mirror of
https://github.com/FEX-Emu/FEX.git
synced 2025-02-24 16:51:32 +00:00
e8aaadb2d0
Somewhere between kernel 5.15 and 6.0 this syscall's behaviour has changed. The unit test expects -ESPIPE result but new kernels will return 0 (noop_llseek in the source). The new solidrun board is our first CI machine to be running kernel 6.1 so this needs to be merged before that CI machine can be enabled.
228 lines
7.0 KiB
Plaintext
228 lines
7.0 KiB
Plaintext
# these fail on x86 CI
|
|
proc_pid_oomscore_test
|
|
|
|
# The behaviour of this is different between x86 and ARM
|
|
# SIGSEGV kills a thread successfully on ARM but not on x86
|
|
time_test
|
|
|
|
# these are disabled
|
|
# never terminates (even though we use timeout)
|
|
kill_test
|
|
|
|
# erratic test results
|
|
xattr_test
|
|
mlock_test
|
|
|
|
# These used to crash and fail its test, now hang
|
|
#mremap_test
|
|
|
|
# these are flaky on x86
|
|
stat_times_test
|
|
|
|
# Spins forever
|
|
ptrace_test
|
|
|
|
# The behaviour of this test changes depending on if running inside the test harness or not
|
|
# Works outside of the test harness but fails inside of it
|
|
pause_test
|
|
|
|
# This takes advantage of CLONE_VM in a way that breaks FEX's TLS usage
|
|
# Need to workaround this somehow
|
|
aio_test
|
|
|
|
# The Solidrun CI board isn't running a kernel with namespaces enabled
|
|
# gvisor doesn't error check correctly in this case and returns hard failure instead of correctly exiting
|
|
proc_pid_uid_gid_map_test
|
|
|
|
# This test is expecting the kernel to update the timeout variable but by default the kernel keeps time outs "sticky"
|
|
# See personality flag STICKY_TIMEOUTS
|
|
select_test
|
|
|
|
# This test uses syscalls that a userspace process doesn't have access to
|
|
syslog_test
|
|
|
|
# This very specifically is testing bad pointers passed in to the kernel to test EFAULT
|
|
access_test
|
|
sigprocmask_test
|
|
|
|
# Relies on some utility executables to be installed
|
|
exec_test
|
|
|
|
# This relies on splice behaviour that doesn't work even on x86 host
|
|
eventfd_test
|
|
|
|
# Relies on tricky behaviour
|
|
# MMapDeathTest.TruncateAfterCOWBreak : Needs to capture SIGBUS. Test fails, handwritten test works
|
|
# MMapTest.ProtNoneDeath : Needs to capture SIGBUS. Test fails, handwritten test works
|
|
# MMapTest.NoProtExecDeath : Needs to support SIGSEGV from memory mapped without PROT_EXEC but WITH READ + WRITE
|
|
# NoExceedLimitData : Requires brk to respect RLIMIT_DATA
|
|
mmap_test
|
|
|
|
# Fails even on host x86 device
|
|
# Expects the tmp folder to be empty and cleaned up between CI runs by a system process
|
|
# Does tests that removes user priveleges to touch those files
|
|
open_create_test
|
|
|
|
# Fails even on host x86 device
|
|
# Expects the /tmp folder to be cleaned up between INDIVIDUAL tests
|
|
# OpenTest.OTrunc creates /tmp/truncd
|
|
# OpenTest.OTruncAndReadOnlyDir attempts to create the same directory just afterwards
|
|
# Neither clean up after themselves
|
|
open_test
|
|
|
|
# This test doesn't correctly check to see if seccomp is enabled
|
|
# Attempts using regardless of enabled or not
|
|
seccomp_test
|
|
|
|
# Relies on ptrace to execute very simple binaries
|
|
exec_binary_test
|
|
|
|
# Relies on ptrace to send the parent process a signal
|
|
prctl_test
|
|
|
|
# Doesn't event pass on real x86 host
|
|
# GetrusageTest.IgnoreSIGCHLD is expecting memory usage to be zero, but it isn't
|
|
getrusage_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
sendfile_test
|
|
|
|
# Does a couple of things we don't support currently
|
|
# ForkTest.SigAltStack : Fork should inherit altstack
|
|
# ForkTest.Affinity : Affinity masks should be inherited
|
|
# CloneTest.NonCanonicalTLS : Non-Canonical TLS addresses should be rejecte with -EPERM
|
|
fork_test
|
|
|
|
# Requires accurate exit status from vfork
|
|
# Which we currently don't achieve
|
|
vfork_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
# ProcSelfFd.GetdentsDuplicates : Expects fcntl F_DUPFD, 1024 to work. Returns -1
|
|
# ProcSelfFdInfo.GetdentsDuplicates : Same as above
|
|
# ProcSelfAuxv.EntryPresence expects AT_SYSINFO_EHDR to exist. We don't support VDSO atm
|
|
# ProcPidCmdline.SubprocessForkSameCmdline expects to read parent pid's /proc/{pid}/cmdline and get executable name. Expecting a match.
