Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII;
 format=flowed
Content-Transfer-Encoding: 7bit
Date:   Mon, 11 Jun 2018 18:52:12 +0530
From:   Abhishek Sahu <absahu@codeaurora.org>
To:     Miquel Raynal <miquel.raynal@bootlin.com>
Cc:     Boris Brezillon <boris.brezillon@bootlin.com>,
        David Woodhouse <dwmw2@infradead.org>,
        Brian Norris <computersforpeace@gmail.com>,
        Marek Vasut <marek.vasut@gmail.com>,
        Richard Weinberger <richard@nod.at>,
        Cyrille Pitchen <cyrille.pitchen@wedev4u.fr>,
        linux-arm-msm@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mtd@lists.infradead.org, Andy Gross <andy.gross@linaro.org>,
        Archit Taneja <architt@codeaurora.org>
Subject: Re: [PATCH v3 13/16] mtd: rawnand: qcom: minor code reorganization
 for bad block check
In-Reply-To: <20180607145326.339170fd@xps13>
References: <1527250904-21988-1-git-send-email-absahu@codeaurora.org>
 <1527250904-21988-14-git-send-email-absahu@codeaurora.org>
 <20180526104629.74315561@xps13>
 <90ae248edf8a06a1d35e2da458f75af5@codeaurora.org>
 <20180528090352.254022ed@xps13>
 <53caff8799d30b6a81ac10f63a7c56c4@codeaurora.org>
 <20180607145326.339170fd@xps13>
Message-ID: <7179eb382dbd3d49e2066178914636fd@codeaurora.org>
User-Agent: Roundcube Webmail/1.2.5
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 2018-06-07 18:23, Miquel Raynal wrote:
> Hi Abhishek,
> 
> On Mon, 28 May 2018 15:40:52 +0530, Abhishek Sahu
> <absahu@codeaurora.org> wrote:
> 
>> On 2018-05-28 12:33, Miquel Raynal wrote:
>> > Hi Abhishek,
>> > >> >>  /* implements ecc->read_page() */
>> >> >>  static int qcom_nandc_read_page(struct mtd_info *mtd, struct >> nand_chip *chip,
>> >> >>  				uint8_t *buf, int oob_required, int page)
>> >> >> @@ -2118,6 +2083,7 @@ static int qcom_nandc_block_bad(struct mtd_info >> *mtd, loff_t ofs)
>> >> >>  	struct nand_ecc_ctrl *ecc = &chip->ecc;
>> >> >>  	int page, ret, bbpos, bad = 0;
>> >> >>  	u32 flash_status;
>> >> >> +	u8 *bbm_bytes_buf = chip->data_buf;
>> >> >> >>  	page = (int)(ofs >> chip->page_shift) & chip->pagemask;
>> >> >> >> @@ -2128,11 +2094,31 @@ static int qcom_nandc_block_bad(struct >> mtd_info *mtd, loff_t ofs)
>> >> >>  	 * that contains the BBM
>> >> >>  	 */
>> >> >>  	host->use_ecc = false;
>> >> >> +	bbpos = mtd->writesize - host->cw_size * (ecc->steps - 1);
>> >> > > Are we sure there is no layout with only 1 step?
>> >> >>   All the layouts are such that, the BBM will come in
>> >>   first byte of spare area.
>> >> >>   For 4 bit ECC, the cw_size is 528 so for 2K page
>> >> >>    2048 - 528 * 3 = 464
>> > > My question was more about small page NANDs. But I suppose it works
>> > too if ecc->steps == 1.
>> >
>>   Correct Miquel.
>> 


