Hi Jagan, Adam,
On Thu, 3 Jan 2019, Jagan Teki wrote:
On Thu, Jan 3, 2019 at 6:57 PM Zoltan HERPAI <wigy...@uid0.hu> wrote:
Hi all,
The DTS resync between 2018.09 and 2018.11 seems to have broken the MMC
support for the Linksprite pcDuino (A10) and pcDuino v3 (A20) boards.
The resync happened in b9d59d0 [1] and 3c92cca [2], after which u-boot
doesn't recognize the MMC controller and freezes the board (exactly the
same happens on the v3 board).
---- CUT HERE ----
U-Boot SPL 2018.11 (Dec 31 2018 - 14:36:52 +0000)
DRAM: 1024 MiB
CPU: 1008000000Hz, AXI/AHB/APB: 3/2/2
Trying to boot from MMC1
U-Boot 2018.11 (Dec 31 2018 - 14:36:52 +0000) Allwinner Technology
CPU: Allwinner A10 (SUN4I)
Model: LinkSprite pcDuino
I2C: ready
DRAM: 1 GiB
MMC:
---- CUT HERE ----
Reverting these commits solve the problem and the boards boot correctly.
Initially I thought this might be due to removing the
mmc0_cd_pin_reference_design (syncing that from the kernel into u-boot),
which was discussed here [3] and was considered as a move that might
break MMC on some boards, but re-adding that reference pin only in the
pcduino DTSes did not resolve the freeze.
Questions:
- A similar board - where the reference pin is used for CD - is the
Cubieboard 2. Can someone test 2018.11 on it to see if it freezes as
well? Out of A20, I have a Bananapro, but that's using a non-reference
pin for CD, and boots fine.
A10, enable DM_MMC so it can effect DT. or enable CONFIG_MMC0_CD_PIN="PH1"
Added Adam, (who actually tested DM_MMC on A10)
I tested with "either" and "both", neither worked - see below. Along with
CONFIG_DM_MMC, I've also enabled CONFIG_DM_DEBUG to see what's happening -
here is the output:
[...]
MMC: uclass_find_device_by_seq: 0 0
- -1 -1 'mmc@1c0f000'
- not found
uclass_find_device_by_seq: 1 0
- -1 -1 'mmc@1c0f000'
- not found
uclass_find_device_by_seq: 0 -1
uclass_find_device_by_seq: 0 0
- -1 -1 'mmc@1c0f000'
- not found
Entering sunxi_mmc_probe
ofnode_read_u32: bus-width: 0x4 (4)
sunxi_mmc: priv->reg: 1c0f000
sunxi_mmc: priv->mclkreg: 1c20000
ofnode_read_bool: cd-inverted: false
--------
A20, can boot as it is, it doesn't effect mmc node on DT, since DM_MMC
is not available.
- There are some clock-related changes in this DTS resync, however I
have less than optimal experience with clocks. Can someone help check if
these changes might cause the issue?
OK, better dump the clock reg value at the end of sunxi_mmc_probe
For A20 it looks like this:
--------
MMC: Entering sunxi_mmc_probe
sunxi_mmc: priv->reg: 1c0f000
sunxi_mmc: priv->mclkreg: 1c20000
--------
EDIT: on the second run, I've added some debug around the reset call,
and it seems to die there:
+ printf("sunxi_mmc: before reset\n");
/* Reset controller */
writel(SUNXI_MMC_GCTRL_RESET, &priv->reg->gctrl);
udelay(1000);
+ printf("sunxi_mmc: after reset\n");
--------
MMC: Entering sunxi_mmc_probe
sunxi_mmc: priv->reg: 1c0f000
sunxi_mmc: priv->mclkreg: 1c20000
sunxi_mmc: before reset
--------
Thanks,
Zoltan H
