fiery turtle Apr 21, 2024, 10:06 PM

#

Is it all in free_package?

hollow elm Apr 21, 2024, 10:06 PM

#

not all

fiery turtle Apr 21, 2024, 10:06 PM

#

Hmm

hollow elm Apr 21, 2024, 10:06 PM

#

another one is ```
0x50300009ac34 is located 4 bytes inside of 32-byte region [0x50300009ac30,0x50300009ac50)
freed by thread T0 here:
#0 0x5631214da536 in free /var/tmp/portage/sys-libs/compiler-rt-sanitizers-17.0.6/work/compiler-rt/lib/asan/asan_malloc_linux.cpp:52:3
#1 0x563121524d68 in uacpi_kernel_free /tmp/uacpi/tests/runner/interface_impl.cpp:217:5
#2 0x56312154d6e4 in free_object /tmp/uacpi/source/types.c:666:5
#3 0x56312154d6e4 in free_chain /tmp/uacpi/source/types.c:711:13
#4 0x56312154d6e4 in uacpi_object_unref /tmp/uacpi/source/types.c:749:9

previously allocated by thread T0 here:
#0 0x5631214da7de in malloc /var/tmp/portage/sys-libs/compiler-rt-sanitizers-17.0.6/work/compiler-rt/lib/asan/asan_malloc_linux.cpp:69:3
#1 0x563121525082 in uacpi_kernel_alloc /tmp/uacpi/tests/runner/interface_impl.cpp:188:17
#2 0x563121525082 in uacpi_kernel_calloc /tmp/uacpi/tests/runner/interface_impl.cpp:222:17
#3 0x56312154c571 in uacpi_create_object /tmp/uacpi/source/types.c:313:11

fiery turtle Apr 21, 2024, 10:07 PM

#

Interesting

#

Send that over too

hollow elm Apr 21, 2024, 10:07 PM

#

📎 id000011sig06src000846000051time4264324execs1093266opsplicerep4

fiery turtle Apr 21, 2024, 10:08 PM

#

Thx

hollow elm Apr 21, 2024, 10:08 PM

#

i think i'll let the fuzzer run over night

#

due to the larger input corpus

fiery turtle Apr 21, 2024, 10:08 PM

#

That would be great

hollow elm Apr 21, 2024, 11:42 PM

#

curious ```
[uACPI][WARN] invalid table 'DSDT' checksum!
[uACPI][TRACE] load of 'DSDT' (OEM ID 'ATC' OEM Table ID 'Cofbej')
[uACPI][TRACE] write to [_SB_.TP80] (2 bytes) TableData[0x0000000000000001] = 0x0
AddressSanitizer:DEADLYSIGNAL

==1950122==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000001 (pc 0x556a1f6d71e9 bp 0x7ffcff823501 sp 0x7ffcff823490 T0)
==1950122==The signal is caused by a WRITE memory access.
==1950122==Hint: address points to the zero page.
#0 0x556a1f6d71e9 in memory_write /tmp/uacpi/source/default_handlers.c:270:35
#1 0x556a1f6d71e9 in table_data_region_do_rw /tmp/uacpi/source/default_handlers.c:322:8

#

lmao another aml blob that just segfaults iasl

#

Program received signal SIGSEGV, Segmentation fault.
AcpiDmCheckForSymbolicOpcode (Op=Op@entry=0x5555556cc4c0, Info=Info@entry=0x7fffffffd030) at ../../../source/components/disassembler/dmcstyle.c:429
429           Target = Argument2->Common.Next;
(gdb) p Argument2
$1 = (ACPI_PARSE_OBJECT *) 0x0

hasty plinth Apr 22, 2024, 1:19 AM

#

sure is, its here https://github.com/DeanoBurrito/northport, although if you're looking for how I'm using uacpi its not in the repo yet. I'm using it as a loadable driver, and am still playing around with how I want the api to look on the kernel side.

north holly Apr 22, 2024, 1:20 AM

#

I wanted to do that in my previous kernel, I don't think it's a good idea

#

The code was going to be messy

hasty plinth Apr 22, 2024, 1:21 AM

#

I think its a good idea meme

north holly Apr 22, 2024, 1:21 AM

#

ok then gl

hasty plinth Apr 22, 2024, 1:21 AM

#

ty

fiery turtle Apr 22, 2024, 6:29 AM

#

hollow elm curious ``` [uACPI][WARN] invalid table 'DSDT' checksum! [uACPI][TRACE] load of ...

Huh

fiery turtle Apr 22, 2024, 6:30 AM

#

hollow elm lmao another aml blob that just segfaults iasl

Classic

fiery turtle Apr 22, 2024, 8:47 AM

#

okay yeah this isnt a use after free

#

this is use of uninitialized pointer

#

the fuzzer made it so _STA returned nothing

#

in that case apparently nothing in the code overwrites the pointer to NULL

#

which is unfortunate

#

log with the fix applied

fiery turtle Apr 22, 2024, 8:58 AM

#

hollow elm

same thing here

#

#

these are some insane blobs

north holly Apr 22, 2024, 9:51 AM

#

fiery turtle this is use of uninitialized pointer

This is why you should've listened to my static analyzer

#

Funnily enough, if the code path was simple enough the static analyzer might've been able to tell you what happened

fiery turtle Apr 22, 2024, 9:56 AM

#

uacpi_execute_control_method must set the pointer to null if no return value was provided and the method exited successfully

#

that's just what any sane api would do

fiery turtle Apr 22, 2024, 9:57 AM

#

north holly This is why you should've listened to my static analyzer

probably

north holly Apr 22, 2024, 9:58 AM

#

Tbf this path would've been buried in the 100s of one's like it

fiery turtle Apr 22, 2024, 9:59 AM

#

yup

hollow elm Apr 22, 2024, 11:38 AM

#

https://uwu.foundation/8rX4rUp.png

fiery turtle Apr 22, 2024, 11:40 AM

#

i wonder how many of these is just uninitialized pointer because of missing return value

hollow elm Apr 22, 2024, 11:40 AM

#

a good chunk i presume

fiery turtle Apr 22, 2024, 11:40 AM

#

probably

hollow elm Apr 22, 2024, 11:40 AM

#

[uACPI][TRACE] load of 'DSDT' (OEM ID 'DELL\x' OEM Table ID 'UsbCTabl')
AddressSanitizer:DEADLYSIGNAL
=================================================================
==3975158==ERROR: AddressSanitizer: SEGV on unknown address (pc 0x55def6b6cdc0 bp 0x000000000007 sp 0x7ffea5f45180 T0)
==3975158==The signal is caused by a READ memory access.
==3975158==Hint: this fault was caused by a dereference of a high value address (see register values below).  Disassemble the provided pc to learn which register was used.
    #0 0x55def6b6cdc0 in bit_copy /tmp/uacpi/source/io.c:80:21

fiery turtle Apr 22, 2024, 11:40 AM

#

but cool nonetheless

hollow elm Apr 22, 2024, 11:40 AM

#

wym unknown address

fiery turtle Apr 22, 2024, 11:41 AM

#

lol

hollow elm Apr 22, 2024, 11:41 AM

#

a good chunk of the crashes are also my fault

fiery turtle Apr 22, 2024, 11:41 AM

#

wdym?

hollow elm Apr 22, 2024, 11:41 AM

#

on the assert here

#

by trying to map >512 pages at once

fiery turtle Apr 22, 2024, 11:42 AM

#

ohh

hollow elm Apr 22, 2024, 11:43 AM

#

[uACPI][TRACE] read from [\_SB_.PCI0.LPCB.LPC1] (1 bytes) PCI_Config[0x000000000000006B] = 0xFF
/tmp/uacpi/source/uacpi.c:634:23: runtime error: shift exponent 65 is too large for 32-bit type 'int'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /tmp/uacpi/source/uacpi.c:634:23 in 
invalid free size: originally allocated 56 bytes, freeing as 32

fiery turtle Apr 22, 2024, 11:43 AM

#

that's an interesting one

hollow elm Apr 22, 2024, 11:43 AM

#

indeed

#

hm it's in uacpi_eval_sta

#

so possibly related to the uninit pointer

fiery turtle Apr 22, 2024, 11:44 AM

#

almost certainly

#

diff --git a/source/interpreter.c b/source/interpreter.c
index d6693a0..1aa24a8 100644
--- a/source/interpreter.c
+++ b/source/interpreter.c
@@ -5680,6 +5680,9 @@ uacpi_status uacpi_execute_control_method(
     }

 out:
+    if (out_obj != UACPI_NULL)
+        *out_obj = UACPI_NULL;
+
     if (ctx->ret != UACPI_NULL &&
         ctx->ret->type != UACPI_OBJECT_UNINITIALIZED) {
         uacpi_object_ref(ctx->ret);

#

this is the fix i made locally

#

see if that fixes like all of those

hollow elm Apr 22, 2024, 11:47 AM

#

i wonder if i could teach afl that crashes in memory_read/memory_write are uninteresting

fiery turtle Apr 22, 2024, 11:48 AM

#

probably

hollow elm Apr 22, 2024, 11:56 AM

#

okay that seems to fix all the interesting crashes

fiery turtle Apr 22, 2024, 11:57 AM

#

unlucky

#

but good find nonetheless

#

Does it still try to shift by 65 after that?

hollow elm Apr 22, 2024, 11:57 AM

#

no that went away too

fiery turtle Apr 22, 2024, 11:57 AM

#

oh ok

hollow elm Apr 22, 2024, 11:58 AM

#

or at least i don't see it

fiery turtle Apr 22, 2024, 11:58 AM

#

yeah that shifts the type member

#

since it was a bogus pointer it was shfting by a bogus number

hollow elm Apr 22, 2024, 11:59 AM

#

i wonder how we come up with a read from 0x000000ff

#

i can't see how is_physical_pointer can say it's not physical

fiery turtle Apr 22, 2024, 11:59 AM

#

where did you see that?

hollow elm Apr 22, 2024, 11:59 AM

#

[uACPI][WARN] invalid table 'DSDT' checksum!
[uACPI][TRACE] load of 'DSDT' (OEM ID 'DUC' OEM Table ID 'DkcVfdy')
AddressSanitizer:DEADLYSIGNAL
=================================================================
==263676==ERROR: AddressSanitizer: SEGV on unknown address 0x0000000000ff (pc 0x55cc0f480538 bp 0x7ffeba639cb0 sp 0x7ffeba639bc0 T0)
==263676==The signal is caused by a READ memory access.
==263676==Hint: address points to the zero page.
    #0 0x55cc0f480538 in memory_read /tmp/uacpi/source/default_handlers.c:254:16

fiery turtle Apr 22, 2024, 12:00 PM

#

out of bounds write because of a previous mapping being smaller and then that trashed this pointer?

#

potentially

#

because 0xFF seems like what IO reads return

hollow elm Apr 22, 2024, 12:02 PM

#

hmm

fiery turtle Apr 22, 2024, 12:02 PM

#

 void* uacpi_kernel_map(uacpi_phys_addr addr, uacpi_size size)
 {
+    std::printf("Mapping 0x%016" PRIX64 " with size %zu\n", addr, size);
+

#

try this

#

see if it maps the same thing multiple times

hollow elm Apr 22, 2024, 12:04 PM

#

oh yeah the same addr is mapped twice

#

second time with a larger size

fiery turtle Apr 22, 2024, 12:04 PM

#

yup..

hollow elm Apr 22, 2024, 12:04 PM

#

so that's the realloc thing you mentioned

fiery turtle Apr 22, 2024, 12:04 PM

#

yup

hollow elm Apr 22, 2024, 12:06 PM

#

bit underwhelming to only find 1 real bug but that's still 1 bug less

fiery turtle Apr 22, 2024, 12:06 PM

#

its a pretty important bug so yeah

#

i wonder if that unblocks more bugs

#

i also want to make aborts in methods unwind back to table level code so that will unlock even more bugs potentially

vast kestrel Apr 22, 2024, 12:53 PM

#

Afl looks cool ngl

#

I should try and use it for my dotnet runtime

fiery turtle Apr 22, 2024, 1:05 PM

#

yeah its pretty nice

north holly Apr 22, 2024, 4:03 PM

#

@fiery turtle I get some warnings in uACPI related to variables possibly being used uninitialized.

uACPI/source/interpreter.c:2844:9: warning: 'arg1_buf.ptr' may be used uninitialized [-Wmaybe-uninitialized]
 2844 |         uacpi_memcpy(dst_buf + arg0_buf->size, arg1_buf.ptr, arg1_buf.len);
      |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
uACPI/source/interpreter.c:2834:41: note: 'arg1_buf.ptr' was declared here
 2834 |         struct object_storage_as_buffer arg1_buf;
      |                                         ^~~~~~~~
uACPI/source/interpreter.c:2837:18: warning: 'arg1_buf.len' may be used uninitialized [-Wmaybe-uninitialized]
 2837 |         buf_size = arg0_buf->size + arg1_buf.len;
      |         ~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
uACPI/source/interpreter.c:2834:41: note: 'arg1_buf.len' was declared here
 2834 |         struct object_storage_as_buffer arg1_buf;```

#

same warning around interpreter.c:1714:17

#

actually I might as well give you the entire log

#

📎 analyzer_output.txt

#

you can ignore all analyzer warnings, those are for my kernel

#

everything in the log until it says [27%] Built target uacpi

fiery turtle Apr 22, 2024, 4:26 PM

#

nah its a false positive

#

i can add a default switch case to make it shut up tho

north holly Apr 22, 2024, 4:31 PM

#

I just love it though when the static analyzer yells at me because of nullptr dereferences as a result of a new call

