octeontx2-af: Clean-up some bitmap related usage
Subsystems (4)
cover octeontx2-af: Clean-up some bitmap related usage
The main point is patch 4.
The other ones are just minor things spoted while looking at the way the 'id_bmap' bitmap was used.
Christophe JAILLET (5):
octeontx2-af: Fix a potentially spurious error message
octeontx2-af: Slightly simplify rvu_npc_exact_init()
octeontx2-af: Use the bitmap API to allocate bitmaps
octeontx2-af: Fix the size of memory allocated for the 'id_bmap'
bitmap
octeontx2-af: Simplify a size computation in rvu_npc_exact_init()
.../marvell/octeontx2/af/rvu_npc_hash.c | 17 ++++++++--------- 1 file changed, 8 insertions(+), 9 deletions(-)
Signature · 1 line
2.34.1
PATCH 1/5 octeontx2-af: Fix a potentially spurious error message
When this error message is displayed, we know that the all the bits in the bitmap are set.
So, bitmap_weight() will return the number of bits of the bitmap, which is 'table->tot_ids'.
It is unlikely that a bit will be cleared between mutex_unlock() and dev_err(), but, in order to simplify the code and avoid this possibility, just take 'table->tot_ids'.
--- drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c +1 -1
@@ -490,7 +490,7 @@ static bool rvu_npc_exact_alloc_id(struct rvu *rvu, u32 *seq_id)if (idx == table->tot_ids) {mutex_unlock (&table->lock);dev_err (rvu->dev,"%s: No space in id bitmap (%d)\n" , - __func__,bitmap_weight (table->id_bmap, table->tot_ids)); + __func__, table->tot_ids);return false; }
Signature · 1 line
2.34.1
Re: [PATCH 1/5] octeontx2-af: Fix a potentially spurious error message
This series was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (master) by Jakub Kicinski <[email protected]>:
Quoted 11 lines · On Fri, 25 Nov 2022 13:23:57 +0100 you wrote:
When this error message is displayed, we know that the all the bits in the bitmap are set.
So,
bitmap_weight()will return the number of bits of the bitmap, which is 'table->tot_ids'.It is unlikely that a bit will be cleared between
mutex_unlock()anddev_err(), but, in order to simplify the code and avoid this possibility, just take 'table->tot_ids'.[...]
Here is the summary with links:
- [1/5] octeontx2-af: Fix a potentially spurious error message
https://git.kernel.org/netdev/net-next/c/2450d7d93fd2 - [2/5] octeontx2-af: Slightly simplify rvu_npc_exact_init() https://git.kernel.org/netdev/net-next/c/b6a0ecaee2e6 - [3/5] octeontx2-af: Use the bitmap API to allocate bitmaps https://git.kernel.org/netdev/net-next/c/05a7b52ee5e4 - [4/5] octeontx2-af: Fix the size of memory allocated for the 'id_bmap' bitmap https://git.kernel.org/netdev/net-next/c/6d135d9e2b00 - [5/5] octeontx2-af: Simplify a size computation in rvu_npc_exact_init() https://git.kernel.org/netdev/net-next/c/450f06505396
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
PATCH 2/5 octeontx2-af: Slightly simplify rvu_npc_exact_init()
In reply to Christophe JAILLETUse kzalloc() instead of kmalloc()/memset().
--- drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c +1 -2
@@ -1870,12 +1870,11 @@ int rvu_npc_exact_init(struct rvu *rvu)/* Set capability to true */ rvu->hw->cap.npc_exact_match_enabled = true; - table =kmalloc (sizeof (*table),GFP_KERNEL ); + table =kzalloc (sizeof (*table),GFP_KERNEL );if (!table)return -ENOMEM ;dev_dbg (rvu->dev,"%s: Memory allocation for table success\n" , __func__); -memset (table,0 ,sizeof (*table)); rvu->hw->table = table;/* Read table size, ways and depth */
Signature · 1 line
2.34.1
PATCH 3/5 octeontx2-af: Use the bitmap API to allocate bitmaps
In reply to Christophe JAILLETUse devm_bitmap_zalloc() instead of hand-writing it.
