- 15 Mar, 2017 5 commits
-
-
Dmitry Safonov authored
The kernel does touch only relevant union member on x86_32. travis-ci: success for zdtm/sigpending/32: check only 12 bytes of _si_fields Cc: Andrei Vagin <avagin@virtuozzo.com> Signed-off-by:
Dmitry Safonov <dsafonov@virtuozzo.com> Signed-off-by:
Pavel Emelyanov <xemul@virtuozzo.com> Signed-off-by:
Andrei Vagin <avagin@virtuozzo.com>
-
Dmitry Safonov authored
It's library plugin to CRIU - it needs to be 64-bit. travis-ci: success for zdtm/test/rtc: build 64-bit criu-rtc.so Signed-off-by:
-
Dmitry Safonov authored
For 32-bit fcntl() Glibc function calls sys_fcntl64(), which needs struct flock64, otherwise the kernel gets a wrong struct. For 64-bit, it's all the same. Also unset errno before fcntl() and check return value of the call. Cc: Qiang Huang <h.huangqiang@huawei.com> Cc: Begunkov Pavel <asml.silence@gmail.com> Cc: Pavel Emelyanov <xemul@virtuozzo.com> travis-ci: success for test/zdtm: use flock64 instead of flock Signed-off-by:
-
Dmitry Safonov authored
Raw sys_mmap() returns address or error like -ENOMEM. Don't check MAP_FAILED, check that result is aligned by page. travis-ci: success for x86/compat: fix error-check for compat mmap() Signed-off-by:
-
Dmitry Safonov authored
Initialy, I thought to name it "compat_restore", but after I've dropped the second 32-bit parasite (which surely made compat code lesser and easier), our parasite works in 64-bit in 32-bit task and ptrace() for setting registers in this long-jumped situation will work correctly only after v4.9 kernel. Maybe it can be work-arounded if needed, but yet no compatible dump for pre-v4.9 kernels. Requested-by:
-
- 14 Mar, 2017 35 commits
-
-
Dmitry Safonov authored
As threads restore in parallel, stack32 may be reused concurrently leading to reusing others thread's data. So, let it lay on stack. It would still worth making 32-bit stack per-task reusing it in threads but at this moment introducing such complexity looks like premature optimization. It does not affect 64-bit C/R. Fixes: file_aio, sigaltstack, clone_fs, socket_aio, different_creds, futex travis-ci: success for 32-bit tests fixes Signed-off-by:
-
Dmitry Safonov authored
It's a workaround to clang-3.4, which doesn't handle numbers in asm macros rightly: https://llvm.org/bugs/show_bug.cgi?id=21500 Which resulted in: CC compel/arch/x86/plugins/std/parasite-head.o <instantiation>:3:2: error: too few operands for instruction pushq ^ compel/arch/x86/plugins/std/parasite-head.S:26:2: note: while in macro instantiation PARASITE_ENTRY ^ Fixes: https://travis-ci.org/0x7f454c46/criu/jobs/186099057 travis-ci: success for 32-bit tests fixes Signed-off-by:
-
Dmitry Safonov authored
Otherwise we'll try to set 32-bit register set to 64-bit task, which is not possible with ptrace - it uses register set size, according to processes mode. So we should set 32-bit regset only to tasks those are in 32-bit mode already. Please, see inline comment in the patch for more info. travis-ci: success for 32-bit tests fixes Cc: Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by:
Cyrill Gorcunov <gorcunov@openvz.org> Signed-off-by:
-
Dmitry Safonov authored
We can be on v4.9 or newer kernel, but have no 32-bit multilib toolchain, which will result in !CONFIG_COMPAT but kdat_compat_sigreturn_test() will return true as we have new arch_prctls in kernel needed for compatible C/R. Fail compat test in this case. travis-ci: success for 32-bit tests fixes Reported-by:
-
Dmitry Safonov authored
As I've united k_rtsigset_t between native and compat ucontext's, it's 8-bytes aligned now. We don't care about align of this as we fill it always with memcpy()/memset(0). So after those changes, ucontext_ia32 has received new padding between uc_mcontext and uc_sigmask (4 bytes to align to 8-byte sized sigmask). Because of this, mask of blocked signals was restored with shift (wrongly). Fixes: signalfd00, file_attr tests. travis-ci: success for 32-bit tests fixes Signed-off-by:
-
Andrei Vagin authored
In this case it will work for any make commangs. For example: make COMPAT_TEST=y -C test/zdtm/lib travis-ci: success for zdtm: handle COMPAT_TEST from Makefile.inc Cc: Dmitry Safonov <dsafonov@virtuozzo.com> Signed-off-by:
-
Kir Kolyshkin authored
Problem: if we do touch criu/arch/x86/cpu.c nothing is being rebuilt. For detailed analisys, see the previous commit ("criu/Makefile: fix rebuilding criu/pie/pie.lib.a"). Note there are other targets in criu/Makefile where one might want to add .FORCE to, but it's not needed as they all depend on another target which itself is marked as .PHONY. travis-ci: success for Makefiles: fix deps checking Signed-off-by:Kir Kolyshkin <kir@openvz.org> Reviewed-by:
-
Kir Kolyshkin authored
