Now that GenFw converts small code model ADRP instructions to ADR on
the fly, we can reduce the alignment for XIP modules, where large
alignment values may cause considerable waste of flash space due to
excessive padding. This limits the module size to 1 MB, but this is
not a concern in practice.
So set the XIP section alignment to 0x20 for DEBUG_GCC49, DEBUG_GCC5
and *_CLANG35, all of which use the small code model.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Leif Lindholm <leif.lindholm@linaro.org>
The ADRP instruction in the AArch64 ISA requires the link time and load time
offsets of a binary to be equal modulo 4 KB. The reason is that this instruction
always produces a multiple of 4 KB, and relies on a subsequent ADD or LDR
instruction to set the offset into the page. The resulting symbol reference
only produces the correct value if the symbol in question resides at that
exact offset into the page, and so loading the binary at arbitrary offsets
is not possible.
Due to the various levels of padding when packing FVs into FVs into FDs, this
alignment is very costly for XIP code, and so we would like to relax this
alignment requirement if possible.
Given that symbols that are sufficiently close (within 1 MB) of the reference
can also be reached using an ADR instruction which does not suffer from this
alignment issue, let's replace ADRP instructions with ADR after linking if
the offset can be encoded in this instruction's immediate field. Note that
this only makes sense if the section alignment is < 4 KB. Otherwise,
replacing the ADRP has no benefit, considering that the subsequent ADD or
LDR instruction is retained, and that micro-architectures are more likely
to be optimized for ADRP/ADD pairs (i.e., via micro op fusing) than for
ADR/ADD pairs, which are non-typical.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Leif Lindholm <leif.lindholm@linaro.org>
This adds support for GCC 5.x in LTO mode for IA32, X64, ARM and
AARCH64. Due to the fact that the GCC project switched to a new
numbering scheme where the first digit is now incremented for every
major release, the new toolchain is simply called 'GCC5', and is
intended to support all GCC v5.x releases.
Since IA32 and X64 enable compiler optimizations (-Os) for both DEBUG
and RELEASE builds, LTO support is equally enabled for both targets.
On ARM and AARCH64, DEBUG builds are not optimized, and so the LTO
optimizations are only enabled for RELEASE.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Recent versions of GNU ld automatically emit a .notes section into
the ELF binary containing a build id. Since this is an allocatable
section by default, it will be identified by GenFw as a section
that requires PE/COFF conversion, which may cause sections to be
moved around unexpectedly.
So retain the section, but tag it as INFO, which tells the linker
that it should not be accounted for in the binary's memory layout.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
To accommodate upcoming GCCx toolchain versions that require 'gcc' to
be used as the linker in order to support LTO, switch GCC44 and later
(including CLANG35) to a new DLINK build rule that invokes 'gcc' as the
linker instead of 'ld'. Since gcc expects its command line arguments in
a different format, and expects arguments that it needs to pass to the
linker to be prefixed with '-Wl,', this involves changes to most of the
DLINK_FLAGS definitions in tools_def.template, as well as some changes to
module .INF files that set their own linker options.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Before we can make non-backward compatible changes to the GCC build rules
regarding the use of the 'gcc' binary as the linker, clone the existing
GCC build rules into a 'GCCLD' build rule family, and move the legacy
toolchains UNIXGCC, CYGGCC, CYGGCCxASL and ELFGCC over to it.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Some versions of Clang fail on every input file when using the
-save-temps options, and produces the following heplful error message:
<unknown>:0: error: Undefined temporary symbol
Simply dropping the option for CLANG35 is the simplest way around this,
since the value of storing .i and .s files is dubious anyway.
Also, drop the arm-use-movt option, which does not appear to be
supported anymore by recent versions of clang.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Acked-by: Jordan Justen <jordan.l.justen@intel.com>
If the binary module is list in the FDF file but not list in the DSC
file, current build report would not include these binary module's info
in the report "Module section". The patch fix this issue.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
If source module INF is not listed in DSC, it will not be built. And it
is listed in FDF, GenFds will fail to find its build output. To reminder
user this issue early, build tool should report failure to user in early
phase.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
In batch script files, setting a variable in an 'if' block will only take
effect after the 'if' block.
