Hi all: I am using Genode to do some SSL encryption stuff. My dev-board is i.MX6Quad running Genode 15.02. The OpenSSL version is 1.0.1i. I test the Crypto speed of OpenSSL in Genode. It's very slow( 7MB/s for aes-128-gcm) comparing to OpenSSL in Linux (16MB/s for aes-128-gcm). Genode's OpenSSL doesn't use the assembly code version of some critical crypto function. It disable this option in libcrypto.mk. If I do the same to Linux OpenSSL, the performance goes down to 9MB/s. So I enable this option in libcrypto.mk. use Genode compiler to compile it. Everything seems ok. No compiler/linker errors. But the program can not run. MMU exception occurs before the main program start. It seems that something goes wrong during the library loading. I am really confused. Why does Genode disable this assembly code option? ( I can only find a 1-sentence explaination on Genode release notes 12.08 http://genode.org/documentation/release-notes/12.08
Because libcrypto provides certain optimized assembler functions, which
unfortunately are not expressed with position-independent code, we removed this assembler code and build libcrypto with -DOPENSSL_NO_ASM
I guess the loader of Genode can not handle load-time relocation for shared library so all shared library must be position independent. Am I right?
Thank you.
-- Best wishes Vincent
Hi Vincent,
just a short remark: when you do performance measurements on top of Genode's i.MX6 base-hw platform, you should definitely upgrade to our recent release 16.02. The initial contribution that enabled i.MX6 for Genode did not encounter the L2 cache of this platform. So your measurement comparisons will be misleading anyway as far as you did not enabled the L2 cache yourself. The recent release enables all four cores and the L2 cache for the Wandboard target platform.
Regards Stefan
On 02/29/2016 03:36 AM, Vincent Stone wrote:
Hi all: I am using Genode to do some SSL encryption stuff. My dev-board is i.MX6Quad running Genode 15.02. The OpenSSL version is 1.0.1i. I test the Crypto speed of OpenSSL in Genode. It's very slow( 7MB/s for aes-128-gcm) comparing to OpenSSL in Linux (16MB/s for aes-128-gcm). Genode's OpenSSL doesn't use the assembly code version of some critical crypto function. It disable this option in libcrypto.mk. If I do the same to Linux OpenSSL, the performance goes down to 9MB/s. So I enable this option in libcrypto.mk. use Genode compiler to compile it. Everything seems ok. No compiler/linker errors. But the program can not run. MMU exception occurs before the main program start. It seems that something goes wrong during the library loading. I am really confused. Why does Genode disable this assembly code option? ( I can only find a 1-sentence explaination on Genode release notes 12.08 http://genode.org/documentation/release-notes/12.08
Because libcrypto provides certain optimized assembler functions, which
unfortunately are not expressed with position-independent code, we removed this assembler code and build libcrypto with -DOPENSSL_NO_ASM
I guess the loader of Genode can not handle load-time relocation for shared library so all shared library must be position independent. Am I right?
Thank you.
-- Best wishes Vincent
Site24x7 APM Insight: Get Deep Visibility into Application Performance APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month Monitor end-to-end web transactions and take corrective actions now Troubleshoot faster and improve end-user experience. Signup Now! http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
genode-main mailing list genode-main@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/genode-main
Hello Vincent,
* Vincent Stone <shihang7422166@...9...> [2016-02-29 10:36:29 +0800]:
Why does Genode disable this assembly code option? ( I can only find a 1-sentence explaination on Genode release notes 12.08 http://genode.org/documentation/release-notes/12.08
Because libcrypto provides certain optimized assembler functions, which
unfortunately are not expressed with position-independent code, we removed this assembler code and build libcrypto with -DOPENSSL_NO_ASM
Only specific assembler implementations for x86 were enabled when we did the initial libcrypto port. When the port was updated to a newer version the generated asm files did contain text relocations, which did not work when compiling with -fpic, IIRC. Since potential performance gains from using the asm implementation were not important I decided to remove the files and to use the generic C implementation instead. We never tested or used the asm implementations on other platforms, e.g. ARM.
I guess the loader of Genode can not handle load-time relocation for shared library so all shared library must be position independent. Am I right?
The dynamic loader does support load-time relocations for data but code relocations, e.g. function calls, are not supported [1].
[1] see repos/base/src/lib/ldso/spec/*/relocation.h
Regards, Josef
Thanks for answering my question. Yeah. These assembly file containing text relocation will cause the MMU exception, which leave me 2 choices: 1. rewrite the assembly code to be position independent. 2. add load-time relocations for code in Genode dynamic loader for arm.
And based on my test, the performance of the latest OpenSSL (1.0.2f) is way more better than 1.0.1i(Genode currently using) due to more optimized assembly code. 26 MB/s vs 16 MB/s. But if we disable the assembly code. The performance is almost identical(7~8 MB/s).
On Feb 29, 2016, at 6:28 PM, Josef Söntgen <josef.soentgen@...1...> wrote:
Hello Vincent,
- Vincent Stone <shihang7422166@...9...> [2016-02-29 10:36:29 +0800]:
Why does Genode disable this assembly code option? ( I can only find a 1-sentence explaination on Genode release notes 12.08 http://genode.org/documentation/release-notes/12.08
Because libcrypto provides certain optimized assembler functions, which
unfortunately are not expressed with position-independent code, we removed this assembler code and build libcrypto with -DOPENSSL_NO_ASM
Only specific assembler implementations for x86 were enabled when we did the initial libcrypto port. When the port was updated to a newer version the generated asm files did contain text relocations, which did not work when compiling with -fpic, IIRC. Since potential performance gains from using the asm implementation were not important I decided to remove the files and to use the generic C implementation instead. We never tested or used the asm implementations on other platforms, e.g. ARM.
I guess the loader of Genode can not handle load-time relocation for shared library so all shared library must be position independent. Am I right?
The dynamic loader does support load-time relocations for data but code relocations, e.g. function calls, are not supported [1].
[1] see repos/base/src/lib/ldso/spec/*/relocation.h
Regards, Josef
Site24x7 APM Insight: Get Deep Visibility into Application Performance APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month Monitor end-to-end web transactions and take corrective actions now Troubleshoot faster and improve end-user experience. Signup Now! http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140 _______________________________________________ genode-main mailing list genode-main@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/genode-main
-
Josef Söntgen -
Stefan Kalkowski -
Vincent Stone