[edk2-devel] Casting i128 into f64 in UEFI Rust pagefaults
Andrew Fish via groups.io
afish=apple.com at groups.io
Tue Jul 26 21:38:16 UTC 2022
Ayush,
Well bugs in the runtime are a classic case when have good debugging infrastructure pays off and that class of issue is what started this email thread.
It is not hard to try out gdb to see what happens if you are using QEMU.
>From the root of the edk2 repo:
$ OvmfPkg/build.sh qemu -gdb tcp::9000
Then from another terminal
$ cd BaseTools/Scripts
$ gdb -ex "target remote localhost:9000" -ex "source efi_gdb.py"
That should get you a symbolicated stack frame.
The QEMU -gdb works like a JTAG debugger. You can put dead loops in your code, let it run, and then attach.
Thanks,
Andrew Fish
> On Jul 26, 2022, at 12:15 PM, Ayush Singh <ayushdevel1325 at gmail.com> wrote:
>
> Hi Andrew. I do agree with having better debugging support for Rust-UEFI. However, I am not much experienced with even the C side of UEFI debugging, so it has been difficult for me to do much on the Rust side either. It isn't like Rust UEFI debugging is not possible. There are some examples ([1]), but as you might guess, they are quite few. Additionally, most of them seem to be concentrated around Windows. Finally, there is actually quite good debugging support in uefi-rs [2], but well, that's licensed under MPL-2.0, so I have stayed clear of it.
>
>
>
> Another reason I have gone so long without using any actual debugger is because, well, Rust makes it quite explicit where wired errors can occur. Most functions, even inside the std, don't use pointers but rather use `NonNull`, `MaybeUninint` or some other safe abstraction. Even on the worst failure, the program will abort instead of crash. On aborting, Rust also gives a nice message on stderr stating exactly where the error occurred along with all the details of the error. Basically, it was always so clear what caused the error and where it occurred that I simply didn't feel the need to attach a debugger, till now.
>
>
>
> There is some work going on to actually document using Rust for UEFI in the rustc docs, sponsored by Red Hat [3], so the situation should improve soon. What I do know is that the `.pdb` files generated by Rust (which are always generated) should contain all the symbols for debugging. I can work on debugging once I finish up with solving all other errors which are due to my implementation of std rather than the ones that have been caused by the rust intrinsic.
>
>
>
> Ayush Singh
>
>
>
> [1]: https://xitan.me/posts/rust-uefi-runtime-driver/
>
> [2]: https://github.com/rust-osdev/uefi-rs
>
> [3]: https://github.com/rust-lang/rust/pull/99760
>
>
>
> On 7/26/22 23:20, Andrew Fish wrote:
>>> On Jul 25, 2022, at 10:43 PM, Ayush Singh <ayushdevel1325 at gmail.com> <mailto:ayushdevel1325 at gmail.com> wrote:
>>>
>>> Hi Andrew. Thanks for all your work. The more I look at this, the more it feels like it might be a problem on the LLVM side instead of Rust. I also found some more tests (all related to numbers btw) which can cause different types of exceptions, so I think I will try filing bugs upstream.
>>>
>>
>> Ayush,
>>
>> In general If we want to move to Rust we are going to need a way to debug issues like this down to the root cause. I think figuring out how to debug will make it easier to move forward with the Rust port in general. It will time well spent.
>>
>> The best way to get LLVM fixed, if it is even broken, is to provide a simple test case that reproduces the behavior. I don’t think at this point we know what is going on. It is very unlikely that some random LLVM developer is going to invest the time in trying to setup some UEFI environment to try and root cause this bug. I general find I have to create a simple at desk example and then I get stuff fixed quickly. Basically a test case the LLVM developer can compile at their desk and see the error in assembler, or at least run it at desk and have bogus output.
>>
>> I’m not 100% sure what toolchain you are using. Can you `objdump -d hello_world_std.efi`and get some symbols with the disassembly? For VC++ I think it would be `DUMPBIN /DISASM`.
>>
>> What are you planning on using for source level debugging Rust? I wrote some gdb[1] and lldb[2] debugging commands in Python. I’m guessing loading Rust symbols from PE/COFF images should be similar, as long as the debugger knows about rust.
>>
>> I’m happy to help you figure out stuff related to debugging Rust.