This commit fixes the issue of using the variable 'CONF_PATH' right after
it is being set in an 'if' block.
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Hao Wu <hao.a.wu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
When CONF_PATH is already set, toolsetup.bat overwrites its value.
This is not the case on Linux platforms (BuildEnv) and
contra-productive when using the same Workspace across multiple
Operating Systems.
With this patch, a check is performed prior to setting the variable.
Furthermore, it will not be scanned for Conf directories in
PACKAGES_PATH directories to respect the user's choice.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Marvin Haeuser <Marvin.Haeuser@outlook.com>
Reviewed-by: Yonghong Zhu <yonghong.zhu@intel.com>
According to UEFI spec:
Once an image is loaded, LoadImage() installs
EFI_HII_PACKAGE_LIST_PROTOCOL on the handle if the image contains a
custom PE/COFF resource with the type 'HII'. The protocol's
interface pointer points to the HII package list which is contained
in the resource's data.
This is controlled by the UEFI_HII_RESOURCE_SECTION define in the INF
file. When present the HII resource is linked with the module
binary.
Unfortunately GCC-built binaries have been stripping the .hii section
entirely. See "[edk2] HII gEfiHiiPackageListProtocolGuid problem
with GCC48(VS2012x86 works)"
http://thread.gmane.org/gmane.comp.bios.tianocore.devel/13438http://thread.gmane.org/gmane.comp.bios.tianocore.devel/14899
This patch tells the linker to preserve the .hii sections
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Thomas Palmer <thomas.palmer@hpe.com>
Acked-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Bruce Cran <bruce.cran@gmail.com>
Reviewed-by: Bruce Cran <bruce.cran@gmail.com>
Reviewed-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
The ordinary small code model for x86_64 cannot be used in UEFI, since
it assumes the executable is loaded in the first 2 GB of memory.
Therefore, we use the large model instead, which can execute anywhere,
but uses absolute 64-bit wide quantities for all symbol references,
which is costly in terms of code size.
So switch to the PIE small code model, this uses 32-bit relative
references where possible, but does not make any assumptions about the
load address (i.e., all absolute symbol references are 64-bits wide).
Note that, due to the 'protected' visibility pragma introduced in an
earlier patch, there is no need for the EDK2 build system to deal with
GOT related ELF relocation types.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Laszlo Ersek <lersek@redhat.com>
Tested-By: Liming Gao <liming.gao@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Now that we switched to the __builtin_ms_va_list VA_LIST type for
GCC/X64, we can trust the compiler to do the right thing even under
optimization, and so we can enable -Os optimization all the way back
to GCC44, and drop the -D define that prevents the use of the __builtin
VA_LIST types. Note that this requires the -maccumulate-outgoing-args
switch as well.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Tested-by: Laszlo Ersek <lersek@redhat.com>
Tested-By: Liming Gao <liming.gao@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
NASM had been unable to assemble segment register operations before the
following git commit:
http://repo.or.cz/nasm.git/commitdiff/21d4ccc3c338
That commit was first released in NASM 2.10:
http://repo.or.cz/nasm.git/commitdiff/ff62f33da0a2
This makes NASM 2.07 unusable for edk2 in general, because now we have a
lot of X64 assembly code that works with segment registers. For example
in:
UefiCpuPkg/Library/CpuExceptionHandlerLib/X64/ExceptionHandlerAsm.nasm
Bump the minimum required version to 2.10, for use with GCC toolchains.
Furthermore, list NASM 2.12.01 as a requirement for all other toolchains.
In particular, for source level debugging, VS20xx requires CodeView 8
debug symbols, and only NASM 2.12.01 and later produce those. (Suggested
by Liming, Mike, and Andrew.)