#

It's so annoying

fiery turtle Apr 22, 2024, 9:05 PM

#

@hollow elm I think I came up with a genius idea

diff --git a/tests/runner/interface_impl.cpp b/tests/runner/interface_impl.cpp
index cee0874..ae387f4 100644
--- a/tests/runner/interface_impl.cpp
+++ b/tests/runner/interface_impl.cpp
@@ -116,22 +116,7 @@ uacpi_status uacpi_kernel_pci_write(
     return UACPI_STATUS_OK;
 }

-static bool is_physical_address(uacpi_phys_addr addr, uacpi_size size)
-{
-#ifdef _WIN32
-    return IsBadReadPtr((void*)addr, size);
-#else
-    // https://stackoverflow.com/questions/4611776/isbadreadptr-analogue-on-unix
-    int nullfd = open("/dev/random", O_WRONLY);
-    bool ok = true;
-
-    if (write(nullfd, (void*)addr, size) < 0)
-        ok = false;
-
-    close(nullfd);
-    return !ok;
-#endif
-}
+bool g_expect_virtual_addresses = true;

 static
 std::unordered_map<void*, std::pair<uacpi_phys_addr, size_t>>
@@ -141,7 +126,7 @@ static std::unordered_map<uacpi_phys_addr, void*> phys_to_virt;

 void* uacpi_kernel_map(uacpi_phys_addr addr, uacpi_size size)
 {
-    if (is_physical_address(addr, size)) {
+    if (!g_expect_virtual_addresses) {
         auto it = phys_to_virt.find(addr);
         if (it != phys_to_virt.end()) {
             virt_to_phys_and_refcount[it->second].second++;
diff --git a/tests/runner/test_runner.cpp b/tests/runner/test_runner.cpp
index 891257e..d36daa6 100644
--- a/tests/runner/test_runner.cpp
+++ b/tests/runner/test_runner.cpp
@@ -131,6 +131,8 @@ static void run_test(
     uacpi_status st = uacpi_initialize(&params);
     ensure_ok_status(st);

+    g_expect_virtual_addresses = false;
+
     st = uacpi_install_notify_handler(
         uacpi_namespace_root(), handle_notify, nullptr
     );

hollow elm Apr 22, 2024, 9:06 PM

#

incredible

fiery turtle Apr 22, 2024, 9:06 PM

#

lmao

fiery turtle Apr 22, 2024, 9:39 PM

#

@hollow elm i made an insanely overengineered system that should be safe against everything

#

we should try fuzzing that

#

im about to merge it in a bit

#

@@ -118,31 +120,60 @@ uacpi_status uacpi_kernel_pci_write(

 bool g_expect_virtual_addresses = true;

+struct mapping {
+    void *virt;
+    uacpi_size size;
+
+    bool operator==(const mapping& m) const
+    {
+        return size == m.size;
+    }
+};
+
+template <>
+struct std::hash<mapping>
+{
+    size_t operator()(const mapping& m) const
+    {
+        return m.size;
+    }
+};
+
 static
 std::unordered_map<void*, std::pair<uacpi_phys_addr, size_t>>
 virt_to_phys_and_refcount;

-static std::unordered_map<uacpi_phys_addr, void*> phys_to_virt;
+static std::unordered_map<uacpi_phys_addr, std::unordered_set<mapping>>
+phys_to_virt;

 void* uacpi_kernel_map(uacpi_phys_addr addr, uacpi_size size)
 {
     if (!g_expect_virtual_addresses) {
         auto it = phys_to_virt.find(addr);
         if (it != phys_to_virt.end()) {
-            virt_to_phys_and_refcount[it->second].second++;
-            return it->second;
+            auto mapping_it = it->second.find({ nullptr, size });
+
+            if (mapping_it != it->second.end()) {
+                virt_to_phys_and_refcount[mapping_it->virt].second++;
+                return mapping_it->virt;
+            }
+
+            std::printf("WARN: remapping physical 0x%016" PRIX64
+                        " with size %zu\n", addr, size);
         }

         void *virt = std::calloc(size, 1);
+        mapping m = { virt, size };
+
         virt_to_phys_and_refcount[virt] = { addr, 1 };
-        phys_to_virt[addr] = virt;
+        phys_to_virt[addr].insert(m);
         return virt;
     }

     return reinterpret_cast<void*>(addr);
 }

-void uacpi_kernel_unmap(void* addr, uacpi_size)
+void uacpi_kernel_unmap(void* addr, uacpi_size size)
 {
     auto it = virt_to_phys_and_refcount.find(addr);
     if (it == virt_to_phys_and_refcount.end())
@@ -151,7 +182,18 @@ void uacpi_kernel_unmap(void* addr, uacpi_size)
     if (--it->second.second > 0)
         return;

-    phys_to_virt.erase(it->second.first);
+    auto phys_it = phys_to_virt.find(it->second.first);
+    auto mapping_it = phys_it->second.find({ nullptr, size });
+    if (mapping_it == phys_it->second.end()) {
+        std::printf("WARN: cannot identify mapping virt=%p phys=0x%016" PRIX64
+                    " with size %zu\n", addr, phys_it->first, size);
+        return;
+    }
+
+    phys_it->second.erase(mapping_it);
+    if (phys_it->second.empty())
+        phys_to_virt.erase(it->second.first);
+
     std::free(it->first);
     virt_to_phys_and_refcount.erase(it);
 }

#

basically this

#

it now successfully runs that blob you sent yesterday

#

okay merged

#

now I expect 0 crashes in the fuzzer

hollow elm Apr 22, 2024, 9:55 PM

#

nice

fiery turtle Apr 22, 2024, 9:57 PM

#

inb4 instant crash

hollow elm Apr 22, 2024, 11:01 PM

#

for this round of fuzzing i'm enabling cmplog

#

which hooks comparisons and sends the values being compared against to afl which then tries to inject them into the fuzzer output

#

hopefully this should make it so more valid instructions appear

fiery turtle Apr 22, 2024, 11:06 PM

#

Hmm

#

Im not sure what that means but if it makes the fuzzing more efficient sounds great!

#

Thank you

lilac knoll Apr 22, 2024, 11:09 PM

#

ig it means that if you have a branch if (x == 42) then instead of afl randomly trying values for x until it hits 42, it'd get a hint that it should shove the value 42 in there

hollow elm Apr 22, 2024, 11:09 PM

#

yeah

fiery turtle Apr 22, 2024, 11:09 PM

#

Ooh nice

hollow elm Apr 22, 2024, 11:09 PM

#

i'm thinking it'll figure that it should shove more valid aml opcodes into the input

fiery turtle Apr 22, 2024, 11:10 PM

#

Yeah thats definitely interesting

left orbit Apr 22, 2024, 11:14 PM

#

personally i think part of what made fuzzing great was the fuzzer throwing random opcodes everywhere so you can find weird ass edge cases with uacpi

#

also i kinda forgot but thanks for sized frees :^)

fiery turtle Apr 22, 2024, 11:14 PM

#

Npnp

left orbit Apr 22, 2024, 11:14 PM

#

not my uacpi kernel api free is no longer a stub

#

although i do not believe it was called either way

fiery turtle Apr 22, 2024, 11:15 PM

#

left orbit not my uacpi kernel api free is no longer a stub

Nice

fiery turtle Apr 22, 2024, 11:15 PM

#

left orbit although i do not believe it was called either way

Why is that lol

left orbit Apr 22, 2024, 11:15 PM

#

i absolutely do not utilize any of the public apis just yet

#

ill replace table iteration with uacpi shortly

fiery turtle Apr 22, 2024, 11:15 PM

#

Not even init?

#

Lol

left orbit Apr 22, 2024, 11:16 PM

#

oh no i do call init

fiery turtle Apr 22, 2024, 11:16 PM

#

And namespace load as well?

left orbit Apr 22, 2024, 11:16 PM

#

i call uacpi_initialize and uacpi_install_notify_handler

#

uhhhh

#

thinkies

fiery turtle Apr 22, 2024, 11:17 PM

#

Yeah there's uh

left orbit Apr 22, 2024, 11:17 PM

#

oh shit i totally did not read

#

the test runner fully

fiery turtle Apr 22, 2024, 11:17 PM

#

U dont need the notify handler too

left orbit Apr 22, 2024, 11:17 PM

#

yeah let me throw in some more calls and see what happens

fiery turtle Apr 22, 2024, 11:17 PM

#

Its mostly for testing

#

Look at oberrows kernel or managarm init sequence

#

I should really make a tutorial

left orbit Apr 22, 2024, 11:18 PM

#

is eval("\MAIN") required?

fiery turtle Apr 22, 2024, 11:18 PM

#

Nope its the test runner abi

hollow elm Apr 22, 2024, 11:18 PM

#

left orbit personally i think part of what made fuzzing great was the fuzzer throwing rando...

well it's still gonna do that, but it's not gonna throw in invalid opcodes which just abort execution

left orbit Apr 22, 2024, 11:18 PM

#

yeah i figured, just wanted to ask

left orbit Apr 22, 2024, 11:18 PM

#

hollow elm well it's still gonna do that, but it's not gonna throw in invalid opcodes which...

oh i see

hollow elm Apr 22, 2024, 11:18 PM

#

or at least that's the idea

left orbit Apr 22, 2024, 11:18 PM

#

idk why i thought it was gonna generate fully compliant aml

#

lmao

hollow elm Apr 22, 2024, 11:19 PM

#

no way in hell it's doing that

fiery turtle Apr 22, 2024, 11:19 PM

#

Well sometimes it does

hollow elm Apr 22, 2024, 11:19 PM

#

did you see the kind of output it makes

fiery turtle Apr 22, 2024, 11:19 PM

#

Else { else { else {

hollow elm Apr 22, 2024, 11:19 PM

#

if it is even decompilable with iasl

lilac knoll Apr 22, 2024, 11:19 PM

#

KEKW

#

i wonder how much acpica has been fuzzed

hollow elm Apr 22, 2024, 11:20 PM

#

i am inadvertently fuzzing iasl as well :^)

fiery turtle Apr 22, 2024, 11:20 PM

#

It died almost instantly as well I think?

fiery turtle Apr 22, 2024, 11:20 PM

#

lilac knoll i wonder how much acpica has been fuzzed

Considering it dies if u return a reference to a local object

#

Probably not at all

hollow elm Apr 22, 2024, 11:22 PM

#

i was about to say "i wonder how hard it is to go from aml -> taking over linux due to bugs in acpica" but then i remembered aml can poke at arbitrary memory already

fiery turtle Apr 22, 2024, 11:22 PM

#

Yeah its trusted code basically

left orbit Apr 22, 2024, 11:26 PM

#

[  0] (main) err: Kernel panic: switch on corrupt value
[  0] (debug) err: Stack backtrace:
[  0] (debug) err:   0xFFFFFFFF800827F7: <no symbol info> at kernel/src/mutex.zig:17
[  0] (debug) err:   0xFFFFFFFF800B96B4: uacpi_kernel_acquire_mutex at kernel/src/uacpi.zig:306
[  0] (debug) err:   0xFFFFFFFF8001F8ED: enter_method at kernel/./uacpi/source/interpreter.c:4232
[  0] (debug) err:   0xFFFFFFFF800167F6: prepare_method_call at kernel/./uacpi/source/interpreter.c:4892
[  0] (debug) err:   0xFFFFFFFF80015F97: uacpi_execute_control_method at kernel/./uacpi/source/interpreter.c:5635
[  0] (debug) err:   0xFFFFFFFF8000E896: uacpi_eval at kernel/./uacpi/source/uacpi.c:618
[  0] (debug) err:   0xFFFFFFFF8000E8E7: uacpi_eval_typed at kernel/./uacpi/source/uacpi.c:629
[  0] (debug) err:   0xFFFFFFFF80012813: uacpi_eval_sta at kernel/./uacpi/source/utilities.c:620
[  0] (debug) err:   0xFFFFFFFF8000F1E1: sta_eval at kernel/./uacpi/source/uacpi.c:460
[  0] (debug) err:   0xFFFFFFFF8000E54D: do_sta_ini at kernel/./uacpi/source/uacpi.c:504
[  0] (debug) err:   0xFFFFFFFF8005288A: uacpi_namespace_for_each_node_depth_first at kernel/./uacpi/source/namespace.c:476
[  0] (debug) err:   0xFFFFFFFF8000DFF0: uacpi_namespace_initialize at kernel/./uacpi/source/uacpi.c:0
[  0] (debug) err:   0xFFFFFFFF8008F0AF: init at kernel/src/uacpi.zig:41
[  0] (debug) err:   0xFFFFFFFF8008EE73: main at kernel/src/main.zig:155
[  0] (debug) err:   0xFFFFFFFF8008F2FA: _start at kernel/src/main.zig:111
[  0] (debug) err:   0x0000000000000000: <no symbol info> at ??:0```

#

something seems to be going really wrong meme

#

i'll have to fix my mutex bullshit, but thats for tomorrow

hollow elm Apr 22, 2024, 11:26 PM

#

oh oops lmoa ```
[+] Found 44 unique tuples across 4578 files.
[!] WARNING: All test cases had the same traces, check syntax!
[+] Narrowed down to 1 files, saved in 'fuzz-in/'.

