NetBSD/doc/TODO.compat-module

124 lines
6.0 KiB
Plaintext

/* $NetBSD: TODO.compat-module,v 1.5 2019/02/04 22:00:51 mrg Exp $ */
DONE
----
1. Removed the building of the compat library - it is no longer needed.
2. Reverted some intentional auto-load breakage for loading the sysv_ipc
module; the breakage was introduced as the fix for a build break.
3. Split the sysv_ipc compat routines into their own compat_sysv module.
4. Resolved some inter-module dependencies.
5. Extracted some net/if.c compat routines into the compat module, and
replaced the originals with indirect (vectored) function calls.
6. Reconfirmed existing compat-module dependencies, and update the
defopt/defflag lines in the config files* as needed, to insure that
built-in dependencies get resolved.
7. Fixed limits on the number of module dependencies and maximum
recursion level (for auto-loading) have been removed. Previous code
for reporting module status to userland has been versioned and moved
to the (new) compat_80 module.
8. The old monolithic compat module has been broken into multiple
modules, one for each old NetBSD version. The monolithic compat
module is no longer available.
Similarly, the compat_sysv and compat_netbsd32 modules have also
been split into several version-specific modules, and the mini-
monolithic versions of these modules are no longer provided.
9. syscalls.master has been updated to autoload the version-specific
compat modules rather than the monolithic modules.
10. Separated COMPAT_BSDPTY stuff, allowing the COMPAT_60 module to be
built regardless.
11. Implemented a MP-safe mechanism for installing and removing function
pointers, preventing them from being unloaded (via modunload) while
in use. Thanks to riastradh@ for the template code.
12. Finished splitting the vnd_30 and vnd_50 compat code into separate
modules.
13. Cleaned up some previous vectored routines (related to if_43.c) to
use the MP-safe mechanism.
14. Organized (some of) the netbsd32 machine-dependent code to fit a
common build framework, and split version-specific code from baseline
code as needed. More work may be needed here (see #18 below).
15. The rtsock.c code has been split into two separate source files,
one for use in -current and one which is shared with COMPAT_50 (the
code is shared with -current, but macros are used to define version-
specific routine names and variable types). Version-specific parts
of rtsock.c for compat_14 and compat_70 have also been split out and
included in the relevant version-specific compat modules.
TODO - Not required for branch merge
------------------------------------
16. Audit the entire code base for any remaining embedded #ifdef's for
COMPAT_xx. When found, move the actual compat code into the compat
hierarchy and replace originals with indirect (vectored) calls.
17. The compat_60 module still needs some work for XEN systems. We
probably need some build infrastructure changes to ensure that
XEN (and, for i386, XEN-PAE) modules are build with the correct
macros defined and with -I directories specified in the same order
as for building kernels. See PR port-xen/53130. This currently
prevents loading of micro-code updates for amd64 processors running
XEN kernels. This limitation also exists on HEAD.
18. There seems to be quite a bit of MD compat_xx code, in the various
sys/arch/ directories. I haven't yet looked at any of this. But it
seems to me that the MI compat build infrastructure should have some
mechanism to "reach over" to the MD code, #include a Makefile.inc file,
and perhaps define something to enable the MI modcmd code to call a
compat_xx_MD_init() routine.
Note also that there are a few bits of MD code that is COMPAT_44
related. (The only bit of MI COMPAT_44 code is in the single module
shared by COMPAT_43 and COMPAT_09.) This affects the cesfic, hp300,
news68k, and x68k platforms, all in their respective machdep.c
source file. Additionally, the zaurus platform defines COMPAT_44 in
its INSTALL kernel configuration - but no other configuration files!
As far as I can tell, none of the MD compat code is currently built
into the monolithic COMPAT module on HEAD. Thus, its absence from
any of the version-specific modules is not a regression.
19. For compat_50, there are some things in dev/gpio and dev/wscons/wsmux
that I haven't been able to cleanly separate. These items are not
currently included in the monolithic COMPAT module on HEAD, so lack of
integration on the branch is not a regression.
20. Find all the remaining dependencies on the compat_utils routines and
deal with them appropriately. For now, we simply ensure that they
are included in every kernel via 'options COMPAT_UTILS' in file
sys/conf/std
21. The netbsd32_machine32_hook should be moved out of the main kernel
and into the compat_netbsd32 module. Unfortunately there are some
machines which include the consumer of this hook but do not have a
compat_netbsd32 module (specifically, i386 and sgimips). This
should be sorted out sometime soon, but does not block merging.
22. Note that the MPSAFE kernel option is currently not specified for
building modules, nor is it included in any standard kernel
configuration files. If you build a custom kernel with the MPSAFE
option set, and you also use modules (especially those modules for
network interface device drivers), you'll need to build custom
modules, too. The MPSAFE stuff needs to be extracted out and made
into "hooks".
23. The raidframe-netbsd32 compat code needs to be better separated
from the main raidframe module. The current mechanism requires us
to include compat/netbsd32/netbsd32.h in rf_netbsdkintf.c to get
various structure definitions. This should all be handled in the
compat module, but requires that the code in the ioctl switch be
moved into a function so the compat code can call it directly and
handle the ioctl commands entirely.