yeah, there's a reason it is fffffe, and not ffff8, but memory fails me. See this for details of the address hole: https://en.wikipedia.org/wiki/X86-64
Charles may remember, but it goes back to the port they did in 2005. On Tue, Jan 28, 2025 at 5:08 PM Ron Minnich <rminn...@p9f.org> wrote: > btw, if you > acid 9pc64 > you can paste this right into acid > src(0xfffffffff011cdee); // dumpstack+0x10 > src(0xfffffffff013d50f); // panic+0x133 > src(0xfffffffff0116a3b); // KADDR+0x55 > src(0xfffffffff012fe55); // sigsearch+0xc8 > src(0xfffffffff012fec9); // mpsinit+0x14 > src(0xfffffffff011622a); // main+0x30b > src(0xfffffffff0110204); // ndnr > and see the source. > > Also, the ndnr is a jmk-ism: it means "no deposit, no return" > > so, let's see, I can't tell if we went over this before. > What is KADDR and KADDR2? They relate to TMFM, another jmk-ism I > believe: Too Many F-ing Megabytes, where too many is "more than 2G" -- > why 2G? well ... > > basically, amd64, like lots of things (risc-v) uses this one simple > trick: if you sign-extend a 32-bit pointer, you get something anchored > either at the top 2G (kernel va) or the low 2G (user code). > > i.e. 0x80000000 -> 0xffffffff_80000000 -> this is convenient. You can > use 32-bit pointers for lots of things, and, since the amd64 is a > pretty half-way 64-bit CPU (lots of 64-bit instructions only > completely work with RAX), this is helpful. > > And it works great until you get CPUs with TMFM. Then you need to > split memory up: > physical 0 -> 2 Gb becomes virtual 0xfffffff_80000000 > physical 2Gb and up becomes ... fffffe0000000000 > Why fffffe0000000000? the first amd64 only had something like 41(?) > bits of virtual address:there's this giant hole in the middle,and > kernel virtual HAD to start at that address -- 64 bits - whatever gets > you to 23 bits. [I can't find the actual documents on this, I am out > of time to look, so you'll need to fill in my likely errors here] > It's a hardware mandate from opteron land. > > OK, so KADDR2 is fffffe0000000000, and that error is saying code > called KADDR2 with something that's not in KADDR2. That va > fffffffffffffc00 is in the low 2GiB physical, which is KADDR. > > This is a side effect of the real problem: you don't have the table it > wants. So you need to fix that, OR, start using qemu for your testing. > > I hope I did not mess the details up too much here.... > > On Tue, Jan 28, 2025 at 1:24 PM ron minnich <rminn...@gmail.com> wrote: > > > > I'd be happier to remove the mps dependency actually. the mps is long > dead. But that's a bigger story. > > > > > > On Tue, Jan 28, 2025 at 11:24 AM Paul Lalonde <paul.a.lalo...@gmail.com> > wrote: > >> > >> Ah, that's the code path that sent me to QEMU. > >> Vmx doesn't have any MP tables, which leads to this fault in mpsinit. > >> Ron provided this minimal one for me, which I think we could learn from > to adapt into vmx. The hacky version of pointing the code directly at > something like this baked in didn't excite me. > >> > >> 50 43 4D 50 ; "PCMP" > >> 00 00 ; Table Length (placeholder) > >> 04 ; Spec Revision > >> 00 ; Checksum (placeholder) > >> 42 4F 43 48 53 43 50 55 ; "BOCHSCPU" > >> 30 2E 31 20 20 20 20 20 20 20 20 ; "0.1 " > >> 00 00 00 00 ; OEM Table Pointer > >> 00 00 ; OEM Table Size > >> 14 00 ; Entry Count (2 CPUs + 18 = 20, > little-endian) > >> 00 00 E0 FE ; Local APIC Address (0xfee00000) > >> 00 00 ; Ext Table Length > >> 00 ; Ext Table Checksum > >> 00 ; Reserved > >> > >> On Tue, Jan 28, 2025 at 11:15 AM <tlaro...