#

i threw out all blobs that time out after 0.8s

fiery turtle Apr 22, 2024, 11:28 PM

#

left orbit ``` [ 0] (main) err: Kernel panic: switch on corrupt value [ 0] (debug) err: S...

Lol

fiery turtle Apr 22, 2024, 11:28 PM

#

hollow elm oh oops lmoa ``` [+] Found 44 unique tuples across 4578 files. [!] WARNING: All ...

What does this mean?

hollow elm Apr 22, 2024, 11:28 PM

#

i forgot to tell it to pass the filename to the test runner so it reduced all the inputs to just one

fiery turtle Apr 22, 2024, 11:29 PM

#

Lol

left orbit Apr 22, 2024, 11:29 PM

#

holy SHIT i think i might just be retarded !!!

hollow elm Apr 22, 2024, 11:29 PM

#

$ afl-cmin -i hw-blobs/ -o fuzz-in/ -T all -- _build-nocmplog/test-runner 
``` vs ```
$ afl-cmin -i hw-blobs/ -o fuzz-in/ -T all -- _build-nocmplog/test-runner @@

fiery turtle Apr 22, 2024, 11:29 PM

#

Makes sense

hollow elm Apr 22, 2024, 11:29 PM

#

left orbit holy SHIT i think i might just be retarded !!!

lol

fiery turtle Apr 22, 2024, 11:29 PM

#

Who knows maybe test runner segfaults at startup sometimes

left orbit Apr 22, 2024, 11:29 PM

#

/// Returns a pointer to undefined memory.
/// Call `destroy` with the result to free the memory.
pub fn create(self: Allocator, comptime T: type) Error!*T {```

#

HJAHA!!!!

#

WOW

#

WHO WOULD HAVE THOUGHT

#

[ 0] (phys) info: uacpi-info: namespace initialization done: 35 devices, 0 thermal zones, 1 processors

hollow elm Apr 22, 2024, 11:30 PM

#

fiery turtle Who knows maybe test runner segfaults at startup sometimes

well then the chance is less than 1 in like 30 million

fiery turtle Apr 22, 2024, 11:30 PM

#

left orbit `[ 0] (phys) info: uacpi-info: namespace initialization done: 35 devices, 0 the...

Nice

left orbit Apr 22, 2024, 11:31 PM

#

uninitialized allocations strike again

fiery turtle Apr 22, 2024, 11:31 PM

#

hollow elm well then the chance is less than 1 in like 30 million

Lol

hollow elm Apr 22, 2024, 11:31 PM

#

and less than 1 in 12 thousand on master

fiery turtle Apr 22, 2024, 11:32 PM

#

Fuzzing the clone syscall

lilac knoll Apr 22, 2024, 11:37 PM

#

left orbit uninitialized allocations strike again

r-rust?

fiery turtle Apr 22, 2024, 11:37 PM

#

Zig I think

hollow elm Apr 22, 2024, 11:43 PM

#

hm i fucked something up

#

https://uwu.foundation/8Las2t2.png

fiery turtle Apr 22, 2024, 11:43 PM

#

Everything just hangs?

hollow elm Apr 22, 2024, 11:44 PM

#

it's running very slow

fiery turtle Apr 22, 2024, 11:44 PM

#

Huh

hollow elm Apr 22, 2024, 11:45 PM

#

which is confusing because i threw out all the hw blobs that timed out

#

even before feeding it into afl-cmin

fiery turtle Apr 22, 2024, 11:46 PM

#

Are you compiling llvm perhaps KEKW

hollow elm Apr 22, 2024, 11:46 PM

#

nah lol

fiery turtle Apr 22, 2024, 11:47 PM

#

Strange

hollow elm Apr 22, 2024, 11:50 PM

#

am i losing it, the cmplog variant runs fine

fiery turtle Apr 22, 2024, 11:50 PM

#

Lol

hollow elm Apr 23, 2024, 12:12 AM

#

okay no it also experienced slowdowns

#

im gonna run afl-tmin on the input corpus

#

and then maybe cmin again on the minimized files

#

i love gnu parallel ```
$ find ./ -type f | parallel afl-tmin -i {} -o ../fuzz-in-min/{/} -- ../_build-nocmplog/test-runner @@
Academic tradition requires you to cite works you base your article on.
If you use programs that use GNU Parallel to process data for an article in a
scientific publication, please cite:

Tange, O. (2024, January 22). GNU Parallel 20240122 ('Frederik X').
Zenodo. https://doi.org/10.5281/zenodo.10558745

This helps funding further development; AND IT WON'T COST YOU A CENT.
If you pay 10000 EUR you should feel free to use GNU Parallel without citing.

More about funding GNU Parallel and the citation notice:
https://www.gnu.org/software/parallel/parallel_design.html#citation-notice

To silence this citation notice: run 'parallel --citation' once.

fiery turtle Apr 23, 2024, 12:14 AM

#

Lmao

hollow elm Apr 23, 2024, 12:15 AM

#

damn i should've passed --bar to parallel

#

to get a progress bar

fiery turtle Apr 23, 2024, 12:17 AM

#

Lol yes

hollow elm Apr 23, 2024, 12:34 AM

#

okay i think i'm gonna look into it tomorrow

vagrant hull Apr 23, 2024, 1:21 AM

#

hollow elm i love gnu parallel ``` $ find ./ -type f | parallel afl-tmin -i {} -o ../fuzz-i...

what in the name of fuck

fiery turtle Apr 23, 2024, 11:22 AM

#

@hollow elm did it crash?

hollow elm Apr 23, 2024, 11:23 AM

#

havent had any crashes yesterday but i didnt do any long campaigns

fiery turtle Apr 23, 2024, 11:27 AM

#

well thats progress already anyway

#

pretty nice

fiery turtle Apr 23, 2024, 7:27 PM

#

Made a basic uACPI page on osdev wiki: https://wiki.osdev.org/UACPI

#

for now this only has initialization tutorial but ill add some code examples later

north holly Apr 23, 2024, 7:28 PM

#

Take some from my previous kernel :^)

fiery turtle Apr 23, 2024, 7:29 PM

#

I might add a list of kernels to look at for examples

fiery turtle Apr 23, 2024, 7:54 PM

#

Bro

#

Don't u just love when github ci runners randomly break every fucking week for no reason

jaunty fox Apr 23, 2024, 7:56 PM

#

fiery turtle Don't u just love when github ci runners randomly break every fucking week for n...

typical fickle python

fiery turtle Apr 23, 2024, 7:57 PM

#

that's okay i fixed it caret

hollow elm Apr 23, 2024, 8:03 PM

#

didn't know there was an env var

rustic compass Apr 24, 2024, 10:28 AM

#

fiery turtle Made a basic uACPI page on osdev wiki: https://wiki.osdev.org/UACPI

maybe add example code to read some tables and maybe add some docs that list all api headers with a description on what functions they expose and their purpose

fiery turtle Apr 24, 2024, 10:29 AM

#

yeah thats on the todo list

tired crater Apr 24, 2024, 11:45 AM

#

fiery turtle Made a basic uACPI page on osdev wiki: https://wiki.osdev.org/UACPI

🎉

hasty plinth Apr 25, 2024, 12:30 AM

#

whats the recommended course of action for handling the fatal type of firmware request? should that result in a panic or is it safe to keep running for a bit?

fiery turtle Apr 25, 2024, 6:51 AM

#

hasty plinth whats the recommended course of action for handling the fatal type of firmware r...

Linux just logs it and then ignores it

hasty plinth Apr 25, 2024, 6:56 AM

#

OK cheers, I saw managarm doing the same

fiery turtle Apr 26, 2024, 10:39 PM

#

Added a few more code examples in osdev wiki: https://wiki.osdev.org/uACPI

#

Also fixed a few misc things in uacpi, including namespace node attach order

jaunty fox Apr 26, 2024, 10:46 PM

#

fiery turtle Added a few more code examples in osdev wiki: https://wiki.osdev.org/uACPI

nice

#

i will implement shutdown at some point

#

uacpi will make the doing it easy it's the preparing that won't be

fiery turtle Apr 26, 2024, 10:54 PM

#

yeah, unmounting everything and stuff is definitely not easy

fiery turtle Apr 26, 2024, 10:55 PM

#

jaunty fox nice

thank you

jaunty fox Apr 26, 2024, 10:56 PM

#

proper shutdown is one of those features which i don't think i have ever seen a hobby OS

#

i'm not sure about unmounting either, which goes with it

fiery turtle Apr 26, 2024, 10:57 PM

#

yup

#

best u get is just how mangarm currently does it

#

power button press == instant shutdown

hollow elm Apr 26, 2024, 11:00 PM

#

we should probably implement sync sometime :^)

fiery turtle Apr 26, 2024, 11:00 PM

#

just sync probably won't be enough

#

but yeah

hollow elm Apr 26, 2024, 11:01 PM

#

well yeah ideally we'd wire the power button to tell userspace to shutdown

fiery turtle Apr 26, 2024, 11:01 PM

#

yeah

#

also usually power button is configured to actually trigger sleep instead

#

or bring up a gui prompt

hollow elm Apr 26, 2024, 11:02 PM

#

but in the current situation having all managed memory objects (objects used to implement page caches) be told to do a writeback at least would be a good idea

#

if we want to just power off immediately

fiery turtle Apr 26, 2024, 11:02 PM

#

but yeah all of that comes down to routing it to userspace

fiery turtle Apr 26, 2024, 11:02 PM

#

hollow elm but in the current situation having all managed memory objects (objects used to ...

thats a start for sure

wind fiber Apr 26, 2024, 11:30 PM

#

Yeah

#

That reminds me I have some shutdown related work still standing. I should get back to work and get that shit fixed

fiery turtle Apr 27, 2024, 1:33 PM

#

Made method aborts while loading tables non-fatal (aka only the call chain is aborted and the table keeps loading)

hollow elm Apr 27, 2024, 4:04 PM

#

[uACPI][TRACE] write to [\PCI0.UHCI.TEST.PCIR] (1 bytes) TableData[0x0000000000000000] = 0x7
AddressSanitizer:DEADLYSIGNAL
=================================================================
==1140720==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x5655286ea9d3 bp 0x7fff6db14201 sp 0x7fff6db14170 T0)
==1140720==The signal is caused by a WRITE memory access.
==1140720==Hint: address points to the zero page.
    #0 0x5655286ea9d3 in memory_write /tmp/uacpi/source/default_handlers.c:267:34
    #1 0x5655286ea9d3 in table_data_region_do_rw /tmp/uacpi/source/default_handlers.c:322:8

#

i suppose there's not much you can do about this?

fiery turtle Apr 27, 2024, 4:08 PM

#

hmm

#

how the f is the pointer 0

hollow elm Apr 27, 2024, 4:09 PM

#

well the connection between TableData[0x0000000000000000] and the pointer being 0x000000000000 seems obvious? but i don't know how it actually works

fiery turtle Apr 27, 2024, 4:10 PM

#

oh actually i know exactly how this happened

#

so table data opregions are created via TableRegionDataOp

#

under the hood it uses the same opregion structure

#

but

#

this aml created it via OpregionOp by hijacking the address space type

#

address space type happened to match internal enum that corresponds to tabledata

#

and table data is mapped as just a plain virtual address

#

so it did not further validation on that

#

#

basically this "Internal type"

hollow elm Apr 27, 2024, 4:12 PM

#

hm yeah ```
OperationRegion (PCIR, 0xFE, 0x00, Arg0)
Field (PCIR, AnyAcc, NoLock, Preserve)
{
REG0, 38
}

fiery turtle Apr 27, 2024, 4:12 PM

#

lmao

hollow elm Apr 27, 2024, 4:12 PM

#

iasl managed to actually decompile it :^)

fiery turtle Apr 27, 2024, 4:12 PM

#

technically this is invalid i think, because its a reserved address space

#

but yeah i guess i should add validation for this in the code

#

to at least check that its not 0xFE

hollow elm Apr 27, 2024, 4:14 PM

#

that's the only crash so far, after about an hour

#

but i'm only using the tests as input

fiery turtle Apr 27, 2024, 4:14 PM

#

technically this is a real bug, so nice

#

#

hmm well i guess its OEM defined

#

so maybe a u8 is not enough to represent it

#

idk

hollow elm Apr 27, 2024, 4:16 PM

#

what's the TableData type used for?

fiery turtle Apr 27, 2024, 4:17 PM

#

hollow elm Apr 27, 2024, 4:17 PM

#

ah

#

literally what it says on the tin

fiery turtle Apr 27, 2024, 4:18 PM

#

ye

#

these are basically system memory, but the base is a virtual address

#

at least in uacpi

hollow elm Apr 27, 2024, 4:19 PM

#

makes sense

fiery turtle Apr 27, 2024, 4:35 PM

#

Now that I think about why am I even doing _REG for it, thats not even a real opregion type

#

Good thing this bug was found

hollow elm Apr 27, 2024, 4:42 PM

#

another crash found ```
[uACPI][WARN] error during _REG execution for operation region \PCI0.UHCI.TEST.PCIR@0x506000000b60: invalid argument
AddressSanitizer:DEADLYSIGNAL

==2406755==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000 (pc 0x562631cbcd83 bp 0x7fff8eebe010 sp 0x7fff8eebdf20 T0)
==2406755==The signal is caused by a READ memory access.
==2406755==Hint: address points to the zero page.
#0 0x562631cbcd83 in memory_read /tmp/uacpi/source/default_handlers.c:245:16
#1 0x562631cc1063 in dispatch_field_io /tmp/uacpi/source/io.c:229:11
#2 0x562631cc1063 in access_field_unit /tmp/uacpi/source/io.c:288:12
#3 0x562631cbf7fd in uacpi_read_field_unit /tmp/uacpi/source/io.c:364:15

#

i can't disassemble it, it has invalid opcodes, and if i just patch them to 0x00 iasl segfaults instead :^)

#

but it seems related, \PCI0.UHCI.TEST.PCIR is what used to be the TableData opregion in the previous crashing input