As reported by Andrei Vagin: touch criu/pie/parasite-vdso.c does not lead to rebuild pie.lib.a and its dependencies. This happened because the real dependencies of criu/pie/pie.lib.a are listed in criu/pie/Makefile.library, which is never included or called with sub-make, because the criu/Makefile says: criu/pie/pie.lib.a: $(ARCH-LIB) $(Q) $(MAKE) $(call build-as,Makefile.library,criu/pie) all essentially saying we only need to call a sub-make with Makefile.library if the target is absent, or $(ARCH-LIB) is newer than the target. A workaround is to use .FORCE so that the Makefile.library is always called and so the dependencies are checked. Note the above is also true for any target that involves calling a sub-make -- it should either be used with .FORCE or otherwise depend on a phony target. I haven't checked all the CRIU makefiles but suspect there might be more cases like this one. travis-ci: success for Makefiles: fix deps checking Cc: Cyrill Gorcunov <gorcunov@openvz.org> Reported-by:
-
Cyrill Gorcunov authored
travis-ci: success for .gitignore: Add compel and common asm symlinks Signed-off-by:
-
Kir Kolyshkin authored
Yet another nail in the coffin of compel cli usage complexity. Since commit cd1c9d9 ("compel: simplify usage wrt ids") landed, it became obvious that the newly introduced -p option can be made optional. First, prefix value is not very important; second, it can be easily generated from the input (or output) file name. This is what this commit does, also trying to check that the resulting prefix is adeqate for using in C code. In case it is not (say, file names used are 1.po 1.h), an error is printed, suggesting to use --prefix. The commit also makes use of this functionality by removing -p option from compel/test/infect and criu/pie Makefiles. While at it, let's also remove -l 4 and the kludge of hiding it. If the output generated by this is ever needed, one can easily run "compel hgen -l4 ..." manually. travis-ci: success for More polishing for compel cli Signed-off-by: -
Kir Kolyshkin authored
Add "compel libs" that prints the list of libraries needed to link the parasite loader. Make compel/test/ and criu/ to use it. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
A compel user should not be aware of compel internals, including the paths to various files. This commit introduces a command to get the plugin files (well, currently just one plugin, "std"). Example for uninstalled compel: $ cd compel/test/infect/ $ ../../../compel/compel-host plugins ../../../compel/plugins/std.built-in Example for installed compel: $ compel plugins /usr/libexec/compel/std.built-in.o The commit also makes use of this command by compel/test/infect and criu/pie. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
Commit 8173ea2 ("compel/test/infect/Makefile: don't add .h to sources") was a bad one as it removed dependency between parasite.h and spy binary. Fix it. travis-ci: success for More polishing for compel cli Signed-off-by: -
Kir Kolyshkin authored
Since commit 656710e the list of prerequisites are the same as the list of objects, so we can use $^ to avoid repetition. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
compel tool doesn't need to be run as root, so it makes no sense to install it to SBINDIR. Fix to use BINDIR. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
This commit adds -T path/to/linker_script to the output of "compel ldflags", so compel user does not have to specify one manually. This commit also makes use of this functionality in criu/pie and compel/test. NOTE this commit also drops the linker script dependency in criu/pie/Makefile, meaning if it will be changed that won't cause a rebuild. I hope it's not a big issue, and it is sort of inevitable as compel is becoming a separate tool. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
Instead of hardcoding the path to compel uapi includes, call compel tool to get one. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
Build of criu binary depends on many other things that needs to be built before it. Let's clean these deps a bit by using criu-deps variable. This also removes wrong "$(VERSION_HEADER): include/common/asm" dependency -- one can certainly succeed in generating criu/include/version.h file without creating include/common/asm symlink fist! travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
Working on CRIU Makefilest feels like walking on a minefield. This is not an attempt to demine it, but to make some mines a bit easier to notice. OK, this is what we're trying to do here: 1. Drop the $(SRC_DIR)/ prefix from the target: as this is a top-level Makefile, we can just use paths relative to top-level source dir. 2. Drop the $(SRC_DIR) from the symlink, use relative one. Relative symlinks are always better -- say, if a source directory is moved, everything will still work. 3. The "compel/include/asm: compel/arch/$(ARCH)/src/lib/include" dependency is useless. Yes, the left side is created as a symlink to the right side, but that doesn't mean that "make" should compare the timestamps of both to decide whether to remake the target. 4. The "$(COMPEL_VERSION_HEADER): compel/include/asm" dependency is wrong, the compel/include/asm symlink is not needed for $(COMPEL_VERSION_HEADER) generation. Remove it. 5. Move compel/plugins/std.built-in.o prerequisite from the rule to compel-plugins variable, and use it. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