@kergis.com> wrote: > >>> > >>> On Tue, Jan 28, 2025 at 09:18:29AM -0800, Paul Lalonde wrote: > >>> > ktrace can generate a stack for you from that dump. The line > starting with > >>> > "ktrace" is the command line (you might change 9k8cpu to the path to > the > >>> > kernel file in you're not in the directory where you built it). > >>> > Then the following lines up to but not including the "cpu0: exiting" > can be > >>> > dropped into ktrace's stdin to have it generate a stack trace. > You'll need > >>> > to add the ^d at the end if you're cut-and-pasting. > >>> > > >>> > Though it looks like it's just triggering the page fault trap on that > >>> > 0xfffffffffffffc00 address, which itself looks like a victim of > >>> > sign-extension. So back up to the fault and find the source of that > >>> > address? > >>> > >>> Yes: > >>> > >>> src(0xfffffffff011cdee); // dumpstack+0x10 > >>> src(0xfffffffff013d50f); // panic+0x133 > >>> src(0xfffffffff0116a3b); // KADDR+0x55 > >>> src(0xfffffffff012fe55); // sigsearch+0xc8 > >>> src(0xfffffffff012fec9); // mpsinit+0x14 > >>> src(0xfffffffff011622a); // main+0x30b > >>> src(0xfffffffff0110204); // ndnr > >>> > >>> this doesn't tell me much more than what I knew already: it panics in > >>> mpsinit, calling KADDR in map.c. > >>> > >>> During my next wandering under Nix, I will try to track back from > >>> where the offending address is taken or with what it is constructed. > >>> > >>> > > >>> > On Tue, Jan 28, 2025 at 9:09?AM <tlaro...@kergis.com> wrote: > >>> > > >>> > > On Tue, Jan 28, 2025 at 07:49:02AM -0800, Paul Lalonde wrote: > >>> > > > Do you have a stack for the assert, from the ktrace? > >>> > > > > >>> > > > >>> > > Yes, and I was wrong: it fails relatively "late" in main.c: at > >>> > > mpsinit. > >>> > > > >>> > > Here is the info (I added a bunch of print() before each function > call > >>> > > to know where it stumbled upon an incorrect address): > >>> > > > >>> > > term% nix/test_vmx > >>> > > > >>> > > NIX > >>> > > mmunit...mmuinit: vmstart 0xfffffffff0000000 vmunused > 0xfffffffff023d000 > >>> > > vmunmapped 0xfffffffff0400000 vmend 0xfffffffff4000000 > >>> > > sys->pd 0x108003 0x108023 > >>> > > cpu0: mmu l3 pte 0xfffffffff0106ff8 = 107023 > >>> > > cpu0: mmu l2 pte 0xfffffffff0107ff8 = 108023 > >>> > > cpu0: mmu l1 pte 0xfffffffff0108c00 = e3 > >>> > > cpu0: mmu l1 pte 0xfffffffff0108c00 = e3 > >>> > > ioinit... multibootmemassert... kbdinit... meminit...asm: addr > >>> > > 0x0000000004000000 end 0x0000000004000000 type 1 size 0 > >>> > > cm 0: addr 0x4000000 npage 0 > >>> > > 0 0 0 > >>> > > npage 0 upage 0 kpage 16384 > >>> > > confinit... archinit... mallocinit...base 0xfffffffff023d000 ptr > >>> > > 0xfffffffff023d000 nunits 4047617 > >>> > > acpiinit... umeminit... trapinit... printinit... i8259init... > procinit... > >>> > > mpsinit...panic: cpu0: map.c:KADDR() passed addr fffffffffffffc00 > >>> > > fffffe0000000000 > >>> > > panic: cpu0: map.c:KADDR() passed addr fffffffffffffc00 >= > fffffe0000000000 > >>> > > > >>> > > dumpstack > >>> > > ktrace 9k8cpu 0xfffffffff011cdee 0xfffffffff0105d58 > >>> > > estackx 0xfffffffff0106000 > >>> > > 0xfffffffff0105c70=0xfffffffff0105da8 > >>> > > 0xfffffffff0105c78=0xfffffffff011cb91 > >>> > > 0xfffffffff0105c80=0xfffffffff0105c98 > >>> > > 0xfffffffff0105c98=0xfffffffff013cff7 > >>> > > 0xfffffffff0105cb0=0xfffffffff0105cd0 > >>> > > 0xfffffffff0105cc0=0xfffffffff0105ea7 > >>> > > 0xfffffffff0105cc8=0xfffffffff0105df3 > >>> > > 0xfffffffff0105ce0=0xfffffffff013d14d > >>> > > 0xfffffffff0105d08=0xfffffffff0105d90 > >>> > > 0xfffffffff0105d28=0xfffffffff011cdee > >>> > > 0xfffffffff0105d30=0xfffffffff0105da8 > >>> > > 0xfffffffff0105d40=0xfffffffff0105d58 > >>> > > 0xfffffffff0105d48=0xfffffffff0105da8 > >>> > > 0xfffffffff0105d50=0xfffffffff011cdee > >>> > > 0xfffffffff0105d58=0xfffffffff011cb99 > >>> > > 0xfffffffff0105d68=0xfffffffff013d50f > >>> > > 0xfffffffff0105d88=0xfffffffff0105ed0 > >>> > > 0xfffffffff0105d90=0xfffffffff013cff7 > >>> > > 0xfffffffff0105d98=0xfffffffff0105db5 > >>> > > 0xfffffffff0105e08=0xfffffffff013d1b8 > >>> > > 0xfffffffff0105e10=0xfffffffff0105e00 > >>> > > 0xfffffffff0105e20=0xfffffffff0105ea3 > >>> > > 0xfffffffff0105e28=0xfffffffff0105e98 > >>> > > 0xfffffffff0105e38=0xfffffffff013d1b8 > >>> > > 0xfffffffff0105e40=0xfffffffff0105e98 > >>> > > 