fiery turtle Apr 27, 2024, 5:39 PM

#

can u send both crashes pls?

#

but yeah read is traced after the read so

#

there isnt like "about to read"

hollow elm Apr 27, 2024, 5:40 PM

#

makes sense

#

so far there's 5 saved crashes but afaics they are all on a TableData opregion

fiery turtle Apr 27, 2024, 5:41 PM

#

--- a/source/interpreter.c
+++ b/source/interpreter.c
@@ -1080,6 +1080,9 @@ static uacpi_status handle_create_op_region(struct execution_context *ctx)
     op_region = obj->op_region;

     op_region->space = item_array_at(&ctx->cur_op_ctx->items, 1)->immediate;
+    if (op_region->space == UACPI_ADDRESS_SPACE_TABLE_DATA)
+        return UACPI_STATUS_AML_BAD_ENCODING;
+

#

try this

hollow elm Apr 27, 2024, 5:44 PM

#

yeah that fixes all the crashes

fiery turtle Apr 27, 2024, 5:44 PM

#

ok nice

fiery turtle Apr 28, 2024, 11:43 AM

#

I'm adding core API that should be used during initial namespace device discovery to bind corresponding drivers

#

Im thinking ill add a mode to the test runner to run this against every node, perhaps this will add better coverage for the fuzzer @hollow elm

hollow elm Apr 28, 2024, 11:44 AM

#

sounds like a good idea

fiery turtle Apr 28, 2024, 11:44 AM

#

maybe if its a device ill also run _CRS and _PRS against it

hollow elm Apr 28, 2024, 11:45 AM

#

it'd probably also be useful for managarm for exposing all the legacy devices over mbus :^)

fiery turtle Apr 28, 2024, 11:45 AM

#

yeah definitely

#

this api i basically stole from acpica

#

linux uses it for initial bus enumeration

#

Also i want to add a CI mode that decodes all 500 AML blobs from that repo and runs the test runner against all of them

#

sounds like useful regression testing

vast kestrel Apr 28, 2024, 1:11 PM

#

Is there a reason why the EC driver is not part of uacpi

fiery turtle Apr 28, 2024, 1:52 PM

#

yes

#

because a proper implementation requires very tight coupling with kernel primitives

#

as well as being able to handle platform specific EC quirks

#

uacpi is not gonna take on that responsibility and provide a shitty half-working implementation

fiery turtle Apr 28, 2024, 1:53 PM

#

fiery turtle because a proper implementation requires very tight coupling with kernel primiti...

e.g. being able to asynchronously process transactions, or switch to poll mode as needed

#

https://github.com/torvalds/linux/blob/2c8159388952f530bd260e097293ccc0209240be/drivers/acpi/ec.c#L1883

GitHub

linux/drivers/acpi/ec.c at 2c8159388952f530bd260e097293ccc0209240be...

Linux kernel source tree. Contribute to torvalds/linux development by creating an account on GitHub.

#

this is completely outside of scope of what uacpi is doing

#

and uacpi philosophy is either do it well or dont do it at all

north holly Apr 28, 2024, 1:57 PM

#

I thought you said an EC driver would be easy

fiery turtle Apr 28, 2024, 1:57 PM

#

it is trivial for a blocking EC driver that works on 80% of hardware

#

a proper async one that handles everything is quite a bit more complex tho

fiery turtle Apr 28, 2024, 1:58 PM

#

fiery turtle it is trivial for a blocking EC driver that works on 80% of hardware

see: managarm ec driver

#

like: https://github.com/torvalds/linux/blob/2c8159388952f530bd260e097293ccc0209240be/drivers/acpi/ec.c#L2223

GitHub

linux/drivers/acpi/ec.c at 2c8159388952f530bd260e097293ccc0209240be...

Linux kernel source tree. Contribute to torvalds/linux development by creating an account on GitHub.

#

this file has so many bugzilla links

vast kestrel Apr 28, 2024, 3:33 PM

#

Makes sense

#

I just always see the same EC driver in hobby osdev and am like "this looks tiny"

#

But seeing Linux explains everything :^)

#

Same vibe as xhci

fiery turtle Apr 28, 2024, 3:45 PM

#

Yup

fiery turtle Apr 29, 2024, 12:46 PM

#

Today i decided that i will add more tables to acpi.h

#

So that users dont have to roll their own

#

ill start with madt

#

but ill also do hpet and a few other

north holly Apr 29, 2024, 1:13 PM

#

I got it, now what? /j

fiery turtle Apr 29, 2024, 1:16 PM

#

stabilize it so it doesnt segfault when u blow on it

north holly Apr 29, 2024, 1:16 PM

#

bet

#

will do

#

(after school)

fiery turtle Apr 29, 2024, 1:17 PM

#

also when rewrite 5

north holly Apr 29, 2024, 1:17 PM

#

When I figure out if the kernel is truly unfixable

#

When it's damaged beyond repair

#

I think I might try testing the VMM a bit (after school)

#

With large page allocations

#

and frees (those are largely untested)

#

SetProtection is so trivial there is no point in testing

#

But I use that in the driver loader and it works fine

#

Actually, the reason it took me so long to finish the VMM for this rewrite is because I wasn't working on the kernel

#

So if I work on the kernel more, a rewrite could be feasible

#

Anyway enough littering this thread

fiery turtle Apr 29, 2024, 1:21 PM

#

sure sounds good

north holly Apr 29, 2024, 1:23 PM

#

I also kind of made a mistake where IRQs are non-preemptible like I originally wanted them to

#

(because otherwise the kernel shits itself)

fiery turtle Apr 29, 2024, 5:38 PM

#

jesus madt has so many entry types now

#

took a while to write all of them out

jaunty fox Apr 29, 2024, 5:49 PM

#

are the RISC-V entries official now?

#

i know drwe de vault was very upset that ACPI was being pushed for risc-v

tired crater Apr 29, 2024, 5:52 PM

#

i know i know dumb question but who is acpi talking to when someone says it interacts with firmware? Like does it talk to some seperate controller running some system control code or does it talk to some special device or what does it really do

fiery turtle Apr 29, 2024, 6:09 PM

#

jaunty fox are the RISC-V entries official now?

not yet i think

fiery turtle Apr 29, 2024, 6:09 PM

#

tired crater i know i know dumb question but who is acpi talking to when someone says it inte...

multiple answers:

for more modern osses, mostly the firmware in terms of the embedded controller
for older osses, AML invokes SMM

#

there's a data structure in ACPI NVS that has SMM opcode & args, that is configured and invoked by the AML bytecode

#

then bytecode waits for the smm to reply

#

and propagates the reply to the os etc

tired crater Apr 29, 2024, 6:11 PM

#

I see thanks for clarifying

fiery turtle Apr 29, 2024, 6:12 PM

#

np

lilac knoll Apr 29, 2024, 6:14 PM

#

jaunty fox i know drwe de vault was very upset that ACPI was being pushed for risc-v

lol

#

where was that?

jaunty fox Apr 29, 2024, 6:15 PM

#

lilac knoll where was that?

at FOS.DEM 2023

#

https://archive.fosdem.org/2023/schedule/event/heliosuk/attachments/slides/5402/export/events/attachments/heliosuk/slides/5402/helios.pdf

#

fiery turtle Apr 29, 2024, 6:15 PM

#

i think the riscv structures might get added in acpi 6.6

#

6.5 is 3 years old at this point

fiery turtle Apr 29, 2024, 6:16 PM

#

jaunty fox

why so mad

jaunty fox Apr 29, 2024, 6:16 PM

#

fiery turtle why so mad

being a linux user, it is unlikely he is accustomed to concepts like "suspend" or "power management"

fiery turtle Apr 29, 2024, 6:16 PM

#

lmao

lilac knoll Apr 29, 2024, 6:17 PM

#

probably mad because it makes the kernel complicated

fiery turtle Apr 29, 2024, 6:17 PM

#

yes, for a hobby os like linux a fully static description of the state of the system at boot is totally enough

lilac knoll Apr 29, 2024, 6:17 PM

#

but kernels are complicated so that's a cringe complaint

jaunty fox Apr 29, 2024, 6:17 PM

#

lilac knoll probably mad because it makes the kernel complicated

it's why one ought to use a pleasant library instead

#

granted it won't be doing much for you unless you lean in to what it offers

fiery turtle Apr 29, 2024, 6:18 PM

#

tbh if u use something like uacpi u can implement suspend/resume in like 1000 loc

#

its not crazy difficult

#

plus some assembly for wake vector

jaunty fox Apr 29, 2024, 6:19 PM

#

i would like to try it at some point

fiery turtle Apr 29, 2024, 6:19 PM

#

i might do a PoC example for managarm evetually

#

for S1 suspend u dont even need assembly wake vector or saving ACPI NVS

#

it just resumes from where it left off

#

granted u dont save much power that way

lilac knoll Apr 29, 2024, 6:20 PM

#

jaunty fox

it just triggers me because it reminds me of the casey muratori rant where he talks about hardware complexity

fiery turtle Apr 29, 2024, 6:21 PM

#

yeah id say that complaint is more of a skill issue

jaunty fox Apr 29, 2024, 6:21 PM

#

lilac knoll it just triggers me because it reminds me of the casey muratori rant where he ta...

what's he upset about?

lilac knoll Apr 29, 2024, 6:21 PM

#

he's upset that hardware is complicated

jaunty fox Apr 29, 2024, 6:21 PM

#

"the world was better before branch predictors" sort of take?

lilac knoll Apr 29, 2024, 6:21 PM

#

it's an insane rant

#

let me find the vid though it's 3h long

#

honestly kinda

jaunty fox Apr 29, 2024, 6:22 PM

#

i will not watch a 3hr rant

lilac knoll Apr 29, 2024, 6:22 PM

#

casey's take on acpi would be: 'oh my god shutting down the computer should just be a single memory access!!!!!'

fiery turtle Apr 29, 2024, 6:22 PM

#

jaunty fox i know drwe de vault was very upset that ACPI was being pushed for risc-v

didnt he port a huge library to implement ext4

#

and he complains about acpi complexity

lilac knoll Apr 29, 2024, 6:22 PM

#

https://youtu.be/kZRE7HIO3vk

YouTube

Molly Rocket

The Thirty Million Line Problem

A historical argument for creating a stable instruction set architecture (ISA) for entire system-on-a-chip (SoC) packages.

Please note that although it's never mentioned in the audio there was a typo in the slide deck that makes the situation even more ludicrous when comparing 1990's machine to 2015's: the hard drive line erroneously said "gig"...

▶ Play video

jaunty fox Apr 29, 2024, 6:23 PM

#

fiery turtle didnt he port a huge library to implement ext4

lwext4, it's of a reasonable size

#

it's designed to be low in memory use though

#

i looked at it before

fiery turtle Apr 29, 2024, 6:23 PM

#

how many loc is it

#

i bet its liek 30k

lilac knoll Apr 29, 2024, 6:23 PM

#

lilac knoll https://youtu.be/kZRE7HIO3vk

TL;DR:

'operating systems are very complicated. that's because hardware and drivers are complicated. what if we just hits blunt made hardware not complicated?'

jaunty fox Apr 29, 2024, 6:24 PM

#

it wouldn't work well for a kernel unless you reconstructed a lot of it (drop its custom caching, integrate it with your kernel's, drop its vfs, make it possible to use without holding a global lock around every read/write)

fiery turtle Apr 29, 2024, 6:24 PM

#

yeah from what im seeing its targeting embedded

#

❯ cloc --vcs=git .
      90 text files.
      90 unique files.
       3 files ignored.

github.com/AlDanial/cloc v 1.90  T=0.09 s (916.7 files/s, 305922.9 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
C                               30           3411           2940          14145
C/C++ Header                    32            988           2819           3301
make                             2            191              2            555
CMake                           19             68             29            296
Markdown                         1             45              0            199
YAML                             3              5              1             38
-------------------------------------------------------------------------------
SUM:                            87           4708           5791          18534
-------------------------------------------------------------------------------

#

okay yeah that library is the size of uacpi

hollow elm Apr 29, 2024, 6:26 PM

#

i imagine drew went for it since bunnix is just a unix-like os speedrun

jaunty fox Apr 29, 2024, 6:27 PM

#

the question is why he has abandoned helix

#

his microkernel-based system

fiery turtle Apr 29, 2024, 6:28 PM

#

I also sometiems get tired of complex code and having to do everything properly and just end up wanting to speedrun something like that

jaunty fox Apr 29, 2024, 6:28 PM

#

i fell into that mode and regretted it

#

well

#

i don't regret it as such, it's very inspiring

#

running x11 was fun but i need to make this thing elegant, teachable

