Re: [RFC PATCH v2 2/2] mtd: rawnand: meson: add support for Amlogic NAND flash controller

[Liang Yang <liang.yang@amlogic.com>]

Hi Boris,

On 8/17/2018 9:56 PM, Boris Brezillon wrote:
> On Fri, 17 Aug 2018 21:03:59 +0800
> Liang Yang <liang.yang@amlogic.com> wrote:
> 
>> Hi Boris,
>> On 2018/8/2 5:50, Boris Brezillon wrote:
>>
>>> Hi Yixun,
>>>
>>> On Thu, 19 Jul 2018 17:46:12 +0800
>>> Yixun Lan <yixun.lan@amlogic.com> wrote:
>>>
>>> I haven't finished reviewing the driver yet (I'll try to do that later
>>> this week), but I already pointed a few things to fix/improve.
>>>   
>>>> +
>>>> +static int meson_nfc_exec_op(struct nand_chip *chip,
>>>> +			     const struct nand_operation *op, bool check_only)
>>>> +{
>>>> +	struct mtd_info *mtd = nand_to_mtd(chip);
>>>> +	struct meson_nfc *nfc = nand_get_controller_data(chip);
>>>> +	const struct nand_op_instr *instr = NULL;
>>>> +	int ret = 0, cmd;
>>>> +	unsigned int op_id;
>>>> +	int i;
>>>> +
>>>> +	for (op_id = 0; op_id < op->ninstrs; op_id++) {
>>>> +		instr = &op->instrs[op_id];
>>>> +		switch (instr->type) {
>>>> +		case NAND_OP_CMD_INSTR:
>>>> +			cmd = nfc->param.chip_select | NFC_CMD_CLE;
>>>> +			cmd |= instr->ctx.cmd.opcode & 0xff;
>>>> +			writel(cmd, nfc->reg_base + NFC_REG_CMD);
>>>> +			meson_nfc_cmd_idle(nfc, NAND_TWB_TIME_CYCLE);
>>> This is not necessarily TWB you have to wait after a CMD cycle. It can
>>> be tWHR. And you should definitely not hardcode the value, since,
>>> AFAIR, it depends on the selected SDR timings. Probably something you
>>> should calculate in ->setup_data_interface().
>>
>> Indeed. TWB is not necessarily. And tWHR will be promised by NFC.
>> so I will delete it.
> 
> Are you sure the engine always applies a tWHR delay, even when tWB is
> expected? tWB should be applied everytime you are about to wait for a
> R/B transition. tWHR is about switching IO pins from input to output on
> the NAND chip side.
> 

it seems work well even do not add tWHR, but software needs to promise 
tWHR, also the same as TWB. I will check the code and add them.

>>
>>>> +			meson_nfc_drain_cmd(nfc);
>>> I don't know exactly how the NAND controller works, but it's usually
>>> not a good idea to execute the operation right away, especially if you
>>> have address/cmd/data cycles following this cmd and those can be
>>> packed in the same controller operation.
>>
>> it doesn't need meson_nfc_drain_cmd(nfc) here. i will delete it next version
> 
> What's the CMD queue depth? I think you'll have to ensure the requested
> op fits in the CMD FIFO and split things in several sub-ops if it does
> not.
> 

there are maximum 32 commands.
I think it should be enough to promise ONE maximum number of 
operations(cmd - addr - dma - 2 ilde commands).