>> >> >>   So for last CW, the 464 is BBM (i.e 2048th byte) in
>> >>   full page.
>> >> >> > >> >>  	clear_bam_transaction(nandc);
>> >> >> -	ret = copy_last_cw(host, page);
>> >> >> -	if (ret)
>> >> >> +	clear_read_regs(nandc);
>> >> >> +
>> >> >> +	set_address(host, host->cw_size * (ecc->steps - 1), page);
>> >> >> +	update_rw_regs(host, 1, true);
>> >> >> +
>> >> >> +	/*
>> >> >> +	 * The last codeword data will be copied from NAND device to NAND
>> >> >> +	 * controller internal HW buffer. Copy only required BBM size bytes
>> >> >> +	 * from this HW buffer to bbm_bytes_buf which is present at
>> >> >> +	 * bbpos offset.
>> >> >> +	 */
>> >> >> +	nandc_set_read_loc(nandc, 0, bbpos, host->bbm_size, 1);
>> >> >> +	config_nand_single_cw_page_read(nandc);
>> >> >> +	read_data_dma(nandc, FLASH_BUF_ACC + bbpos, bbm_bytes_buf,
>> >> >> +		      host->bbm_size, 0);
>> >> >> +
>> >> >> +	ret = submit_descs(nandc);
>> >> >> +	free_descs(nandc);
>> >> >> +	if (ret) {
>> >> >> +		dev_err(nandc->dev, "failed to copy bad block bytes\n");
>> >> >>  		goto err;
>> >> >> +	}
>> >> >> >>  	flash_status = le32_to_cpu(nandc->reg_read_buf[0]);
>> >> >> >> @@ -2141,12 +2127,10 @@ static int qcom_nandc_block_bad(struct >> mtd_info *mtd, loff_t ofs)
>> >> >>  		goto err;
>> >> >>  	}
>> >> >> >> -	bbpos = mtd->writesize - host->cw_size * (ecc->steps - 1);
>> >> >> -
>> >> >> -	bad = nandc->data_buffer[bbpos] != 0xff;
>> >> >> +	bad = bbm_bytes_buf[0] != 0xff;
>> >> > > This is suspect as it still points to the beginning of the data buffer.
>> >> > Can you please check you did not meant bbm_bytes_buf[bbpos]?
>> >> >
>> >>   The main thing here is
>> >>   nandc_set_read_loc(nandc, 0, bbpos, host->bbm_size, 1);
>> >> >>   After reading one complete CW from NAND, the data will be still
>> >>   in NAND HW buffer.
>> >> >>   The above register tells that we need to read data from
>> >>   bbpos of size host->bbm_size (which is 1 byte for 8 bus witdh
>> >>   and 2 byte for 16 bus width) in bbm_bytes_buf.
>> > > I see: idx 0 in bbm_bytes_buf is the data at offset bbpos. Then
>> > it's ok.
>> > >> >>   So bbm_bytes_buf[0] will contain the BBM first byte.
>> >>   and bbm_bytes_buf[1] will contain the BBM second byte.
>> >> >>   Regards,
>> >>   Abhishek
>> >> >> >> >>  	if (chip->options & NAND_BUSWIDTH_16)
>> >> >> -		bad = bad || (nandc->data_buffer[bbpos + 1] != 0xff);
>> >> >> +		bad = bad || (bbm_bytes_buf[1] != 0xff);
>> > > Sorry, my mistake, I did not see the above line.
>> > > However, technically, the BBM could be located in the first, second or
>> > last page of the block. You should check the three of them are 0xFF
>> > before declaring the block is not bad.
>> > > The more I look at the function, the more I wonder if you actually need
>> > it. Why does the generic nand_block_bad() implementation in the core
>> > do not fit?
>> 
>>   The BBM bytes can be accessed in raw mode only for QCOM NAND
>>   Contoller. We started with following patch for initial patches
>> 
>>   http://patchwork.ozlabs.org/patch/508565/
>> 
>>   I am also not very much sure, how can we go ahead now.
>>   Ideally we need to use generic function only which
>>   requires raw_read.
>> 
> 
> I see, thanks for pointing this thread.
> 
> Well for now then let's keep our driver-specific implementation.
> 
> I will just ask you to do a consistent check as requested above (you
> can copy code from the core) and add a comment above this function
> explaining why it is needed (what you just told me).
> 

  Hi Miquel,

  I explored more regarding making custom bad block functions in this
  thread and it looks like, we can move to generic block_bad function
  by small changes in QCOM NAND driver
  only. The main problem was, in read page with ECC, the bad block
  byte was skipped.

  But controller is copying the bad block bytes in another register
  with following status bytes.

  BAD_BLOCK_STATUS : With every page read operation, when the controller
  reads a page with a bad block, it writes the bad block status data into
  this register.

  We can update the BBM bytes at start of OOB data in read_oob function
  with these status bytes. It will help in getting rid of driver-specific
  implementation for chip->block_bad.

  For chip->block_markbad, if we want to get rid of
  driver-specific implementation then we can have
  following logic

  in write_oob function check for bad block bytes in oob
  and do the raw write for updating BBM bytes alone in
  flash if BBM bytes are non 0xff.

  Thanks,
  Abhishek