Re: [PATCH v2 0/2] mtd: nand: wait for tWHR, and fix the setup_data_interface of Denali

[Boris Brezillon]

On Fri, 29 Sep 2017 23:06:42 +0900
Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote:

> 2017-09-29 21:26 GMT+09:00 Boris Brezillon <boris.brezillon@xxxxxxxxxxxxxxxxxx>:
> > On Fri, 29 Sep 2017 19:38:38 +0900
> > Masahiro Yamada <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> >  
> >> 1/2 : add NAND_WAIT_TWHR and nand_whr_delay().
> >>       You can set this new flag if you want nand_command(_lp)
> >>       to insert tWHR delay where needed.
> >>
> >> 2/2 : Fix Denali setup_data_interface.
> >>       Boris' suggestion in v1 was a good reminder that
> >>       made me realize tCCS was missing in the driver.  Fix it now.
> >>
> >>
> >> Changes in v2:
> >>   - Add nand_whr_delay() helper
> >>     Wait for tWHR only for drivers that explicitly set NAND_WAIT_TWHR flag
> >>   - newly added
> >>
> >> Masahiro Yamada (2):
> >>   mtd: nand: wait for tWHR after NAND_CMD_STATUS / NAND_CMD_READID  
> >
> > Hm, I thought you were introducing this to then use it in the denali
> > driver. Sorry, but I don't want to apply something that nobody needs.
> > If someone ever complains about a missing delay I'll point him to your
> > patch, but until then I'll keep the core unchanged.  
> 
> 
> At first, I thought this was necessary for me,
> but I realized it was my misunderstanding.
> 
> 
> Please let me explain one more.
> 
> See commit 3158fa0e739615769cc047d2428f30f4c3b6640e.
> 
> Prior to that commit, READID waited for #R/B transition,
> it was wrong, so I fixed it.
> 
> However, it dropped the delay completely.
> If somebody was implicitly relying on the delay of chip->dev_ready,
> the first byte might be read out before the valid data
> is available.
> 
> This was motivation of v1, where inserted ndelay(200)
> unconditionally.

Okay. Anyway, this extra delay is activated with an opt-in flag (I
know, I'm the one who asked that), and noone set this flag in
chip->options, so, if there's a bug, it's still here even after
applying "mtd: nand: wait for tWHR after NAND_CMD_STATUS /
NAND_CMD_READID".

Honestly, I think all advanced controllers have the tWHR/tRHW timings
enforced in the HW logic (configurable through a reg). This leaves
basic controllers like the nand-gpio one, and even for these ones, the
delay between the chip->cmd_ctrl(ADDR) and chip->read_buf() calls is
probably long enough to hide the problem.

Note that I'm absolutely not against this patch, it's just that I'd
like to have a real user before merging this logic.

> 
> 
> 
> 
> 
> >>   mtd: nand: denali: fix setup_data_interface to meet tCCS delay  
> >
> > This one is valid. I'll queue it to nand/next soon.  
> 
> If you drop 1/2, please let me do v3.
> 
> V2 mentions NAND_WAIT_TWHR, this is strange.
> 

Sure.

> 
>