Sorry about the crosspost but I estimate that this reaches those who need
to see this..
There was a change in the 4.x kernel.h on June 15 that broke backwards
compatibility for binary distributed driver files
(distributed as .o files) It was an MFC of a patch by peter..
but we didn't understand the ramifications in 4.x.
my fix (a reversion in part) is as follows:
Index: kernel.h
===================================================================
RCS file: /repos/cvs/mod/freebsd/src/sys/sys/kernel.h,v
retrieving revision 1.63.2.5
diff -u -r1.63.2.5 kernel.h
--- kernel.h 2001/07/26 23:27:53 1.63.2.5
+++ kernel.h 2001/10/11 21:30:03
@@ -113,13 +113,13 @@
SI_SUB_VM = 0x1000000, /* virtual memory system
init*/
SI_SUB_KMEM = 0x1800000, /* kernel memory*/
SI_SUB_KVM_RSRC = 0x1A00000, /* kvm operational
limits*/
- SI_SUB_CPU = 0x2000000, /* CPU resource(s)*/
- SI_SUB_KLD = 0x2100000, /* KLD and module setup */
- SI_SUB_INTRINSIC = 0x2200000, /* proc 0*/
- SI_SUB_VM_CONF = 0x2300000, /* config VM, set limits*/
- SI_SUB_RUN_QUEUE = 0x2400000, /* the run queue*/
- SI_SUB_CREATE_INIT = 0x2500000, /* create the init process
*/
- SI_SUB_DRIVERS = 0x3100000, /* Let Drivers initialize
*/
+ SI_SUB_CPU = 0x1e00000, /* CPU resource(s)*/
+ SI_SUB_KLD = 0x1f00000, /* KLD and module setup */
+ SI_SUB_INTRINSIC = 0x2000000, /* proc 0*/
+ SI_SUB_VM_CONF = 0x2100000, /* config VM, set limits*/
+ SI_SUB_RUN_QUEUE = 0x2200000, /* the run queue*/
+ SI_SUB_CREATE_INIT = 0x2300000, /* create the init process
*/
+ SI_SUB_DRIVERS = 0x2400000, /* Let Drivers initialize
*/
SI_SUB_CONFIGURE = 0x3800000, /* Configure devices */
SI_SUB_VFS = 0x4000000, /* virtual file system*/
SI_SUB_CLOCKS = 0x4800000, /* real time and stat
clocks*/
The result of this was that old .o drivers are initialise at sequence
0x2400000 but the device framework is not initialised until 0x3100000.
my patch shifts all the numbers back to below that that binary
,o files would be compiled with, thus ensuring that
the world is ready for them when they initialise.
The hardest hit would be any drivers using the PCI compatibility shims
which core-dump on a null reference because the pci compatibility
shim devclass pointer is not yet initialised..
I'd like to check this in to 4.4++ as itwill allow anyone else with binary
.o drivers (No idea if there are any others) to use their
dirvers without having to go back to their suppliers and pay "lots-o-cash"
to get them to recompile it. (If they still have the source).
Julian
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-hackers" in the body of the message
