Skip to contentSkip to conversation

percpu: simplify the logic of pcpu_alloc_first_chunk()

PATCH Superseded Yuntao Wang 2 participants 6 messages Apr 23, 2024 – May 14, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 percpu: simplify the logic of pcpu_alloc_first_chunk()

In the logic for hiding the end of the bitmap, there are several places where the same value 'region_bits - offset_bits' is calculated over and over again using different methods. Eliminate these redundant calculations to improve code readability.

---
 mm/percpu.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
mm/percpu.c +5 -7
index 4e11fc1e6def..2a051f00f68d 100644
@@ -1421,15 +1421,13 @@ static struct pcpu_chunk * __init pcpu_alloc_first_chunk(unsigned long tmp_addr,
 	if (chunk->end_offset) {
 		/* hide the end of the bitmap */
 		offset_bits = chunk->end_offset / PCPU_MIN_ALLOC_SIZE;
-		bitmap_set(chunk->alloc_map,
-			   pcpu_chunk_map_bits(chunk) - offset_bits,
-			   offset_bits);
-		set_bit((start_offset + map_size) / PCPU_MIN_ALLOC_SIZE,
-			chunk->bound_map);
+		start_offset = region_bits - offset_bits;
+
+		bitmap_set(chunk->alloc_map, start_offset, offset_bits);
+		set_bit(start_offset, chunk->bound_map);
 		set_bit(region_bits, chunk->bound_map);
 
-		pcpu_block_update_hint_alloc(chunk, pcpu_chunk_map_bits(chunk)
-					     - offset_bits, offset_bits);
+		pcpu_block_update_hint_alloc(chunk, start_offset, offset_bits);
 	}
 
 	return chunk;
Signature · 1 line
2.44.0
maintainer
#1 lore raw

Hi Yuntao,

Quoted 15 lines · Yuntao Wang

In the logic for hiding the end of the bitmap, there are several places where the same value 'region_bits - offset_bits' is calculated over and over again using different methods. Eliminate these redundant calculations to improve code readability.

---
 mm/percpu.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
mm/percpu.c +0 -0
index 4e11fc1e6def..2a051f00f68d 100644

mm/percpu.c static struct pcpu_chunk * __init pcpu_alloc_first_chunk(unsigned long tmp_addr, · line 1424

Quoted 9 lines
mm/percpu.c +1 -5
@@ -1421,15 +1421,13 @@ static struct pcpu_chunk * __init pcpu_alloc_first_chunk(unsigned long tmp_addr,
 	if (chunk->end_offset) {
 		/* hide the end of the bitmap */
 		offset_bits = chunk->end_offset / PCPU_MIN_ALLOC_SIZE;
-		bitmap_set(chunk->alloc_map,
-			   pcpu_chunk_map_bits(chunk) - offset_bits,
-			   offset_bits);
-		set_bit((start_offset + map_size) / PCPU_MIN_ALLOC_SIZE,
-			chunk->bound_map);
+		start_offset = region_bits - offset_bits;

Generally I think this makes sense, but I'm less inclined to mix the start_offset variable name in here.

A helper function might make this cleaner, consolidating the start_offset and end_offset logic.

static void pcpu_chunk_hide_region(chunk, bit_off, bits);

Thanks,
Dennis

Quoted 14 lines
mm/percpu.c +4 -2
+
+		bitmap_set(chunk->alloc_map, start_offset, offset_bits);
+		set_bit(start_offset, chunk->bound_map);
 		set_bit(region_bits, chunk->bound_map);
 
-		pcpu_block_update_hint_alloc(chunk, pcpu_chunk_map_bits(chunk)
-					     - offset_bits, offset_bits);
+		pcpu_block_update_hint_alloc(chunk, start_offset, offset_bits);
 	}
 
 	return chunk;
Signature · 1 line
2.44.0
#2 lore raw

PATCH v2 1/1 percpu: simplify the logic of pcpu_alloc_first_chunk()

In the logic for hiding the end region of the chunk, there are several places where the same value 'region_bits - offset_bits' is calculated over and over again using different methods. Eliminating these redundant calculations can improve code readability.

Additionally, there is a lot of repetitive code when hiding the start and end regions of the chunk. We can consolidate this logic into a function, making the final code cleaner and more concise.

---
v1 -> v2: Optimize the code based on Dennis's suggestion

 mm/percpu.c | 27 ++++++++++++---------------
 1 file changed, 12 insertions(+), 15 deletions(-)
mm/percpu.c +12 -15
index 4e11fc1e6def..d22b317f3d41 100644
@@ -1329,6 +1329,15 @@ static void pcpu_init_md_blocks(struct pcpu_chunk *chunk)
 		pcpu_init_md_block(md_block, PCPU_BITMAP_BLOCK_BITS);
 }
 
+static void pcpu_chunk_hide_region(struct pcpu_chunk *chunk, int bit_off,
+				   int bits)
+{
+	bitmap_set(chunk->alloc_map, bit_off, bits);
+	set_bit(bit_off, chunk->bound_map);
+	set_bit(bit_off + bits, chunk->bound_map);
+	pcpu_block_update_hint_alloc(chunk, bit_off, bits);
+}
+
 /**
  * pcpu_alloc_first_chunk - creates chunks that serve the first chunk
  * @tmp_addr: the start of the region served
@@ -1409,27 +1418,15 @@ static struct pcpu_chunk * __init pcpu_alloc_first_chunk(unsigned long tmp_addr,
 	if (chunk->start_offset) {
 		/* hide the beginning of the bitmap */
 		offset_bits = chunk->start_offset / PCPU_MIN_ALLOC_SIZE;
-		bitmap_set(chunk->alloc_map, 0, offset_bits);
-		set_bit(0, chunk->bound_map);
-		set_bit(offset_bits, chunk->bound_map);
-
 		chunk->chunk_md.first_free = offset_bits;
-
-		pcpu_block_update_hint_alloc(chunk, 0, offset_bits);
+		pcpu_chunk_hide_region(chunk, 0, offset_bits);
 	}
 
 	if (chunk->end_offset) {
 		/* hide the end of the bitmap */
 		offset_bits = chunk->end_offset / PCPU_MIN_ALLOC_SIZE;
-		bitmap_set(chunk->alloc_map,
-			   pcpu_chunk_map_bits(chunk) - offset_bits,
-			   offset_bits);
-		set_bit((start_offset + map_size) / PCPU_MIN_ALLOC_SIZE,
-			chunk->bound_map);
-		set_bit(region_bits, chunk->bound_map);
-
-		pcpu_block_update_hint_alloc(chunk, pcpu_chunk_map_bits(chunk)
-					     - offset_bits, offset_bits);
+		pcpu_chunk_hide_region(chunk, region_bits - offset_bits,
+				       offset_bits);
 	}
 
 	return chunk;
Signature · 1 line
2.44.0
maintainer
#4 lore raw

Hi Yuntao,

Quoted 6 lines · Yuntao Wang

Hi Dennis,

Can this v2 version of the patch be merged? Or what else do I need to do?

Thanks,
Yuntao

I'm going to massage it just a little bit to format a bit more like the rest of percpu.c. If it's a little too much, I'll send out a v3 and keep you as the author, then apply it to for-6.10.

Thanks,
Dennis

#5 lore raw

Hi Dennis,

Quoted 10 lines · Dennis Zhou

Hi Yuntao,

Quoted 6 lines · Yuntao Wang

Hi Dennis,

Can this v2 version of the patch be merged? Or what else do I need to do?

Thanks,
Yuntao

Quoted 6 lines

I'm going to massage it just a little bit to format a bit more like the rest of percpu.c. If it's a little too much, I'll send out a v3 and keep you as the author, then apply it to for-6.10.

Thanks,
Dennis

Thanks a lot. :-)