Skip to contentSkip to conversation

lib/atomic64_test: do not build on non-atomic64 systems

PATCH In review Mike Frysinger 3 participants 13 messages Oct 16, 2010 – Oct 25, 2010 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lib/atomic64_test: do not build on non-atomic64 systems

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

---
 lib/atomic64_test.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)
lib/atomic64_test.c +4 -0
index 250ed11..0ac1a66 100644
@@ -12,6 +12,8 @@
 #include <linux/kernel.h>
 #include <asm/atomic.h>
 
+#ifdef ATOMIC64_INIT
+
 #define INIT(c) do { atomic64_set(&v, c); r = c; } while (0)
 static __init int test_atomic64(void)
 {
@@ -164,3 +166,5 @@ static __init int test_atomic64(void)
 }
 
 core_initcall(test_atomic64);
+
+#endif
Signature · 1 line
1.7.3.1
maintainer
#1 lore raw
Quoted 3 lines · On Sat, 16 Oct 2010 13:27:15 -0400 Mike Frysinger <[email protected]> wrote:

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

I don't get it. If the arch doesn't implement atomic64 then this file will get zillions of build errors, won't it?

lib/atomic64_test.c line 15

Quoted 9 lines
lib/atomic64_test.c +1 -0
index 250ed11..0ac1a66 100644
@@ -12,6 +12,8 @@
 #include <linux/kernel.h>
 #include <asm/atomic.h>
 
+#ifdef ATOMIC64_INIT

hm, that's a bit lazy. It should really use a CONFIG_HAVE_ thing. What a pita.

Quoted 11 lines
lib/atomic64_test.c +2 -0
 #define INIT(c) do { atomic64_set(&v, c); r = c; } while (0)
 static __init int test_atomic64(void)
 {
@@ -164,3 +166,5 @@ static __init int test_atomic64(void)
 }
 
 core_initcall(test_atomic64);
+
+#endif
Signature · 1 line
1.7.3.1
#2 lore raw
Quoted 6 lines · On Thursday, October 21, 2010 18:02:50 Andrew Morton wrote:
Quoted 2 lines · On Sat, 16 Oct 2010 13:27:15 -0400 Mike Frysinger wrote:

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

I don't get it. If the arch doesn't implement atomic64 then this file will get zillions of build errors, won't it?

... which is why i added the ifdef protection

Quoted 6 lines
Quoted 6 lines
lib/atomic64_test.c +0 -0
index 250ed11..0ac1a66 100644
@@ -12,6 +12,8 @@
 
Quoted 7 lines

lib/atomic64_test.c line 15

Quoted 4 lines
lib/atomic64_test.c +1 -0
 #include <linux/kernel.h>
 #include <asm/atomic.h>
 
+#ifdef ATOMIC64_INIT

hm, that's a bit lazy. It should really use a CONFIG_HAVE_ thing. What a pita.

ATOMIC64_INIT is required for atomic64 headers to provide, and having a Kconfig knob doesnt gain anything else
-mike

maintainer
#3 lore raw
Quoted 9 lines · On Thu, 21 Oct 2010 18:23:37 -0400 Mike Frysinger <[email protected]> wrote:
Quoted 6 lines · On Thursday, October 21, 2010 18:02:50 Andrew Morton wrote:
Quoted 2 lines · On Sat, 16 Oct 2010 13:27:15 -0400 Mike Frysinger wrote:

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

I don't get it. If the arch doesn't implement atomic64 then this file will get zillions of build errors, won't it?

... which is why i added the ifdef protection

So the changelog was poor. Please write complete changelogs so I need to have this sort of conversation less often?

Why doesn't this cause lots of you-broke-my-build complaints? Nobody's running allmodonfig?

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
lib/atomic64_test.c +0 -0
index 250ed11..0ac1a66 100644
@@ -12,6 +12,8 @@
 
 #include <linux/kernel.h>
 #include <asm/atomic.h>
 
Quoted 7 lines
Quoted 4 lines

lib/atomic64_test.c line 15

Quoted 1 line
lib/atomic64_test.c +1 -0
+#ifdef ATOMIC64_INIT

hm, that's a bit lazy. It should really use a CONFIG_HAVE_ thing. What a pita.

ATOMIC64_INIT is required for atomic64 headers to provide, and having a Kconfig knob doesnt gain anything else

I know that. But the standard way for an architecture to indicate to the core that it impements a feature is for it to define CONFIG_HAVE_*. Picking some related #define which architectures happen to implement is atypical and unexpected.

Will it cause problems? Probably not, unless the arch goes and defines ATOMIC64_INIT without actually implementing atomic64. But it's atypical and unexpected and, yes, lazy!