>>
>>>> +			break;
>>>> +
>>>> +		case NAND_OP_ADDR_INSTR:
>>>> +			for (i = 0; i < instr->ctx.addr.naddrs; i++) {
>>>> +				cmd = nfc->param.chip_select | NFC_CMD_ALE;
>>>> +				cmd |= instr->ctx.addr.addrs[i] & 0xff;
>>>> +				writel(cmd, nfc->reg_base + NFC_REG_CMD);
>>>> +			}
>>>> +			break;
>>>> +
>>>> +		case NAND_OP_DATA_IN_INSTR:
>>>> +			meson_nfc_read_buf(mtd, instr->ctx.data.buf.in,
>>>> +					   instr->ctx.data.len);
>>>> +			break;
>>>> +
>>>> +		case NAND_OP_DATA_OUT_INSTR:
>>>> +			meson_nfc_write_buf(mtd, instr->ctx.data.buf.out,
>>>> +					    instr->ctx.data.len);
>>> Well, I'm not entirely sure what happens when you call
>>> read/write_buf(), but it seems you're doing that one byte at a time,
>>> and that sounds not so efficient given the operation you do for each
>>> byte read/written. Don't you have a way to tell the engine that you
>>> want to read/write X bytes?
>>
>> As i known, there is no way to read/write X bytes once.
> 
> Okay, then maybe you can queue several byte read/write reqs before
> flushing the queue (meson_nfc_drain_cmd() +
> meson_nfc_wait_cmd_finish()).
> 
>>
>>>> +			break;
>>>> +
>>>> +		case NAND_OP_WAITRDY_INSTR:
>>>> +			mdelay(instr->ctx.waitrdy.timeout_ms);
>>>> +			ret = nand_soft_waitrdy(chip,
>>>> +						instr->ctx.waitrdy.timeout_ms);
>>> Hm, i'd be surprised if the controller does not have a way to optimize
>>> waits on R/B transitions.
>>
>> When i delete the delay here, erasing operation will be failed.
>> Does it mean NFC send 0x70 to nand device when rb is busy(low)?
> 
> I was not even talking about the delay, but yes, mdelay() seems way too
> big. Remember that it's a timeout, and you usually don't have to wait
> that much. You can do ndelay(instr->ctx.delay_ns) before calling
> nand_soft_waitrdy() to make sure tWB is enforced.
> 
> Anyway, that's not what I was initially referring to. What I meant is
> that nand_soft_waitrdy() should be replaced by native R/B pin or status
> polling wait logic so that the CPU is released while waiting for a R/B
> transition.
> 
>> If so, i will ask our NFC designer for comfirmation or grasping the waveform.
> 
> You have to wait tWB, that's for sure.
> 

Indeed.

>>
>>>> +			break;
>>>> +		}
>>>> +	}
>>>> +	return ret;
>>>> +}
>>>> +
>>>> +static int meson_ooblayout_ecc(struct mtd_info *mtd, int section,
>>>> +			       struct mtd_oob_region *oobregion)
>>>> +{
>>>> +	struct nand_chip *chip = mtd_to_nand(mtd);
>>>> +	int free_oob;
>>>> +
>>>> +	if (section >= chip->ecc.steps)
>>>> +		return -ERANGE;
>>>> +
>>>> +	free_oob = (section + 1) * 2;
>>>> +	oobregion->offset = section * chip->ecc.bytes + free_oob;
>>> Hm, this offset calculation looks weird. Are you sure it's correct?
>>> I'd bet on something like:
>>>
>>> 	oobregion->offset = 2 + (section * (chip->ecc.bytes + 4));
>>
>> Each ecc page have 2 user bytes. Assume one 2KB+64B page size nand
>> flash using ECC8/1KB which ecc parity bytes is 14B.
>>       _ _ _ _ _ _ _ _ _ _ _ _  _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
>>      |             |  |      |             |  |     |  not  |
>>      |	1KB      |2B| 14B  |     1KB     |2B| 14B | used  |  (layout on nand)
>>      |_ _ _ _ _ _ _|_ |_ _ _ | _ _ _ _ _ _ |_ |_ _ _|_ _ _ _|
>> 		    (2KB + 64B)
>> when reading from nand, I will format the page as follow:
>>       _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _  _ _ _ _
>>      |             |             |  |     |  |      |  not  |
>>      |	1KB      |    1KB      |2B| 14B |2B|  14B | used  |(layout on ddr)
>>      |_ _ _ _ _ _ _|_ _ _ _ _ _ _|_ |_ _ _|_ |_ _ _ |_ _ _ _|
>> 		    (2KB + 64B)
>> So i get "oobregion->offset = section * chip->ecc.bytes + free_oob".
> 
> Okay, but I prefer when it's written like that:
> 
> 	oobregion->offset = 2 + (section * (2 + chip->ecc.bytes));

em, I prefer too. it looks better.
> >> Maybe i don't get what does 'section' mean. i think it means the ecc 
page number.
> 
> Section is just the free OOB or ECC section number. It starts at 0 and
> goes up to N - 1, where N usually is the number of ECC steps you have in
> a page.
> 
> .
>