Skip to contentSkip to conversation

ata: replace macro with static inline in libata.h

PATCH In review Harvey Harrison 2 participants 8 messages Feb 15, 2008 – Feb 16, 2008 libata

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ata: replace macro with static inline in libata.h

Move to using a static inline which will force the same typechecking that min_t/max_t do (in this case, short). As a bonus, avoid a ton of sparse warnings like:

drivers/ata/pata_ali.c:176:14: warning: symbol '__x' shadows an earlier one drivers/ata/pata_ali.c:176:14: originally declared here

Due to nesting min_t macro inside max_t macro which both use a __x identifier internally.

---
 include/linux/libata.h |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)
include/linux/libata.h +8 -1
index bc5a8d0..b5590fb 100644
@@ -764,7 +764,14 @@ struct ata_timing {
 	unsigned short udma;		/* t2CYCTYP/2 */
 };
 
-#define FIT(v, vmin, vmax)	max_t(short, min_t(short, v, vmax), vmin)
+static inline short FIT(short v, short vmin, short vmax)
+{
+	if (v >= vmax)
+		return vmax;
+	if (v <= vmin)
+		return vmin;
+	return v;
+}
 
 extern const unsigned long sata_deb_timing_normal[];
 extern const unsigned long sata_deb_timing_hotplug[];
Signature · 1 line
1.5.4.1.1278.gc75be
#1 lore raw
Quoted 9 lines · On Fri, 15 Feb 2008 14:06:55 -0800 Harvey Harrison <[email protected]>…

Move to using a static inline which will force the same typechecking that min_t/max_t do (in this case, short). As a bonus, avoid a ton of sparse warnings like:

drivers/ata/pata_ali.c:176:14: warning: symbol '__x' shadows an earlier one drivers/ata/pata_ali.c:176:14: originally declared here

Due to nesting min_t macro inside max_t macro which both use a __x identifier internally.

NAK. This is a sparse bug, fix sparse.

#2 lore raw
Quoted 7 lines · Alan Cox
Quoted 4 lines · On Fri, 15 Feb 2008 14:06:55 -0800 Harvey Harrison <[email protected]>…

Move to using a static inline which will force the same typechecking that min_t/max_t do (in this case, short). As a bonus, avoid a ton of sparse warnings like:

Quoted 7 lines
Quoted 5 lines

drivers/ata/pata_ali.c:176:14: warning: symbol '__x' shadows an earlier one drivers/ata/pata_ali.c:176:14: originally declared here

Due to nesting min_t macro inside max_t macro which both use a __x identifier internally.

NAK. This is a sparse bug, fix sparse.

Yes, fair enough, but that's not all the patch is about.

1) it's using a max_t and min_t to force the comparisons as shorts, why not just make it a static inline?

2) the static inline is a little clearer about the intent here.

3) the sparse warnings are entirely secondary (and technically correct when the macros expand, __x is shadowed)

4) I may be mistaken, but I thought then when something can be written as a static inline instead of a macro it was preferred. At least I've seen akpm say so, but I'll let him speak for himself (added to CC:)

Cheers,

Harvey

From: Harvey Harrison <[email protected]>
Subject: [PATCH] ata: replace macro with static inline in libata.h

Avoid a metric ton of sparse warnings like:
drivers/ata/pata_ali.c:176:14: warning: symbol '__x' shadows an earlier one drivers/ata/pata_ali.c:176:14: originally declared here

Due to nesting min_t macro inside max_t macro which both use a __x identifier internally.

---
 include/linux/libata.h |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)
include/linux/libata.h +8 -1
index bc5a8d0..b5590fb 100644
@@ -764,7 +764,14 @@ struct ata_timing {
 	unsigned short udma;		/* t2CYCTYP/2 */
 };
 
-#define FIT(v, vmin, vmax)	max_t(short, min_t(short, v, vmax), vmin)
+static inline short FIT(short v, short vmin, short vmax)
+{
+	if (v >= vmax)
+		return vmax;
+	if (v <= vmin)
+		return vmin;
+	return v;
+}
 
 extern const unsigned long sata_deb_timing_normal[];
 extern const unsigned long sata_deb_timing_hotplug[];
Signature · 1 line
1.5.4.1.1278.gc75be
#3 lore raw
Quoted 6 lines
Quoted 1 line

NAK. This is a sparse bug, fix sparse.

Yes, fair enough, but that's not all the patch is about.

1) it's using a max_t and min_t to force the comparisons as shorts, why not just make it a static inline?

Because max_t and min_t also force the comparsion types

Quoted 1 line

2) the static inline is a little clearer about the intent here.

Why ?

Quoted 2 lines

3) the sparse warnings are entirely secondary (and technically correct when the macros expand, __x is shadowed)

In a controlled manner. I guess you could make min and max use __x and __y

Quoted 3 lines

4) I may be mistaken, but I thought then when something can be written as a static inline instead of a macro it was preferred. At least I've seen akpm say so, but I'll let him speak for himself (added to CC:)

gcc still sometimes seems to optimise macros better than inlines.

Alan

#4 lore raw
Quoted 8 lines · Alan Cox
Quoted 6 lines
Quoted 1 line

NAK. This is a sparse bug, fix sparse.

Yes, fair enough, but that's not all the patch is about.

1) it's using a max_t and min_t to force the comparisons as shorts, why not just make it a static inline?

Because max_t and min_t also force the comparsion types

Umm, maybe I'm missing something then, but how does the static inline not do this?

