| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1376432265-5497-2-git-send-email-neepatel@cisco.com>
Date: Tue, 13 Aug 2013 15:17:43 -0700
From: Neel Patel <neepatel@...co.com>
To: netdev@...r.kernel.org
Cc: umalhi@...co.com, Neel Patel <neepatel@...co.com>,
Nishank Trivedi <nistrive@...co.com>,
Christian Benvenuti <benve@...co.com>
Subject: [PATCH v2 1/3] drivers/net: enic: Adding support for Cisco Low Latency NIC
This patch,
- Adds new firmware commands for the new Cisco Low Latency NIC
(aka. USNIC).
- Enables descriptor prefetch on the hardware. This feature
reduces latency for small packet transmit.
Signed-off-by: Neel Patel <neepatel@...co.com>
Signed-off-by: Nishank Trivedi <nistrive@...co.com>
Signed-off-by: Christian Benvenuti <benve@...co.com>
---
drivers/net/ethernet/cisco/enic/enic.h | 2 +-
drivers/net/ethernet/cisco/enic/enic_res.h | 9 +-
drivers/net/ethernet/cisco/enic/vnic_dev.c | 5 +
drivers/net/ethernet/cisco/enic/vnic_dev.h | 1 +
drivers/net/ethernet/cisco/enic/vnic_devcmd.h | 176 +++++++++++++++++++++++++-
drivers/net/ethernet/cisco/enic/vnic_rq.c | 5 +-
drivers/net/ethernet/cisco/enic/vnic_rq.h | 5 +-
drivers/net/ethernet/cisco/enic/vnic_wq.c | 3 -
drivers/net/ethernet/cisco/enic/vnic_wq.h | 86 ++++++++++++-
9 files changed, 275 insertions(+), 17 deletions(-)
diff --git a/drivers/net/ethernet/cisco/enic/enic.h b/drivers/net/ethernet/cisco/enic/enic.h
index 2e37c63..75e842d 100644
--- a/drivers/net/ethernet/cisco/enic/enic.h
+++ b/drivers/net/ethernet/cisco/enic/enic.h
@@ -33,7 +33,7 @@
#define DRV_NAME "enic"
#define DRV_DESCRIPTION "Cisco VIC Ethernet NIC Driver"
#define DRV_VERSION "2.1.1.39"
-#define DRV_COPYRIGHT "Copyright 2008-2011 Cisco Systems, Inc"
+#define DRV_COPYRIGHT "Copyright 2008-2013 Cisco Systems, Inc"
#define ENIC_BARS_MAX 6
diff --git a/drivers/net/ethernet/cisco/enic/enic_res.h b/drivers/net/ethernet/cisco/enic/enic_res.h
index 25be273..69f60af 100644
--- a/drivers/net/ethernet/cisco/enic/enic_res.h
+++ b/drivers/net/ethernet/cisco/enic/enic_res.h
@@ -47,6 +47,9 @@ static inline void enic_queue_wq_desc_ex(struct vnic_wq *wq,
int offload_mode, int cq_entry, int sop, int eop, int loopback)
{
struct wq_enet_desc *desc = vnic_wq_next_desc(wq);
+ u8 desc_skip_cnt = 1;
+ u8 compressed_send = 0;
+ u64 wrid = 0;
wq_enet_desc_enc(desc,
(u64)dma_addr | VNIC_PADDR_TARGET,
@@ -59,7 +62,8 @@ static inline void enic_queue_wq_desc_ex(struct vnic_wq *wq,
(u16)vlan_tag,
(u8)loopback);
- vnic_wq_post(wq, os_buf, dma_addr, len, sop, eop);
+ vnic_wq_post(wq, os_buf, dma_addr, len, sop, eop, desc_skip_cnt,
+ (u8)cq_entry, compressed_send, wrid);
}
static inline void enic_queue_wq_desc_cont(struct vnic_wq *wq,
@@ -120,6 +124,7 @@ static inline void enic_queue_rq_desc(struct vnic_rq *rq,
dma_addr_t dma_addr, unsigned int len)
{
struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
+ u64 wrid = 0;
u8 type = os_buf_index ?