#4 lore raw
Quoted 6 lines · Andrew Morton
Quoted 5 lines · On Thu, 21 Oct 2010 18:23:37 -0400 Mike Frysinger wrote:
Quoted 4 lines · On Thursday, October 21, 2010 18:02:50 Andrew Morton wrote:
Quoted 2 lines · On Sat, 16 Oct 2010 13:27:15 -0400 Mike Frysinger wrote:

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

I don't get it.  If the arch doesn't implement atomic64 then this file will get zillions of build errors, won't it?

... which is why i added the ifdef protection

So the changelog was poor.  Please write complete changelogs so I need to have this sort of conversation less often?

the changelog seems pretty clear to me. arch doesnt provide atomic64, so dont build code that uses atomic64.

Quoted 2 lines

Why doesn't this cause lots of you-broke-my-build complaints?  Nobody's running allmodonfig?

that's how i noticed. perhaps the other arches that dont provide atomic64 arent doing as many build tests as i am.

Quoted 17 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
lib/atomic64_test.c +0 -0
index 250ed11..0ac1a66 100644
@@ -12,6 +12,8 @@
 
  #include <linux/kernel.h>
  #include <asm/atomic.h>
 
Quoted 7 lines
Quoted 4 lines

lib/atomic64_test.c line 15

Quoted 1 line
lib/atomic64_test.c +1 -0
+#ifdef ATOMIC64_INIT

hm, that's a bit lazy.  It should really use a CONFIG_HAVE_ thing.  What a pita.

ATOMIC64_INIT is required for atomic64 headers to provide, and having a Kconfig knob doesnt gain anything else

Quoted 8 lines

I know that.  But the standard way for an architecture to indicate to the core that it impements a feature is for it to define CONFIG_HAVE_*. Picking some related #define which architectures happen to implement is atypical and unexpected.

Will it cause problems?  Probably not, unless the arch goes and defines ATOMIC64_INIT without actually implementing atomic64.  But it's atypical and unexpected and, yes, lazy!

you can say "lazy" all you like. i dont see the point in going that route. -mike

maintainer
#5 lore raw
Quoted 10 lines · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger <[email protected]> wrote:
Quoted 9 lines · Andrew Morton
Quoted 8 lines · On Thu, 21 Oct 2010 18:23:37 -0400 Mike Frysinger wrote:
Quoted 6 lines · On Thursday, October 21, 2010 18:02:50 Andrew Morton wrote:
Quoted 2 lines · On Sat, 16 Oct 2010 13:27:15 -0400 Mike Frysinger wrote:

If the arch doesn't provide atomic64 functionality (there are quite a few), then don't bother trying to build this test.

I don't get it. __If the arch doesn't implement atomic64 then this file will get zillions of build errors, won't it?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

... which is why i added the ifdef protection

So the changelog was poor. __Please write complete changelogs so I need to have this sort of conversation less often?

the changelog seems pretty clear to me. arch doesnt provide atomic64, so dont build code that uses atomic64.

That the patch fixes build errors is rather important information.

Quoted 10 lines
Quoted 8 lines

I know that. __But the standard way for an architecture to indicate to the core that it impements a feature is for it to define CONFIG_HAVE_*. Picking some related #define which architectures happen to implement is atypical and unexpected.

Will it cause problems? __Probably not, unless the arch goes and defines ATOMIC64_INIT without actually implementing atomic64. __But it's atypical and unexpected and, yes, lazy!

you can say "lazy" all you like. i dont see the point in going that route.

Try

grep HAVE arch/x86/Kconfig

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

There's your point.

#6 lore raw
Quoted 10 lines · Andrew Morton
Quoted 1 line · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger wrote:

you can say "lazy" all you like.  i dont see the point in going that route.

Try

       grep HAVE arch/x86/Kconfig

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

fun times. new tact.

Luca: your new atomic64_t test build fails on all arches that lack atomic64_t. please fix.
-mike

maintainer
#7 lore raw
Quoted 8 lines · On Fri, 22 Oct 2010 16:14:49 -0400 Mike Frysinger <[email protected]> wrote:
Quoted 7 lines · Andrew Morton
Quoted 1 line · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger wrote:

you can say "lazy" all you like. __i dont see the point in going that route.

Try

__ __ __ __grep HAVE arch/x86/Kconfig

Quoted 8 lines
Quoted 3 lines

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

fun times. new tact.

Luca: your new atomic64_t test build fails on all arches that lack atomic64_t. please fix.

That's only part of the problem. The following won't build also:

net/rds
kernel/perf_event.c
drivers/staging/octeon
drivers/infiniband/hw

with more to come. These things should be made dependent upon CONFIG_HAVE_ATOMIC64 in Kconfig. (Can't use #ifdef ATOMIC64_INIT for this!)

