Now that someone has implementented resource alignment in the resource
allocator, someone could review and integrate the attached patch.
Background:
I do have an old system with several PnP devices. Two of the request the
following IO ports:
first device: port range 0x100-0x3ff size=1 align=1
second device: port range 0x100-0x3f0 size=8 align=8
The first device gets port 0x100-0x100 allocated. Then the code
in isa_common.c tries to allocate the ports for the second device.
0x100 is already used, so it gets the next free range: 0x101-0x108,
ignoring the alignment constraints.
The general problem in the code /sys/isa_common.c
isa_find_port(), isa_find_memory(), etc.
The loops in these routines try to honor the alignment constraints but
the real work is done in /sys/subr_rman.c. Regardless of resource usage
the for(...)-look in above functions is only run once.
I already filed a PR for this problem but my first solution was a real
hack (kern/21461).
[another solution would be to introduce another flag for
rman_reserve_resource() not to search for alternate regions.
Daniel
Index: isa/isa_common.c
===================================================================
RCS file: /data/cvs/src/sys/isa/isa_common.c,v
retrieving revision 1.18
diff -u -r1.18 isa_common.c
--- isa/isa_common.c 2000/07/12 00:42:08 1.18
+++ isa/isa_common.c 2000/11/09 20:11:31
@@ -207,10 +207,10 @@
start, size);
res[i] = bus_alloc_resource(child,
SYS_RES_IOPORT, &i,
- 0, ~0, 1, 0 /* !RF_ACTIVE */);
+ 0, ~0, 1,
+rman_make_alignment_flags(align)/* !RF_ACTIVE */);
if (res[i]) {
- result->ic_port[i].ir_start = start;
- result->ic_port[i].ir_end = start + size - 1;
+ result->ic_port[i].ir_start = res[i]->r_start;
+ result->ic_port[i].ir_end = res[i]->r_start + size - 1;
result->ic_port[i].ir_size = size;
result->ic_port[i].ir_align = align;
break;
