[U-Boot] [PATCH 3/7] Add timing information to printf's for use with bootgraph.pl
Andrew Murray
amurray at theiet.org
Thu Sep 1 01:42:28 CEST 2011
On 1 September 2011 00:38, Graeme Russ <graeme.russ at gmail.com> wrote:
> Hi Mike, Andrew,
>
> On Thu, Sep 1, 2011 at 8:47 AM, Mike Frysinger <vapier at gentoo.org> wrote:
>> On Wednesday, August 31, 2011 18:20:57 Andrew Murray wrote:
>>> va_list args;
>>> uint i;
>>> char printbuffer[CONFIG_SYS_PBSIZE];
>>> + char *buf = printbuffer;
>>>
>>> va_start(args, fmt);
>>>
>>> +#if defined(CONFIG_BOOT_TRACE)
>>> + unsigned long long ticks = get_ticks();
>>> + int secs = ticks / get_tbclk();
>>> + int msec = ((ticks * 1000000) / get_tbclk()) - (secs * 1000000);
>>> +
>>> + i += sprintf(buf, "[%5lu.%06lu] ", secs, msec);
>>> + buf += i;
>>> +#endif
>>> +
>>> /* For this to work, printbuffer must be larger than
>>> * anything we ever want to print.
>>> */
>>> - i = vsprintf(printbuffer, fmt, args);
>>> + i += vsprintf(buf, fmt, args);
>>> va_end(args);
>>
>> NAK for a few reasons:
>> - i dont see how this could possibly compile warning free
>> - you never initialize "i", only added to it
>> - you call va_start() inbetween variable decls
>
> And correct me if I'm wrong, but EVERY printf() will get the timing info
> tacked on - Even the ones without \n which are intermediate prints in
> larger messages which is going to lead to very ugly outputs
>
> I think instead we should look at another 'printf() with timestamp'
> function which can be used on an as-needed basis
Yes that's exactly the case ... e.g.
[ 3.122000] initcall 0x9ff864cc returned
[ 3.126000] calling 0x9ff78174
[ 3.129000] ## Booting kernel from Legacy Image at 82000000 ...
[ 3.135000] Image Name: Angstrom/2.6.32/beagleboard
[ 3.141000] Image Type: [ 3.143000] ARM Linux Kernel Image
(uncompressed)
[ 3.148000] Data Size: [ 3.151000] 3194192 Bytes = [
3.154000] 3[ 3.155000]
MiB
It's certainty ugly. A dedicated printf seems very sensible.
Andrew Murray
More information about the U-Boot
mailing list