RE: [LINUX PATCH v14] mtd: rawnand: pl353: Add basic driver for arm pl353 smc nand interface

[Naga Sureshkumar Relli <nagasure@xilinx.com>]

Hi Helmut,

> -----Original Message-----
> From: Helmut Grohne <helmut.grohne@intenta.de>
> Sent: Monday, April 29, 2019 5:48 PM
> To: Naga Sureshkumar Relli <nagasure@xilinx.com>
> Cc: bbrezillon@kernel.org; miquel.raynal@bootlin.com; richard@nod.at;
> dwmw2@infradead.org; computersforpeace@gmail.com; marek.vasut@gmail.com; linux-
> mtd@lists.infradead.org; linux-kernel@vger.kernel.org; Michal Simek <michals@xilinx.com>;
> nagasureshkumarrelli@gmail.com
> Subject: Re: [LINUX PATCH v14] mtd: rawnand: pl353: Add basic driver for arm pl353 smc
> nand interface
> 
> On Mon, Apr 29, 2019 at 11:31:14AM +0000, Naga Sureshkumar Relli wrote:
> > But just wanted to know, do you see issues with these __force and __iomem castings?
> 
> I only see a minor issue: They're (deliberately) lengthy. Using many of them diverts attention
> of the reader. Therefore, my proposal attempted to reduce their frequency. The only issue I see
> here is readability.
Ok then, I will update it.
> 
> > >
> > > > +	u8 addr_cycles;
> > > > +	struct clk *mclk;
> > >
> > > All you need here is the memory clock frequency. Wouldn't it be
> > > easier to extract that frequency once during probe and store it
> > > here? That assumes a constant frequency, but if the frequency isn't constant, you have a
> race condition.
> > That is what we are doing in the probe.
> > In the probe, we are getting mclk using of_clk_get() and then we are
> > getting the actual frequency Using clk_get_rate().
> > And this is constant frequency only(getting from dts)
> 
> Not quite. You're getting a clock reference in probe and then repeatedly access the frequency
> elswhere. I am suggesting that you get the clock frequency during probe and never save the
> clock reference to a struct.
Ok. got it. Will update.
> 
> > > > +		case NAND_OP_ADDR_INSTR:
> > > > +			offset = nand_subop_get_addr_start_off(subop, op_id);
> > > > +			naddrs = nand_subop_get_num_addr_cyc(subop, op_id);
> > > > +			addrs = &instr->ctx.addr.addrs[offset];
> > > > +			nfc_op->addrs = instr->ctx.addr.addrs[offset];
> > > > +			for (i = 0; i < min_t(unsigned int, 4, naddrs); i++) {
> > > > +				nfc_op->addrs |= instr->ctx.addr.addrs[i] <<
> > >
> > > I don't quite understand what this code does, but it looks strange
> > > to me. I compared it to other drivers. The code here is quite
> > > similar to marvell_nand.c. It seems like we are copying a varying
> > > number (0 to 6) of addresses from the buffer instr->ctx.addr.addrs.
> > > However their indices are special: 0, 1, 2, 3, offset + 4, offset + 5. This is non-consecutive
> and different from marvell_nand.c in this regard. Could it be that you really meant index
> offset+i here?
> > I didn't get, what you are saying here.
> > It is about updating page and column addresses.
> > Are you asking me to remove nfc_op->addrs = instr->ctx.addr.addrs[offset]; before for
> loop?
> 
> I compared this code to marvell_nand.c and noticed a subtle difference.
> Both snippets read 6 address bytes and consume them in a driver-specific way. Now which
> address bytes are consumed differs.
> 
> marvell_nand.c consumes instr->ctx.addr.addrs at indices offset,
> offset+1, offset+2, offset+3, offset+4, offset+5. pl353_nand.c consumes
> instr->ctx.addr.addrs at indices 0, 1, 2, 3, offset, offset+4, offset+5.
> (In my previous mail, I didn't notice that it was also consuming the offset index.)
> 
> I would have expected this behaviour to be consistent between different drivers. If I assume
> marvell_nand.c to do the right thing and pl353_nand.c to be wrong (which is not necessarily a
> correct assumption), then the code woule likely becom:
> 
> 	addrs = &instr->ctx.addr.addrs[offset];
> 	for (i = 0; i < min_t(unsigned int, 4, naddrs); i++) {
> 		nfc_op->addrs |= addrs[i] << (8 * i);
> 		              // ^^^^^
> 	}
> 
> Hope this helps.
Ok. let me re check this and I will update this accordingly.

Thanks,
Naga Sureshkumar Relli
> 
> Helmut