Or, much better, we implement atomic64 on the offending architectures.

#8 lore raw
Quoted 13 lines · Andrew Morton
Quoted 12 lines · On Fri, 22 Oct 2010 16:14:49 -0400 Mike Frysinger wrote:
Quoted 10 lines · Andrew Morton
Quoted 1 line · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger wrote:

you can say "lazy" all you like. __i dont see the point in going that route.

Try

__ __ __ __grep HAVE arch/x86/Kconfig

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

Quoted 8 lines
Quoted 4 lines

fun times.  new tact.

Luca: your new atomic64_t test build fails on all arches that lack atomic64_t.  please fix.

That's only part of the problem.  The following won't build also:

net/rds

not true. that code base is already using my suggestion:
net/rds/rds.h:
#ifdef ATOMIC64_INIT
#define KERNEL_HAS_ATOMIC64
#endif

but this isnt a matter of "use atomic64_t or atomic_t" ... this code manually takes care of doing a spinlock around a u64 member. i imagine if you'd notice this before it was merged you'd have made them fix this cleanly.

Quoted 1 line

kernel/perf_event.c

also not true -- this requires arches to opt in to HAVE_PERF_EVENTS and only arches which have validated it works (i.e. they have atomic64) have done that

Quoted 1 line

drivers/staging/octeon

not an issue -- this depends on CPU_CAVIUM_OCTEON which is only provided by mips which provides atomic64

Quoted 1 line

drivers/infiniband/hw

the only code usage of atomic64 is in code that already depends on the Kconfig symbol 64BIT

as for why it depends on this, i dont know ... maybe it's because of atomic64_t

Quoted 1 line

Or, much better, we implement atomic64 on the offending architectures.

i dont want to give people the impression that 64 atomics are free to use if in reality they're pretty expensive. on a Blackfin system, i'd need to implement every access with basically a spinlock.

Quoted 2 lines

with more to come. These things should be made dependent upon CONFIG_HAVE_ATOMIC64 in Kconfig. (Can't use #ifdef ATOMIC64_INIT for this!)

these are actually compelling arguments unlike the original one. even if all these examples ended up not really being true (only the new atomic64_t test code is available to build on arches that lack atomic64).
-mike

maintainer
#9 lore raw
Quoted 21 lines · On Fri, 22 Oct 2010 16:47:36 -0400 Mike Frysinger <[email protected]> wrote:
Quoted 20 lines · Andrew Morton
Quoted 8 lines · On Fri, 22 Oct 2010 16:14:49 -0400 Mike Frysinger wrote:
Quoted 7 lines · Andrew Morton
Quoted 1 line · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger wrote:

you can say "lazy" all you like. __i dont see the point in going that route.

Try

__ __ __ __grep HAVE arch/x86/Kconfig

Quoted 8 lines
Quoted 3 lines

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

fun times. __new tact.

Luca: your new atomic64_t test build fails on all arches that lack atomic64_t. __please fix.

That's only part of the problem. __The following won't build also:

Quoted 7 lines
Quoted 1 line

net/rds

not true. that code base is already using my suggestion:
net/rds/rds.h:
#ifdef ATOMIC64_INIT
#define KERNEL_HAS_ATOMIC64
#endif

IOW, your suggestion led to a nasty local hack. One which would be unneeded had we implemented this properly via Kconfig.

Quoted 4 lines

but this isnt a matter of "use atomic64_t or atomic_t" ... this code manually takes care of doing a spinlock around a u64 member. i imagine if you'd notice this before it was merged you'd have made them fix this cleanly.

I'd have suggested that they use Kconfig.

I'd also have suggested that they implement a generic spinlocked atomic64 library rather than open-coding stuff down in net/rds/.

People do tend to prefer to do localised expedient things rather than sticking their necks out and implementing proper, generic kernel-wide functions. If I see it happen, I'll tell them. Usually I don't see it until months after it's merged :(

Quoted 5 lines
Quoted 1 line

kernel/perf_event.c

also not true -- this requires arches to opt in to HAVE_PERF_EVENTS and only arches which have validated it works (i.e. they have atomic64) have done that

kernel/perf_event.c has a dependency on the missing CONFIG_HAVE_ATOMIC64.

Quoted 9 lines
Quoted 1 line

drivers/staging/octeon

not an issue -- this depends on CPU_CAVIUM_OCTEON which is only provided by mips which provides atomic64

Quoted 1 line

drivers/infiniband/hw

the only code usage of atomic64 is in code that already depends on the Kconfig symbol 64BIT

Again, that's just wrong and it's a fluke. Will break if a 64-bit arch doesn't implement atomic64.