>>
>> [1] https://github.com/tianocore/edk2/blob/master/BaseTools/Scripts/efi_gdb.py
>> [2] https://github.com/tianocore/edk2/blob/master/BaseTools/Scripts/efi_lldb.py
>>
>> Thanks,
>>
>> Andrew Fish
>>
>>> Yours Sincerely,
>>>
>>> Ayush Singh
>>>
>>>
>>>
>>> On 7/26/22 00:24, Andrew Fish wrote:
>>>> I guess I could at least dump to the end (req)…. Going backwards is a bit painful in x86.
>>>>
>>>> (lldb) dis -s 0x0000000140001B60 -b -c 30
>>>> hello_world_std.efi[0x140001b60]: 48 8b 09 movq (%rcx), %rcx
>>>> hello_world_std.efi[0x140001b63]: 48 01 c1 addq %rax, %rcx
>>>> hello_world_std.efi[0x140001b66]: 4c 89 c2 movq %r8, %rdx
>>>> hello_world_std.efi[0x140001b69]: 48 11 c2 adcq %rax, %rdx
>>>> hello_world_std.efi[0x140001b6c]: 48 31 c1 xorq %rax, %rcx
>>>> hello_world_std.efi[0x140001b6f]: 48 31 c2 xorq %rax, %rdx
>>>> hello_world_std.efi[0x140001b72]: 48 be 00 00 00 00 00 00 00 80 movabsq $-0x8000000000000000, %rsi ; imm = 0x8000000000000000
>>>> hello_world_std.efi[0x140001b7c]: 4c 21 c6 andq %r8, %rsi
>>>> hello_world_std.efi[0x140001b7f]: e8 5c 55 00 00 callq 0x1400070e0
>>>> hello_world_std.efi[0x140001b84]: 48 09 f0 orq %rsi, %rax
>>>> hello_world_std.efi[0x140001b87]: 48 83 c4 20 addq $0x20, %rsp
>>>> hello_world_std.efi[0x140001b8b]: 5e popq %rsi
>>>> hello_world_std.efi[0x140001b8c]: c3 retq
>>>> hello_world_std.efi[0x140001b8d]: cc int3
>>>> hello_world_std.efi[0x140001b8e]: cc int3
>>>> hello_world_std.efi[0x140001b8f]: cc int3
>>>> hello_world_std.efi[0x140001b90]: e9 db 55 00 00 jmp 0x140007170
>>>> hello_world_std.efi[0x140001b95]: cc int3
>>>> …
>>>>
>>>> Then we can guess based on how functions get aligned to find the start….
>>>>
>>>> hello_world_std.efi[0x140001b50]: 56 pushq %rsi
>>>> hello_world_std.efi[0x140001b51]: 48 83 ec 20 subq $0x20, %rsp
>>>> hello_world_std.efi[0x140001b55]: 4c 8b 41 08 movq 0x8(%rcx), %r8
>>>> hello_world_std.efi[0x140001b59]: 4c 89 c0 movq %r8, %rax
>>>> hello_world_std.efi[0x140001b5c]: 48 c1 f8 3f sarq $0x3f, %rax
>>>> hello_world_std.efi[0x140001b60]: 48 8b 09 movq (%rcx), %rcx
>>>> hello_world_std.efi[0x140001b63]: 48 01 c1 addq %rax, %rcx
>>>> hello_world_std.efi[0x140001b66]: 4c 89 c2 movq %r8, %rdx
>>>> hello_world_std.efi[0x140001b69]: 48 11 c2 adcq %rax, %rdx
>>>> hello_world_std.efi[0x140001b6c]: 48 31 c1 xorq %rax, %rcx
>>>> hello_world_std.efi[0x140001b6f]: 48 31 c2 xorq %rax, %rdx
>>>> hello_world_std.efi[0x140001b72]: 48 be 00 00 00 00 00 00 00 80 movabsq $-0x8000000000000000, %rsi ; imm = 0x8000000000000000
>>>> hello_world_std.efi[0x140001b7c]: 4c 21 c6 andq %r8, %rsi
>>>> hello_world_std.efi[0x140001b7f]: e8 5c 55 00 00 callq 0x1400070e0
>>>> hello_world_std.efi[0x140001b84]: 48 09 f0 orq %rsi, %rax
>>>> hello_world_std.efi[0x140001b87]: 48 83 c4 20 addq $0x20, %rsp
>>>> hello_world_std.efi[0x140001b8b]: 5e popq %rsi
>>>> hello_world_std.efi[0x140001b8c]: c3 retq
>>>>
>>>> So the faulting function is getting passed a bad pointer as its 1st arg.