Cc: Andrew Fish <afish@apple.com>
Cc: Liming Gao <liming.gao@intel.com>
Cc: Michael Kinney <michael.d.kinney@intel.com>
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Ref: http://thread.gmane.org/gmane.comp.bios.edk2.devel/14612
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
If the library is listed in [Components] section for build only, its
used FixedPcd Value is not generated into AutoGen code. This patch
cover this case to generate the FixedPcd Value in AutoGen file.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Since in the GenFds phase, the FV is generated as upper letter. This
patch update the FV region name as upper letter, it can fix the build
report generate failure on case sensitive file system.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Reviewed-by: Andrew Fish <afish@apple.com>
In its current form, Region.PadBuffer() fills every second byte with 0x20,
the default separator string of Python's string.join():
https://docs.python.org/2/library/string.html#string.join
This corrupts some firmware because (a) 0x20 never corresponds to any
ErasePolarity, (b) the PadData produced are actually longer than Size.
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Liming Gao <liming.gao@intel.com>
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Reported-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Fixes: bd907fb638
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Tested-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
The current implementation calls both pack() and Buffer.write() Size
times. The new implementation calls both of these methods only once; the
full data to write are constructed locally [1]. The range() function is
replaced by xrange() because the latter is supposed to be faster / lighter
weight [2].
On my laptop, I tested the change as follows: I pre-built the series at
[3] with
build -a X64 -p OvmfPkg/OvmfPkgX64.dsc -t GCC48 -b DEBUG \
-D HTTP_BOOT_ENABLE -D SECURE_BOOT_ENABLE
(The series at [3] is relevant because it increases the size of one of the
padded regions by 8.5 MB, slowing down the build quite a bit.)
With all source code already compiled, repeating the above command takes
approximately 45 seconds. With the patch applied, it goes down to 29
seconds.
[1] http://stackoverflow.com/questions/27384093/fastest-way-to-write-huge-data-in-file
[2] https://docs.python.org/2/library/functions.html?highlight=xrange#xrange
[3] http://thread.gmane.org/gmane.comp.bios.edk2.devel/14214
We can also measure the impact with a synthetic test:
> import timeit
>
> test_old = """
> import struct, string, StringIO
> Size = (8 * 1024 + 512) * 1024
> Buffer = StringIO.StringIO()
> PadData = 0xFF
> for i in range(0, Size):
> Buffer.write(struct.pack('B', PadData))
> """
>
> test_new = """
> import struct, string, StringIO
> Size = (8 * 1024 + 512) * 1024
> Buffer = StringIO.StringIO()
> PadByte = struct.pack('B', 0xFF)
> PadData = string.join(PadByte for i in xrange(0, Size))
> Buffer.write(PadData)
> """
>
> print(timeit.repeat(stmt=test_old, number=1, repeat=3))
> print(timeit.repeat(stmt=test_new, number=1, repeat=3))
The output is
[8.231637001037598, 8.81188416481018, 8.948754072189331]
[0.5503702163696289, 0.5461571216583252, 0.578315019607544]
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
The same logic is used in five places; factor it out to a common method.
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Current MemoryProfileSymbolGen.py assumes the rva is 32bits,
the patch is to remove the restriction to match any lengths
of rva.
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Liming Gao <liming.gao@intel.com>
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Star Zeng <star.zeng@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
Build spec mentions $(FAMILY) macro be used in DSC/FDF to specify the tool
chain family, like GCC, MSFT. This patch add the support for this macro.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
This tool depends on DIA2Dump.exe (VS) or nm (gcc) to parse debug entry.
Usage: MemoryProfileSymbolGen.py [--version] [-h] [--help] [-i inputfile
[-o outputfile]]
Copyright (c) 2016, Intel Corporation. All rights reserved.
Options:
--version show program's version number and exit
-h, --help show this help message and exit
-i INPUTFILENAME, --inputfile=INPUTFILENAME
The input memory profile info file output from
MemoryProfileInfo application in MdeModulePkg
-o OUTPUTFILENAME, --outputfile=OUTPUTFILENAME
The output memory profile info file with symbol,
MemoryProfileInfoSymbol.txt will be used if it is not
Cc: Jiewen Yao <jiewen.yao@intel.com>
Cc: Liming Gao <liming.gao@intel.com>
Cc: Yonghong Zhu <yonghong.zhu@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Star Zeng <star.zeng@intel.com>
Reviewed-by: Jiewen Yao <jiewen.yao@intel.com>
We now check to see if the destination .nasm file already exists. If
it does, then we don't try to convert the .asm to .nasm.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
In the first stage of conversion, we need to preserve the AT&T style
.s assembly files for use with OS X toolchains.
