On Sat, Apr 6, 2013 at 9:16 AM, Mark Cave-Ayland mark.cave-ayland@ilande.co.uk wrote:
On 05/04/13 21:08, Tarl Neustaedter wrote:
On 2013-Apr-5 15:40 , Artyom Tarasenko wrote:
[...] OFMEM: ofmem_map_page_range ffc7d000 -> 006f7d000 00001000 mode 000000bc OFMEM: ofmem_claim phys=ffffffffffffffff size=00014000 align=00020000 OFMEM: ofmem_claim_virt virt=00000000 size=00014000 align=00020000 OFMEM: ofmem_map_page_range ffc60000 -> 006f60000 00014000 mode 000000bc
^^^^ this one
Wherever that's coming from, that looks like an allocator doing descending addresses.
OFMEM: ofmem_map_page_range ffc5c000 -> 006f78000 00001000 mode 000000bc OFMEM: ofmem_claim phys=ffffffffffffffff size=00002000 align=00002000 OFMEM: ofmem_claim_virt virt=ffc5d000 size=00002000 align=00000000 OFMEM: ofmem_map_page_range ffc5d000 -> 006f5e000 00002000 mode 000000bc ...OFMEM: ofmem_claim phys=ffffffffffffffff size=00002000 align=00002000 OFMEM: ofmem_claim_virt virt=ffc5f000 size=00002000 align=00000000 OFMEM: Non-free virtual memory claimed!
^^^^ overlaps with this one.
This allocator, whatever it is has been doing ascending addresses. This actually looks like you simply ran out of room.
Ah now I see. It looks as if kadb assumes that it has free use of the region from 0xffc00000 but the OFMEM allocator starts auto-allocating memory from beneath the internal heap which is below the ROM image at 0xffd00000. As Tarl rightly points out, as the allocations grow then kadb tries to grab an area of memory we have already claimed which causes the overlap error.
Does the attached patch against SVN trunk enable you to get into kadb? It's the same as my last patch except for the following addition:
--- a/openbios-devel/arch/sparc32/ofmem_sparc32.c +++ b/openbios-devel/arch/sparc32/ofmem_sparc32.c @@ -61,7 +61,7 @@ ucell ofmem_arch_get_heap_top(void)
ucell ofmem_arch_get_virt_top(void) {
return (ucell)TOP_OF_RAM;
return (ucell)0xff000000;}
So when we are auto-allocating from OFMEM, rather than starting from beneath the internal heap, we simply start auto-allocating from 0xff000000 downwards which should leave the 0xffc00000 region being used by kadb well alone. A quick test with the patch shows that my Solaris 8 image can still boot both normally and into kadb.
Yes! In kadb I get!
FMEM: ofmem_map_page_range ffcb0000 -> 006ea0000 00008000 mode 000000bc .vmunix loaded - 0x7c000 bytes used
Unfortunately it's not as helpful as it became under Solaris 2.6+ :
kadb> :c OFMEM: ofmem_claim phys=ffffffffffffffff size=00001000 align=00001000 OFMEM: ofmem_claim_virt virt=00000000 size=00001000 align=00001000 OFMEM: ofmem_map_page_range fef65000 -> 006f77000 00001000 mode 000000bc Begin traceback... sp = f0186b20 Called from ffc017cc, fp=f0186b80, args=9 f013f870 f013f874 0 200 0 Called from f013f83c, fp=f0186c08, args=f01e4000 1 f01e2bf0 fd020000 f0186d88 f01e2ff4 Called from f0140a38, fp=f0186c68, args=f01e563c 1cc5 f01e5652 ff000008 ff000008 4400fe1 Called from f0054f44, fp=f0186cc8, args=f0186d8c 22 ffffffff a f0186d8c 22 Called from f0054724, fp=f0186d28, args=f0186d8c 22 5 0 ffffffff 0 Called from f0054494, fp=f0186e18, args=f01e2173 f0186ec0 5 0 f0186e10 0 Called from f0132e7c, fp=f0186e78, args=f01e2173 0 0 0 f0186e98 f0186ec0 Called from f01375e0, fp=f0186ef8, args=f01e43c2 ffd2e31c f01e28f4 0 f01e2b1c ffffffff Called from f0005250, fp=f0186f58, args=400 f01aa000 f9fbe000 0 200 f0178570 Called from ffc11c94, fp=0, args=4000 3ffb78 0 ffc37570 ffc259d0 1 End traceback... fault and calling cmd: trap 9 sp f0186b20 pc f013f870 npc f013f874 stopped at 0: unimp 0x0 kadb> bt symbol not found kadb>
But still it's definitely a progress!
-- Regards, Artyom Tarasenko
linux/sparc and solaris/sparc under qemu blog: http://tyom.blogspot.com/search/label/qemu