|
From: | Subhashish |
Subject: | Re: GSOC - valgrind-hurd queries |
Date: | Sun, 01 Jun 2014 18:26:25 +0530 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.5.0 |
On Saturday 31 May 2014 06:32 AM, Samuel Thibault wrote:
Um I think I'll proceed with the darwin style now, since it's a time-consuming process analysing all the #defines and typedefs and putting them together - we need a vki-gnu.h and a vki-x86-gnu.h - the latter holds arch specific definitions and is included into the former. This is nice, since multiple architectures of an os may exist. However this requires careful analysis.Subhashish, le Sat 31 May 2014 06:23:39 +0530, a écrit :Now since I've #include-ed "i386-gnu/bits/resource.h", do I need to #define the RLIMIT_* symbols with VKI_ prefixes? vki-linux.h doesn't but darwin's does.vki-linux does, in the vki-*-linux.h files.Moreover, ours are inside eums, so maybe I think they will be automatically picked up by rlimit and friends.But valgrind uses VKI_RLIMIT_* macros, not RLIMIT_* macros, so you have to define the former. Now, AIUI valgrind redefines everything from scratch, without using system headers. It seems darwin does use system headers, but AIUI it's not the way it's supposed to be done: vki-*.h redefines everything that coregrind needs from scratch. Samuel
I'm sure this using of system headers is not the way, but it can be easily swapped out when the redefining-from-scratch implementation is ready. I'm going with the darwin one now for that seems quick. I'll work on the other implementation later.
Now then a question - Do we use struct sigaction in ....bits/sigaction.h while passing sigactions both to and from the kernel? (emphasis on both)
Subhashish
[Prev in Thread] | Current Thread | [Next in Thread] |