|
|
# ProcPidExe.Subprocess : Same as above
|
|
# ProcPidEnviron.MatchesEnviron Doesn't expect us injecting environment variables in to its container space
|
|
proc_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
# Takes a long time to run as well
|
|
pty_test
|
|
|
|
# This syscall isn't implemented on ARM64
|
|
mempolicy_test
|
|
|
|
# Relies on namespaces to exist on the host (and user accessible) and isn't guaranteed by our CI
|
|
prctl_setuid_test
|
|
|
|
# This does 32-bit syscalls from a 64-bit process. Which we don't support yet
|
|
32bit_test
|
|
|
|
# We are lying to the guest application about true affinity. Missing some edges here
|
|
affinity_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
# Uses some private sandbox socket system?
|
|
# Seems like it needs some environment variables set that is google specific
|
|
connect_external_test
|
|
|
|
# Tests a bunch of different exception exits and we don't support them all
|
|
exceptions_test
|
|
|
|
# These search for folders in /
|
|
# Might need to rbind some folders to make it happy
|
|
# This currently passes on x86-64 hosts
|
|
# This fails on AArch64 because getdents isn't emulated
|
|
getdents_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
# Needs some helper executable to work
|
|
sigaltstack_test
|
|
|
|
# Sets up an alarm to fire every 500ms
|
|
# Sets up rcx and r11 to a known good value
|
|
# sets a boolean in memory and spins until the signal handler sets another boolean
|
|
# Stores the resulting values to a memory address
|
|
# Then checks to ensure the signal handler picked up the correct context values
|
|
# Seems like FEX currently doesn't?
|
|
sigiret_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
# This relies on splice behaviour that doesn't work even on x86 host
|
|
# Hangs after failing
|
|
splice_test
|
|
|
|
# Relies on ptrace to run
|
|
wait_test
|
|
sysret_test
|
|
|
|
# Has race conditions in the code and takes a long time to run
|
|
tcp_socket_test
|
|
|
|
# Doesn't even pass on real x86 host
|
|
socket_bind_to_device_test
|
|
socket_netlink_route_test
|
|
socket_bind_to_device_distribution_test
|
|
socket_bind_to_device_sequence_test
|
|
socket_inet_loopback_nogotsan_test
|
|
socket_ipv4_udp_unbound_loopback_nogotsan_test
|
|
socket_ipv4_udp_unbound_loopback_test
|
|
|
|
# This takes forever even on x86 host
|
|
socket_inet_loopback_test
|
|
socket_ip_tcp_generic_loopback_test
|
|
socket_stress_test
|
|
alarm_test
|
|
itimer_test
|
|
futex_test
|
|
fcntl_test
|
|
|
|
# these take a long time
|
|
ppoll_test
|
|
flock_test
|
|
pipe_test
|
|
epoll_test
|
|
concurrency_test
|
|
|
|
# This requires partial read support in the kernel
|
|
# Tries reading a page's worth of bytes from an FD across a page with no permissions
|
|
# Expecting a partial read to succeed
|
|
# Arm64 devices don't have this
|
|
partial_bad_buffer_test
|
|
|
|
# Depending on board can work or not, depending on kernel configuration
|
|
# IPv6 maybe?
|
|
socket_ipv6_udp_unbound_external_networking_test
|
|
|
|
# This test is flaky
|
|
proc_net_test
|
|
rtsignal_test
|
|
sigstop_test
|
|
stat_test
|
|
|
|
# Flaky
|
|
udp_socket_test
|
|
|
|
# This test checks for the header layout from `/proc/net/udp` which has changed slightly with newer kernels
|
|
proc_net_udp_test
|
|
|
|
# This test is broken since kernel commit: 36e2c7421f02
|
|
# Upstream gvisor has the offending tests removed
|
|
inotify_test
|
|
|
|
# Needs testing
|
|
semaphore_test
|
|
|
|
# This test does an unsafe timer test. (NoTimeout subtest specifically).
|
|
# Sets a timer for 100ms then starts a pselect with unlimited timeout.
|
|
# If the timer fires before the pselect starts then pselect will hang forever.
|
|
# Since we are running these tests on a fully loaded system, we tend to context switch,
|
|
# causing the chance of a hang to occur quite frequently.
|
|
pselect_test
|
|
|
|
# Somewhere between kernel versions 5.15 and 6.0, timerfd lseek behaviour has changed.
|
|
# On older kernels it would return -ESPIPE, on new kernels it returns 0.
|
|
# This test expects -ESPIPE behaviour.
|
|
timerfd_test
|