On Tue Mar 10, 2026 at 4:40 PM IST, Neha Francis wrote: > Hi Anshul > > On 3/10/2026 2:46 PM, Anshul Dalal wrote: >> The dram_init_banksize API configures the dram banks inside gd->bd, so >> ensure bd is initialized by calling spl_set_bd as part of >> spl_enable_cache. >> >> Signed-off-by: Anshul Dalal <[email protected]> >> --- >> arch/arm/mach-k3/common.c | 3 +++ >> 1 file changed, 3 insertions(+) >> >> diff --git a/arch/arm/mach-k3/common.c b/arch/arm/mach-k3/common.c >> index 2f3df5519c5..39467c21679 100644 >> --- a/arch/arm/mach-k3/common.c >> +++ b/arch/arm/mach-k3/common.c >> @@ -418,6 +418,9 @@ void spl_enable_cache(void) >> gd->ram_top = CFG_SYS_SDRAM_BASE; >> int ret = 0; >> >> + /* dram_init_banksize populates bd, so init it first */ >> + spl_set_bd(); >> + > > I think we really need to evaluate where all we are calling > dram_init_banksize, > else we keep finding related debugs confusing [0] > > We have to stick as closely as possible to the common SPL flow: board_init_r > --> > spl_set_bd .. dram_init_banksize
spl_set_bd should not re-init bd if it has been initialized already so the second call to spl_set_bd as part of board_init_r shouldn't effect anything. > > Problem is we need dram_init_banksize calls even before relocation because of > spl_enable_cache. For my knowledge, do we have a specific reason why we are > calling spl_enable_cache before relocation? I am guessing some MMU setup > happening before we relocate partly to DDR as being the reason? > SPL does not relocate to my knowledge, we largely need spl_enable_cache to speed up boot especially in A-Core falcon mode where the large kernel image needs to be authenticated and uncompressed by the SPL. > I think it's better you move calling spl_set_bd() up a level into board_init_f > for now (similar to what arch/arm/cpu/armv8/fsl-layerscape/spl.c does). We We could do that too but it'd have to be change in every SoC's init file and the dependency between spl_set_bd and dram_init_banksize would not be enforced (i.e someone can call spl_enable_cache while not calling spl_set_bd first). > should be calling this before we call dram_init (or in our case k3_mem_init) > so > that we are sure it is setup prior to any dram_init_banksize calls. > >> dram_init(); >> dram_init_banksize(); >> >> >> --- >> base-commit: 1e240f7206fccde4ec73ea432ab8360d321c5fe5 >> change-id: 20260310-init_bd_before_dram_init-7a9c3ad0b311 >> >> Best regards, > > > [0] https://lore.kernel.org/u-boot/[email protected]/
