Hi Tim and Andrey,
I find the root cause, it is because during the 1.8v voltage switch process, SD clock do not gate off/ on according to the SD spec. For i.MX usdhc, if want to gate off the sd card clock, need to clear the bit 8 of VEND_SPEC, and set this bit to gate on the sd card clock. For the bit[14:11] of VEND_SPEC, our IP do not implement the function, just mark as reserved bits. So for current logic, we'll see the second mmc_wait_dat0() in mmc_switch_voltage() always return timeout.
Let me double confirm with our IC team, and will send a formal fix patch tomorrow.
Haibo
-----Original Message----- From: Bough Chen Sent: 2021年1月22日 20:42 To: 'ZHIZHIKIN Andrey' andrey.zhizhikin@leica-geosystems.com; tharvey@gateworks.com Cc: Adam Ford aford173@gmail.com; Fabio Estevam festevam@gmail.com; Peng Fan peng.fan@nxp.com; u-boot u-boot@lists.denx.de; Stefano Babic sbabic@denx.de; Jaehoon Chung jh80.chung@samsung.com Subject: RE: IMX8MM SD UHS support
Hi Tim and Andrey
I can reproduce this issue on my side, after revert my patch which you point out, SD can work on SDR104 mode. Till now, I do not know why wait for data0 status will cause this issue. Give me more time, I will try to dig that out.
Again, thanks for report this issue.
Haibo
-----Original Message----- From: ZHIZHIKIN Andrey [mailto:andrey.zhizhikin@leica-geosystems.com] Sent: 2021年1月20日 4:48 To: tharvey@gateworks.com Cc: Bough Chen haibo.chen@nxp.com; Adam Ford
Fabio Estevam festevam@gmail.com; Peng Fan peng.fan@nxp.com; u-boot u-boot@lists.denx.de; Stefano Babic sbabic@denx.de; Jaehoon Chung jh80.chung@samsung.com Subject: RE: IMX8MM SD UHS support
Hello Tim,
-----Original Message----- From: Tim Harvey tharvey@gateworks.com Sent: Tuesday, January 19, 2021 6:32 PM To: ZHIZHIKIN Andrey andrey.zhizhikin@leica-geosystems.com Cc: haibo.chen@nxp.com; Adam Ford aford173@gmail.com; Fabio
Estevam
festevam@gmail.com; Peng Fan Peng.Fan@nxp.com; u-boot <u- boot@lists.denx.de>; Stefano Babic sbabic@denx.de; Jaehoon Chung jh80.chung@samsung.com Subject: Re: IMX8MM SD UHS support
On Mon, Jan 18, 2021 at 11:38 AM ZHIZHIKIN Andrey andrey.zhizhikin@leica-geosystems.com wrote:
Hello Tim,
Sorry it took me quite some time to get this sorted out, but I believe I was able
to identify an offending commit that is preventing the USDHC to switch to higher speed modes.
It is in fact b5874b552f ("mmc: fsl_esdhc_imx: add wait_dat0() support"),
reverting it makes SD Card to properly report capabilities and switch to high speed modes.
Can you try to revert this on your end to see if the SD Card would start to
operate in high speed mode?
I'm still investigating why this addition of wait_dat0() caused this, I believe this
is due to the fact that the same wait is already performed while voltage switch to handle the errata, thus this addition wait might
erroneously timeout.
++ Haibo Chen haibo.chen@nxp.com
Haibo,
Can you please explain the purpose of adding wait_dat0() Introduced with
commit b5874b552f? It is not clear from the commit message what was the purpose of adding it. Have you tested the USDHC switch to higher modes with this change?
Andrey,
Reverting b5874b552f ("mmc: fsl_esdhc_imx: add wait_dat0() support") does not resolve the issue. That function waits for a specified timeout for the card to assert DAT[0] either high or low depending on arg and is used to check for command busy or failure. The patch in question adds that function so that the common mmc code can utilize it. If the function does not exist in the host controller driver any call to mmc_wait_dat0 returns -ENOSYS so it must be there to support UHS anyway.
Perhaps, this is due to the fact that the same wait cycle is already executed during the invocation of mmc_send_cmd above the mmc_wait_dat0() in drivers/mmc/mmc.c.
mmc_send_cmd calls esdhc_send_cmd_common in drivers/mmc/fsl_esdhc_imx.c, which already has a wait loop implemented to cover the "Workaround for ESDHC errata ENGcm03648" (as seen from the comment), which might explain why consecutive wait fails to return within timeout value.
What is not clear to me is why the card would hold DAT[0] low on the voltage switch indicating a failure as it does not in the Linux driver which appears to me to be doing the same thing.
This behavior might be explained by the implementation I traced above.
Again, if we ignore the return code UHS works fine and I'm not at all clear why you wouldn't run into this issue: diff --git a/drivers/mmc/mmc.c b/drivers/mmc/mmc.c index a6394bc..5ea3cd2 100644 --- a/drivers/mmc/mmc.c +++ b/drivers/mmc/mmc.c @@ -579,10 +579,12 @@ static int mmc_switch_voltage(struct mmc
*mmc,
int signal_voltage) * dat[0:3] low. Wait for at least 1 ms according to spec */ err = mmc_wait_dat0(mmc, 1, 1000); +#if 0 // hack: not clear why card always holds DAT[0] high here on fsl_esdhc_imx if (err == -ENOSYS) udelay(1000); else if (err) return -ETIMEDOUT; +#endif
return 0;}
This is expected. When the wait_dat0 callback is undefined in dm_mmc_ops structure - it defaults to return -ENOSYS, which effectively just inhibit a delay in mmc_switch_voltage and returns 0 indicating that call was successful. This all can be seen from the code snippet you've posted above and had commented out under #if0 block.
Looking at the change you've posted, it seems to me that you haven't attempted to revert the patch I mentioned, as by reverting it - the "if (err == -ENOSYS)" path would've been taken and the desired return 0; would've been called.
Honestly I don't have the time to keep delving into this. I don't have any reason to believe that UHS has ever worked with fsl_esdhc_imx and because my boards boot from eMMC (and HS200/HS400 works) I'm not missing out on much by having a slow microSD.
I still believe (and witnessed it myself) that the original implementation was indeed capable of successfully switching the USDHC to
high speed modes.
At this point in time it might not be relevant for your implementation, but could help those who has a severe impact from the microSD RW performance in U-Boot.
Anyways, thanks a lot for writing back on the findings you have!
As for me, it would still be beneficial if the patch author (Haibo) could comment on the intention of its introduction, because I clearly see that reverting it on the current master branch does improve the
behavior.
Best regards,
Tim
Cheers, Andrey