Re: [PATCH v2] net: ethernet: ti: cpdma: switch to use genalloc

[Mugunthan V N]

On Friday 24 June 2016 05:13 PM, Grygorii Strashko wrote:
> TI CPDMA currently uses a bitmap for tracking descriptors alloactions
> allocations, but The genalloc already handles the same and can be used
> as with special memory (SRAM) as with DMA cherent memory chank
> (dma_alloc_coherent()). Hence, switch to using genalloc and add
> desc_num property for each channel for limitation of max number of
> allowed descriptors for each CPDMA channel. This patch do not affect
> on net throuput.
> 
> Tested-by: Ivan Khoronzhuk  
> Signed-off-by: Grygorii Strashko 

Acked-by: Mugunthan V N 

Regards
Mugunthan V N

[PATCH v2] net: ethernet: ti: cpdma: switch to use genalloc

[Grygorii Strashko]

TI CPDMA currently uses a bitmap for tracking descriptors alloactions
allocations, but The genalloc already handles the same and can be used
as with special memory (SRAM) as with DMA cherent memory chank
(dma_alloc_coherent()). Hence, switch to using genalloc and add
desc_num property for each channel for limitation of max number of
allowed descriptors for each CPDMA channel. This patch do not affect
on net throuput.

Tested-by: Ivan Khoronzhuk  
Signed-off-by: Grygorii Strashko 
---
Testing
TCP window: 256K, bandwidth in Mbits/sec:
 host: iperf -s
 device: iperf -c  172.22.39.17 -t600 -i5 -d -w128K

AM437x-idk, 1Gbps link
 before: : 341.60, after: 232+123=355
am57xx-beagle-x15, 1Gbps link
 before: : 1112.80, after: 814+321=1135
am335x-boneblack, 100Mbps link
 before: : 162.40, after: 72+93=165

changes in v2:
 - reverted change in desc_phys() to keep am3517 which has separate CPPI
   addresses from EMAC and CPU perspective
 - minor format changes.

link on v1:
 https://lkml.org/lkml/2016/6/23/353

 drivers/net/ethernet/ti/davinci_cpdma.c | 132 +++-
 1 file changed, 60 insertions(+), 72 deletions(-)

diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c 
b/drivers/net/ethernet/ti/davinci_cpdma.c
index 63b3009..b40a402 100644
--- a/drivers/net/ethernet/ti/davinci_cpdma.c
+++ b/drivers/net/ethernet/ti/davinci_cpdma.c
@@ -21,7 +21,7 @@
 #include 
 #include 
 #include 
-
+#include 
 #include "davinci_cpdma.h"
 
 /* DMA Registers */
@@ -87,9 +87,8 @@ struct cpdma_desc_pool {
void*cpumap;/* dma_alloc map */
int desc_size, mem_size;
int num_desc, used_desc;
-   unsigned long   *bitmap;
struct device   *dev;
-   spinlock_t  lock;
+   struct gen_pool *gen_pool;
 };
 
 enum cpdma_state {
@@ -117,6 +116,7 @@ struct cpdma_chan {
int chan_num;
spinlock_t  lock;
int count;
+   u32 desc_num;
u32 mask;
cpdma_handler_fnhandler;
enum dma_data_direction dir;
@@ -145,6 +145,20 @@ struct cpdma_chan {
 (directed << CPDMA_TO_PORT_SHIFT));\
} while (0)
 
+static void cpdma_desc_pool_destroy(struct cpdma_desc_pool *pool)
+{
+   if (!pool)
+   return;
+
+   WARN_ON(pool->used_desc);
+   if (pool->cpumap) {
+   dma_free_coherent(pool->dev, pool->mem_size, pool->cpumap,
+ pool->phys);
+   } else {
+   iounmap(pool->iomap);
+   }
+}
+
 /*
  * Utility constructs for a cpdma descriptor pool.  Some devices (e.g. davinci
  * emac) have dedicated on-chip memory for these descriptors.  Some other
@@ -155,24 +169,25 @@ static struct cpdma_desc_pool *
 cpdma_desc_pool_create(struct device *dev, u32 phys, dma_addr_t hw_addr,
int size, int align)
 {
-   int bitmap_size;
struct cpdma_desc_pool *pool;
+   int ret;
 
pool = devm_kzalloc(dev, sizeof(*pool), GFP_KERNEL);
if (!pool)
-   goto fail;
-
-   spin_lock_init(>lock);
+   goto gen_pool_create_fail;
 
pool->dev   = dev;
pool->mem_size  = size;
pool->desc_size = ALIGN(sizeof(struct cpdma_desc), align);
pool->num_desc  = size / pool->desc_size;
 
-   bitmap_size  = (pool->num_desc / BITS_PER_LONG) * sizeof(long);
-   pool->bitmap = devm_kzalloc(dev, bitmap_size, GFP_KERNEL);
-   if (!pool->bitmap)
-   goto fail;
+   pool->gen_pool = devm_gen_pool_create(dev, ilog2(pool->desc_size), -1,
+ "cpdma");
+   if (IS_ERR(pool->gen_pool)) {
+   dev_err(dev, "pool create failed %ld\n",
+   PTR_ERR(pool->gen_pool));
+   goto gen_pool_create_fail;
+   }
 
if (phys) {
pool->phys  = phys;
@@ -185,24 +200,22 @@ cpdma_desc_pool_create(struct device *dev, u32 phys, 
dma_addr_t hw_addr,
pool->phys = pool->hw_addr; /* assumes no IOMMU, don't use this 
value */
}
 
-   if (pool->iomap)
-   return pool;
-fail:
-   return NULL;
-}
-
-static void cpdma_desc_pool_destroy(struct cpdma_desc_pool *pool)
-{
-   if (!pool)
-   return;
+   if (!pool->iomap)
+   goto gen_pool_create_fail;
 
-   WARN_ON(pool->used_desc);
-   if (pool->cpumap) {
-   dma_free_coherent(pool->dev, pool->mem_size, pool->cpumap,
- pool->phys);
-   } else {
-   iounmap(pool->iomap);
+   ret = gen_pool_add_virt(pool->gen_pool, (unsigned long)pool->iomap,
+