[CMAKE] Fix system boot crash when compiled as NT6+ #6869
Conversation
There was a problem hiding this comment.
What is the problem here? That 0x600 dlls need more space than 0x502 ones?
Shouldn't we use a separate set of files for each version?
If not, I think we should had a reminder comment, in the very files in git, that, from now on, they must be regenerated with 0x600 dlls only, in order to support both versions.
And I hope 0x600 builds do not miss any 0x502 filenames...
| @@ -1,121 +1,121 @@ | |||
| # Generated from C:\Users\katahiromz\reactos-1\output-MinGW-i386 | |||
| # Generated from ../../output-MinGW-i386 | |||
There was a problem hiding this comment.
I suggest to remove this line from git repository: updating it each time someone uses a different path is only noise.
There was a problem hiding this comment.
I suggest to preserve the directory name output-MinGW-i386 which tells the build configuration.
Currently, yes, 0x600 needs more space. |
Perhaps indeed it would be of interest to use new "baseaddress[_msvc]_0600.cmake" files for such builds, until at least we are able to auto-generate these at build time. |
binarymaster
added
bugfix
binarymaster
changed the title
There was a problem hiding this comment.
Please restore "baseaddress_custom" which you deleted.
Please restore "baseaddress_notifyhook" which you deleted.
and so on...
Here is the output of GCC dbg i386 0x502 buildbot after your changes:
Proving that you can use the github actions to verify the correctness of your changes.
RatinCN
force-pushed
the
Update_Cmake_Base_Address
branch
from
4666163
to
170d076
Compare
|
Thanks, it improved. But still 5 (instead of formerly 9) are missing: If also adding |
RatinCN
force-pushed
the
Update_Cmake_Base_Address
branch
2 times, most recently
from
456bb5d
to
230d962
Compare
|
RatinCN
force-pushed
the
Update_Cmake_Base_Address
branch
from
230d962
to
dc9f076
Compare
|
Build failure seems unrelated: Try again... |
Fix system boot crash with `STATUS_ILLEGAL_DLL_RELOCATION` when compiled with `-DDLL_EXPORT_VERSION=0x600`
RatinCN
force-pushed
the
Update_Cmake_Base_Address
branch
from
dc9f076
to
fd2c186
Compare
There was a problem hiding this comment.
We are almost there: latest gcc x86 dbgNT502
I am not sure why it does still complain about those 2 since they are now both mentioned in the GCC baseaddress.cmake.
And in releases/0.4.7 to releases/0.4.14 that works and is enough, and leaves us with zero warnings in this context.
In the old branches I had a final empty-line in the file.... Not sure whether that helps to cut off one more... ???
That's a pic from releases/0.4.11 for example, see the blinking cursor...
But I am now giving my approval already, because the status after your PR is now already definitely better than the status of master head.
For MSVC dbg, the current state warns about a few more modules still, but I don't remember what the status quo on master head is currently with that, because I cannot build with that myself anymore since the VS2010-dropping:
EDIT: Ahh now I know why it still warns about notifyhook and vfd for GCC: because those 2 are only mentioned in MSVC, and not in the GCC file. I missed that detail, when I quickly searched for them in the changed files.
So yes, theoretically those could still be fixed as well.
Purpose
Fix system boot crashed with
STATUS_ILLEGAL_DLL_RELOCATION (0xC0000269)when compiled with-DDLL_EXPORT_VERSION=0x600.JIRA issue: None
Proposed changes
Run
gen_baseaddress.pyto update the baseaddress. Test DLLs are not involved, lemme know if-DENABLE_ROSTESTS=1is needed.Before:
After:
In Debug mode, another error logged:
which is not point the problem out, thanks DarkFire01 who give me hint to find the problem around base address, all are work fine now.