>>>>
>>>> Thanks,
>>>>
>>>> Andrew Fish
>>>>
>>>>> On Jul 25, 2022, at 11:45 AM, Andrew Fish <afish at apple.com> <mailto:afish at apple.com> wrote:
>>>>>
>>>>> Ops… Looks like your PE/COFF is linked at 0x0000000140000000, so 0x140001b60 is the interesting bit.
>>>>>
>>>>> (lldb) dis -s 0x0000000140001B60 -b
>>>>> hello_world_std.efi[0x140001b60]: 48 8b 09 movq (%rcx), %rcx
>>>>> hello_world_std.efi[0x140001b63]: 48 01 c1 addq %rax, %rcx
>>>>> hello_world_std.efi[0x140001b66]: 4c 89 c2 movq %r8, %rdx
>>>>> hello_world_std.efi[0x140001b69]: 48 11 c2 adcq %rax, %rdx
>>>>> hello_world_std.efi[0x140001b6c]: 48 31 c1 xorq %rax, %rcx
>>>>> hello_world_std.efi[0x140001b6f]: 48 31 c2 xorq %rax, %rdx
>>>>> hello_world_std.efi[0x140001b72]: 48 be 00 00 00 00 00 00 00 80 movabsq $-0x8000000000000000, %rsi ; imm = 0x8000000000000000
>>>>> hello_world_std.efi[0x140001b7c]: 4c 21 c6 andq %r8, %rsi
>>>>>
>>>>> RCX - FFFFFFFFFFFFFFFF
>>>>>
>>>>> So yea that looks like the fault.
>>>>>
>>>>> I don’t see that pattern in your .s file….
>>>>>
>>>>> Can you figure out what function is @ 0x140001b60 in the PE/COFF image. Do you have a map file from the linker?
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Andrew Fish
>>>>>
>>>>> PS Again sorry I don’t have anything installed to crack PDB files.
>>>>>
>>>>> Thanks,
>>>>>
>>>>> Andrew Fish
>>>>>
>>>>>> On Jul 25, 2022, at 10:51 AM, Andrew Fish via groups.io <afish=apple.com at groups.io> <mailto:afish=apple.com at groups.io> wrote:
>>>>>>
>>>>>> Ayush,
>>>>>>
>>>>>> CR2 is the fault address so 0xFFFFFFFFFFFFFFFF. Given for EFI Virt == Physical the fault address looks like a bad pointer.
>>>>>>
>>>>>> Sorry I’ve not used VC++ in a long time so I don’t know how to debug with VC++, but If I was using clang/lldb I’d look at the source and assembly for the fault address.
>>>>>>
>>>>>> The image base is: 0x000000000603C000
>>>>>> The fault PC/RIP is: 000000000603DB60
>>>>>>
>>>>>> So the faulting code is at 0x1B60 in the image. Given the images are linked at zero you should be able to load the build product into the debugger and look at what code is at offset 0x1B60. The same should work for any tools that dump the binary.
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Andrew Fish
>>>>>>
>>>>>>> On Jul 25, 2022, at 10:33 AM, Ayush Singh <ayushdevel1325 at gmail.com> <mailto:ayushdevel1325 at gmail.com> wrote:
>>>>>>>
>>>>>>> Hello everyone.While running Rust tests in UEFI environment, I have come across a numeric test that causes a pagefault. A simple reproducible example for this is given below:
>>>>>>>
>>>>>>> ```rust
>>>>>>>
>>>>>>> fn main() {
>>>>>>> use std::hint::black_box as b;
>>>>>>>
>>>>>>> let z: i128 = b(1);
>>>>>>> assert!((-z as f64) < 0.0);
>>>>>>> }
>>>>>>>
>>>>>>> ```
>>>>>>>
>>>>>>>
>>>>>>> The exception output is as follows:
>>>>>>>
>>>>>>> ```
>>>>>>>
>>>>>>> !!!! X64 Exception Type - 0E(#PF - Page-Fault) CPU Apic ID - 00000000 !!!!
