Re: [RFC] dynamic_debug: introduce debug_hex_dump()

On Sunday, November 18, 2012 11:47:58 AM Vladimir Kondratiev wrote:

On Wednesday, November 14, 2012 07:27:47 PM Vladimir Kondratiev wrote:

quoted

+#define dynamic_hex_dump(prefix_str, prefix_type, rowsize,     \
+                        groupsize, buf, len, ascii)            \
+do {                                                           \
+       DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, prefix_str);  \
+       if (unlikely(descriptor.flags & _DPRINTK_FLAGS_PRINT))  \
+               print_hex_dump(KERN_DEBUG, prefix_str,          \
+                              prefix_type, rowsize, groupsize, \
+                              buf, len, ascii);                \
+} while (0)

There is some problem with the code above. There are existing drivers
that use print_hex_dump_bytes() with the prefix_str built dynamically,
like this (see drivers/tty/ipwireless/hardware.c:362)

static void dump_data_bytes(const char *type, const unsigned char *data,
			    unsigned length)
{
	char prefix[56];

	sprintf(prefix, IPWIRELESS_PCCARD_NAME ": %s %s ",
			type, data_type(data, length));
	print_hex_dump_bytes(prefix, 0, (void *)data,
			length < DUMP_MAX_BYTES ? length : DUMP_MAX_BYTES);
}

In this case, prefix_str is not constant, and this solution will not
work.

I see 2 drivers that use dynamic prefix:

- mentioned above drivers/tty/ipwireless/hardware.c
- drivers/isdn/hardware/mISDN/

I see several options:

1) require prefix_str to be constant. Patch code mentioned. For example,
code above may reads:

static void dump_data_bytes(const char *type, const unsigned char *data,
			    unsigned length)
{
	pr_debug(IPWIRELESS_PCCARD_NAME ": %s %s\n",
			type, data_type(data, length));
	print_hex_dump_bytes(IPWIRELESS_PCCARD_NAME, 0, (void *)data,
			length < DUMP_MAX_BYTES ? length : DUMP_MAX_BYTES);
}

2) introduce fixed format string in DEFINE_DYNAMIC_DEBUG_METADATA, like

DEFINE_DYNAMIC_DEBUG_METADATA((descriptor, "hexdump");

3) remove print_hex_dump_bytes from this patch

I think 1) is the best, opinions?

Thanks, Vladimir

In case of option 1), patches follows. Note there is change in the initial patch as well (I forgot to exclude print_hex_dump_bytes() from compilation in hexdump.c. Please, comment.