Thanks Glenn for suggestion. But unfortunately it dint work.
*Issue must be related to USB host driver (ehci-omap.c) because USB mass storage - panda being the host, usb stick being the device - also has transfer issue (Write). An interrupt is not fired properly in TX, RX is fine.* Some analysis here, https://groups.google.com/forum/?fromgroups=#!search/android$20$2B$20disable$20CPU$20idle/pandaboard/f45T_Q9a2R0/7QMSssJfUrgJ still searching for way to enable interrupt for writing to ethernet as the interrupt for read works well.... BR On Tuesday, October 2, 2012 5:54:53 AM UTC+3, Glenn Kasten wrote: > > I'm not familiar with panda, but I've used this in the kernel > configuration file for another device: > CONFIG_CPU_IDLE=n > But watch out because it will use more power of course. > > On Monday, October 1, 2012 7:59:40 AM UTC-7, Sagar Shinde wrote: >> >> Hi, >> >> We were trying to figure out the network throughput on eth0 using iperf >> to send data over Ethernet. >> We are using ICS 4.0.4 with kernel 3.0.8 on Pandaboard. Eth0 gives very >> bad throughput in fact it almost does not work. >> >> One strange thing we noticed is that when we keep the display (connected >> to pandaboard) busy (eg. keep moving mouse or have GIF in browser), then >> the throughput suddenly improves a lot. Infact it seems that the networking >> over eth0 starts working. Same was the result when I run a script in the >> background which just does "ls" in a loop after every 1ms interval. >> >> Only conclusion, I could draw was that CPU goes idle in ICS and thus the >> throughput suffers. >> I searched and found out that WFI (wait for interrupt) when enabled makes >> CPU goes idle until given an interrupt. >> I tried to disable it by adding "nohlt" in kernel boot args, but the >> throughput din't improve. >> >> Is there something more we can do prevent CPU from going idle? >> >> BR >> Sagar >> > -- unsubscribe: android-kernel+unsubscr...@googlegroups.com website: http://groups.google.com/group/android-kernel