This also makes the comment "Allocate bitmap for 32 entry mcam" more explicit because now 32 is really used in the allocation function, instead of an obscure 'sizeof(long)'.
--- drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c +3 -3
@@ -1898,15 +1898,15 @@ int rvu_npc_exact_init(struct rvu *rvu) table_size = table->mem_table.depth * table->mem_table.ways;/* Allocate bitmap for 4way 2K table */ - table->mem_table.bmap =devm_kcalloc (rvu->dev,BITS_TO_LONGS (table_size), -sizeof (long ),GFP_KERNEL ); + table->mem_table.bmap =devm_bitmap_zalloc (rvu->dev, table_size, +GFP_KERNEL );if (!table->mem_table.bmap)return -ENOMEM ;dev_dbg (rvu->dev,"%s: Allocated bitmap for 4way 2K entry table\n" , __func__);/* Allocate bitmap for 32 entry mcam */ - table->cam_table.bmap =devm_kcalloc (rvu->dev,1 ,sizeof (long ),GFP_KERNEL ); + table->cam_table.bmap =devm_bitmap_zalloc (rvu->dev,32 ,GFP_KERNEL );if (!table->cam_table.bmap)return -ENOMEM ;
Signature · 1 line
2.34.1
PATCH 4/5 octeontx2-af: Fix the size of memory allocated for the 'id_bmap' bitmap
In reply to Christophe JAILLETThis allocation is really spurious.
The size of the bitmap is 'tot_ids' and it is used as such in the driver.
So we could expect something like:
table->id_bmap = devm_kcalloc(rvu->dev, BITS_TO_LONGS(table->tot_ids),
sizeof(long), GFP_KERNEL);
However, when the bitmap is allocated, we allocate:
BITS_TO_LONGS(table->tot_ids) * table->tot_ids ~= table->tot_ids / 32 * table->tot_ids ~= table->tot_ids^2 / 32
It is proportional to the square of 'table->tot_ids' which seems to potentially be big.
Allocate the expected amount of memory, and switch to the bitmap API to have it more straightforward.
---
This patch is speculative.
If I'm right, I'm curious to know if 'able->tot_ids' can really get big (I'm just guessing) , and if yes, how much.
--- drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c +2 -2
@@ -1914,8 +1914,8 @@ int rvu_npc_exact_init(struct rvu *rvu)dev_dbg (rvu->dev,"%s: Allocated bitmap for 32 entry cam\n" , __func__); table->tot_ids = (table->mem_table.depth * table->mem_table.ways) + table->cam_table.depth; - table->id_bmap =devm_kcalloc (rvu->dev,BITS_TO_LONGS (table->tot_ids), - table->tot_ids,GFP_KERNEL ); + table->id_bmap =devm_bitmap_zalloc (rvu->dev, table->tot_ids, +GFP_KERNEL );if (!table->id_bmap)return -ENOMEM ;
Signature · 1 line
2.34.1
PATCH 5/5 octeontx2-af: Simplify a size computation in rvu_npc_exact_init()
In reply to Christophe JAILLETWe know that table_size = table->mem_table.depth * table->mem_table.ways, so use it instead, it is less verbose.
--- drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/net/ethernet/marvell/octeontx2/af/rvu_npc_hash.c +1 -1
@@ -1913,7 +1913,7 @@ int rvu_npc_exact_init(struct rvu *rvu)dev_dbg (rvu->dev,"%s: Allocated bitmap for 32 entry cam\n" , __func__); - table->tot_ids = (table->mem_table.depth * table->mem_table.ways) + table->cam_table.depth; + table->tot_ids = table_size + table->cam_table.depth; table->id_bmap =devm_bitmap_zalloc (rvu->dev, table->tot_ids,GFP_KERNEL );
Signature · 1 line
2.34.1