This change allows '--keep=s' to be used with the script to preserve
these files.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Python 3's filter and map functions returns an iterator which you
can't call len() on. Since we'll want to use len() later, we put the
filter results into a tuple.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
For single module build, it would call gen_libs target. then if it use
binary LIB file, it cause build failure.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Fix the bug that always use the '-e' as OPTROM_FLAGS even the .INF file
has statement 'PCI_COMPRESS = TRUE'.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Add the error handling to cover the case that current_dir is not exist.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
this is something I missed in 8ac46e4
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Michael Zimmermann <sigmaepsilon92@gmail.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
GenFds GenSection will search the output file based on the file extension.
If the output file has no extension, it should be skip.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Liming Gao <liming.gao@intel.com>
Reviewed-by: Andrew Fish <afish@apple.com>
when user don't set TOOL_CHAIN_TAG and TARGET by –D Flag, then GenFds
would report failure for format:
FILE DATA = $(OUTPUT_DIRECTORY)/$(TARGET)_$(TOOL_CHAIN_TAG)/testfile
so this patch enhance to get the TOOL_CHAIN_TAG and TARGET value by
following priority (high to low): 1. the Macro value set by -D Flag;
2. Get the value by the -t/-b option. 3. get the value from target.txt
file. Besides, this patch also remove the error checking for missing
-t/-b option.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
This patch adds two error conditions:
1) if one path in PACKAGES_PATH doesn't exist.
2) if the space exists in the PACKAGES_PATH.
In V2, highlight one path in PACKAGES_PATH env doesn't exist.
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Reviewed by: Andrew Fish <afish@apple.com>
EDKII build spec and DEC spec updated to support private package
definition.
If GUID, Protocol or PPI is listed in a DEC file, where the Private
modifier is used in the section tag ([Guids.common.Private] for example),
only modules within the package are permitted to use the GUID, Protocol
or PPI. If a module or library instance outside of the package attempts
to use the item, the build must fail with an appropriate error message.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
The bug is use FILE_GUID override to build the same module more than
once, GenFds report warning "xxx NOT found in DSC file; Is it really
a binary module?". The root cause is the module path with FILE_GUID
overridden has the file name FILE_GUIDmodule.inf, then
PlatformDataBase.Modules use FILE_GUIDmodule.inf as key which cause
__GetPlatformArchList__ return empty.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
in the generated Asbuilt inf would include the driver's complete
dependency expression, and it would be wrote as comment format. Original
bug is mix the depex expression with real comment in the depex section.
this patch is ignore the real comment, and list the depex expression.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
It should start from the last '\x' position + 1 to find next '\x'
character.
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Description:
When building for any specific architecture, the build script today is loading
DSC sections for other architectures not in the build. The build process should
disregard DSC sections that are not relevant to the build.
My previous patch only fixed issue for one section type (Components). This
patch will handle all section types by updating the MetaFileParser class, which
now takes a Arch argument and will filter the DSC table results as they are
returned from the database. The database still contains all information from
DSCs for when builds support multiple arch's
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Thomas Palmer <thomas.palmer@hpe.com>
Reviewed-by: Yonghong Zhu <yonghong.zhu@intel.com>
Fix the bug for FMP image to support to use Macro as path description,
eg: FILE DATA = $(OUTPUT_DIRECTORY)/$(TARGET)_$(TOOL_CHAIN_TAG)/test.efi
Cc: Liming Gao <liming.gao@intel.com>
Contributed-under: TianoCore Contribution Agreement 1.0
Signed-off-by: Yonghong Zhu <yonghong.zhu@intel.com>
Reviewed-by: Liming Gao <liming.gao@intel.com>
Ard Biesheuvel
Hess Chen
Yonghong Zhu
Liming Gao
Hao Wu
Marvin Haeuser
Thomas Palmer
Laszlo Ersek
Star Zeng
Jordan Justen
Michael Zimmermann
Zenith432
Samer El-Haj-Mahmoud