[U-Boot] Query on timeouts with NetSetTimeout
Hi,
Recently I ran into some timeout issues with Ethernet on U-Boot 2009.01 for OMAP3 EVM boards.
NetSetTimeout() function provide a way to get timeouts for network related operations. I understand
From the code now that the NetSetTimeout() function takes 2 params , the 1st being the timeout value
and the second one being the function to be invoked when timeout expires. The issue is now the NetSetTimeout() takes timeout in milliseconds only i.e the first parameter has to be a count in milliseconds.
The NetSetTimeout() invokes get_timer() to do its operations. The get_timer() should return the counter value. It's not always true that the counter runs at millisecond clock. I believe the earlier versions of the NetSetTimeout calls in u-boot/net directory used to have a multiple of CONFIG_SYS_HZ for timeout that could easily be used to get required timeout for different platforms.
Regards Manikandan
Hi Minikandan,
Pillai, Manikandan wrote:
Hi,
Recently I ran into some timeout issues with Ethernet on U-Boot 2009.01 for OMAP3 EVM boards.
NetSetTimeout() function provide a way to get timeouts for network related operations. I understand From the code now that the NetSetTimeout() function takes 2 params , the 1st being the timeout value and the second one being the function to be invoked when timeout expires. The issue is now the NetSetTimeout() takes timeout in milliseconds only i.e the first parameter has to be a count in milliseconds.
The NetSetTimeout() invokes get_timer() to do its operations. The get_timer() should return the counter value. It's not always true that the counter runs at millisecond clock. I believe the earlier versions of the NetSetTimeout calls in u-boot/net directory used to have a multiple of CONFIG_SYS_HZ for timeout that could easily be used to get required timeout for different platforms.
get_timer() is expected to return milliseconds. If your architecture isn't doing this it should be fixed. You'll find much about this in the archives.
regards, Ben
Dear "Pillai, Manikandan",
In message 19F8576C6E063C45BE387C64729E739403FA90441B@dbde02.ent.ti.com you wrote:
The NetSetTimeout() invokes get_timer() to do its operations. The get_timer() should return the counter value. It's not always true that the counter runs at millisecond clock. I believe the earlier
We call this a bug, then. get_timer() is specified to return tick os one millisecond resolution.
If your code behaves differntly, it needs to be fixed.
versions of the NetSetTimeout calls in u-boot/net directory used to have a multiple of CONFIG_SYS_HZ for timeout that could easily be used to get required timeout for different platforms.
That code was cleaned up some time ago. If there are board ports which define CONFIG_SYS_HZ to anything else byt the mandatory value of 1000 (like OMAP seems to do), then such ports are broken and need to be fixed.
Best regards,
Wolfgang Denk
Hi Mani,
Wolfgang Denk wrote:
Dear "Pillai, Manikandan",
In message 19F8576C6E063C45BE387C64729E739403FA90441B@dbde02.ent.ti.com you wrote:
The NetSetTimeout() invokes get_timer() to do its operations. The get_timer() should return the counter value. It's not always true that the counter runs at millisecond clock. I believe the earlier
We call this a bug, then. get_timer() is specified to return tick os one millisecond resolution.
If your code behaves differently, it needs to be fixed.
versions of the NetSetTimeout calls in u-boot/net directory used to have a multiple of CONFIG_SYS_HZ for timeout that could easily be used to get required timeout for different platforms.
That code was cleaned up some time ago. If there are board ports which define CONFIG_SYS_HZ to anything else by the mandatory value of 1000 (like OMAP seems to do), then such ports are broken and need to be fixed.
Do you kindly like to look into the two issues Wolfgang mentions here, test and send a patch? Sounds like you understood the issue and can test it ;)
Many thanks
Dirk
P.S.: Looping in BeagleBoard list.
participants (4)
-
Ben Warren -
Dirk Behme -
Pillai, Manikandan -
Wolfgang Denk