Quoted 4 lines
Quoted 2 lines

2) the static inline is a little clearer about the intent here.

Why ?

OK, maybe not much clearer. But isn't the inline easier to see at a glance that it is returning a value constrained to be

vmin <= v <= vmax

I suppose the variable names make it clear, but the macro construction is (slightly) less obvious.

Quoted 6 lines
Quoted 3 lines

3) the sparse warnings are entirely secondary (and technically correct when the macros expand, __x is shadowed)

In a controlled manner. I guess you could make min and max use __x and __y

__mint __maxt...but I'm not proposing that.

Quoted 5 lines
Quoted 3 lines

4) I may be mistaken, but I thought then when something can be written as a static inline instead of a macro it was preferred. At least I've seen akpm say so, but I'll let him speak for himself (added to CC:)

gcc still sometimes seems to optimise macros better than inlines.

OK, I didn't realize that, any pointers?

Harvey

#5 lore raw
Quoted 5 lines · On Fri, 15 Feb 2008 15:08:50 -0800 Harvey Harrison <[email protected]>…
Quoted 4 lines · Alan Cox
Quoted 4 lines
Quoted 1 line

NAK. This is a sparse bug, fix sparse.

Yes, fair enough, but that's not all the patch is about.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

1) it's using a max_t and min_t to force the comparisons as shorts, why not just make it a static inline?

Because max_t and min_t also force the comparsion types

Umm, maybe I'm missing something then, but how does the static inline not do this?

You claimed it was an advantage of the static inline earlier but both do anyway

Quoted 7 lines

OK, maybe not much clearer. But isn't the inline easier to see at a glance that it is returning a value constrained to be

vmin <= v <= vmax

I suppose the variable names make it clear, but the macro construction is (slightly) less obvious.

Perhaps then clamp_t()

Quoted 1 line

__mint __maxt...but I'm not proposing that.

I am - as I bet there are other examples of that construct in the tree.

Quoted 3 lines
Quoted 1 line

gcc still sometimes seems to optimise macros better than inlines.

OK, I didn't realize that, any pointers?

Not offhand, there is discussion in the archives but it may be somewhat out of date for the latest gcc.

I'm not arguing your change is -wrong- I just think the original is tidier and clearer. Its up to Jeff anyway

#6 lore raw
Quoted 11 lines · Alan Cox
Quoted 8 lines · On Fri, 15 Feb 2008 15:08:50 -0800 Harvey Harrison <[email protected]>…
Quoted 7 lines · Alan Cox
Quoted 6 lines
Quoted 1 line

NAK. This is a sparse bug, fix sparse.

Yes, fair enough, but that's not all the patch is about.

1) it's using a max_t and min_t to force the comparisons as shorts, why not just make it a static inline?

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

Because max_t and min_t also force the comparsion types

Umm, maybe I'm missing something then, but how does the static inline not do this?

You claimed it was an advantage of the static inline earlier but both do anyway

Oh, I thought I said it accomplished the _same_ typechecking, my bad.

Quoted 6 lines
Quoted 6 lines

OK, maybe not much clearer. But isn't the inline easier to see at a glance that it is returning a value constrained to be

vmin <= v <= vmax

Quoted 8 lines
Quoted 2 lines

I suppose the variable names make it clear, but the macro construction is (slightly) less obvious.

Perhaps then clamp_t()

Quoted 1 line

__mint __maxt...but I'm not proposing that.

I am - as I bet there are other examples of that construct in the tree.

Lots, but form what I've seen, most could use a helper inline that is a lot cleaner than what is currently there. This case is about the same either way.

Quoted 10 lines
Quoted 4 lines
Quoted 2 lines

gcc still sometimes seems to optimise macros better than inlines.

OK, I didn't realize that, any pointers?

Not offhand, there is discussion in the archives but it may be somewhat out of date for the latest gcc.

I'm not arguing your change is -wrong- I just think the original is tidier and clearer. Its up to Jeff anyway

Fair enough. This change would make sparse a whole lot more useful for libata, as it's down to 6 warnings with this and my 3/3 patch removing another min/max nesting. Going forward this would probably make it easier on the maintainer to script up some automated checking.

Cheers,

Harvey

#7 lore raw

From: Harvey Harrison <[email protected]>
Subject: [PATCH 3/3] ata: simplify clock divisor logic in pata_amd.c

Current code is essentially choosing between dividing by 1 or dividing by two, make the conditions a little more obvious.

As a bonus, removes a sparse error:
drivers/ata/pata_amd.c:59:11: warning: symbol '__x' shadows an earlier one drivers/ata/pata_amd.c:59:11: originally declared here

---
Thought I'd follow up with just what the 3/3 patch I referenced was.

 drivers/ata/pata_amd.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
drivers/ata/pata_amd.c +3 -1
index 4b8d9b5..7ef4847 100644
@@ -56,7 +56,9 @@ static void timing_setup(struct ata_port *ap, struct ata_device *adev, int offse
 	u8 t;
 
 	T = 1000000000 / amd_clock;
-	UT = T / min_t(int, max_t(int, clock, 1), 2);
+	UT = T;
+	if (clock >= 2)
+		UT = T / 2;
 
 	if (ata_timing_compute(adev, speed, &at, T, UT) < 0) {
 		dev_printk(KERN_ERR, &pdev->dev, "unknown mode %d.\n", speed);
Signature · 1 line
1.5.4.1.1278.gc75be