Load IRT in dynamic text region

[zhiting zhu]

Dear group members:

I'm studying the native client source code for my research project. I find that IRT is loaded into dynamic text region. Is there any reason IRT.nexe needs to be backed by shared memory? Can I load IRT.nexe into static text region? In my use case, I only supports static linking and don't need to support JIT. 

Best,

Zhiting

[Mark Seaborn]

On 10 March 2016 at 13:23, zhiting zhu <zzt...@gmail.com> wrote:

Dear group members:

I'm studying the native client source code for my research project. I find that IRT is loaded into dynamic text region. Is there any reason IRT.nexe needs to be backed by shared memory? Can I load IRT.nexe into static text region? In my use case, I only supports static linking and don't need to support JIT.

In principle, the IRT could be loaded using mmap() rather than being copied into the shared memory segment.  But if I remember correctly (and from looking at NaClElfImageLoadDynamically() in elf_util.c), the codebase does not implement that at the moment.

What's your reason for wanting to avoid using the dynamic code region?  Performance, or a different reason?

Cheers,

Mark

[zhiting zhu]

I'm running NaCl in a environment that shared mapping cannot be implemented. 

在 2016年3月10日星期四 UTC-6下午4:13:37，Mark Seaborn写道：

[Mark Seaborn]

Can you do mmap() of files at all in the context you're running NaCl in?

If so, changing IRT loading to mmap() the IRT's code segment would be a good solution.

Otherwise, you'd need a change that loads the IRT's code segment into anonymous memory.  Or alternatively, you could use NaCl without the IRT -- that might be an option, depending on your use case.

Cheers,

Mark

[zhiting zhu]

Yes, I can mmap() files with MAP_PRIVATE. 

I'm interested in using NaCl without irt. As a off browser environment, I don't need to support PPAPI and imc calls. But I do need support for pthread. Is that a possible option? 

Best,

Zhiting
在 2016年3月15日星期二 UTC-5下午8:00:37，Mark Seaborn写道：

[Mark Seaborn]

On 15 March 2016 at 19:55, zhiting zhu <zzt...@gmail.com> wrote:

Yes, I can mmap() files with MAP_PRIVATE. 

I'm interested in using NaCl without irt. As a off browser environment, I don't need to support PPAPI and imc calls. But I do need support for pthread. Is that a possible option?

Yes, you can build nexes that don't depend on the IRT.  Just link against libnacl_sys_private and (if necessary) libpthread_private -- these call NaCl syscalls directly.

The resulting nexes won't be suitable for running in Chrome, because they won't be using ABI-stable interfaces, but they are fine for other use cases for NaCl that don't require ABI stability for nexes.

Cheers,

Mark

[zhiting zhu]

Hi Mark,

After building some application successfully with libnacl_sys_private and libpthread_private, I encounter one application which gives out signal 4 immediately at startup. I trace through the program execution. It fails at __libnacl_fatal. I think it fails in libnacl_irt_init which calls the __libnacl_mandatory_irt_query. As I don't have irt loaded in, the __nacl_irt_query is NULL which calls the __libnacl_fatal. It's quiet strange that their behavior are different. Is there anything I forget to do?  

Best,

Zhiting

Mark Seaborn <msea...@chromium.org>于2016年3月18日周五 下午1:17写道：

[Mark Seaborn]

On 25 April 2016 at 17:32, zhiting zhu <zzt...@gmail.com> wrote:

Hi Mark,

After building some application successfully with libnacl_sys_private and libpthread_private, I encounter one application which gives out signal 4 immediately at startup. I trace through the program execution. It fails at __libnacl_fatal. I think it fails in libnacl_irt_init which calls the __libnacl_mandatory_irt_query. As I don't have irt loaded in, the __nacl_irt_query is NULL which calls the __libnacl_fatal. It's quiet strange that their behavior are different. Is there anything I forget to do?

libnacl_sys_private is supposed to override the definition of libnacl_irt_init() with an empty definition.  So when you say you're getting a failure in libnacl_irt_init(), that suggests that the overriding isn't working successfully.  Maybe there's some link ordering problem?

How does the application with the problem get built -- i.e. what type of build system does it use?  How are you adding in libnacl_sys_private at link time?

Cheers,

Mark

[zhiting zhu]

We figure out a solution. The problem we have is on c++ application built with clang-newlib toolchain. We find that we need to put c++ before libnacl_sys_private in the link order. Not sure why but we find that after multiple trials. 
在 2016年4月26日星期二 UTC-5下午1:08:01，Mark Seaborn写道：

[Victor Khimenko]

On Thu, May 5, 2016 at 3:12 PM, zhiting zhu <zzt...@gmail.com> wrote:

We figure out a solution. The problem we have is on c++ application built with clang-newlib toolchain. We find that we need to put c++ before libnacl_sys_private in the link order. Not sure why but we find that after multiple trials. 

You need to put libnacl_sys_private after libc++ or libstdc++ because C/C++ linking was designed that way decades ago. I mean: HOW ELSE would you override definition of functions?

The process is described in all books I've seen which teach C starting from the venerable K&R ( https://en.wikipedia.org/wiki/The_C_Programming_Language ), but if you somehow managed to forget about such basic things there are short description of the whole mechanism on stack overflow ( http://stackoverflow.com/questions/45135/why-does-the-order-in-which-libraries-are-linked-sometimes-cause-errors-in-gcc ) and also more precise description of what libnacl_sys_private is doing, too ( http://stackoverflow.com/questions/19023018/overriding-c-library-functions-calling-original ).

Again: that's not NaCl-specific at all, Unix did that for longer I've been alive and all popular OSes (including Mac, Windows and so on) work like this - thus, of course, Mark felt that there are no need to describe all the details.

--
You received this message because you are subscribed to the Google Groups "Native-Client-Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to native-client-di...@googlegroups.com.
To post to this group, send email to native-cli...@googlegroups.com.
Visit this group at https://groups.google.com/group/native-client-discuss.
For more options, visit https://groups.google.com/d/optout.

[zhiting zhu]

Well, I find the solution before I see the answer. I just report what I did. I don't know about the overriding definition part when I find the answer. I only look at the nacl source code but not the libc/c++ code. There is no any description on how libc/libc++ interect with the libnacl_sys_private. All I can find is without irt is not officially support and nothing else.