Skip to contentSkip to conversation

Virt MM DEBUG_LL support arm32

DISCUSSION Mario Smarduch 3 participants 7 messages Oct 7, 2015 – Oct 12, 2015 KVM arm64

lore
Share & tools
lore
#0 lore raw

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

Thanks,
- Mario

#1 lore raw
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

thanks
-- PMM

#2 lore raw
Quoted 14 lines · Peter Maydell
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Quoted 7 lines

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

thanks
-- PMM

No actually I was thinking of limiting choices to the ones that work for Virt MM during kernel config. But thought that maybe others may be supported through some options. Form me PL011 alone is enough.

- Mario

#3 lore raw
Quoted 15 lines · Mario Smarduch
Quoted 14 lines · Peter Maydell
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Quoted 7 lines
Quoted 3 lines

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

No actually I was thinking of limiting choices to the ones that work for Virt MM during kernel config. But thought that maybe others may be supported through some options. Form me PL011 alone is enough.

8250 I think is what kvmtool uses, which is why that is a permitted option. Not sure who's implementing ICE DCC (though as I say QEMU might eventually).

thanks
-- PMM

#4 lore raw
Quoted 20 lines · Peter Maydell
Quoted 18 lines · Mario Smarduch
Quoted 17 lines · Peter Maydell
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

Quoted 7 lines
Quoted 3 lines

No actually I was thinking of limiting choices to the ones that work for Virt MM during kernel config. But thought that maybe others may be supported through some options. Form me PL011 alone is enough.

8250 I think is what kvmtool uses, which is why that is a permitted option. Not sure who's implementing ICE DCC (though as I say QEMU might eventually).

I see thanks for pointing that out.

Quoted 4 lines

thanks
-- PMM

I came across the issue of selecting wrong UART for DEBUG_LL while trying to boot arm32 on arm64. ARCH_VIRT system type defaults for 'Kernel low-level debugging port' are not too clear.

I'm wondering if it would make sense to add entries for virtual system types (like QEMU Virt or kvmtool virt) and enhance help section with possible UART_PHYS address, also set a default port. It appears like hard coding is going to be deprecated.

This is not a major enhancement but when you're out of options and you just see a blank screen anything helps. Especially for virtual host where you may start suspecting the host system as well. Since I tripped over it, I took a closer look.

Thanks,
Mario

#5 lore raw

Hi Mario,

Quoted 47 lines · Mario Smarduch
Quoted 20 lines · Peter Maydell
Quoted 18 lines · Mario Smarduch
Quoted 17 lines · Peter Maydell
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

Quoted 7 lines
Quoted 3 lines

No actually I was thinking of limiting choices to the ones that work for Virt MM during kernel config. But thought that maybe others may be supported through some options. Form me PL011 alone is enough.

8250 I think is what kvmtool uses, which is why that is a permitted option. Not sure who's implementing ICE DCC (though as I say QEMU might eventually).

I see thanks for pointing that out.

Quoted 4 lines

thanks
-- PMM

I came across the issue of selecting wrong UART for DEBUG_LL while trying to boot arm32 on arm64. ARCH_VIRT system type defaults for 'Kernel low-level debugging port' are not too clear.

I'm wondering if it would make sense to add entries for virtual system types (like QEMU Virt or kvmtool virt) and enhance help section with possible UART_PHYS address, also set a default port. It appears like hard coding is going to be deprecated.

Quoted 5 lines

This is not a major enhancement but when you're out of options and you just see a blank screen anything helps. Especially for virtual host where you may start suspecting the host system as well. Since I tripped over it, I took a closer look.

I agree that it is particularly discouraging when you just get a blank screen, and it is in fact one of the top support questions I get from new users in private as well.

I think there were some discussions with Rob and Grant in the past about some way for Linux to discover the location of the UART for early printk, but I have forgotten the details and where that landed.

I personally think it would be a good idea with *any* viable easy method to configure a kernel to support earlyprintk for our two virtual
platforms: QEMU's virt board and kvmtool's platform.

-Christoffer

#6 lore raw
Quoted 50 lines · Christoffer Dall

Hi Mario,

Quoted 47 lines · Mario Smarduch
Quoted 20 lines · Peter Maydell
Quoted 18 lines · Mario Smarduch
Quoted 17 lines · Peter Maydell
Quoted 6 lines · Mario Smarduch

Hi Peter,
I noticed that icedcc, and 8250 don't work
with DEBUG_LL early debug print. And the kernel dies if these are selected. Besides PL011 is there any other serial devices that can be used for early debug with Virt MM? Maybe some additional options are needed?

PL011 and semihosting are the obvious choices for early debug output. There's no 8250 in the virt board so that definitely isn't going to work. We don't currently implement the ICE DCC channel, though I think Edgar was considering connecting it up to a chardev backend.

Really though, I think the UART and semihosting should be enough for whatever the guest wants to do in the way of early debug. Why do you need more options?

Quoted 7 lines
Quoted 3 lines

No actually I was thinking of limiting choices to the ones that work for Virt MM during kernel config. But thought that maybe others may be supported through some options. Form me PL011 alone is enough.

8250 I think is what kvmtool uses, which is why that is a permitted option. Not sure who's implementing ICE DCC (though as I say QEMU might eventually).

I see thanks for pointing that out.

Quoted 4 lines

thanks
-- PMM

I came across the issue of selecting wrong UART for DEBUG_LL while trying to boot arm32 on arm64. ARCH_VIRT system type defaults for 'Kernel low-level debugging port' are not too clear.

I'm wondering if it would make sense to add entries for virtual system types (like QEMU Virt or kvmtool virt) and enhance help section with possible UART_PHYS address, also set a default port. It appears like hard coding is going to be deprecated.

Quoted 8 lines
Quoted 5 lines

This is not a major enhancement but when you're out of options and you just see a blank screen anything helps. Especially for virtual host where you may start suspecting the host system as well. Since I tripped over it, I took a closer look.

I agree that it is particularly discouraging when you just get a blank screen, and it is in fact one of the top support questions I get from new users in private as well.

Yes a momentum killer :)

Quoted 4 lines

I think there were some discussions with Rob and Grant in the past about some way for Linux to discover the location of the UART for early printk, but I have forgotten the details and where that landed.

I'll double check but I believe ARM reference models support this and have a nice detailed interface

Quoted 4 lines

I personally think it would be a good idea with *any* viable easy method to configure a kernel to support earlyprintk for our two virtual
platforms: QEMU's virt board and kvmtool's platform.

Will do and get the kernel folks input, it's just mostly menu selection that needs a little cleanup.

Thanks
- Mario

Quoted 3 lines

-Christoffer