Re: [PATCH] i2ctools: Add capability to write block command

[Guenter Roeck <guenter.roeck-IzeFyvvaP7pWk0Htik3J/w@public.gmane.org>]

On Thu, Jan 27, 2011 at 11:59:08AM -0500, Jean Delvare wrote:
> Hi Guenter,
> 
> On Thu, 27 Jan 2011 08:00:58 -0800, Guenter Roeck wrote:
> > This patch adds support to write block data to i2cset.
> > 
> > I tried to limit the changes as much as possible. Detecting new write modes
> > is a bit tricky since the command supports an undocumented parameter (mask)
> > after the mode.
> 
> This can go away if it bothers you. This was the old way to pass the
> mask value. I have implemented -m meanwhile, and this is what people
> should be using by now. I added it 2 years ago, so I think it's
> acceptable to stop supporting the legacy way.
> 
Would be a separate patch. Also, the current code accepts nonsense parameters 
(such as comments) after the mode parameter. Not sure if I want to change that.

> > So I decided to handle block data first and bypass the rest
> > of the parameter handling code.
> > 
> > Guenter
> > 
> > --
> > Index: tools/i2cset.c
> > ===================================================================
> > --- tools/i2cset.c	(revision 5909)
> > +++ tools/i2cset.c	(working copy)
> > @@ -35,13 +35,15 @@
> >  static void help(void)
> >  {
> >  	fprintf(stderr,
> > -	        "Usage: i2cset [-f] [-y] [-m MASK] I2CBUS CHIP-ADDRESS DATA-ADDRESS [VALUE] [MODE]\n"
> > +	        "Usage: i2cset [-f] [-y] [-m MASK] I2CBUS CHIP-ADDRESS DATA-ADDRESS [VALUE] ... [MODE]\n"
> >  		"  I2CBUS is an integer or an I2C bus name\n"
> >  		"  ADDRESS is an integer (0x03 - 0x77)\n"
> >  		"  MODE is one of:\n"
> >  		"    c (byte, no value)\n"
> >  		"    b (byte data, default)\n"
> >  		"    w (word data)\n"
> > +		"    i (I2C block data)\n"
> > +		"    s (SMBus block data)\n"
> >  		"    Append p for SMBus PEC\n");
> >  	exit(1);
> >  }
> > @@ -78,6 +80,19 @@
> >  			return -1;
> >  		}
> >  		break;
> > +
> > +	case I2C_SMBUS_BLOCK_DATA:
> > +		if (!(funcs & I2C_FUNC_SMBUS_READ_BLOCK_DATA)) {
> > +			fprintf(stderr, MISSING_FUNC_FMT, "SMBus block read");
> > +			return -1;
> > +		}
> > +		break;
> > +	case I2C_SMBUS_I2C_BLOCK_DATA:
> > +		if (!(funcs & I2C_FUNC_SMBUS_READ_I2C_BLOCK)) {
> > +			fprintf(stderr, MISSING_FUNC_FMT, "I2C block read");
> > +			return -1;
> > +		}
> > +		break;
> 
> Why are you testing READ functionalities when what you want to do is
> WRITE?
> 
Typical cut-and-paste error.

> >  	}
> >  
> >  	if (pec
> > @@ -90,7 +105,7 @@
> >  }
> >  
> >  static int confirm(const char *filename, int address, int size, int daddress,
> > -		   int value, int vmask, int pec)
> > +		   int value, int vmask, unsigned char *block, int len, int pec)
> 
> The block pointer could be const.
> 
Ok.

> >  {
> >  	int dont = 0;
> >  
> > @@ -109,7 +124,16 @@
> >  		"0x%02x, data address\n0x%02x, ", filename, address, daddress);
> >  	if (size == I2C_SMBUS_BYTE)
> >  		fprintf(stderr, "no data.\n");
> > -	else
> > +	else if (size == I2C_SMBUS_BLOCK_DATA ||
> > +		 size == I2C_SMBUS_I2C_BLOCK_DATA) {
> > +		int i;
> > +
> > +		fprintf(stderr, "data");
> > +		for (i = 0; i < len; i++)
> > +			fprintf(stderr, " 0x%02x", block[i]);
> > +		fprintf(stderr, ", mode %s.\n", size == I2C_SMBUS_BLOCK_DATA
> > +			? "smbus block" : "i2c block");
> > +	} else
> >  		fprintf(stderr, "data 0x%02x%s, mode %s.\n", value,
> >  			vmask ? " (masked)" : "",
> >  			size == I2C_SMBUS_BYTE_DATA ? "byte" : "word");
> > @@ -136,6 +160,8 @@
> >  	int pec = 0;
> >  	int flags = 0;
> >  	int force = 0, yes = 0, version = 0, readback = 0;
> > +	unsigned char block[32];
> 
> It might make sense to use I2C_SMBUS_BLOCK_MAX instead of hard-coding
> 32?
> 
Ok.

