This is the mail archive of the glibc-bugs@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[Bug build/21601] glibc-2.25: unknown type name 'lll_futex_timed'


https://sourceware.org/bugzilla/show_bug.cgi?id=21601

Andrej Valek <andrej.valek at siemens dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #10147|0                           |1
        is obsolete|                            |

--- Comment #4 from Andrej Valek <andrej.valek at siemens dot com> ---
Created attachment 10150
  --> https://sourceware.org/bugzilla/attachment.cgi?id=10150&action=edit
proposal patch V2 to fix "unknown type name 'lll_futex_timed'"

Some words for introduction into problematic.

I have a legacy java JVM, which can not be rebuild with new gblic (I have only
a binary). My application uses this JVM binary for Thread.sleep method calling
to wait some time. The main idea was to have waiting in thread time
independent. From the strace analysis/debugging have found, that binary uses
semaphores for waiting. The main problem is, that the semaphores uses
CLOCK_REALTIME as clock source.

My java application was working well with glibc_2.11 . Now, I have upgraded the
glibc to version 2.23 and it has stopped working. So I have created
C-application for testing (which I have had under control). The application
uses CLOCK_MONOTONIC and CLOCK_REALTIME as selective clock sources. So I could
compare the time spared by waiting. I was changing a system time over 'date'
command back and forward. Time spared by waiting had to be equal with the set
timeout.

C-app
- glibc_2.11 and 2.20
 - MONO and REAL was not influenced by system time changing
- glibc_2.23
 - MONO same as 2.11 and 2.20
 - REAL - was influenced by system time changing what is correct

Java-app
- only glibc_2.11 was time independent

So I was trying to make some workaround, how to solve the problem without JVM
binary changing. I have undefined __ASSUME_FUTEX_CLOCK_REALTIME . Now the java
uses lll_futex_timed_wait function not the lll_futex_timed_wait_bitset, where
is FUTEX_CLOCK_REALTIME as default.

I don't know, if this is a right solution. I didn't want to add some "ugly"
callers checking or something else into the glibc.

And now, I was trying to make same thing in 2.25.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]