Re: [PATCH v2 4/4] cache/slow-path: reduce cache align requirement for 128-byte cache targets

On Fri, Dec 11, 2015 at 12:55:57PM +0000, Ananyev, Konstantin wrote:

Hi Jerin,

quoted

-----Original Message-----
From: Jerin Jacob [mailto:jerin.jacob@caviumnetworks.com]
Sent: Thursday, December 10, 2015 4:36 PM
To: dev@dpdk.org
Cc: thomas.monjalon@6wind.com; Ananyev, Konstantin; viktorin@rehivetech.com; jianbo.liu@linaro.org; Jerin Jacob
Subject: [dpdk-dev] [PATCH v2 4/4] cache/slow-path: reduce cache align requirement for 128-byte cache targets

slow-path data structures need not be 128-byte cache aligned.
Reduce the alignment to 64-byte to save the memory.

No behavior change for 64-byte cache aligned systems as minimum
cache line size as 64.

Signed-off-by: Jerin Jacob <redacted>
---
 lib/librte_ether/rte_ethdev.h    | 4 ++--
 lib/librte_mempool/rte_mempool.h | 2 +-
 lib/librte_ring/rte_ring.h       | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
index bada8ad..4dbf73b 100644
--- a/lib/librte_ether/rte_ethdev.h
+++ b/lib/librte_ether/rte_ethdev.h

@@ -863,7 +863,7 @@ struct rte_eth_rxq_info {
 	struct rte_eth_rxconf conf; /**< queue config parameters. */
 	uint8_t scattered_rx;       /**< scattered packets RX supported. */
 	uint16_t nb_desc;           /**< configured number of RXDs. */
-} __rte_cache_aligned;
+} __rte_cache_min_aligned;

 /**
  * Ethernet device TX queue information structure.

@@ -872,7 +872,7 @@ struct rte_eth_rxq_info {
 struct rte_eth_txq_info {
 	struct rte_eth_txconf conf; /**< queue config parameters. */
 	uint16_t nb_desc;           /**< configured number of TXDs. */
-} __rte_cache_aligned;
+} __rte_cache_min_aligned;

 /** Maximum name length for extended statistics counters */
 #define RTE_ETH_XSTATS_NAME_SIZE 64

diff --git a/lib/librte_mempool/rte_mempool.h b/lib/librte_mempool/rte_mempool.h
index 6e2390a..8e5d10c 100644
--- a/lib/librte_mempool/rte_mempool.h
+++ b/lib/librte_mempool/rte_mempool.h

@@ -92,7 +92,7 @@ struct rte_mempool_debug_stats {
 	uint64_t get_success_objs; /**< Objects successfully allocated. */
 	uint64_t get_fail_bulk;    /**< Failed allocation number. */
 	uint64_t get_fail_objs;    /**< Objects that failed to be allocated. */
-} __rte_cache_aligned;
+} __rte_cache_min_aligned;
 #endif

 #if RTE_MEMPOOL_CACHE_MAX_SIZE > 0

diff --git a/lib/librte_ring/rte_ring.h b/lib/librte_ring/rte_ring.h
index de036ce..33166aa 100644
--- a/lib/librte_ring/rte_ring.h
+++ b/lib/librte_ring/rte_ring.h

@@ -123,7 +123,7 @@ struct rte_ring_debug_stats {
 	uint64_t deq_success_objs; /**< Objects successfully dequeued. */
 	uint64_t deq_fail_bulk;    /**< Failed dequeues number. */
 	uint64_t deq_fail_objs;    /**< Objects that failed to be dequeued. */
-} __rte_cache_aligned;
+} __rte_cache_min_aligned;
 #endif

I think we better keep both struct rte_ring_debug_stats and rte_mempool_debug_stats
as __rte_cache_aligned.
Both are on a  per core basis and can be used at run-time
(when RTE_LIBRTE_RING_DEBUG/RTE_LIBRTE_MEMPOOL_DEBUG=y),
and not supposed to be shared by different cores.
All other things in the series look good to me.

OK, I will fix the alignment of rte_ring_debug_stats and
rte_mempool_debug_stats and send the next revision with your ACK.

Jerin

BTW, by some reason I can't find that series in the patchworks.
Konstantin

quoted

 #define RTE_RING_NAMESIZE 32 /**< The maximum length of a ring name. */
--
2.1.0