Quoted 7 lines

as for why it depends on this, i dont know ... maybe it's because of atomic64_t

Quoted 1 line

Or, much better, we implement atomic64 on the offending architectures.

i dont want to give people the impression that 64 atomics are free to use if in reality they're pretty expensive. on a Blackfin system, i'd need to implement every access with basically a spinlock.

spin_lock_irqsave(), really. Becomes local_irq_save() on UP.

But what's the alternative? Either entire features become unavailable on those architectures or we grow local hacks like rds_ib_get_ack() which, yup, uses spin_lock_irqsave().

Quoted 7 lines
Quoted 2 lines

with more to come. These things should be made dependent upon CONFIG_HAVE_ATOMIC64 in Kconfig. (Can't use #ifdef ATOMIC64_INIT for this!)

these are actually compelling arguments unlike the original one. even if all these examples ended up not really being true (only the new atomic64_t test code is available to build on arches that lack atomic64).

Really, the atomic64_t implementation was imcomplete. It should have provided the generic arch-neutral fallback code.

#10 lore raw
Quoted 24 lines · Andrew Morton
Quoted 23 lines · On Fri, 22 Oct 2010 16:47:36 -0400 Mike Frysinger wrote:
Quoted 21 lines · Andrew Morton
Quoted 8 lines · On Fri, 22 Oct 2010 16:14:49 -0400 Mike Frysinger wrote:
Quoted 7 lines · Andrew Morton
Quoted 1 line · On Thu, 21 Oct 2010 19:04:36 -0400 Mike Frysinger wrote:

you can say "lazy" all you like. __i dont see the point in going that route.

Try

__ __ __ __grep HAVE arch/x86/Kconfig

Quoted 8 lines
Quoted 3 lines

If all of those were instead to use some random #define which the particular feature happened to define in some header file then we would have a mess on our hands.

fun times. __new tact.

Luca: your new atomic64_t test build fails on all arches that lack atomic64_t. __please fix.

That's only part of the problem. __The following won't build also:

net/rds

Quoted 8 lines
Quoted 5 lines

not true.  that code base is already using my suggestion:
net/rds/rds.h:
#ifdef ATOMIC64_INIT
#define KERNEL_HAS_ATOMIC64
#endif

IOW, your suggestion led to a nasty local hack.  One which would be unneeded had we implemented this properly via Kconfig.

no. mine was specific to the atomic64 api and not random consumers. the test code is tightly coupled to the atomic64 api by nature, so my proposal as constrained by the single instance makes perfect sense. as soon as you start talking about more things needing to know "is atomic64 available" and doing so at the config dependency level, then a Kconfig option makes sense. thus i didnt and still dont buy your original argument against the original patch and constraint. but with real tree data, your suggestion makes sense. perhaps i'll implement it if i find time.
-mike

#11 lore raw
In reply to Andrew Morton
Quoted 3 lines

That's only part of the problem. The following won't build also:

drivers/infiniband/hw

Interesting... I hadn't looked at that usage before. Both drivers that seem to use atomic64 already depend on 64BIT in Kconfig, so I suspect the intersection of 64BIT and !ATOMIC64 is empty?

But eg drivers/infiniband/hw/qib/qib_rc.c does essentially:

	atomic64_t *maddr;
	u64 sdata = <something>
	...
	maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
	atomic64_add_return(sdata, maddr)

is it legit to cast some random address to atomic64_t and expect it to work across archs that implement atomic64? (I expect x86_64 is probably OK but I wonder if other architectures might not be so understanding)

- R.

maintainer
#12 lore raw
Quoted 10 lines · On Sun, 24 Oct 2010 09:20:34 -0700 Roland Dreier <[email protected]> wrote:
Quoted 3 lines

That's only part of the problem. The following won't build also:

drivers/infiniband/hw

Interesting... I hadn't looked at that usage before. Both drivers that seem to use atomic64 already depend on 64BIT in Kconfig, so I suspect the intersection of 64BIT and !ATOMIC64 is empty?

But eg drivers/infiniband/hw/qib/qib_rc.c does essentially:

Quoted 8 lines
	atomic64_t *maddr;
	u64 sdata = <something>
	...
	maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
	atomic64_add_return(sdata, maddr)

is it legit to cast some random address to atomic64_t and expect it to work across archs that implement atomic64?

Not really. If someone implements atomic64 on 32-bit they may do it by putting a spinlock in the atomic64_t. Or they might use hashed spinlocks, in which case that'll work.

Probably hashed spinlocks, given the (realtively new) convention that the all-zeroes pattern is a legit way of initialising an atomic_t.