On Tue, Apr 23, 2024 at 01:34:42PM +0800, Leo Liang wrote:

On Mon, Apr 22, 2024 at 04:43:59PM -0300, Daniel Henrique Barboza wrote:

...

[EXTERNAL MAIL]

Hi,

In QEMU we have a 'max' type CPU that implements (almost) all extensions that QEMU is able to emulate. Recently, in QEMU commit 249e0905d05, we bumped the extensions for this CPU.

And after this commit this CPU is now unable to boot a guest using upstream u-boot. Here's the error being thrown:

qemu-system-riscv64 \ -machine virt -nographic -m 8G -smp 8 \ -cpu max -kernel uboot.elf (...) (...)

initcall sequence 000000008027c3e8 failed at call 000000008021259e (err=-28) ### ERROR ### Please RESET the board ###

I can get the guest to boot if I disable the following extensions from the 'max' CPU:

-cpu max,zfbfmin=false,zvfbfmin=false,zvfbfwma=false

Due to QEMU extension dependencies I'm not able to disable these individually. What I can say is that u-boot isn't playing ball to at least one of them.

Is this an u-boot bug? Up to this point I was assuming that u-boot would silently ignore hart extensions that it doesn't support.

Hi Daniel,

Which u-boot version are you using?

I think this issue is fixed by the following patch set sent by Conor.

f39b1b77d8 riscv: support extension probing using riscv, isa-extensions b90edde701 riscv: don't read riscv, isa in the riscv cpu's get_desc()

I've tested and can reproduce the issue you mentioned if these two patches are reverted.

Could you try with the lastest u-boot master branch again?

For reference, my testing commands are as follows:

1. cd ${u-boot} && make qemu-riscv64_defconfig && make -j`nproc`
2. ./${qemu}/build/qemu-system-riscv64 -nographic -machine virt -cpu max -bios u-boot.bin -m 8G -smp 8

• u-boot branch (commit): master (38ea74d6d5c0 "Prepare v2024.07-rc1")
• qemu branch (commit): master (62dbe54c24db "Update version for v9.0.0-rc4 release")

I'll go take a look at this, it's possible that my patches only hide the problem due to the new property being prioritised.

On Tue, Apr 23, 2024 at 09:52:06AM -0300, Daniel Henrique Barboza wrote:

On 4/23/24 09:41, Conor Dooley wrote:

...

On Tue, Apr 23, 2024 at 01:34:42PM +0800, Leo Liang wrote:

...

On Mon, Apr 22, 2024 at 04:43:59PM -0300, Daniel Henrique Barboza wrote:

...

[EXTERNAL MAIL]

Hi,

In QEMU we have a 'max' type CPU that implements (almost) all extensions that QEMU is able to emulate. Recently, in QEMU commit 249e0905d05, we bumped the extensions for this CPU.

And after this commit this CPU is now unable to boot a guest using upstream u-boot. Here's the error being thrown:

qemu-system-riscv64 \ -machine virt -nographic -m 8G -smp 8 \ -cpu max -kernel uboot.elf (...) (...)

initcall sequence 000000008027c3e8 failed at call 000000008021259e (err=-28) ### ERROR ### Please RESET the board ###

I can get the guest to boot if I disable the following extensions from the 'max' CPU:

-cpu max,zfbfmin=false,zvfbfmin=false,zvfbfwma=false

Due to QEMU extension dependencies I'm not able to disable these individually. What I can say is that u-boot isn't playing ball to at least one of them.

Is this an u-boot bug? Up to this point I was assuming that u-boot would silently ignore hart extensions that it doesn't support.

Hi Daniel,

Which u-boot version are you using?

I think this issue is fixed by the following patch set sent by Conor.

f39b1b77d8 riscv: support extension probing using riscv, isa-extensions b90edde701 riscv: don't read riscv, isa in the riscv cpu's get_desc()

I've tested and can reproduce the issue you mentioned if these two patches are reverted.

Could you try with the lastest u-boot master branch again?

For reference, my testing commands are as follows:

1. cd ${u-boot} && make qemu-riscv64_defconfig && make -j`nproc`
2. ./${qemu}/build/qemu-system-riscv64 -nographic -machine virt -cpu max -bios u-boot.bin -m 8G -smp 8

• u-boot branch (commit): master (38ea74d6d5c0 "Prepare v2024.07-rc1")
• qemu branch (commit): master (62dbe54c24db "Update version for v9.0.0-rc4 release")

I'll go take a look at this, it's possible that my patches only hide the problem due to the new property being prioritised.

Don't bother. I just checked with most recent u-boot master and I can't reproduce the problem, as Leo said.

My "fear" was that new QEMU + new U-Boot meant that the riscv,isa-extensions codepath was in use and there was something lurking in the riscv,isa parsing which had a problem. Fortunately, I think that's not the case, as the fix seems to be b90edde701 "riscv: don't read riscv,isa in the riscv cpu's get_desc()" rather than f39b1b77d8 "riscv: support extension probing using riscv, isa-extensions".

I am, however, not going to look into why exactly it was failing before, I'm satisfied once the riscv,isa parsing isn't broken in master.

I apologize for the noise. I failed to fetch the latest upstream and do a last test before posting it here.

We were discussing here and there about disabling these extensions in the 'max' CPU in QEMU if u-boot wasn't able to handle them. I'm happy to see that u-boot is now able to do so and we can keep the 'max' CPU as is.

Thanks for the help,

Daniel