>>>>>>> ExceptionData - 0000000000000000 I:0 <i:0> R:0 <r:0> U:0 <u:0> W:0 <w:0> P:0 <p:0> PK:0 <pk:0> SS:0 <ss:0> SGX:0 <sgx:0>
>>>>>>> RIP - 000000000603DB60, CS - 0000000000000038, RFLAGS - 0000000000000246
>>>>>>> RAX - 0000000000000000, RCX - FFFFFFFFFFFFFFFF, RDX - FFFFFFFFFFFFFFFF
>>>>>>> RBX - 0000000000000000, RSP - 0000000007EDF1D0, RBP - 0000000007EDF4C0
>>>>>>> RSI - 0000000007EDF360, RDI - 0000000007EDF3C0
>>>>>>> R8 - 0000000000000000, R9 - 0000000000000038, R10 - 0000000000000000
>>>>>>> R11 - 0000000000000000, R12 - 00000000060C6018, R13 - 0000000007EDF520
>>>>>>> R14 - 0000000007EDF6A8, R15 - 0000000005FA9490
>>>>>>> DS - 0000000000000030, ES - 0000000000000030, FS - 0000000000000030
>>>>>>> GS - 0000000000000030, SS - 0000000000000030
>>>>>>> CR0 - 0000000080010033, CR2 - FFFFFFFFFFFFFFFF, CR3 - 0000000007C01000
>>>>>>> CR4 - 0000000000000668, CR8 - 0000000000000000
>>>>>>> DR0 - 0000000000000000, DR1 - 0000000000000000, DR2 - 0000000000000000
>>>>>>> DR3 - 0000000000000000, DR6 - 00000000FFFF0FF0, DR7 - 0000000000000400
>>>>>>> GDTR - 00000000079DE000 0000000000000047, LDTR - 0000000000000000
>>>>>>> IDTR - 0000000007418018 0000000000000FFF, TR - 0000000000000000
>>>>>>> FXSAVE_STATE - 0000000007EDEE30
>>>>>>> !!!! Find image based on IP(0x603DB60) /var/home/ayush/Documents/Programming/Rust/uefi/hello_world_std/target/x86_64-unknown-uefi/debug/deps/hello_world_std-338028f9369e2d42.pdb (ImageBase=000000000603C000, EntryPoint=000000000603D8C0) !!!!
>>>>>>>
>>>>>>> ```
>>>>>>>
>>>>>>>
>>>>>>> From my testing, the exception only occurs when a few conditions are met.
>>>>>>>
>>>>>>> 1. The binary is compiled in Debug mode. No error in Release mode.
>>>>>>>
>>>>>>> 2. `i128` is in a black_box [1]. Does not occur if `black_box` is not present.
>>>>>>>
>>>>>>> 3. It has to be `i128`. `i64` or something else work fine.
>>>>>>>
>>>>>>> 4. The cast has to be done on `-z`. Doing the same with `+z` is fine.
>>>>>>>
>>>>>>>
>>>>>>> I have also been discussing this in the Rust zulipchat [2], so feel free to chime in there.
>>>>>>>
>>>>>>>
>>>>>>> Additionally, here are links for more information about this program:
>>>>>>>
>>>>>>> 1. Assembly: https://rust-lang.zulipchat.com/user_uploads/4715/od51Y9Dkfjahcg9HHcOud8Fm/hello_world_std-338028f9369e2d42.s
>>>>>>>
>>>>>>> 2. EFI Binary: https://rust-lang.zulipchat.com/user_uploads/4715/CknqtXLR8SaJZmyOnXctQkpL/hello_world_std.efi
>>>>>>>
>>>>>>> 3. PDB file: https://rust-lang.zulipchat.com/user_uploads/4715/zV4i6DsjgQXotp_gS1naEsU0/hello_world_std-338028f9369e2d42.pdb
>>>>>>>
>>>>>>>
>>>>>>> Yours Sincerely,
>>>>>>>
>>>>>>> Ayush Singh
>>>>>>>
>>>>>>>
>>>>>>> [1]: https://doc.rust-lang.org/std/hint/fn.black_box.html
>>>>>>>
>>>>>>> [2]: https://rust-lang.zulipchat.com/#narrow/stream/182449-t-compiler.2Fhelp/topic/Casting.20i128.20to.20f64.20in.20black_box.20causes.20exception.20in.20UEFI
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>
>
>
-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#91864): https://edk2.groups.io/g/devel/message/91864
Mute This Topic: https://groups.io/mt/92610101/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/edk2-devel-archive/attachments/20220726/b82aa9db/attachment-0001.htm>
More information about the edk2-devel-archive
mailing list