Commit 021230d4 authored by Ayyappan Veeraiyan's avatar Ayyappan Veeraiyan Committed by Jeff Garzik

ixgbe: Introduce MSI-X queue vector code

This code abstracts the per-queue MSI-X interrupt vector into
a queue vector layer. This abstraction is needed since there can
be many more queues than available MSI-X vectors in a machine.

The MSI-X irq vectors are remapped to a shared queue vector which
can point to several (both RX and TX) hardware queues. The NAPI
algorithm then cleans the appropriate ring/queues on interrupt
or poll.

The remapping is a delicate and complex calculation to make sure
that we're not unbalancing the irq load, and spreads the irqs
as much as possible, and may combine RX and TX flows onto the
same queue vector.

This effectively enables receive flow hashing across vectors
and helps irq load balance across CPUs.
Signed-off-by: default avatarAyyappan Veeraiyan <ayyappan.veeraiyan@intel.com>
Signed-off-by: default avatarAuke Kok <auke-jan.h.kok@intel.com>
Acked-by: default avatarJesse Brandeburg <jesse.brandeburg@intel.com>
Acked-by: default avatarWaskiewicz Jr, Peter P <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: default avatarJeff Garzik <jeff@garzik.org>
parent 53e7c46b
......@@ -120,7 +120,6 @@ struct ixgbe_queue_stats {
};
struct ixgbe_ring {
struct ixgbe_adapter *adapter; /* backlink */
void *desc; /* descriptor ring memory */
dma_addr_t dma; /* phys. address of descriptor ring */
unsigned int size; /* length in bytes */
......@@ -128,6 +127,7 @@ struct ixgbe_ring {
unsigned int next_to_use;
unsigned int next_to_clean;
int queue_index; /* needed for multiqueue queue management */
union {
struct ixgbe_tx_buffer *tx_buffer_info;
struct ixgbe_rx_buffer *rx_buffer_info;
......@@ -137,7 +137,13 @@ struct ixgbe_ring {
u16 tail;
u16 reg_idx; /* holds the special value that gets the hardware register
* offset associated with this ring, which is different
* for DCE and RSS modes */
struct ixgbe_queue_stats stats;
u8 v_idx; /* maps directly to the index for this ring in the hardware
* vector array, can also be used for finding the bit in EICR
* and friends that represents the vector for this ring */
u32 eims_value;
u16 itr_register;
......@@ -146,6 +152,31 @@ struct ixgbe_ring {
u16 work_limit; /* max work per interrupt */
};
#define RING_F_VMDQ 1
#define RING_F_RSS 2
#define IXGBE_MAX_RSS_INDICES 16
#define IXGBE_MAX_VMDQ_INDICES 16
struct ixgbe_ring_feature {
int indices;
int mask;
};
#define MAX_RX_QUEUES 64
#define MAX_TX_QUEUES 32
/* MAX_MSIX_Q_VECTORS of these are allocated,
* but we only use one per queue-specific vector.
*/
struct ixgbe_q_vector {
struct ixgbe_adapter *adapter;
struct napi_struct napi;
DECLARE_BITMAP(rxr_idx, MAX_RX_QUEUES); /* Rx ring indices */
DECLARE_BITMAP(txr_idx, MAX_TX_QUEUES); /* Tx ring indices */
u8 rxr_count; /* Rx ring count assigned to this vector */
u8 txr_count; /* Tx ring count assigned to this vector */
u32 eitr;
};
/* Helper macros to switch between ints/sec and what the register uses.
* And yes, it's the same math going both ways.
*/
......@@ -166,6 +197,14 @@ struct ixgbe_ring {
#define IXGBE_MAX_JUMBO_FRAME_SIZE 16128
#define OTHER_VECTOR 1
#define NON_Q_VECTORS (OTHER_VECTOR)
#define MAX_MSIX_Q_VECTORS 16
#define MIN_MSIX_Q_VECTORS 2
#define MAX_MSIX_COUNT (MAX_MSIX_Q_VECTORS + NON_Q_VECTORS)
#define MIN_MSIX_COUNT (MIN_MSIX_Q_VECTORS + NON_Q_VECTORS)
/* board specific private data structure */
struct ixgbe_adapter {
struct timer_list watchdog_timer;
......@@ -173,10 +212,11 @@ struct ixgbe_adapter {
u16 bd_number;
u16 rx_buf_len;
struct work_struct reset_task;
struct ixgbe_q_vector q_vector[MAX_MSIX_Q_VECTORS];
char name[MAX_MSIX_COUNT][IFNAMSIZ + 5];
/* TX */
struct ixgbe_ring *tx_ring; /* One per active queue */
struct napi_struct napi;
u64 restart_queue;
u64 lsc_int;
u64 hw_tso_ctxt;
......@@ -192,22 +232,26 @@ struct ixgbe_adapter {
u64 non_eop_descs;
int num_tx_queues;
int num_rx_queues;
int num_msix_vectors;
struct ixgbe_ring_feature ring_feature[3];
struct msix_entry *msix_entries;
u64 rx_hdr_split;
u32 alloc_rx_page_failed;
u32 alloc_rx_buff_failed;
/* Some features need tri-state capability,
* thus the additional *_CAPABLE flags.
*/
u32 flags;
#define IXGBE_FLAG_RX_CSUM_ENABLED (u32)(1)
#define IXGBE_FLAG_RX_CSUM_ENABLED (u32)(1 << 0)
#define IXGBE_FLAG_MSI_ENABLED (u32)(1 << 1)
#define IXGBE_FLAG_MSIX_ENABLED (u32)(1 << 2)
#define IXGBE_FLAG_RX_PS_ENABLED (u32)(1 << 3)
#define IXGBE_FLAG_IN_NETPOLL (u32)(1 << 4)
/* Interrupt Throttle Rate */
u32 rx_eitr;
u32 tx_eitr;
#define IXGBE_FLAG_MSIX_ENABLED (u32)(1 << 2)
#define IXGBE_FLAG_RX_PS_ENABLED (u32)(1 << 3)
#define IXGBE_FLAG_IN_NETPOLL (u32)(1 << 4)
#define IXGBE_FLAG_IMIR_ENABLED (u32)(1 << 5)
#define IXGBE_FLAG_RSS_ENABLED (u32)(1 << 6)
#define IXGBE_FLAG_VMDQ_ENABLED (u32)(1 << 7)
/* OS defined structs */
struct net_device *netdev;
......@@ -218,7 +262,10 @@ struct ixgbe_adapter {
struct ixgbe_hw hw;
u16 msg_enable;
struct ixgbe_hw_stats stats;
char lsc_name[IFNAMSIZ + 5];
/* Interrupt Throttle Rate */
u32 rx_eitr;
u32 tx_eitr;
unsigned long state;
u64 tx_busy;
......
This diff is collapsed.
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment