Skip to contentSkip to conversation

ate_resource->lowest_free_index signed or unsigned?

DISCUSSION Roel Kluin 2 participants 6 messages Apr 20, 2009 – Apr 23, 2009

lore
Share & tools
lore

Roel Kluin

author
#0 lore raw

In arch/ia64/include/asm/sn/pcibr_provider.h:94

struct ate_resource{
        u64 *ate;
        u64 num_ate;
        u64 lowest_free_index;
};

as you see lowest_free_index is unsigned,

but in alloc_ate_resource() and free_ate_resource() it is treated as signed:

static inline void free_ate_resource(struct ate_resource *ate_resource,
int start)

{
        mark_ate(ate_resource, start, ate_resource->ate[start], 0);
        if ((ate_resource->lowest_free_index > start) ||
            (ate_resource->lowest_free_index < 0))
                ate_resource->lowest_free_index = start;
}

/*
* alloc_ate_resource: Allocate the requested number of ATEs. */
static inline int alloc_ate_resource(struct ate_resource *ate_resource,
int ate_needed)

{
        int start_index;

        /*
         * Check for ate exhaustion.
         */
        if (ate_resource->lowest_free_index < 0)
                return -1;

        /*
         * Find the required number of free consecutive ates.
         */
        start_index             find_free_ate(ate_resource, ate_resource->lowest_free_index,
                          ate_needed);
        if (start_index >= 0)
                mark_ate(ate_resource, start_index, ate_needed, ate_needed);

        ate_resource->lowest_free_index             find_free_ate(ate_resource, ate_resource->lowest_free_index, 1);

        return start_index;
}

should it be signed instead?

#1 lore raw
Quoted 8 lines · Roel Kluin

In arch/ia64/include/asm/sn/pcibr_provider.h:94

struct ate_resource{
        u64 *ate;
        u64 num_ate;
        u64 lowest_free_index;
};
Quoted 4 lines

as you see lowest_free_index is unsigned,

but in alloc_ate_resource() and free_ate_resource() it is treated as signed:

Hi Roel,

I don't thing it's a big issue to be honest, it just means the tests for lowest_free_index < 0 are no-ops. I would be fine with a patch that changes it to signed.

Cheers,
Jes

Quoted 37 lines

static inline void free_ate_resource(struct ate_resource *ate_resource,
int start)

{
        mark_ate(ate_resource, start, ate_resource->ate[start], 0);
        if ((ate_resource->lowest_free_index > start) ||
            (ate_resource->lowest_free_index < 0))
                ate_resource->lowest_free_index = start;
}

/*
* alloc_ate_resource: Allocate the requested number of ATEs. */
static inline int alloc_ate_resource(struct ate_resource *ate_resource,
int ate_needed)

{
        int start_index;

        /*
         * Check for ate exhaustion.
         */
        if (ate_resource->lowest_free_index < 0)
                return -1;

        /*
         * Find the required number of free consecutive ates.
         */
        start_index >             find_free_ate(ate_resource, ate_resource->lowest_free_index,
                          ate_needed);
        if (start_index >= 0)
                mark_ate(ate_resource, start_index, ate_needed, ate_needed);

        ate_resource->lowest_free_index >             find_free_ate(ate_resource, ate_resource->lowest_free_index, 1);

        return start_index;
}

should it be signed instead?

author
#2 lore raw
In reply to Roel Kluin
Quoted 7 lines · Jes Sorensen
Quoted 2 lines · Roel Kluin

lowest_free_index is unsigned, but in alloc_ate_resource() and free_ate_resource() it is treated as signed

I don't thing it's a big issue to be honest, it just means the tests for lowest_free_index < 0 are no-ops. I would be fine with a patch that changes it to signed.

Ok, How's this? If my changelog is not clear feel free to amend it. ------------------------------>8-------------8<--------------------------------- unsigned lowest_free_index was set in alloc_ate_resource() with the return value of find_free_ate(), which is -1 when no free ate is found. Since unsigned, a lowest_free_index of -1, `ate exhaustion', would then not be noticed in a subsequent call to alloc_ate_resource().

