Repository navigation
v0.1.16
A small patch release that settles the G2 question about holes and fixes the daily watch job.
Changes:
- The cpus smoke check is excused for a kernel whose
.configturns SMP off, like tinyconfig on x86, since it only ever sees one of the two CPUs. (#196) - The frontier search runs every point column a signature pins by exact release, such as GCC 14.2, inside the working set it finds. The last points step over those releases, and a failure there is a known hole, so it does not send the row dense. (#198)
gk pinsreads the GCC and binutils listings from mirrors.kernel.org when ftp.gnu.org is down, which is what made the 6 October watch run fail. (#199)
Findings:
- Open question 1 is settled. Over the last-point columns, 1 of 56 dense rows has a real hole, 2.6.31 on x86_64 with GCC 3.4.6, so tier 1 keeps the frontier search. With the point columns it is 5 of 61, and four of those are GCC 14.2.0 and objtool. (#198)
- 2.6.9 and 2.6.10 declare an array of an incomplete type in fb.h, which GCC 4 refuses. On 2.6.31 to 2.6.37, GCC 3.4 drops the
_THIS_IP_label of the inline kmalloc in con_init, fixed by 4a92379bdfb4 in v2.6.38-rc1. The 2.6.33 MCE severity table needs a newer binutils for its bignums. (#197)