fiery turtle Apr 29, 2024, 6:29 PM

#

true

#

garbage just doesnt scale and u get tired of it

jaunty fox Apr 29, 2024, 6:32 PM

#

the benefit apart from the fun of making it run exciting software was perhaps the experience and familiarity it offered

fiery turtle Apr 29, 2024, 6:32 PM

#

makes sense

jaunty fox Apr 29, 2024, 6:32 PM

#

implementing something poorly can be a good prelude to implementing it well if you are too slow to turn out something beautiful the first time

fiery turtle Apr 29, 2024, 6:33 PM

#

oberrow@

fiery turtle Apr 29, 2024, 11:55 PM

#

added a whole bunch of tables

#

anything else people use these days?

hasty plinth Apr 30, 2024, 12:12 AM

#

SLIT would be nice if you've got SRAT, but I don't know how many are actually using those 🤔

fiery turtle Apr 30, 2024, 12:13 AM

#

yeah SLIT should be added def

hollow elm Apr 30, 2024, 12:18 AM

#

there's a pr for lai about adding the tpm tables, but i'm not sure how useful it is or how well that fits

fiery turtle Apr 30, 2024, 12:19 AM

#

any table is useful, i think eventually uacpi should have every table in that header

#

ill take a look at the tpm table

#

Also numa aware allocators in managarm when

hollow elm Apr 30, 2024, 12:21 AM

#

:^)

rustic compass Apr 30, 2024, 7:14 AM

#

fiery turtle anything else people use these days?

I will give you a wishlist once i start working on things that need tables

wind fiber Apr 30, 2024, 7:54 AM

#

jaunty fox

I remember that, I was there. Pretty sure I attended this specific talk lol

jaunty fox Apr 30, 2024, 11:05 AM

#

fiery turtle yeah SLIT should be added def

i will be using the SLIT table eventually

#

from time to time i get interested in extreme scaling on NUMA machines

#

so i want SLIT and SRAT

fiery turtle Apr 30, 2024, 1:01 PM

#

fiery turtle I'm adding core API that should be used during initial namespace device discover...

alright that's done

#

I made a nice namespace dumper in the test runner

#

e.g. heres a dump of pitust's chromebook namespace:

📎 message.txt

#

#

me when GOOG000A ps/2 keyboard

tired crater Apr 30, 2024, 1:22 PM

#

fiery turtle I made a nice namespace dumper in the test runner

Stealing this immediately

fiery turtle Apr 30, 2024, 1:30 PM

#

tired crater Stealing this immediately

its not upstreamed yet lol

#

but i will

rain nimbus Apr 30, 2024, 1:31 PM

#

fiery turtle me when GOOG000A ps/2 keyboard

That's the most HID of all time

fiery turtle Apr 30, 2024, 10:13 PM

#

fiery turtle e.g. heres a dump of pitust's chromebook namespace:

added a new --ns-dump switch for the emulation mode that makes a dump like that at the end

loud ice Apr 30, 2024, 10:14 PM

#

fiery turtle me when GOOG000A ps/2 keyboard

your ps/2 keyboard isn't of the "google" brand?

fiery turtle Apr 30, 2024, 10:47 PM

#

@hollow elm do u remember the argument that i made for why its difficult to implement proper while timeouts?

#

im thiking abut how to design it now and i dont see any problems

#

liek i dont remember what the problem was last time

hollow elm Apr 30, 2024, 10:48 PM

#

that the aml can trigger calls to kernel functions and those take an unknown amount of time?

fiery turtle Apr 30, 2024, 10:48 PM

#

hmm i dont think so?

#

well like an Event wait i guess

#

but thats not related to While timeouts

#

i think

#

anyway i added a start_ticks to every While codeblock, now i just check it at the beginning of that While

#

either when Continue or the end of block

hollow elm Apr 30, 2024, 10:56 PM

#

nice

#

but yeah other than that i can't remember either

fiery turtle Apr 30, 2024, 11:20 PM

#

#

Yay

vagrant hull Apr 30, 2024, 11:20 PM

#

you really should make a uACPI test image

fiery turtle Apr 30, 2024, 11:21 PM

#

managarm caret

vagrant hull Apr 30, 2024, 11:21 PM

#

i wanted to test it on this new board i got but running managarm is tedious lol

#

yeah, tedious

fiery turtle Apr 30, 2024, 11:21 PM

#

yeah but a proper test image that supports everything uacpi needs will be an absolute ton of work

vagrant hull Apr 30, 2024, 11:21 PM

#

is the latest CI image fine?

#

of managarm

fiery turtle Apr 30, 2024, 11:21 PM

#

otherwise it would be a piece of crap

fiery turtle Apr 30, 2024, 11:21 PM

#

vagrant hull is the latest CI image fine?

yeah

vagrant hull Apr 30, 2024, 11:21 PM

#

alright

#

i'll test using that

fiery turtle Apr 30, 2024, 11:21 PM

#

give it a shot

#

if the latest ci image is like within the last week

vagrant hull Apr 30, 2024, 11:22 PM

#

it's from yesterday

fiery turtle Apr 30, 2024, 11:22 PM

#

thats good enough!

fiery turtle Apr 30, 2024, 11:22 PM

#

vagrant hull i wanted to test it on this new board i got but running managarm is tedious lol

whats the board?

#

like a new pc motherboard?

vagrant hull Apr 30, 2024, 11:23 PM

#

yeah

#

it's an MSI PRO B650M-B

#

i got it because it's one of the few AM5 motherboards with separate PS/2 keyboard and mouse

#

(it works well btw)

fiery turtle Apr 30, 2024, 11:24 PM

#

nice

vagrant hull Apr 30, 2024, 11:40 PM

#

ok i tried it

#

it seems to work

#

it gets to "waiting for /dev/sda0" and gets stuck there (of course it does)

#

but pressing the power button makes the machine shut down

fiery turtle Apr 30, 2024, 11:41 PM

#

yeah if it says waiting for sda it made it to userspace

#

and xhci died probably

fiery turtle Apr 30, 2024, 11:41 PM

#

vagrant hull but pressing the power button makes the machine shut down

nice

#

Fuck, never mind

#

The way While works right now is the code block is re-created every time when it gets jumped to

#

so the context of the while that was just being executed is lost

#

this will need some refactoring

lilac knoll Apr 30, 2024, 11:53 PM

#

vagrant hull it gets to "waiting for /dev/sda0" and gets stuck there (of course it does)

managarm moment

verbal herald May 1, 2024, 12:08 AM

#

vagrant hull it gets to "waiting for /dev/sda0" and gets stuck there (of course it does)

live usb moment

fiery turtle May 1, 2024, 12:09 AM

#

Managarm does have USB drivers btw

verbal herald May 1, 2024, 12:09 AM

#

yes but it always default to sda0

#

not sdb0 or something

fiery turtle May 1, 2024, 12:10 AM

#

Thats not why it doesn't work tho

verbal herald May 1, 2024, 12:11 AM

#

sda0 usually corresponds to sata

#

with sdb0 being usb

#

shouldn't there be a boot option to use sdb0 instead like linux

fiery turtle May 1, 2024, 12:12 AM

#

It corresponds to SATA only if u have a SATA drive

#

And thats not a problem in managarm anyway

#

It can boot like this too

verbal herald May 1, 2024, 12:12 AM

#

yes but it always defaults to sda0 all the time

#

which is sata and not usb

#

if you have a sata drive

#

idk what the drive name is if you have nvme

#

maybe just like linux where its nvme0p0

#

so unless you only have usb, it always defauts to sata or whatever

hollow elm May 1, 2024, 12:18 AM

#

i mean ideally we'd just run udev in init-stage1, let it set up /dev/disk/by-uuid and just mount by uuid using symlinks from there

fiery turtle May 1, 2024, 12:19 AM

#

Yup

hollow elm May 1, 2024, 12:20 AM

#

shouldn't even be that hard to do now, since the /dev/disk/by-uuid part was already working, and i recently merged some old patch that removes the size limitation for the initramfs

fiery turtle May 1, 2024, 10:46 AM

#

tired crater Stealing this immediately

https://github.com/UltraOS/uACPI/blob/1b1356523e164fd4c69239f9fa87509331d6a69b/tests/runner/test_runner.cpp#L87

tired crater May 1, 2024, 11:14 AM

#

fiery turtle https://github.com/UltraOS/uACPI/blob/1b1356523e164fd4c69239f9fa87509331d6a69b/t...

fiery turtle May 1, 2024, 1:36 PM

#

fiery turtle this will need some refactoring

nope, i think i managed to implement it in very few loc

#

the idea is to preserve the previous while expiration and code offset before re-evaluating it

#

Here's a blob that infinitely polls some HW register

#

previously it would run forever

#

And another one from @vast kestrel

hollow elm May 3, 2024, 12:26 AM

#

okay so i tried managarm on my laptop

#

and uacpi initialized fine, and there were even some queries happening and the aml was doing some notification (thermal zone related judging by the name), but unfortunately the power button didn't work

hollow elm May 3, 2024, 1:15 AM

#

unfortunately i can't check if the uacpi init just failed since i didn't get into weston

#

looking at a disassembly of the acpi tables, it seems to use the new kind of power button

#

the acpi tables also apparently contain a (partial) list of bios settings? https://uwu.foundation/9SVYeaT.png https://uwu.foundation/7CvN1gC.png

#

oh it seems to be for WMI

fiery turtle May 3, 2024, 5:46 AM

#

hollow elm oh it seems to be for WMI

Yeah this looks like WMI

#

Does it have a power button object in the namespace?

#

Would be interesting to see full logs ofc

#

Also pls send the dump

fiery turtle May 3, 2024, 7:48 AM

#

could also be related to _INI order, which I recently changed (but not yet merged into managarm)

#

so try git fetch && git checkout origin/master in the mangarm ports (although that will break find_by_signature a bit)

hollow elm May 3, 2024, 10:15 AM

#

fiery turtle Does it have a power button object in the namespace?

it does yeah

hollow elm May 3, 2024, 10:15 AM

#

fiery turtle could also be related to _INI order, which I recently changed (but not yet merge...

will try, and i can push the changes

fiery turtle May 3, 2024, 10:30 AM

#

that would be great, and a dsdt dump would be awesome too

hollow elm May 3, 2024, 10:51 AM

#

yeah i can post it later

fiery turtle May 3, 2024, 11:22 AM

#

nice

hollow elm May 3, 2024, 12:13 PM

#

📎 tables.tar.xz

fiery turtle May 3, 2024, 12:18 PM

#

@hollow elm is there an ECDT table as well?

hollow elm May 3, 2024, 12:19 PM

#

no ecdt, linux says ```
[ 0.964029] ACPI: _SB_.PCI0.SBRG.EC0_: Boot DSDT EC used to handle transactions

fiery turtle May 3, 2024, 12:19 PM

#

okay hmm

#

and when you press the power button no query is produced? @hollow elm

hollow elm May 3, 2024, 12:20 PM

#

nop, nothing happens

fiery turtle May 3, 2024, 12:21 PM

#

okay yeah its definitely the AML thinking we're a legacy os

hollow elm May 3, 2024, 12:21 PM

#

there are some queries happening, and a notify that goes ignored for _TZ_.THRM, but nothing for the power button

fiery turtle May 3, 2024, 12:21 PM

#

do the fn+whatever keys do anything?

hollow elm May 3, 2024, 12:22 PM

#

hm good q, didn't think to try them

#

mmm i hate usb

#

qemu-system-x86_64 -m 2G -device qemu-xhci -device usb-host,vendorid=0x13fe,productid=0x4200 boots, but on my laptop the usb msd driver doesn't seem to pick the usb stick up

fiery turtle May 3, 2024, 12:29 PM

#

lol

fiery turtle May 3, 2024, 12:29 PM

#

fiery turtle do the fn+whatever keys do anything?

also lid and AC

hollow elm May 3, 2024, 12:29 PM

#

ac i can't test because the battery is kinda dead meme

fiery turtle May 3, 2024, 12:30 PM

#

anyway its definitely a result of misconfiguration

#

could be order thats currently in managarm

#

could be some opregion is read incorrectly

#

they use indexfields a lot in your laptop

#

can you give the latest uacpi master a try before i do more debugging pls

hollow elm May 3, 2024, 12:31 PM

#

yeah ofc

fiery turtle May 3, 2024, 12:31 PM

#

thanks

#

also pls check that there's 100% no ECDT there ls /sys/firmware/acpi/tables | grep ECDT

#

because we had a similar issue in qwinci's laptop and that was because of that

hollow elm May 3, 2024, 12:41 PM

#

qookie@selenium ~ λ ls /sys/firmware/acpi/tables | grep ECDT
qookie@selenium ~ λ ls /sys/firmware/acpi/tables
APIC  CRAT  dynamic  FIDT  IVRS   SSDT10  SSDT13  SSDT2  SSDT5  SSDT8  VFCT
BGRT  data  FACP     FPDT  MCFG   SSDT11  SSDT14  SSDT3  SSDT6  SSDT9  WSMT
CDIT  DSDT  FACS     HPET  SSDT1  SSDT12  SSDT15  SSDT4  SSDT7  TPM2
qookie@selenium ~ λ

fiery turtle May 3, 2024, 12:41 PM

#

yeah ok

hollow elm May 3, 2024, 12:41 PM

#

/sys/firmware/acpi/tables/dynamic/ and /sys/firmware/acpi/tables/data/ are empty

fiery turtle May 3, 2024, 12:41 PM

#

because it does feature detection in the _REG method

#

just like qwinci

#

so if thats called early it will set the bare minimum features

