_buffer import SerialBuffer
class CrosServoRun:
def __init__(self, cpu, ec, test_timeout):
self.cpu_ser = SerialBuffer(
cpu, "results/serial.txt", "R SERIAL-CPU> ")
# Merge the EC serial into the cpu_ser's line stream so that we can
# effectively poll on both at the same time and not have to worry about
self.ec_ser = SerialBuffer(
ec, "results/serial-ec.txt", "R SERIAL-EC> ", line_queue=self.cpu_ser.line_queue)
self.test_timeout = test_timeout
def close(self):
self.ec_ser.close()
self.cpu_ser.close()
def ec_write(self, s):
print("W SERIAL-EC> %s" % s)
self.ec_ser.serial.write(s.encode())
def cpu_write(self, s):
print("W SERIAL-CPU> %s" % s)
self.cpu_ser.serial.write(s.encode())
def print_error(self, message):
RED = '\033[0;31m'
NO_COLOR = '\033[0m'
print(RED + message + NO_COLOR)
def run(self):
# Flush any partial commands in the EC's prompt, then ask for a reboot.
self.ec_write("\n")
self.ec_write("reboot\n")
bootloader_done = False
tftp_failures = 0
# This is emitted right when the bootloader pauses to check for input.
# Emit a ^N character to request network boot, because we don't have a
# direct-to-netboot firmware on cheza.
for line in self.cpu_ser.lines(timeout=120, phase="bootloader"):
if re.search("load_archive: loading locale_en.bin", line):
self.cpu_write("\016")
bootloader_done = True
break
# The Cheza firmware seems to occasionally get stuck looping in
# this error state during TFTP booting, possibly based on amount of
# network traffic around it, but it'll usually recover after a
# reboot. Currently mostly visible on google-freedreno-cheza-14.
if re.search("R8152: Bulk read error 0xffffffbf", line):
tftp_failures += 1
if tftp_failures >= 10:
self.print_error(
"Detected intermittent tftp failure, restarting run.")
return 1
# If the board has a netboot firmware and we made it to booting the
# kernel, proceed to processing of the test run.
if re.search("Booting Linux", line):
bootloader_done = True
break
# The Cheza boards have issues with failing to bring up power to
# the system sometimes, possibly dependent on ambient temperature
# in the farm.
if re.search("POWER_GOOD not seen in time", line):
self.print_error(
"Detected intermittent poweron failure, abandoning run.")
return 1
if not bootloader_done:
print("Failed to make it through bootloader, abandoning run.")
return 1
for line in self.cpu_ser.lines(timeout=self.test_timeout, phase="test"):
if re.search("---. end Kernel panic", line):
return 1
# There are very infrequent bus errors during power management transitions
# on cheza, which we don't expect to be the case on future boards.
if re.search("Kernel panic - not syncing: Asynchronous SError Interrupt", line):
self.print_error(
"Detected cheza power management bus error, abandoning run.")
return 1
# If the network device dies, it's probably not graphics's fault, just try again.
if re.search("NETDEV WATCHDOG", line):
self.print_error(
"Detected network device failure, abandoning run.")
return 1
# These HFI response errors started appearing with the introduction
# of piglit runs. CosmicPenguin says:
#
# "message ID 106 isn't a thing, so likely what happened is that we
# got confused when parsing the HFI queue. If it happened on only
# one run, then memory corruption could be a possible clue"
#
# Given that it seems to trigger randomly near a GPU fault and then
# break many tests after that, just restart the whole run.
if re.search("a6xx_hfi_send_msg.*Unexpected message id .* on the response queue", line):
self.print_error(
"Detected cheza power management bus error, abandoning run.")
return 1
if re.search("coreboot.*bootblock starting", line):
self.print_error(
"Detected spontaneous reboot, abandoning run.")
return 1
if re.search("arm-smmu 5040000.iommu: TLB sync timed out -- SMMU may be deadlocked", line):
self.print_error("Detected cheza MMU fail, abandoning run.")
return 1
result = re.search("hwci: mesa: (\S*)", line)
if result:
if result.group(1) == "pass":
return 0
else:
return 1
self.print_error(
"Reached the end of the CPU serial log without finding a result")
return 1
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--cpu', type=str,
help='CPU Serial device', required=True)
parser.add_argument(
'--ec', type=str, help='EC Serial device', required=True)
parser.add_argument(
'--test-timeout', type=int, help='Test phase timeout (minutes)', required=True)
args = parser.parse_args()
servo = CrosServoRun(args.cpu, args.ec, args.test_timeout * 60)
retval = servo.run()
# power down the CPU on the device
servo.ec_write("power off\n")
servo.close()
sys.exit(retval)
if __name__ == '__main__':
main()