On Sun, 9 Mar 2025, at 7:31 PM, Jernej Skrabec wrote: > It turns out that checking just one write is not enough. Due to > unexplained reasons scan procedure detected double the size. By making > 16 dword writes and comparisons that never happens. > > New procedure is also inverted. Instead of writing two different values > to base address and some offset and then reading both and comparing > values, simplify this by writing pattern at the base address and then > search for this pattern at some offset. > > Signed-off-by: Jernej Skrabec <[email protected]> > --- > arch/arm/mach-sunxi/dram_sun50i_h616.c | 58 +++++++++++++++++++++++++- > 1 file changed, 56 insertions(+), 2 deletions(-)
Tested with H700 devices with both 1GB (RG35XX-H) and 4GB (RG40XX-V) of LPDDR4; reliable detection with over 50 boots combined on both devices. Previously ~1 in 10 boots would mistakenly incorrectly detect RAM size, in particular 2 or 4 GB RAM on the 1GB device. Tested-by: Ryan Walklin <[email protected]> Regards, Ryan
