[U-Boot] qemu-x86 issues
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Any ideas, or places I should try poking to solve these issues? I'm going to kludge around this for now by disabling these tests locally. Thanks!
Hi Tom,
On Sat, Oct 15, 2016 at 7:32 AM, Tom Rini trini@konsulko.com wrote:
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
I just tried QEMU v2.6.1 with u-boot.rom built from the master head, but did not see such issue.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Typical emulation target issue :) Did you see similar issue on other QEMU platform with U-Boot?
Any ideas, or places I should try poking to solve these issues? I'm going to kludge around this for now by disabling these tests locally. Thanks!
Regards, Bin
On Tue, Oct 18, 2016 at 10:36:41AM +0800, Bin Meng wrote:
Hi Tom,
On Sat, Oct 15, 2016 at 7:32 AM, Tom Rini trini@konsulko.com wrote:
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
I just tried QEMU v2.6.1 with u-boot.rom built from the master head, but did not see such issue.
Ah, I think I see it now. We can't use 0x0 as where to load stuff to and the tests default to 0x0 if you don't provide an address rather than $loadaddr. I'll see about changing the tests.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Typical emulation target issue :) Did you see similar issue on other QEMU platform with U-Boot?
So this one is interesting. I see the same failure for MIPS/PPC for the qemu targets we have as well. I don't see this under QEMU for the ARM targets that I'm emulating. I suspect there's some common problem on the U-Boot side as also don't see this on the VMs I have running under qemu nor when I boot the kernelci rootfs from U-Boot.
On Tue, Oct 18, 2016 at 09:19:59AM -0400, Tom Rini wrote:
On Tue, Oct 18, 2016 at 10:36:41AM +0800, Bin Meng wrote:
Hi Tom,
On Sat, Oct 15, 2016 at 7:32 AM, Tom Rini trini@konsulko.com wrote:
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
I just tried QEMU v2.6.1 with u-boot.rom built from the master head, but did not see such issue.
Ah, I think I see it now. We can't use 0x0 as where to load stuff to and the tests default to 0x0 if you don't provide an address rather than $loadaddr. I'll see about changing the tests.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Typical emulation target issue :) Did you see similar issue on other QEMU platform with U-Boot?
So this one is interesting. I see the same failure for MIPS/PPC for the qemu targets we have as well. I don't see this under QEMU for the ARM targets that I'm emulating. I suspect there's some common problem on the U-Boot side as also don't see this on the VMs I have running under qemu nor when I boot the kernelci rootfs from U-Boot.
Bah, lemme correct myself. MIPS* and PowerPC fail due to sleep taking slightly too long. x86 is the only one where it's just way too fast. But this is not true once Linux itself is up.
Hi Tom,
On Tue, Oct 18, 2016 at 9:35 PM, Tom Rini trini@konsulko.com wrote:
On Tue, Oct 18, 2016 at 09:19:59AM -0400, Tom Rini wrote:
On Tue, Oct 18, 2016 at 10:36:41AM +0800, Bin Meng wrote:
Hi Tom,
On Sat, Oct 15, 2016 at 7:32 AM, Tom Rini trini@konsulko.com wrote:
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
I just tried QEMU v2.6.1 with u-boot.rom built from the master head, but did not see such issue.
Ah, I think I see it now. We can't use 0x0 as where to load stuff to and the tests default to 0x0 if you don't provide an address rather than $loadaddr. I'll see about changing the tests.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Typical emulation target issue :) Did you see similar issue on other QEMU platform with U-Boot?
So this one is interesting. I see the same failure for MIPS/PPC for the qemu targets we have as well. I don't see this under QEMU for the ARM targets that I'm emulating. I suspect there's some common problem on the U-Boot side as also don't see this on the VMs I have running under qemu nor when I boot the kernelci rootfs from U-Boot.
Bah, lemme correct myself. MIPS* and PowerPC fail due to sleep taking slightly too long. x86 is the only one where it's just way too fast. But this is not true once Linux itself is up.
U-Boot uses x86 TSC timer as the delay timer. Right now the timer frequency is hard-coded in device tree. Auto-calibration does not work on QEMU. I need have another check to see if we can do auto-calibration.
Regards, Bin
On Wed, Oct 19, 2016 at 11:18:00AM +0800, Bin Meng wrote:
Hi Tom,
On Tue, Oct 18, 2016 at 9:35 PM, Tom Rini trini@konsulko.com wrote:
On Tue, Oct 18, 2016 at 09:19:59AM -0400, Tom Rini wrote:
On Tue, Oct 18, 2016 at 10:36:41AM +0800, Bin Meng wrote:
Hi Tom,
On Sat, Oct 15, 2016 at 7:32 AM, Tom Rini trini@konsulko.com wrote:
Hey guys,
After making Stephen's scripts work with qemu today, I've started testing out what platforms I can test out via qemu. I've found two problems with qemu-x86. First, there's a real problem with the network. I run qemu with "-nographic -cpu qemu32 -netdev user,id=net0,tftp=/tftpboot -device e1000,netdev=net0" so that I can tftpboot things in and I'm getting crc32 mismatches.
I just tried QEMU v2.6.1 with u-boot.rom built from the master head, but did not see such issue.
Ah, I think I see it now. We can't use 0x0 as where to load stuff to and the tests default to 0x0 if you don't provide an address rather than $loadaddr. I'll see about changing the tests.
Second, the sleep command is not at all accurate. It's returning the 3 second test in less than 1 second.
Typical emulation target issue :) Did you see similar issue on other QEMU platform with U-Boot?
So this one is interesting. I see the same failure for MIPS/PPC for the qemu targets we have as well. I don't see this under QEMU for the ARM targets that I'm emulating. I suspect there's some common problem on the U-Boot side as also don't see this on the VMs I have running under qemu nor when I boot the kernelci rootfs from U-Boot.
Bah, lemme correct myself. MIPS* and PowerPC fail due to sleep taking slightly too long. x86 is the only one where it's just way too fast. But this is not true once Linux itself is up.
U-Boot uses x86 TSC timer as the delay timer. Right now the timer frequency is hard-coded in device tree. Auto-calibration does not work on QEMU. I need have another check to see if we can do auto-calibration.
OK, thanks. So long as the problem is known it's not a high priority thing, esp if you're sure it's not going to cascade to problems elsewhere. I just use -k 'not sleep' in my jenkins job now.
Hi!
Bah, lemme correct myself. MIPS* and PowerPC fail due to sleep taking slightly too long. x86 is the only one where it's just way too fast. But this is not true once Linux itself is up.
U-Boot uses x86 TSC timer as the delay timer. Right now the timer frequency is hard-coded in device tree. Auto-calibration does not work on QEMU. I need have another check to see if we can do auto-calibration.
TSC is not really suitable for delays on x86. (Actually, TSC is pretty much not suitable for anything :-) ). PIT should be easy to use and work everywhere...
Pavel
participants (3)
-
Bin Meng -
Pavel Machek -
Tom Rini