Similar to the previous commit, there is absolutely no need to create/remove this symlink from Makefiles, as it can be made a constant one. Add the symlink to sources and save a few lines in Makefiles. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
There is absolutely no need to create/remove this symlink from Makefiles, as it is constant. Just add the symlink to sources and save a few lines in Makefiles. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
As of commit 3c16dc5 ("compel: plugins -- Merge fds plugin into std") fds plugin is no more. Remove this leftover. travis-ci: success for More polishing for compel cli Signed-off-by: -
Kir Kolyshkin authored
We call compel-host binary from a number of places, so let's add COMPEL_BIN variable and use it. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
I have noticed compel/plugins/include[/uapi] is not needed, not entirely sure why (added symlinks?) but everything compiles just fine without it. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
1. Add "compel includes" command, to be used for parasite *loading* code compilation. 2. Add includes to output of "compel cflags", which is used for parasite code compilation. Now, this patch looks big and complex, this is mostly because we want compel cli to work for both uninstalled (right from the source tree) and installed cases. The paths to be printed are quite different for these two cases, so I had to introduce a wrapper for a non-installed case. The wrapper sets an environment variable, which compel binary uses as a path to non-installed file. If this env var is not set, it means compel is installed so no tricks are needed. Note the wrapper is only provided for the compel-host binary, as compel (which differs from compel-host in case of cross-compiling) is not executed from within the source tree. Because of the wrapper, the original binary had to be renamed, thus the changes to Makefiles and .gitignore. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
This is mainly dictated by the fact that we use stdout from "compel *flags" in a special way, so it should not be garbled by the usage info. Otherwise, for example, the following code in Makefile CFLAGS += $(shell compel --badopt cflags) will lead to the whole usage() output to be added to compiler flags, which looks really really weird. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
There is no need to support all possible architectures for "compel cflags" action. In fact, "compel hgen" can only support the one it was compiled for (with the only exception of 32-bit mode for x86). It looks like if we can use a few #ifdefs, there is no need to specify --arch anymore, let's drop it! Still, for the x86 32-bit mode we need to introduce --compat option. Note that "compel hgen" autodetects 32-bit mode for x86 by looking into ELF header, but in case of "compel clfags" there are no files to look into, so we need this --compat specified explicitly. While at it, - Makefile: define CONFIG_AARCH64 if building for ARM64 - fail to compile on unsupported/unspecified ARCH - make "compel --help" output a bit more compact travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
1 Since all the fields are now initialized to zeroes, we can just say ={}; 2 Since this is static, it is initialized to zeroes anyway, but from my POV being explicit about it is better. travis-ci: success for More polishing for compel cli Signed-off-by: -
Kir Kolyshkin authored
travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
I saw this line in the code unsigned long sret = -ENOSYS; and ended up with this patch. Note syscall(2) man page says return value is long -- who am I to disagree? travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
1. Remove COMMON_IDIR -- there's no need for it, and it makes the build process look more complicated than it is. 2. Use standard CC and CFLAGS defines (so one can use say make CC=clang) 3. Drop the dependency on COMPEL_LIBRARY, as it's supposed to be an external file/tool. 4. Add -Wall to CFLAGS. 5. Simplify ARCH setting in infect test. 6. Drop ARCH from rsys test, as it's not used there. travis-ci: success for More polishing for compel cli Signed-off-by:
-
Kir Kolyshkin authored
travis-ci: success for More polishing for compel cli Signed-off-by:
-
Dmitry Safonov authored
I managed to forgot that kernel erases R8...R11 regitster prior returning to userspace from `int 0x80`: https://lkml.org/lkml/2009/10/1/164 That was the reason for hang on gcc v6.1 on Jenkins/elsewhere: https://ci.openvz.org/job/CRIU/job/CRIU-x86_64/branch/criu-dev/ Tested on gcc v6.1 locally, on Travis-CI: https://travis-ci.org/0x7f454c46/criu/builds/183976899 travis-ci: success for pie: fix clobber registers for int 0x80 (rev2) Signed-off-by:
-
Pavel Emelyanov authored
Signed-off-by:
-
Dmitry Safonov authored
Resolves merge conflict with commit 71abfb84e158 ("compel: simplify usage wrt ids"). Sets start ip for parasite blob according to app mode: native/compat. travis-ci: success for Rectify 32-bit compatible C/R on x86 Signed-off-by:
-