#

but if there's no ECDT that should be correct

fiery turtle May 3, 2024, 1:01 PM

#

yeah basically the power button is unconditionally notified for query 28

            Method (_Q28, 0, NotSerialized)  // _Qxx: EC Query, xx=0x00-0xFF
            {
                Notify (PWRB, 0x80) // Status Change
            }

#

so its 100% EC programmed incorrectly by AML

#

but why is that I currently have no idea

hollow elm May 3, 2024, 2:25 PM

#

i come bearing new findings

fiery turtle May 3, 2024, 2:28 PM

#

Please share!

#

Lol

hollow elm May 3, 2024, 2:29 PM

#

Fn+F2 (brightness down) and Fn+F3 (brightness up) trigger ec queries 16 and 17 respectively, which end up notifying the vga acpi node (which has no listeners), and this only works once (and io-apic irq 7 is left unacknowledged so perhaps it requires the host's ps/2 driver to work),
at random times query 128 happens, which notifies the thermal zone, but again no listeners,
opening and closing the lid triggers query 138, which notifies the lid object, but again no listeners,
holding the power button for a while (but not long enough that it triggers a hard power off, letting go just before it shuts down) triggers query 153, which does nothing (?),
short pressing the power button does nothing

fiery turtle May 3, 2024, 2:30 PM

#

Yeah no listeners just means nothing in managarm hooked it up

#

Long press shutting down makes sense

#

Because it invokes smm

#

But yeah what you said confirms my theory

hollow elm May 3, 2024, 2:31 PM

#

i'd expect it to be implemented in the chipset itself, but the query just before the hard shutdown happens is interesting

fiery turtle May 3, 2024, 2:31 PM

#

Just need to figure out why it thinks were below windows 7

fiery turtle May 3, 2024, 2:32 PM

#

hollow elm i'd expect it to be implemented in the chipset itself, but the query just before...

Its like an emergency notification

hollow elm May 3, 2024, 2:32 PM

#

yeah but it doesn't seem to notify anything

fiery turtle May 3, 2024, 2:32 PM

#

Because managarm doesn't install any notify handlers for those devices

#

Its supposed to be listentened to by a host driver

hollow elm May 3, 2024, 2:32 PM

#

i mean, there was no warning saying there are no listeners

fiery turtle May 3, 2024, 2:33 PM

#

What device was it notifying?

hollow elm May 3, 2024, 2:33 PM

#

oh _Q99 is literally a noop ```
Method (_Q99, 0, NotSerialized) // _Qxx: EC Query, xx=0x00-0xFF
{
}

fiery turtle May 3, 2024, 2:33 PM

#

Lol

#

Didn't hook it up I guess

hollow elm May 3, 2024, 2:34 PM

#

forgor

fiery turtle May 3, 2024, 2:34 PM

#

hollow elm i come bearing new findings

Is this latest master?

hollow elm May 3, 2024, 2:35 PM

#

yes

#

or well latest as of like an hour and a half ago if you pushed anything

fiery turtle May 3, 2024, 2:35 PM

#

Nah I didn't

hollow elm May 3, 2024, 2:36 PM

#

fiery turtle Just need to figure out why it thinks were below windows 7

huh does windows vista and below not deserve the power button?

fiery turtle May 3, 2024, 2:36 PM

#

I think it doesn't support Notify interface for it

hollow elm May 3, 2024, 2:39 PM

#

god bless hp ```
Device (PS2K)
{
Method (_HID, 0, NotSerialized) // _HID: Hardware ID
{
If (W8OS ())
{
Return ("*HPQ8001")
}
Else
{
Return ("PNP0303")
}
}

fiery turtle May 3, 2024, 2:40 PM

#

Lmao

vast kestrel May 3, 2024, 2:40 PM

#

WeightOS

hollow elm May 3, 2024, 2:40 PM

#

ig they made a custom driver for win8+?

#

what for i don't want to even think

fiery turtle May 3, 2024, 2:41 PM

#

They did

vast kestrel May 3, 2024, 2:41 PM

#

EC fun

fiery turtle May 3, 2024, 2:41 PM

#

No idea why

vast kestrel May 3, 2024, 2:41 PM

#

most likely custom memes they do which goes over the PS2 interface

hollow elm May 3, 2024, 2:41 PM

#

also i wonder, do firmware developers have some special asl compiler that accepts longer names and just shortens them or do they actually suffer through 4 character names

vast kestrel May 3, 2024, 2:42 PM

#

afaik they only use 4 letter names

hollow elm May 3, 2024, 2:42 PM

#

ig the fact things like W7OS are even legible makes that pretty clear

vast kestrel May 3, 2024, 2:43 PM

#

WsevenOS

fiery turtle May 3, 2024, 2:43 PM

#

hollow elm also i wonder, do firmware developers have some special asl compiler that accept...

They actually do

#

There isn't any extension or anything

vast kestrel May 3, 2024, 2:43 PM

#

I mean, they could have made a simple custom preprocessor

hollow elm May 3, 2024, 2:43 PM

#

yeah i was thinking as a non-standard thing on top of asl

vast kestrel May 3, 2024, 2:44 PM

#

that just hex(os.urandom(2))

hollow elm May 3, 2024, 2:44 PM

#

that'd make debugging annoying though ig

#

unless you teach the os about the random 4 characters -> name mapping

vast kestrel May 3, 2024, 2:45 PM

#

The thing I wonder about is how they make the WMI memes

hollow elm May 3, 2024, 2:46 PM

#

the wmi bios settings list made me curious

#

because it lists things not exposed in the actual bios

fiery turtle May 3, 2024, 2:47 PM

#

hollow elm because it lists things not exposed in the actual bios

Could be leftover from other laptops

vast kestrel May 3, 2024, 2:47 PM

#

it allows windows (or well, HP prop programs) to change the settings

hollow elm May 3, 2024, 2:47 PM

#

            Package (0x0F)
            {
                "Manufacturing Programming Mode Counter", 
                "Place holder ", 
                "\\Place holder for Path", 
                One, 
                Zero, 
                One, 
                0x0376, 
                0x03, 
                " ", 
                " ", 
                " ", 
                Zero, 
                Zero, 
                0x0A, 
                Zero
            }

vast kestrel May 3, 2024, 2:47 PM

#

Place holder for Path

#

that might actually be a read-only counter and not necesarily a write

hollow elm May 3, 2024, 2:48 PM

#

probably yeah but still kinda interesting

left orbit May 3, 2024, 2:53 PM

#

that is some cursed ass aml

#

but i don't think one should expect laptop firmware manufacturers to put in much effort and/or think too hard

keen marsh May 3, 2024, 2:57 PM

#

If my kernel does not yet support SMP can i just declare the functions from the api headers for locking as always returning true?

jaunty fox May 3, 2024, 2:58 PM

#

keen marsh If my kernel does not yet support SMP can i just declare the functions from the ...

they are supposed to sleepable locks, i think

#

so they are still important

keen marsh May 3, 2024, 2:59 PM

#

Yeah but isnt the sleeping in case they were already locked? ("which cant happen")

left orbit May 3, 2024, 3:00 PM

#

i think the return value specifies whether you could acquire it before the timeout

jaunty fox May 3, 2024, 3:00 PM

#

keen marsh Yeah but isnt the sleeping in case they were already locked? ("which cant happen...

as in they're not supposed to make the thread calling it nonpreemptible

fiery turtle May 3, 2024, 3:00 PM

#

keen marsh If my kernel does not yet support SMP can i just declare the functions from the ...

Should be fine

jaunty fox May 3, 2024, 3:00 PM

#

i defer to the author's advice

fiery turtle May 3, 2024, 3:00 PM

#

As long as you return false

#

If its already acquired

fiery turtle May 3, 2024, 3:01 PM

#

jaunty fox so they are still important

Well technically yes

#

In reality probably won't matter

hollow elm May 3, 2024, 3:01 PM

#

hollow elm ``` Package (0x0F) { "Manufacturing Prog...

            If (SRCM (MFLG, "MPMLock", 0x07))
            {
                If (SCMP (Local0, "Serial Number"))
                {
                    DerefOf (N00D [Arg0]) [0x03] = One
                }

                If (SCMP (Local0, "Product Name"))
                {
                    DerefOf (N00D [Arg0]) [0x03] = One
                }

                If (SCMP (Local0, "UUID"))
                {
                    DerefOf (N00D [Arg0]) [0x03] = One
                }

#

curious

#

SRCM is a strncmp afaics

keen marsh May 3, 2024, 3:02 PM

#

fiery turtle As long as you return false

So i have to implement the locks?

#

And cant just return "aquiring lock was successful" every time

hollow elm May 3, 2024, 3:03 PM

#

hollow elm ``` If (SRCM (MFLG, "MPMLock", 0x07)) { ...

hmm and MFLG is just a 64 bit field at 0xFF007000

#

idk why i am digging into this but this is very interesting

fiery turtle May 3, 2024, 3:08 PM

#

keen marsh And cant just return "aquiring lock was successful" every time

Well

#

I cant promise it won't break, but for truly single threaded workflows yea

#

Since aml mutexes are recursive anyway

hollow elm May 3, 2024, 3:08 PM

#

hollow elm idk why i am digging into this but this is very interesting

lol https://uwu.foundation/7ofXxot.png

hollow elm May 3, 2024, 3:10 PM

#

fiery turtle I cant promise it won't break, but for truly single threaded workflows yea

couldn't you have issues if you do normal acpi work on one thread, and posted work on a separate thread, and query execution is posted from the irq handler?

#

they only mentioned not having SMP

fiery turtle May 3, 2024, 3:11 PM

#

hollow elm couldn't you have issues if you do normal acpi work on one thread, and posted wo...

Yeah of course

fiery turtle May 3, 2024, 3:12 PM

#

hollow elm ``` If (SRCM (MFLG, "MPMLock", 0x07)) { ...

Lol

#

Okay so I would definitely debug this if I had the hardware, there are lots of things to check or dump, but idk how interested you are in that

#

uACPI doesn't have a way to override dsdt but would be interesting to recompile this with debug prints

#

And e.g force write WIN7 in EC as 1

#

Etc

#

Tons of things you could do

hollow elm May 3, 2024, 3:14 PM

#

i'd be up for some debugging

#

i also need to figure out the parameter error i'm getting in the xhci driver when configuring the endpoint

#

but for that i think i have an idea from when i was talking with nooby about his xhci driver

fiery turtle May 3, 2024, 3:14 PM

#

Yeah if we had working xhci we could also properly inspect logs

#

Right now I guess thats impossible as well

hollow elm May 3, 2024, 3:15 PM

#

i am forgetting to set some fields in the device contexts :^)

hollow elm May 3, 2024, 3:15 PM

#

fiery turtle Right now I guess thats impossible as well

well we could just rearrange init and the initramfs a bit to put dmesg and less on it

#

and the ps2 driver

#

although not sure if less would work on the kernel console :^)

fiery turtle May 3, 2024, 3:16 PM

#

Lol yeah

#

Idk how badly this xhci init is broken to tell

hollow elm May 3, 2024, 3:17 PM

#

hollow elm i am forgetting to set some fields in the device contexts :^)

i can try fixing this and seeing if it helps

#

iirc i am missing the ESIT and polling interval fields (oops meme )

#

ofc qemu doesn't care about the polling interval

#

even for hid

fiery turtle May 3, 2024, 3:17 PM

#

Oh if you can fix it that would be amazing, if not we can come up with something else

#

@hollow elm and pressing the power button does do something for your Linux right?

hollow elm May 3, 2024, 3:19 PM

#

yeah on linux it is handled and passed on to user space

fiery turtle May 3, 2024, 3:20 PM

#

Ok ok just checking who knows

hollow elm May 3, 2024, 3:20 PM

#

and triggers the kde shutdown prompt

fiery turtle May 3, 2024, 3:20 PM

#

Yeah makes sense

#

Also I think we unconditionally enable the fixed event for the power button (without checking fadt), maybe this somehow messes with the ec

#

Could comment that out but I doubt thats it

hollow elm May 3, 2024, 3:31 PM

#

i hate the xhci spec

vast kestrel May 3, 2024, 3:32 PM

#

its fine™️

hollow elm May 3, 2024, 3:32 PM

#

"Refer to section 6.2.3.8 for more information."
the 6.2.3.8 is a link, click it, get taken to the section
"Refer to section 4.14.2 for the definition of an “ESIT” [...]"
the 4.14.2 is not a link

#

at the very least 4.14.2 is the correct section

#

there are some references that are off because they added stuff inbetween

fiery turtle May 3, 2024, 3:34 PM

#

Lmao

hollow elm May 3, 2024, 3:39 PM

#

here's an example: in table 6-21:

TD Size. This field provides an indicator of the number of packets remaining in the TD. Refer to
