Opened 15 years ago
Closed 9 years ago
#1 closed task (fixed)
Port VBox driver
| Reported by: | diver | Owned by: | Valery V. Sedletski |
|---|---|---|---|
| Priority: | critical | Milestone: | VBox driver |
| Component: | Driver | Keywords: | |
| Cc: |
Description
Attachments (33)
Change History (46)
comment:1 by , 15 years ago
| Priority: | major → critical |
|---|---|
| Type: | defect → task |
comment:2 by , 9 years ago
| Milestone: | VBox 3.1 → VBox diver |
|---|
comment:3 by , 9 years ago
| Summary: | Port VBox 3.1 driver → Port VBox driver |
|---|
comment:4 by , 9 years ago
| Owner: | set to |
|---|---|
| Status: | new → assigned |
comment:6 by , 9 years ago
The driver was already built successfully a month ago, and doesn't trap in usual cases. But there are some unimplemented functions in R0 Runtime part. At least, the driver is successfully opened, accepts IOCtl commands and about 75% of tests are working (the Runtime and Host Driver testcases). Also, there are some testcases which hang the host machine (need to recall)
comment:7 by , 9 years ago
Uploaded the patches for the Support driver, Runtime, VMM and other parts needed for minimal build, without XPCOM and GUI's. Note that these patches are for gcc-v3.3.5-csd6 at the moment, and need 'configure' and cfg.sh attached to ticket #10
by , 9 years ago
| Attachment: | Config.kmk.diff added |
|---|
by , 9 years ago
| Attachment: | bldprogs.2.diff added |
|---|
comment:8 by , 9 years ago
Here I attached all my latest patches to VBox 5.0.51 sources (applies to svn revision 58105). These are the set of this ticket attachments from additions.2.diff to AutoConfig.2.kmk (20 total). These patches may be applied in any order, because they are intended for independent pieces of file system. Still missing are libvncserver (for VNC VRDP extpack) and libvpx (needed for capture VM video feature). We'll need to make rpm packages for them in the near future. I attached the corresponding version of AutoConfig.kmk, which you could use as an example of what should you get with 'configure' (or, just change the paths in it, to be used without 'configure'). Also note that these patches should be used with latest gcc 4.9.2 from Netlabs RPM repo. I deleted some workarounds for using with gcc3.3.5-csd6. (but I still have them on my hard disk, I could share them if needed).
Note also that other platforms use YASM instead of NASM, and we haven't it. So, there are much warnings. Also, I had to %ifdef some align-related Assert's in VMM sources and some assembler includes.
Also, I changed some network-related code to use sockets instead of pipes in select/poll as we can use sockets only here.
Also, I added "#ifdef OS2_PM_EVENT_QUEUES" in XPCOM code to switch between PM queue-based code and local sockets-based one (aka "unix domain sockets" which OS/2 supports natively, via afos2.sys driver).
The Qt frontend also builds OK here, but some code is still missing, or should be refactored, like hotkeys/keyboard and mouse capture/input events. So, there are lags/PM queue delays in Qt frontend. So, use VBoxHeadless or VBoxSDL for now, they work fine.
comment:9 by , 9 years ago
Hi valerius, I would like to see you're "workarounds" for using with gcc3.3.5-csd6. If it is convienient and you have the time available could you possibly e-mail them to me at rklipp(at)megared.net.mx?
thanks
comment:10 by , 9 years ago
First two patches are applied in r8 and r9 :) @valerius, please check them to see how they my original commits differ from your diff files (you should notice it when merging the SVN with your local copy after svn up though). This is important so that you notice the coding conventions and other tricks and could commit on your own in the future. These two patches are simple of course, not much to check there but further patches are more complex so please check the real commits with attention.
comment:11 by , 9 years ago
Added two kbuild tickets, http://trac.netlabs.org/kbuild/ticket/125 and http://trac.netlabs.org/kbuild/ticket/126, they contain some small kbuild macros patches needed to successfully compile VBox under OS/2.
PS: rklipp, sorry for not answering you so long time :( These old patches for old gcc 3.3.5 are not cleaned up from unneeded code, and they are for VBox 5.0.0. Now our svn is being committed for version 5.0.6 already. And BTW, you'd better use the latest version with latest patches (and these macros patches, you'll need them too). And, the newer VBox versions require new gcc for using .rsp linker scripts (because linker command line already reaches some limit and you'll get linker errors with cut-down command lines). Do you still need these gcc3 patches?
comment:12 by , 9 years ago
Hi valerius, Thanks for the update info, and no, I no longer need the gcc3 patches. Your "Guru Mediation" fix was awesome!
comment:13 by , 9 years ago
| Resolution: | → fixed |
|---|---|
| Status: | assigned → closed |
Most of the above patches were applied in a number of revisions up to (and including) r55. Things left out either turned out not to be necessary or will be committed in further commits (and within separate tickets) after clarifying the details. Also omitted are Qt4 patches (for now). They will be committed within #3.
Now everything is successfully built out of the box (in the RPM environment, of course). I will update build instructions in Wiki as there are more packages to install.
Although VBoxDrv.sys successfully builds now, it fails in the open call with the current set of patches. Valerius can make it work with his local hacks (that cause some RTMp methods to pretend they are running on a single CPU machine) and is currently trying to find why whe generic RTMp implementation (used on other platforms) doesn't work for us. This will be finished within #32.
I'm closing this ticket as the main task is done (and there is also too much irrelevant information). New tickets will be created for the bits that are still to be done for VBoxDrv.
Milestone renamed