On 05/03/2026 07:18, Samuel Thibault wrote:
This time it was math.h:
00000000 00 00 02 ea 01 00 00 00 01 00 00 00 d4 b2 d2 0c |................|
00000010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00001000 00 00 02 ea 01 00 00 00 01 00 00 00 97 cd be 0c |................|
00001010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00002000 20 46 50 5f 49 4e 54 5f 54 4f 4e 45 41 52 45 53 | FP_INT_TONEARES|
00002010 54 20 34 0a 20 20 20 20 20 20 46 50 5f 49 4e 54 |T 4. FP_INT|
00002020 5f 54 4f 4e 45 41 52 45 53 54 2c 0a 20 20 7d 3b |_TONEAREST,. };|
00002030 0a 23 65 6e 64 69 66 0a 0a 23 69 6e 63 6c 75 64 |.#endif..#includ|
Only the two first pages were overwritten, still the same pattern.
Some people have all the luck.
I've been unable to make any progress. I identified a sequence of
packages that boralus had built from boot to first definite corruption.
I tried to build the same sequence locally but ran into problems:
1) My system was crashing with the 'out of memory' panic. It turned out
that the cause of this was my use of a file to provide the storage space
for an ext2 file system for the chroot. The mach-defpager mallocs quite
large chunks of memory whilst syncing a 10G file system used in this way
and soon runs out. I ditched that strategy and just use a directory
within the / file system now.
How is the chroot space for boralus managed?
2) The package versions of those in my sample are moving on and failing
build dependencies. I can remove the odd one but that defeats the point
of reproducing the build sequence that boralus used. Is it realistic for
me to somehow setup a repository that has copies of packages that I can
rely on?
3) I had tried running the build script without a mach-defpager running
at all (like boralus) but my system soon locks awaiting pages. I'm
puzzled how boralus manages to run for longer periods of time.
Still I continue getting nowhere...
Regards,
Mike.