section 4.10.2.4 for how this value is calculated.
section 4.10.2.4: "4.10.2.4 Babble Detected Error"
section 4.11.2.4: "4.11.2.4 TD Size"

fiery turtle May 3, 2024, 3:42 PM

#

Lol

vast kestrel May 3, 2024, 6:35 PM

#

Yeah I had that exact thing happen lol

#

Like exactly with TD size

fiery turtle May 4, 2024, 11:05 AM

#

Added very detailed examples on how to tie acpi into the driver system and enumerate the namespace https://wiki.osdev.org/UACPI#Namespace_Enumeration_.26_Finding_Devices

fiery turtle May 4, 2024, 11:43 AM

#

Method (^BN00, 0, NotSerialized)
{
    Return (Zero)
}

Method (_BBN, 0, NotSerialized)  // _BBN: BIOS Bus Number
{
    Return (BN00 ())
}

least insane AML

verbal herald May 5, 2024, 5:09 AM

#

prob could reuse the code from https://github.com/kora-org/ydin/blob/rewrite-old for an uacpi test kernel

GitHub

GitHub - kora-org/ydin at rewrite-old

A very extendable and versatile hybrid kernel. Contribute to kora-org/ydin development by creating an account on GitHub.

#

slightly rewriting it too with codes from the main zig version translated into c

left orbit May 5, 2024, 5:25 AM

#

managarm is a thing already

#

its much more advanced, basically ruling out kernel being broken at this point

#

no need to maintain a test kernel in addition to uacpi

hasty plinth May 5, 2024, 8:16 AM

#

is there a particular reason detecting pointer size is done this way?

#

as opposed to something like sizeof(void*)?

#

Im asking because it causes my riscv tests to fail nooo

fiery turtle May 5, 2024, 8:49 AM

#

hasty plinth is there a particular reason detecting pointer size is done [this way](https://g...

Probably because sizeof is unusable from macros contexts

#

I dont even remember where this is used, id have to check

fiery turtle May 5, 2024, 8:50 AM

#

hasty plinth Im asking because it causes my riscv tests to fail <:nooo:750466870993223750>

Riscv should be added to the list definitely

hasty plinth May 5, 2024, 8:51 AM

#

fiery turtle Probably because sizeof is unusable from macros contexts

ah I didnt know that, dont do a lot of macro stuff

#

but yeah, riscv support would be helpful (for me lol)

fiery turtle May 5, 2024, 8:52 AM

#

Cant you just define it explicitly for the time being

#

It checks ifndef

hasty plinth May 5, 2024, 8:53 AM

#

ah it does, nice

fiery turtle May 5, 2024, 8:53 AM

#

Ye

hasty plinth May 5, 2024, 9:05 AM

#

fwiw thats the only thing that prevents it from building on riscv

#

cant say if it runs just yet because clang generates relocations I havent implemented 😅

fiery turtle May 5, 2024, 9:06 AM

#

Nice

#

Oh also define UACPI_REDUCED_HARDWARE 1

#

For riscv

jaunty fox May 5, 2024, 9:07 AM

#

very kind of you both to explore risc-v with uacpi

fiery turtle May 5, 2024, 9:08 AM

#

fiery turtle Oh also define UACPI_REDUCED_HARDWARE 1

I think that's the define anyway

jaunty fox May 5, 2024, 9:08 AM

#

i would like to work on a risc-v keyronex against but i decided to wait until risc-v acpi is a bit more established (you needed a custom qemu for it to be any use, last i worked on it)

fiery turtle May 5, 2024, 9:08 AM

#

jaunty fox very kind of you both to explore risc-v with uacpi

Yeah thats nice for sure

#

Yeah riscv stuff is not yet officially there

hasty plinth May 5, 2024, 9:08 AM

#

tbh its a totally selfish goal for me, I want my kernel to be as uniform as it can - will see how that goes when it gets uacpi loaded 😛

#

yeah riscv/acpi is a huge WIP

hasty plinth May 5, 2024, 9:09 AM

#

fiery turtle Oh also define UACPI_REDUCED_HARDWARE 1

will do, thank you 🙂

fiery turtle May 5, 2024, 9:10 AM

#

This strips out huge chunk of code for supporting full acpi hardware

#

Its not mandatory to disable it ofc

#

Its autodetected otherwise

jaunty fox May 5, 2024, 9:11 AM

#

i had a quick google search for risc-v acpi and someone is coping and seething about it

#

you posted cringe mdp_cs

fiery turtle May 5, 2024, 9:11 AM

#

Giant skill issue

jaunty fox May 5, 2024, 9:11 AM

#

there's more to it than acpi.sys and acpi-ca

#

the "easier to implement drivers for each power IC" bit also led me down the garden path, i thought he was being sarcastic at first

fiery turtle May 5, 2024, 9:12 AM

#

This is just wrong on so many levels

jaunty fox May 5, 2024, 9:13 AM

#

no one will ever deny acpi is complex, but such is the price of flexibility

fiery turtle May 5, 2024, 9:14 AM

#

True

jaunty fox May 5, 2024, 9:14 AM

#

it's a far cry from a format loosely inspired by Open Firmware's device tree but turned into a static blob

fiery turtle May 5, 2024, 9:14 AM

#

Yeah

fiery turtle May 5, 2024, 9:22 AM

#

left orbit no need to maintain a test kernel in addition to uacpi

Basically yeah, and features like deferred execution from an interrupt handler are really difficult to get right from scratch

#

Which you must have for the full package

#

At this point the only bad thing about managarm is poor driver stability

#

But that dies in userspace luckily

lilac knoll May 5, 2024, 11:47 AM

#

fiery turtle Probably because sizeof is unusable from macros contexts

why?

fiery turtle May 5, 2024, 11:49 AM

#

try #if sizeof(void*) == 4

lilac knoll May 5, 2024, 11:50 AM

#

ah

#

why do you need that though

tired crater May 5, 2024, 11:51 AM

#

Because pointer size varies

mortal yoke May 5, 2024, 11:51 AM

#

fiery turtle try `#if sizeof(void*) == 4`

why not #if UINTPTR_MAX == UINT32_MAX?

tired crater May 5, 2024, 11:51 AM

#

Have you tested on 32 bit machines copyobject abuser

mortal yoke May 5, 2024, 11:51 AM

#

that would be pretty portable

#

(between 32-bit and 64-bit)

lilac knoll May 5, 2024, 11:52 AM

#

tired crater Because pointer size varies

you don't generally need to branch on this

fiery turtle May 5, 2024, 11:52 AM

#

mortal yoke why not `#if UINTPTR_MAX == UINT32_MAX`?

because that means im relying on some other system header

#

that would have to be replacable by the user etc

fiery turtle May 5, 2024, 11:52 AM

#

tired crater Have you tested on 32 bit machines copyobject abuser

like 32-bit only pre amd64? no, i dont have one that old

#

but uacpi is tested in 32-bit mode in normal ci

mortal yoke May 5, 2024, 11:53 AM

#

fiery turtle because that means im relying on some other system header

I mean you are already using stdint.h in types.h (that can be overwritten by the user as uacpi_types.h)

fiery turtle May 5, 2024, 11:53 AM

#

are those defined in stdint.h?

mortal yoke May 5, 2024, 11:53 AM

#

yes

fiery turtle May 5, 2024, 11:53 AM

#

ig i could do that

#

but then again im not sure thats better than checking the architecture

mortal yoke May 5, 2024, 11:55 AM

#

at least it would work on more architectures without you needing to add them there

#

other than that probably not

fiery turtle May 5, 2024, 11:55 AM

#

mortal yoke at least it would work on more architectures without you needing to add them the...

that is true, yeah

#

but there's only so many architectures with acpi

#

itanium, x86, arm, and risc-v

vast kestrel May 5, 2024, 12:28 PM

#

left orbit managarm is a thing already

Managarm has a pretty big image and it doesn't actually boot on a ton of hardware, so it's not that useful as a debug image....

#

Like, even if the acpi init pass can't really test much when it dies somewhere else down the chain

fiery turtle May 5, 2024, 12:30 PM

#

vast kestrel Managarm has a pretty big image and it doesn't actually boot on a ton of hardwar...

it does boot almost everywhere, but drivers die in userspace so its not a big deal

vast kestrel May 5, 2024, 12:31 PM

#

Does it reach deep enough for testing power button and stuff?

fiery turtle May 5, 2024, 12:31 PM

#

boot as in reach userspace

#

yeah

vast kestrel May 5, 2024, 12:31 PM

#

It doesn't on my PC

fiery turtle May 5, 2024, 12:31 PM

#

thats in the kernel so that almost always works

vast kestrel May 5, 2024, 12:31 PM

#

It dies on USB

hollow elm May 5, 2024, 12:31 PM

#

usb is in userspac

#

e

vast kestrel May 5, 2024, 12:31 PM

#

And on my laptop it dies before even reaching the kernel

fiery turtle May 5, 2024, 12:31 PM

#

vast kestrel It dies on USB

that shouldnt prevent the power button from working

vast kestrel May 5, 2024, 12:31 PM

#

Idk it's in a panic state

#

anyways it's still not a nice test image for quickly checking a bunch of functionality AMshrug

fiery turtle May 5, 2024, 12:33 PM

#

well a proper test image would need a full blown kernel with a sophisticated scheduler, so thats like 10k loc of code

vast kestrel May 5, 2024, 12:33 PM

#

Do you need a sophisticated scheduler tho

fiery turtle May 5, 2024, 12:33 PM

#

can't test events otherwise

vast kestrel May 5, 2024, 12:34 PM

#

I can take the core of my C# kernel very easily

#

It has a scheduler with events, it has a simple memory management

fiery turtle May 5, 2024, 12:35 PM

#

well if it has a higher success rate at booting than managarm and you actually port uacpi there then that can work as a test image as well caret

vast kestrel May 5, 2024, 12:35 PM

#

I might try it if I ever get free time :^)

fiery turtle May 5, 2024, 12:35 PM

#

that would be great

hollow crescent May 5, 2024, 12:36 PM

#

jaunty fox i had a quick google search for risc-v acpi and someone is coping and seething a...

much easier for an OS to implement drivers for each power IC

hell naw

#

no way he said that

jaunty fox May 5, 2024, 12:44 PM

#

hollow crescent no way he said that

that's why i thought he was being sarcastic, but apparently not

fiery turtle May 5, 2024, 1:01 PM

#

finally got around to extending the test suite to run the 500 publically available acpi dumps

#

#

a lot of them timeout because they infinitely poll some MMIO or IO regs

#

thats where my loop timeout detection PR comes in as well

#

hollow elm May 5, 2024, 1:03 PM

#

what's the 20 fails, timeouts?

fiery turtle May 5, 2024, 1:03 PM

#

yeah, all of them are timeouts

hollow elm May 5, 2024, 1:04 PM

#

nice

fiery turtle May 5, 2024, 1:04 PM

#

#

Unfortunately 150 of these 500 don't have the AML to extract for DSDT

#

only its disassembled representation

#

which might not be compilable

#

or at least its not possible to even extract with any existing tools

#

so thats kinda unfortunate

#

#

fuuck this doesnt work on windows lol

#

ill have to modify the repo

#

ok that was just one file, great

#

okay so with the timeout branch only 3 are failing

#

and its really annoying blobs

#

Method (RR_M, 1, Serialized)
{
    If ((Acquire (MUEC, 0xFFFF) == Zero))
    {
        Local0 = Arg0
        Local1 = (Local0 & 0xFF)
        Local0 >>= 0x08
        Local0 &= 0xFF
        CDT2 = Local1
        CDT1 = Local0
        CMD1 = 0x80
        ESEM = 0x0F
        Local0 = 0x7F
        While ((Local0 && CMD1))
        {
            Sleep (One)
            Local0--
        }

        If ((CMD1 == Zero))
        {
            Local0 = EDA1 /* \_SB_.PC00.LPCB.EC0_.EDA1 */
        }
        Else
        {
            Local0 = Ones
        }

        Release (MUEC)
        Return (Local0)
    }

    Return (Ones)
}

#

it does 127 sleeps for 1 ms

#

and this shit is called a lot of times

#

so the blob takes longer than 10 seconds to load and test runner thinks it timed out

north holly May 5, 2024, 1:39 PM

#

Firmware developers are very smart as we can see

fiery turtle May 5, 2024, 1:44 PM

#

#

1 second total sleep time in the end

#

hmm on linux tihs only takes 1.8 seconds to load

#

runner/build-linux-64bits/test-runner   0.33s user 0.25s system 31% cpu 1.837 total

#

with UBSAN+ASAN

#

why is it so slow on windows

#

std::this_thread::sleep_for(std::chrono::milliseconds(msec));

#

like

#

maybe this takes longer on windows for some reason

#

#

Yayy all passes if i port my while timeout commit

#

thats with ubsan and asan

#

the only problem is on windows where sleep takes longer for some reason in debug mode

rustic compass May 5, 2024, 2:01 PM

#

maybe allow longer timouts in debug mode?

fiery turtle May 5, 2024, 2:02 PM

#

the timeout is in the python test runner

#

i could bump it, yeah

#

still annoying

#

its currently set to 10 seconds

hollow crescent May 5, 2024, 4:22 PM

#

fiery turtle a lot of them timeout because they infinitely poll some MMIO or IO regs

