Hi Alexey, On 1/6/2026 7:26 PM, Alexey Charkov wrote: > On Tue, Jan 6, 2026 at 6:09 PM Neil Armstrong <[email protected]> > wrote: >> >> On 1/6/26 14:57, Alexey Charkov wrote: >>> Rockchip RK3576 can boot fully from UFS, but as part of the boot >>> process the SPL needs to place part of the ATF code into PMU SRAM, >>> which is outside the SDRAM range that the UFSHC can access. >>> >>> Enable a quirk to force bounce operation for this case. >>> >>> Signed-off-by: Alexey Charkov <[email protected]> >>> --- >>> drivers/ufs/Kconfig | 1 + >>> drivers/ufs/ufs-rockchip.c | 1 + >>> 2 files changed, 2 insertions(+) >>> >>> diff --git a/drivers/ufs/Kconfig b/drivers/ufs/Kconfig >>> index 6c75bb2a0790..2543b5d47c04 100644 >>> --- a/drivers/ufs/Kconfig >>> +++ b/drivers/ufs/Kconfig >>> @@ -76,6 +76,7 @@ config UFS_RENESAS_GEN5 >>> config UFS_ROCKCHIP >>> bool "Rockchip specific hooks to UFS controller platform driver" >>> depends on UFS >>> + select BOUNCE_BUFFER >>> help >>> This selects the Rockchip specific additions to UFSHCD platform >>> driver. >>> >>> diff --git a/drivers/ufs/ufs-rockchip.c b/drivers/ufs/ufs-rockchip.c >>> index 0384244387da..699460fc4d15 100644 >>> --- a/drivers/ufs/ufs-rockchip.c >>> +++ b/drivers/ufs/ufs-rockchip.c >>> @@ -153,6 +153,7 @@ static int ufs_rockchip_common_init(struct ufs_hba *hba) >>> return err; >>> } >>> >>> + hba->quirks |= UFSHCD_QUIRK_DMA_SDRAM_ONLY; >> >> So you really want to enable this quirk for non-SPL builds aswell ? > > Well I haven't run into any similar issues with the main U-boot, so I > don't have a strong opinion about keeping it in place there. > > However: > 1. Having the quirk enabled doesn't cause all transfers to go via the > bounce buffer. Only those for which the provided destination buffer > happens to be outside SDRAM (which must be extremely rare if at all > possible under U-boot, given that it hasn't caused issues for anyone) > 2. If it turns out that there are indeed cases when the UFSHC needs to > read or write something outside SDRAM while operating under U-boot (I > can't think of a reason now, but then I couldn't think about that ATF > to PMU SRAM thing either, until I ran into it and spent more than half > a day debugging) then I'd rather spend a couple of CPU cycles on a > memmove than end up with a hard lock up, a cryptic error message and a > dump of UFSHC status registers > 3. It's one line less code this way :) > > Happy to add something like "if (xpl_phase() == PHASE_SPL)" though, if > that sounds better to everyone.
You should most likely be able to drop this quirk workaround with the "rockchip: rk3576: Fix DMA transfer of TF-A to PMU_SRAM" series [1][2] that I just sent out. Please try if that also solves the DMA transfer from UFS to PMU SRAM issue, it did for eMMC, SD-card and SPI flash. [1] https://patchwork.ozlabs.org/cover/2181298/ [2] https://lore.kernel.org/r/[email protected]/ Regards, Jonas > Best regards, > Alexey