---

arch/ia64/include/asm/sn/pcibr_provider.h +1 -1
index da205b7..b3133ef 100644
@@ -94,7 +94,7 @@
 struct ate_resource{
 	u64 *ate;
 	u64 num_ate;
-	u64 lowest_free_index;
+	int lowest_free_index;
 };
 
 struct pcibus_info {
#3 lore raw
In reply to Roel Kluin
Quoted 9 lines · Roel Kluin
Quoted 6 lines · Jes Sorensen
Quoted 2 lines · Roel Kluin

lowest_free_index is unsigned, but in alloc_ate_resource() and free_ate_resource() it is treated as signed

I don't thing it's a big issue to be honest, it just means the tests for lowest_free_index < 0 are no-ops. I would be fine with a patch that changes it to signed.

Ok, How's this? If my changelog is not clear feel free to amend it.

Hi Roel,

Please make it 'long' or 's64' to keep the size in the struct.

Cheers,
Jes

Quoted 21 lines

------------------------------>8-------------8<--------------------------------- unsigned lowest_free_index was set in alloc_ate_resource() with the return value of find_free_ate(), which is -1 when no free ate is found. Since unsigned, a lowest_free_index of -1, `ate exhaustion', would then not be noticed in a subsequent call to alloc_ate_resource().

---

arch/ia64/include/asm/sn/pcibr_provider.h +1 -1
index da205b7..b3133ef 100644
@@ -94,7 +94,7 @@
 struct ate_resource{
 	u64 *ate;
 	u64 num_ate;
-	u64 lowest_free_index;
+	int lowest_free_index;
 };
 
 struct pcibus_info {
author
#4 lore raw
In reply to Roel Kluin
Quoted 10 lines · Jes Sorensen
Quoted 7 lines · Roel Kluin
Quoted 6 lines · Jes Sorensen
Quoted 2 lines · Roel Kluin

lowest_free_index is unsigned, but in alloc_ate_resource() and free_ate_resource() it is treated as signed

I don't thing it's a big issue to be honest, it just means the tests for lowest_free_index < 0 are no-ops. I would be fine with a patch that changes it to signed.

Please make it 'long' or 's64' to keep the size in the struct.

------------------------------>8-------------8<--------------------------------- unsigned lowest_free_index was set in alloc_ate_resource() with the return value of find_free_ate(), which is -1 when no free ate is found. Since unsigned, a lowest_free_index of -1, `ate exhaustion', would then not be noticed in a subsequent call to alloc_ate_resource().

---

arch/ia64/include/asm/sn/pcibr_provider.h +1 -1
index da205b7..636d914 100644
@@ -94,7 +94,7 @@
 struct ate_resource{
 	u64 *ate;
 	u64 num_ate;
-	u64 lowest_free_index;
+	s64 lowest_free_index;
 };
 
 struct pcibus_info {
Acked-by
#5 lore raw
In reply to Roel Kluin
Quoted 11 lines · Roel Kluin
Quoted 10 lines · Jes Sorensen
Quoted 7 lines · Roel Kluin
Quoted 6 lines · Jes Sorensen
Quoted 2 lines · Roel Kluin

lowest_free_index is unsigned, but in alloc_ate_resource() and free_ate_resource() it is treated as signed

I don't thing it's a big issue to be honest, it just means the tests for lowest_free_index < 0 are no-ops. I would be fine with a patch that changes it to signed.

Please make it 'long' or 's64' to keep the size in the struct.

Hi Roel,

I ran some basic tests, seems fine.

Cheers,
Jes

Quoted 21 lines

------------------------------>8-------------8<--------------------------------- unsigned lowest_free_index was set in alloc_ate_resource() with the return value of find_free_ate(), which is -1 when no free ate is found. Since unsigned, a lowest_free_index of -1, `ate exhaustion', would then not be noticed in a subsequent call to alloc_ate_resource().

---

arch/ia64/include/asm/sn/pcibr_provider.h +1 -1
index da205b7..636d914 100644
@@ -94,7 +94,7 @@
 struct ate_resource{
 	u64 *ate;
 	u64 num_ate;
-	u64 lowest_free_index;
+	s64 lowest_free_index;
 };
 
 struct pcibus_info {