> > +	int len;
> >  
> >  	/* handle (optional) flags first */
> >  	while (1+flags < argc && argv[1+flags][0] == '-') {
> > @@ -180,6 +206,30 @@
> >  		help();
> >  	}
> >  
> > +	/* check for block data */
> > +	len = 0;
> > +	if (argc > flags + 5) {
> 
> This makes it impossible to write 1-byte blocks, right? This is bad.
> 
No, it is the same check used for 'b' and 'w'. I tested it, and it works.

> > +		switch (argv[argc-1][0]) {
> > +		case 's': size = I2C_SMBUS_BLOCK_DATA; break;
> > +		case 'i': size = I2C_SMBUS_I2C_BLOCK_DATA; break;
> > +		default:
> > +			size = 0;
> > +			break;
> > +		}
> > +		if (size == I2C_SMBUS_BLOCK_DATA || size == I2C_SMBUS_I2C_BLOCK_DATA) {
> > +			pec = argv[argc-1][1] == 'p';
> 
> ip isn't a valid mode. PEC is not defined for non-SMBus transactions
> (and despite its name, I2C_SMBUS_I2C_BLOCK_DATA read and writes are not
> SMBus transactions.
> 
Ok, I added an error check.

> > +			for (len = 0; len < (int)sizeof(block) && len + flags + 5 < argc; len++) {
> 
> Do you actually need the cast?
> 
Yes, at least with recent versions of gcc. sizeof() is an unsigned and len is an int,
so without the cast gcc warns about a signed-unsigned comparison.

I could change len to be unsigned, but then it starts complaining about comparisons 
of len against other variables, so I'd rather stick with the typecast.

> > +				value = strtol(argv[flags + len + 4], &end, 0);
> > +				if (*end || value < 0 || value > 0xff) {
> > +                                	fprintf(stderr, "Error: Block data value invalid!\n");
> > +                                	help();
> > +                        	}
> > +				block[len] = value;
> > +			}
> > +			goto dofile;
> > +		}
> > +	}
> > +
> >  	if (argc > flags + 4) {
> >  		if (!strcmp(argv[flags+4], "c")
> >  		 || !strcmp(argv[flags+4], "cp")) {
> > @@ -236,6 +286,7 @@
> >  		help();
> >  	}
> >  
> > +dofile:
> >  	file = open_i2c_dev(i2cbus, filename, sizeof(filename), 0);
> >  	if (file < 0
> >  	 || check_funcs(file, size, pec)
> > @@ -243,7 +294,7 @@
> >  		exit(1);
> >  
> >  	if (!yes && !confirm(filename, address, size, daddress,
> > -			     value, vmask, pec))
> > +			     value, vmask, block, len, pec))
> >  		exit(0);
> >  
> >  	if (vmask) {
> > @@ -299,11 +350,18 @@
> >  	case I2C_SMBUS_WORD_DATA:
> >  		res = i2c_smbus_write_word_data(file, daddress, value);
> >  		break;
> > +	case I2C_SMBUS_BLOCK_DATA:
> > +		res = i2c_smbus_write_block_data(file, daddress, len, block);
> > +		break;
> > +	case I2C_SMBUS_I2C_BLOCK_DATA:
> > +		res = i2c_smbus_write_i2c_block_data(file, daddress, len, block);
> > +		break;
> >  	default: /* I2C_SMBUS_BYTE_DATA */
> >  		res = i2c_smbus_write_byte_data(file, daddress, value);
> > +		break;
> >  	}
> >  	if (res < 0) {
> > -		fprintf(stderr, "Error: Write failed\n");
> > +		perror("Error: Write failed");
> 
> Hmm, do i2c_smbus_()* calls actually set errno? I didn't expect them
> to. Either way, if this change is wanted, it doesn't belong to this
> patch.
> 
Yes, they do. But you are right, this change doesn't belong into this patch.
Leftover from testing anyway, so I removed it.

> >  		close(file);
> >  		exit(1);
> >  	}
> 
> I tested your patch for the I2C block write case, it worked OK.
> 
> Please also update the manual page.
> 
Ok, will do.

Thanks,
Guenter