Skip to contentSkip to conversation

details, details ...

DISCUSSION Kevin Diggs 2 participants 3 messages Aug 13, 2008 – Aug 14, 2008 2 messages not in the archive

lore
Share & tools
lore
#0 lore raw

Hi,

In cpu exit function of a cpufreq driver:

while (test_bit(cf750gxmChangingPllBit, &cf750gxvStateBits))
msleep(1);

This bit will get cleared by a notifier callback.

In module_exit function of a related module:

        while (test_bit(PLL_LOCK_BIT, (unsigned long *)&boot_ratio)) {
                msleep(1);
        }

This bit will get cleared by a timer. It will also fire the notifiers needed above.

I don't think these are in interrupt context. The sleeps ok here?

Also, from checkpatch:

ERROR: do not initialise externals to 0 or NULL #2483: FILE: arch/powerpc/kernel/cpu/pll_if.c:486:
+int rval = 0;

ERROR: do not initialise statics to 0 or NULL
#2058: FILE: arch/powerpc/kernel/cpu/pll_if.c:61:
+static unsigned int override_bus_clock = 0;

Huh? What? Anyone care to teach an old dog a new trick???

kevin

#1 lore raw
Quoted 3 lines · Kevin Diggs

In cpu exit function of a cpufreq driver:
=20
=A0 =A0 =A0 =A0 =A0while (test_bit(cf750gxmChangingPllBit, &cf750gxvState=

Bits))

Quoted 7 lines

=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0msleep(1);
=20
This bit will get cleared by a notifier callback.
=20
In module_exit function of a related module:
=20
=A0 =A0 =A0 =A0 =A0while (test_bit(PLL_LOCK_BIT, (unsigned long *)&boot_r=

atio)) {

Quoted 7 lines

=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0msleep(1);
=A0 =A0 =A0 =A0 =A0}
=20
This bit will get cleared by a timer. It will also fire the notifiers=20 needed above.
=20
I don't think these are in interrupt context. The sleeps ok here?

Technically ok, but not nice. Besides the coding style issues, it's still a busy loop that should be replaced by wait_for_completion().

Quoted 11 lines

Also, from checkpatch:
=20
ERROR: do not initialise externals to 0 or NULL #2483: FILE: arch/powerpc/kernel/cpu/pll_if.c:486:
+int rval =3D 0;
=20
ERROR: do not initialise statics to 0 or NULL #2058: FILE: arch/powerpc/kernel/cpu/pll_if.c:61:
+static unsigned int override_bus_clock =3D 0; =20
Huh? What? Anyone care to teach an old dog a new trick???

Don't bother. Old gcc variants would put these variables into .data instead of .bss, but with a new (less than 5 years or so) gcc, both will result in .bss storage that is initialized to zero at boot time.

	Arnd <><

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#4 lore raw
Quoted 38 lines · Arnd Bergmann
Quoted 20 lines · Kevin Diggs
Quoted 18 lines · Arnd Bergmann
Quoted 15 lines · Kevin Diggs

In cpu exit function of a cpufreq driver:

while (test_bit(cf750gxmChangingPllBit, &cf750gxvStateBits))
msleep(1);

This bit will get cleared by a notifier callback.

In module_exit function of a related module:

       while (test_bit(PLL_LOCK_BIT, (unsigned long *)&boot_ratio)) {
               msleep(1);
       }

This bit will get cleared by a timer. It will also fire the notifiers needed above.

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

I don't think these are in interrupt context. The sleeps ok here?

Technically ok, but not nice. Besides the coding style issues, it's still a busy loop that should be replaced by wait_for_completion().

I took a brief look at wait_for_completion(). Looks kinda heavy weight. Just to be clear both of these code fragments are in code shutdown (i.e. exit) paths. Is the use of wait_for_completion() still preferred? I thought a "busy loop" used the delay routines?

You should always write code that is easy to understand and tells the reader what you mean. If you want to wait for something to complete, use wait_for_completion. If you look at what msleep does internally, you will find that it isn't simpler than wait_for_completion either.

Quoted 5 lines

The loop doing msleep(1) is not as bad as a loop doing delays, but it can still cause unnecessary wakeups and on average will sleep one milisecond too long. Neither of these is a real problem, but if you can do it correctly, just do.

Please forgive me. I'm not trying to be argumentative. I'm just trying to learn. I found a section in the O'Reilly Linux device driver guide on this completion stuff. If I understand it correctly, I initialize a completion thing. In the code that starts the task I do a complete(). In the exit code I'll do a wait_for_completion(). In my usage paradigm I will VERY rarely ever call wait_for_completion() and have it actually wait. This still match completion's intended use?

Quoted 11 lines
Quoted 2 lines

Can you elaborate on the coding style issues? Variable names? Use of the bit stuff? Those brace thingies?

Variables should be lower-case names, constants should be upper-case. Both should have names that tell you what they are used for. The case in the second code sample is either wrong, or redundant. You should leave out curly braces when you only have a single line in the basic block. Read Documentation/CodingStyle to learn about more things to consider.

Can I post the 2 routines for RFC style comments? Or is that to much trouble?

Quoted 6 lines
Quoted 6 lines
Quoted 5 lines

Don't bother. Old gcc variants would put these variables into .data instead of .bss, but with a new (less than 5 years or so) gcc, both will result in .bss storage that is initialized to zero at boot time.

Quoted 8 lines
Quoted 1 line

??? So ... I can ignore the error? Or I should not be initializing variables to 0 (or NULL)?

Either fix checkpatch.pl not to warn about these, or just don't initialize the variables. Initializing a variable at declaration time is frowned upon by some people, because it is redundant in case of static or global variables, and error-prone for automatic variables.

When you say initializing is frowned upon, do you mean only when you are initializing to zero? Is the redundancy (for the case of 0?) a part of the C spec? Or is it gcc specific? error-prone for automatic variables?

kevin

Quoted 3 lines
	Arnd <><