Re: [PATCH 1/3] dt-bindings: mtd: stm32_fmc2: add STM32 FMC2 NAND controller documentation
From: Boris Brezillon
Date: Mon Sep 24 2018 - 13:17:41 EST
Hi Christophe,
On Mon, 24 Sep 2018 18:36:27 +0200
Christophe Kerello <christophe.kerello@xxxxxx> wrote:
> >> +- st,fmc2_timings: array of 8 bytes for NAND timings. The meanings of
> >> + these bytes are:
> >> + byte 0 TCLR : CLE to RE delay in number of AHB clock cycles, only 4 bits
> >> + are valid. Zero means one clock cycle, 15 means 16 clock
> >> + cycles.
> >> + byte 1 TAR : ALE to RE delay, 4 bits are valid. Same format as TCLR.
> >> + byte 2 THIZ : number of HCLK clock cycles during which the data bus is
> >> + kept in Hi-Z (tristate) after the start of a write access.
> >> + Only valid for write transactions. Zero means 1 cycle,
> >> + 255 means 256 cycles.
> >> + byte 3 TWAIT : number of HCLK clock cycles to assert the command to the
> >> + NAND flash in response to SMWAITn. Zero means 1 cycle,
> >> + 255 means 256 cycles.
> >> + byte 4 THOLD_MEM : common memory space timing
> >> + number of HCLK clock cycles to hold the address (and data
> >> + when writing) after the command deassertion. Zero means
> >> + 1 cycle, 255 means 256 cycles.
> >> + byte 5 TSET_MEM : common memory space timing
> >> + number of HCLK clock cycles to assert the address before
> >> + the command is asserted. Zero means 1 cycle, 255 means 256
> >> + cycles.
> >> + byte 6 THOLD_ATT : attribute memory space timing
> >> + number of HCLK clock cycles to hold the address (and data
> >> + when writing) after the command deassertion. Zero means
> >> + 1 cycle, 255 means 256 cycles.
> >> + byte 7 TSET_ATT : attribute memory space timing
> >> + number of HCLK clock cycles to assert the address before
> >> + the command is asserted. Zero means 1 cycle, 255 means 256
> >> + cycles.
> >
> > Let me review the driver but this array of timings is really
> > suspicious. I am pretty sure you don't need it in the DT.
>
> "st,fmc2-timings" is an optional property that allow the end user to
> overwrite the timings calculated by setup_data_interface callback. By
> setting this property in the NAND flash memory device tree node, the end
> user could have a better throughput. For NON ONFI SLC NAND, timing mode
> 0 is often used.
Exactly the kind of tweaking I'd like to avoid. If the NAND is not ONFI,
the vendor driver (nand_<manufacturer>.c) can overwrite
chip->default_onfi_timing_mode, and if the ONFI timings modes are not
exactly matching the NAND spec and you need the exact timings, then we
should consider adding a manufacturer hook to let the manufacturer
driver tweak the timings. In any case, I'm not willing to accept
timings description in the DT.
Regards,
Boris