Bug #2875
closedPlain models don't work with 3.4 Linux kernel
0%
Description
For all drivers RCV prints in its log:
... include/linux/rcupdate.h:432: Error: Cannot resolve variable rcu_lock_map. include/linux/rcupdate.h:432: Error: Expected lval for ADDROF. Got Const(Int64(0,int,None)) include/linux/rcupdate.h:917: Warning: CALL in constant include/linux/rcupdate.h:917: Error: Length of array is negative include/linux/rcupdate.h:920: Warning: CALL in constant include/linux/rcupdate.h:920: Error: doPureExp: not pure ...
The first issue seems to be connected with incorrect manual kernel configuring. We unset CONFIG_DEBUG_LOCK_ALLOC and CONFIG_PROVE_LOCKING, but don't touch CONFIG_PROVE_RCU, that in fact depends on CONFIG_PROVE_LOCKING. So, I guess that solution is to turn off CONFIG_PROVE_RCU.
The second issue comes from:
BUILD_BUG_ON(!__builtin_constant_p(offset));
that becomes after preprocessing:
((void)sizeof(char[1 - 2*!!(!__builtin_constant_p(offset))]));
That also is a problem for aspect models:
In file included from include/linux/rculist.h:10:0, from include/linux/dcache.h:6, from include/linux/fs.h:387, from include/linux/tty.h:39, from /ldv/launch/work/current--X--drivers/usb/serial/kobil_sct.ko--X--defaultlinux--X--100_1a/linux/csd_deg_dscv/11/dscv_tempdir/dscv/ri/100_1a/drivers/usb/serial/kobil_sct.c.p:2: include/linux/rcupdate.h: In function '__kfree_rcu': include/linux/rcupdate.h:917:2: error: size of unnamed array is negative
Looks like 3.4 kernel isn't intended for build... __kfree_rcu function body is unconditional. May be just configure BUILD_BUG_ON in some way.
Updated by Evgeny Novikov almost 12 years ago
Evgeny Novikov wrote:
The second issue comes from:
[...]
that becomes after preprocessing:
[...]
That also is a problem for aspect models:
[...]
Looks like 3.4 kernel isn't intended for build... __kfree_rcu function body is unconditional. May be just configure BUILD_BUG_ON in some way.
In according with the issue may be related with turning on/off gcc optimizations (-Ox, -O2, -Os, etc.). Some investigation should be done.
Updated by Evgeny Novikov almost 12 years ago
The second issue is due to BUILD_BUG_ON define was moved from include/linux/kernel.h to include/linux/bug.h. But it's strange nevertheless that we need to remove this definition. It seems that the real problem is in our Makefile patches. And it should be investigated separately.
Updated by Evgeny Novikov almost 12 years ago
- Status changed from Open to Resolved
- Published in build set to 0933ee3
0933ee3 fixed both issues, but I don't like both fixes. See #2875 that continues investigation for the second issue. As for the first issue, I hope that we'll give up plain models and all related workarounds soon.
Updated by Evgeny Novikov almost 12 years ago
Updated by Evgeny Novikov almost 12 years ago
- Status changed from Resolved to Closed