man that sucks

hollow crescent May 5, 2024, 4:22 PM

#

fiery turtle ``` Method (RR_M, 1, Serialized) { If ((Acquire (MUEC, 0xFFFF) == Zero)) ...

Is this decompiled from AML?

fiery turtle May 5, 2024, 4:37 PM

#

hollow crescent Is this decompiled from AML?

Yea

fiery turtle May 5, 2024, 8:00 PM

#

me when firmware developers

#

and their error paths

#

Method (_CRS, 0, NotSerialized)  // _CRS: Current Resource Settings
{
    If ((TPTY == One))
    {
        Name (SBFB, ResourceTemplate ()
        {
            I2cSerialBusV2 (0x0015, ControllerInitiated, 0x00061A80,
                AddressingMode7Bit, "\\_SB.I2CA",
                0x00, ResourceConsumer, , Exclusive,
                )
        })
    }

    If ((TPTY == 0x02))
    {
        Name (SBFS, ResourceTemplate ()
        {
            I2cSerialBusV2 (0x002C, ControllerInitiated, 0x00061A80,
                AddressingMode7Bit, "\\_SB.I2CA",
                0x00, ResourceConsumer, , Exclusive,
                )
        })
    }

    Name (SBFI, ResourceTemplate ()
    {
        GpioInt (Level, ActiveLow, ExclusiveAndWake, PullUp, 0x0000,
            "\\_SB.GPIO", 0x00, ResourceConsumer, ,
            )
            {   // Pin list
                0x0009
            }
    })
    If ((TPTY == One))
    {
        Return (ConcatenateResTemplate (SBFB, SBFI))
    }

    If ((TPTY == 0x02))
    {
        Return (ConcatenateResTemplate (SBFS, SBFI))
    }
}

#

brooo are u fucking kidding

#

lmao

#

Method (_CRS, 0, NotSerialized)  // _CRS: Current Resource Settings
{
    If (GTPM ())
    {
        Return (Zero)
    }

    If (GCTP ())
    {
        Return (ResourceTemplate ()
        {
            Memory32Fixed (ReadWrite,
                0xFED40000,         // Address Base
                0x00005000,         // Address Length
                )
            IO (Decode16,
                0x0CB0,             // Range Minimum
                0x0CB0,             // Range Maximum
                0x08,               // Alignment
                0x10,               // Length
                )
        })
    }
    Else
    {
        Return (Zero)
    }
}

Thats not how you _CRS LULW

#

Brooo hand rolled fucking resource template

Method (_CRS, 0, NotSerialized)  // _CRS: Current Resource Settings
{
    Name (NAM, Buffer (0x0F)
    {
        "\\_SB.SDHB.BRCM"
    })
    Name (SPB, Buffer (0x0C)
    {
        /* 0000 */  0x8E, 0x18, 0x00, 0x01, 0x00, 0xC0, 0x02, 0x00,  // ........
        /* 0008 */  0x00, 0x01, 0x00, 0x00                           // ....
    })
    Name (END, Buffer (0x02)
    {
         0x79, 0x00                                       // y.
    })
    Concatenate (SPB, NAM, Local0)
    Concatenate (Local0, END, Local1)
    Return (Local1)
}

#

[uACPI][ERROR] invalid/unsupported serial connection type 192

#

i guess i should support that

#

Well ACPICA doesnt support it

#

so i wont either, whatever

vast kestrel May 5, 2024, 8:33 PM

#

maybe just make it a warning instead

fiery turtle May 5, 2024, 8:38 PM

#

well i have to be able to convert it somehow

#

for it to be a warning

#

because this aborts the _CRS invocation

vast kestrel May 5, 2024, 8:40 PM

#

hmmm, you also want a way for drivers to have custom types in there probably

fiery turtle May 5, 2024, 8:41 PM

#

probably, but acpica managed to not support it for 30 years and be fine

#

i guess no one uses it

vast kestrel May 5, 2024, 8:41 PM

#

is that your inspiration

left orbit May 5, 2024, 8:41 PM

#

ah, so that's (_CRS) what uacpi_get_current_resources calls and gives back to you? (message from nyaux chat, dont wanna pollute it)

vast kestrel May 5, 2024, 8:41 PM

#

smugFace

fiery turtle May 5, 2024, 8:41 PM

#

vast kestrel is that your inspiration

lol

#

like from a user perspective linux doesnt need that and it supports like all hw out there

vast kestrel May 5, 2024, 8:42 PM

#

I do wonder what windows do

fiery turtle May 5, 2024, 8:42 PM

#

probably some proprietary driver that handles this serial connection

fiery turtle May 5, 2024, 8:42 PM

#

left orbit ah, so that's (_CRS) what uacpi_get_current_resources calls and gives back to yo...

yeah

#

there's also possible_resources (_PRS)

#

which is a list of possible values that can be used to override the current resources

left orbit May 5, 2024, 8:43 PM

#

setting everything up properly is quite involved lol

fiery turtle May 5, 2024, 8:44 PM

#

yeah

#

and this resource conversion thing has a bytecode VM for converting them

#

its pretty insane

#

#

here's one for IRQ for example

#

Some of these public blobs do absolutely insane things

#

crazy uacpi survives all of that

#

good fuzzing experience because of qookie i guess

fiery turtle May 5, 2024, 10:15 PM

#

Added a test case for infinite while loops

#

600k iterations with ASAN+UBSAN and O0 is pretty decent

#

(in 3 seconds)

hollow elm May 5, 2024, 10:16 PM

#

can aml integers be arbitrarily large?

fiery turtle May 5, 2024, 10:17 PM

#

only 64 bits

hollow elm May 5, 2024, 10:17 PM

#

or are they just a fixed-size int

#

ah

fiery turtle May 5, 2024, 10:17 PM

#

yeah

#

idk if its worth guarding against 64-bit overflow lmao

#

unless this runs on nasa quatum computer

fiery turtle May 5, 2024, 10:18 PM

#

hollow elm or are they just a fixed-size int

if aml is revision 1 the ints are 32 bits, 64 for >= 2

hollow elm May 5, 2024, 10:19 PM

#

ah right i remember something about 32-bit integers and acpi

fiery turtle May 5, 2024, 10:19 PM

#

yeah its an annoying backwards compat thing

#

also if DSDT is rev 1 and SSDTs are 2 everything is still done as 32 bits

#

this can be seen on real hw too

hollow elm May 5, 2024, 10:19 PM

#

huh

fiery turtle May 5, 2024, 10:20 PM

#

basically only DSDT revision matters

hollow elm May 5, 2024, 10:20 PM

#

do you actually support that?

fiery turtle May 5, 2024, 10:20 PM

#

of course

hollow elm May 5, 2024, 10:21 PM

#

how much do you even need to do to handle it? just & 0xFFFFFFFF on every arith operation?

fiery turtle May 5, 2024, 10:21 PM

#

i have a UACPI_PARSE_OP_TRUNCATE_NUMBER

#

thats basically applied to any math operation

fiery turtle May 5, 2024, 10:22 PM

#

hollow elm how much do you even need to do to handle it? just `& 0xFFFFFFFF` on every arith...

yeah, basically that

hollow elm May 5, 2024, 10:22 PM

#

hm what about a load from a 64-bit field?

fiery turtle May 5, 2024, 10:22 PM

#

its loaded as a buffer

hollow elm May 5, 2024, 10:22 PM

#

ah

fiery turtle May 5, 2024, 10:22 PM

#

if in 32 bit mode

hollow elm May 5, 2024, 10:22 PM

#

not sure what i expected

fiery turtle May 5, 2024, 10:22 PM

#

also on NT if u have a qword immediate its still loaded as a qword

#

even in 32 bit mode

#

so i do that as well

#

and its never truncated unless u do math on it

hollow elm May 5, 2024, 10:23 PM

#

interesting

fiery turtle May 5, 2024, 10:24 PM

#

this affects e.g ToHexString

#

or any compare, greater than etc

hollow elm May 5, 2024, 10:24 PM

#

makes sense

#

i assume it also affects storing to 64-bit fields

fiery turtle May 5, 2024, 10:24 PM

#

yet another acpica incompatibility

#

yup

#

well actually no

#

it would only store the first 4 bytes

#

the rest is zeroed

hollow elm May 5, 2024, 10:25 PM

#

thonk

fiery turtle May 5, 2024, 10:25 PM

#

but i guess its implementation defiend ofc caret

#

i know how nt does it

#

acpica recently fixed ToHexString

#

aka ToHexString(0xF) == "0xF" and not 00000000F

hollow elm May 5, 2024, 10:26 PM

#

god bless

fiery turtle May 5, 2024, 10:26 PM

#

a drop in the bucket but oh well

hollow elm May 5, 2024, 10:26 PM

#

making that one machine where the fw engineer went insane work on linux

fiery turtle May 5, 2024, 10:27 PM

#

lol

#

it was actually from a bug report yeah

#

Some firmware does ToHexString and then Mid to parse the serial number

#

So on linux it was printing out bogus serial

hollow elm May 5, 2024, 10:28 PM

#

ig that use makes sense

fiery turtle May 5, 2024, 10:28 PM

#

yeah

#

not the worst example i can think of

#

Yay large tests pass on CI as well!

#

In both 32-bit and 64-bit builds

#

Big

#

this bumps the ci up to like 6 minutes on ubuntu

#

but its fine i guess

left orbit May 7, 2024, 11:46 AM

#

@fiery turtle are kernels using uacpi supposed to discover all devices using the uacpi methods? like you can find the ps2 keyboard using acpi, how about pci devices? or is that better done by probing manually

fiery turtle May 7, 2024, 11:46 AM

#

Well you can discover PCI roots and from there match the devices

#

But not all devices are guaranteed to be in the namespace

#

So of course a pci subsystem lives on its own

#

And talks to acpi as needed

#

For pci acpi is more of a nice extension, not a replacement

left orbit May 7, 2024, 11:48 AM

#

i see, i was just wondering since my way of enumerating pci devices is pretty janky and i thought that maybe there is some big brain acpi way to put me out of misery :^)

fiery turtle May 7, 2024, 11:48 AM

#

What sort of jankiness?

left orbit May 7, 2024, 11:49 AM

#

ehhh

#

im not sure if its janky

#

it feels like it

fiery turtle May 7, 2024, 11:49 AM

#

How do you do it?

left orbit May 7, 2024, 11:49 AM

#

i can send the code if you want to take a look, its around 30-40 lines

#

ive written it more than a year ago

fiery turtle May 7, 2024, 11:49 AM

#

I mean sure

#

Lol

left orbit May 7, 2024, 11:49 AM

#

i dont understand the way it works lmao

#

https://github.com/48cf/zigux/blob/7435f7d39869a463809956ad15d34eba95a59357/kernel/src/pci.zig#L238-L309

#

it might be that i just dont remember how exactly pci works anymore, but this isnt very good code either way, it doesnt support mmio which i think is preferred way of doing pci

fiery turtle May 7, 2024, 11:51 AM

#

Looks fairly okay barring some magic numbers

fiery turtle May 7, 2024, 11:52 AM

#

left orbit it might be that i just dont remember how exactly pci works anymore, but this is...

Brush up on osdev I guess

left orbit May 7, 2024, 11:52 AM

#

yeahhh

#

i had to do some msi-x stuff yesterday for xhci so it was a nice reintroduction into some parts of my old codebase

fiery turtle May 7, 2024, 11:52 AM

#

Nice

left orbit May 7, 2024, 11:53 AM

#

i discovered some funny bugs

#

like not having a separate stack for the page fault handler

fiery turtle May 7, 2024, 11:53 AM

#

Are you reviving an abandoned project?

left orbit May 7, 2024, 11:53 AM

#

and now when i add a field to my xhci controller struct i get an kvm error :D

fiery turtle May 7, 2024, 11:53 AM

#

Damn

left orbit May 7, 2024, 11:53 AM

#

ehh im not sure what i wanna do, i wanted to start from scratch

fiery turtle May 7, 2024, 11:54 AM

#

left orbit and now when i add a field to my xhci controller struct i get an kvm error :D

Similar bug like @kindred beacon

left orbit May 7, 2024, 11:54 AM

#

with the silly kernel™️

#

but its so much work that im not sure i want to put in

fiery turtle May 7, 2024, 11:54 AM

#

Makes sense

left orbit May 7, 2024, 11:55 AM

#

but zig i just meh, so i guess im working on it for experience so that i can make something better in the future, rather than have the zig kernel be actually good

fiery turtle May 7, 2024, 11:55 AM

#

Why?

#

I thought it was good

left orbit May 7, 2024, 11:55 AM

#

compiler is poo poo

#uACPI - a portable and easy-to-integrate ACPI implementation

curious ``` [uACPI][WARN] invalid table 'DSDT' checksum! [uACPI][TRACE] load of 'DSDT' (OEM ID 'ATC' OEM Table ID 'Cofbej') [uACPI][TRACE] write to [_SB_.TP80] (2 bytes) TableData[0x0000000000000001] = 0x0 AddressSanitizer:DEADLYSIGNAL

another crash found ``` [uACPI][WARN] error during _REG execution for operation region \PCI0.UHCI.TEST.PCIR@0x506000000b60: invalid argument AddressSanitizer:DEADLYSIGNAL

curious ```
[uACPI][WARN] invalid table 'DSDT' checksum!
[uACPI][TRACE] load of 'DSDT' (OEM ID 'ATC' OEM Table ID 'Cofbej')
[uACPI][TRACE] write to [_SB_.TP80] (2 bytes) TableData[0x0000000000000001] = 0x0
AddressSanitizer:DEADLYSIGNAL

another crash found ```
[uACPI][WARN] error during _REG execution for operation region \PCI0.UHCI.TEST.PCIR@0x506000000b60: invalid argument
AddressSanitizer:DEADLYSIGNAL