0xfffffffff0105e60=0xfffffffff013d217 > >>> > > 0xfffffffff0105e68=0xfffffffff015d9c9 > >>> > > 0xfffffffff0105e80=0xfffffffff0105fb8 > >>> > > 0xfffffffff0105e90=0xfffffffff015d5d9 > >>> > > 0xfffffffff0105ea8=0xfffffffff0105ed0 > >>> > > 0xfffffffff0105ec0=0xfffffffff0116a3b > >>> > > 0xfffffffff0105ef8=0xfffffffff012fe55 > >>> > > 0xfffffffff0105f08=0xfffffffff01a1afa > >>> > > 0xfffffffff0105f10=0x0000000000000004 > >>> > > 0xfffffffff0105f18=0x0000000000000046 > >>> > > 0xfffffffff0105f20=0xfffffffff00fffd9 > >>> > > 0xfffffffff0105f28=0x0000000000000006 > >>> > > 0xfffffffff0105f30=0xfffffffff015d5d9 > >>> > > 0xfffffffff0105f38=0xfffffffff0000400 > >>> > > 0xfffffffff0105f40=0x0000000000000000 > >>> > > 0xfffffffff0105f48=0xfffffffff012fec9 > >>> > > 0xfffffffff0105f50=0xfffffffff01a1aff > >>> > > 0xfffffffff0105f58=0x0000000000000208 > >>> > > 0xfffffffff0105f60=0x0000000000000124 > >>> > > 0xfffffffff0105f68=0xfffffffff01149d0 > >>> > > 0xfffffffff0105f70=0x0000000000000006 > >>> > > 0xfffffffff0105f78=0xfffffffff0114ba7 > >>> > > 0xfffffffff0105f80=0xfffffffff0227510 > >>> > > 0xfffffffff0105f88=0xffffffff00000000 > >>> > > 0xfffffffff0105f90=0x0000000000000000 > >>> > > 0xfffffffff0105f98=0xfffffffff0105fb8 > >>> > > 0xfffffffff0105fa0=0x0000000bf0116b0d > >>> > > 0xfffffffff0105fa8=0xfffffffff011622a > >>> > > 0xfffffffff0105fb0=0xffffffff00000400 > >>> > > 0xfffffffff0105fb8=0xffffffff00000000 > >>> > > 0xfffffffff0105fc0=0x0000000000000000 > >>> > > 0xfffffffff0105fc8=0x0000000000000000 > >>> > > 0xfffffffff0105fd0=0x0000000000000000 > >>> > > 0xfffffffff0105fd8=0x0000000000000000 > >>> > > 0xfffffffff0105fe0=0x0000000000000000 > >>> > > 0xfffffffff0105fe8=0xfffffffff0110204 > >>> > > 0xfffffffff0105ff0=0x000000002badb002 > >>> > > 0xfffffffff0105ff8=0x000000000023b000 > >>> > > cpu0: exiting > >>> > > > >>> > > > > >>> > > > > >>> > > > On Tue, Jan 28, 2025 at 6:09?AM <tlaro...@kergis.com> wrote: > >>> > > > > >>> > > > > After fixing problems leading to compiler warnings---legitimate > >>> > > > > warnings, but even the too short binary negated unsigned > 32bits values > >>> > > > > promoted to 64 bits with leading bits hence 0 as mask were > harmless--- > >>> > > > > now I want to look at the stumbing block. > >>> > > > > > >>> > > > > For me, under vmx, this is the assert in map.c:17: > >>> > > > > > >>> > > > > assert(pa < KSEG2); > >>> > > > > > >>> > > > > that triggers, and it should come from a call from multiboot. > >>> > > > > > >>> > > > > My first reflex is to start adding printf() instructions to > track the > >>> > > > > problem, but is there a better way when dealing with the > kernel? > >>> > > > > > >>> > > > > Second question: since, if I'm not mistaken, 9front doesn't use > >>> > > > > multiboot, is vmx usable (i.e. agnostic about) with the > multiboot > >>> > > stuff? > >>> > > > > The embedded boot stuff should handle the thing by itself > without load > >>> > > > > addresses having to be adjusted because of vmx? > >>> > > > > -- > >>> > > > > Thierry Laronde <tlaronde +AT+ kergis +dot+ com> > >>> > > > > http://www.kergis.com/ > >>> > > > > http://kertex.kergis.com/ > >>> > > > > Key fingerprint = 0FF7 E906 FBAF FE95 FD89 250D 52B1 AE95 > 6006 F40C > >>> > > > >>> > > -- > >>> > > Thierry Laronde <tlaronde +AT+ kergis +dot+ com> > >>> > > http://www.kergis.com/ > >>> > > http://kertex.kergis.com/ > >>> > > Key fingerprint = 0FF7 E906 FBAF FE95 FD89 250D 52B1 AE95 6006 > F40C > >>> > >>> -- > >>> Thierry Laronde <tlaronde +AT+ kergis +dot+ com> > >>> http://www.kergis.com/ > >>> http://kertex.kergis.com/ > >>> Key fingerprint = 0FF7 E906 FBAF FE95 FD89 250D 52B1 AE95 6006 F40C > > > > 9fans / 9fans / see discussions + participants + delivery options > Permalink ------------------------------------------ 9fans: 9fans Permalink: https://9fans.topicbox.com/groups/9fans/T8b5b89fcf829819e-Meb14903b2db62f531083c7ae Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
