On 10/2/18 12:49 PM, Peter Maydell wrote: > On 21 September 2018 at 17:19, Cédric Le Goater <[email protected]> wrote: >> The overhead for the OpenBMC firmware images using the a custom U-Boot >> is around 2 seconds, which is fine, but with a U-Boot from mainline, >> it takes an extra 50 seconds or so to reach Linux. A quick survey on >> the number of reads performed on the flash memory region gives the >> following figures : >> >> OpenBMC U-Boot 922478 (~ 3.5 MBytes) >> Mainline U-Boot 20569977 (~ 80 MBytes) >> >> QEMU must be trashing the TCG TBs and reloading text very often. Some >> addresses are read more than 250.000 times. Until we find a solution >> to improve boot time, execution from MMIO is not activated by default. >> >> Setting this option also breaks migration compatibility. >> >> Signed-off-by: Cédric Le Goater <[email protected]> > > Reviewed-by: Peter Maydell <[email protected]>
The DRAM training in mainline u-boot is also significantly slower on real HW. I haven't spotted the exact area but the code is jumping back and forth. This patch should help tracking the culprit. Thanks, C.
