Messages in this thread |  | | Date | Mon, 13 Apr 2015 18:21:41 -0700 | From | Andy Lutomirski <> | Subject | Re: Remove execution domain support |
| |
On 04/11/2015 01:47 PM, Richard Weinberger wrote: > A long time ago there was the idea to support different ABIs on Linux > Such that someone could use for example SCO UNIX binaries on Linux > without emulation. While the linux-abi project existed mostly as > out-of-tree patch it's core component, execution domains, > went mainline. An execution domain allows to specify mappings between > different ABIs, these mappings consist of signal, error, socket type, > socket options and addess familiy mappings. > Mainline has only support for signal mappings. > The signal mapping happens in the architecutre specific signal code. > Some archs support it, some not, most of them copy&pasted from i386. > > While the mapping support is incomplete we have two in-kernel users of > execution domains. > > 1. RISC OS personality on ARM. > The Kconfig help message states that this feature is very experimental > and it is likely that it never worked as expected. > Russel is fine with the removal. > > 2. ia32 support on ia64. > This is a left over from: > commit 32974ad4907cdde6c9de612cd1b2ee0568fb9409 ([IA64] Remove COMPAT_IA32 support) > It registers a dummy Linux/x86 execdomain to make calls to > personality(PER_LINUX32) faster as the execution domain core code > would try to load a module. > > As both users can be removed there is no user left and we can remove > execution domains support at all from the kernel. > The removal will make architecutre specific signal handling easier > and there is no need to keep execuction domains as this feature is > obviously incomplete and abandoned.
I like this.
FWIW, we recently removed restart_block from thread_info, so I'm not too worried about fallout from removing another field.
--Andy
|  |