RQ_ENET_TYPE_NOT_SOP : RQ_ENET_TYPE_ONLY_SOP;
@@ -127,7 +132,7 @@ static inline void enic_queue_rq_desc(struct vnic_rq *rq,
(u64)dma_addr | VNIC_PADDR_TARGET,
type, (u16)len);
- vnic_rq_post(rq, os_buf, os_buf_index, dma_addr, len);
+ vnic_rq_post(rq, os_buf, os_buf_index, dma_addr, len, wrid);
}
struct enic;
diff --git a/drivers/net/ethernet/cisco/enic/vnic_dev.c b/drivers/net/ethernet/cisco/enic/vnic_dev.c
index 97455c5..1b436b3 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_dev.c
+++ b/drivers/net/ethernet/cisco/enic/vnic_dev.c
@@ -970,6 +970,11 @@ err_out:
return NULL;
}
+struct pci_dev *vnic_dev_get_pdev(struct vnic_dev *vdev)
+{
+ return vdev->pdev;
+}
+
int vnic_dev_init_prov2(struct vnic_dev *vdev, u8 *buf, u32 len)
{
u64 a0, a1 = len;
diff --git a/drivers/net/ethernet/cisco/enic/vnic_dev.h b/drivers/net/ethernet/cisco/enic/vnic_dev.h
index f3d9b79..e670029 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_dev.h
+++ b/drivers/net/ethernet/cisco/enic/vnic_dev.h
@@ -127,6 +127,7 @@ int vnic_dev_set_ig_vlan_rewrite_mode(struct vnic_dev *vdev,
struct vnic_dev *vnic_dev_register(struct vnic_dev *vdev,
void *priv, struct pci_dev *pdev, struct vnic_dev_bar *bar,
unsigned int num_bars);
+struct pci_dev *vnic_dev_get_pdev(struct vnic_dev *vdev);
int vnic_dev_init_prov2(struct vnic_dev *vdev, u8 *buf, u32 len);
int vnic_dev_enable2(struct vnic_dev *vdev, int active);
int vnic_dev_enable2_done(struct vnic_dev *vdev, int *status);
diff --git a/drivers/net/ethernet/cisco/enic/vnic_devcmd.h b/drivers/net/ethernet/cisco/enic/vnic_devcmd.h
index 23d5552..b9a0d78 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_devcmd.h
+++ b/drivers/net/ethernet/cisco/enic/vnic_devcmd.h
@@ -281,11 +281,25 @@ enum vnic_devcmd_cmd {
* 0 if no VIF-CONFIG-INFO TLV was ever received. */
CMD_CONFIG_INFO_GET = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ALL, 44),
+ /* INT13 API: (u64)a0=paddr to vnic_int13_params struct
+ * (u32)a1=INT13_CMD_xxx
+ */
+ CMD_INT13_ALL = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ALL, 45),
+
+ /* Set default vlan:
+ * in: (u16)a0=new default vlan
+ * (u16)a1=zero for overriding vlan with param a0,
+ * non-zero for resetting vlan to the default
+ * out: (u16)a0=old default vlan
+ */
+ CMD_SET_DEFAULT_VLAN = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ALL, 46),
+
/* init_prov_info2:
* Variant of CMD_INIT_PROV_INFO, where it will not try to enable
* the vnic until CMD_ENABLE2 is issued.
* (u64)a0=paddr of vnic_devcmd_provinfo
- * (u32)a1=sizeof provision info */
+ * (u32)a1=sizeof provision info
+ */
CMD_INIT_PROV_INFO2 = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 47),
/* enable2:
@@ -339,16 +353,57 @@ enum vnic_devcmd_cmd {
CMD_INTR_COAL_CONVERT = _CMDC(_CMD_DIR_READ, _CMD_VTYPE_ALL, 50),
/*
- * cmd_set_mac_addr
- * set mac address
+ * Set the predefined mac address as default
* in:
* (u48)a0 = mac addr
- *
*/
CMD_SET_MAC_ADDR = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 55),
+
+ /* Update the provisioning info of the given VIF
+ * (u64)a0=paddr of vnic_devcmd_provinfo
+ * (u32)a1=sizeof provision info
+ */
+ CMD_PROV_INFO_UPDATE = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 56),
+
+ /* Add a filter.
+ * in: (u64) a0= filter address
+ * (u32) a1= size of filter
+ * out: (u32) a0=filter identifier
+ */
+ CMD_ADD_FILTER = _CMDC(_CMD_DIR_RW, _CMD_VTYPE_ENET, 58),
+
+ /* Delete a filter.
+ * in: (u32) a0=filter identifier
+ */
+ CMD_DEL_FILTER = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 59),
+
+ /* Enable a Queue Pair in User space NIC
+ * in: (u32) a0=Queue Pair number
+ * (u32) a1= command
+ */
+ CMD_QP_ENABLE = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 60),
+
+ /* Disable a Queue Pair in User space NIC
+ * in: (u32) a0=Queue Pair number
+ * (u32) a1= command
+ */
+ CMD_QP_DISABLE = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 61),
+
+ /* Stats dump Queue Pair in User space NIC
+ * in: (u32) a0=Queue Pair number
+ * (u64) a1=host buffer addr for status dump
+ * (u32) a2=length of the buffer
+ */
+ CMD_QP_STATS_DUMP = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 62),
+
+ /* Clear stats for Queue Pair in User space NIC
+ * in: (u32) a0=Queue Pair number
+ */
+ CMD_QP_STATS_CLEAR = _CMDC(_CMD_DIR_WRITE, _CMD_VTYPE_ENET, 63),
};
/* CMD_ENABLE2 flags */
+#define CMD_ENABLE2_STANDBY 0x0
#define CMD_ENABLE2_ACTIVE 0x1
/* flags for CMD_OPEN */
@@ -364,6 +419,9 @@ enum vnic_devcmd_cmd {
#define CMD_PFILTER_PROMISCUOUS 0x08
#define CMD_PFILTER_ALL_MULTICAST 0x10
+/* Commands for CMD_QP_ENABLE/CM_QP_DISABLE */
+#define CMD_QP_RQWQ 0x0
+
/* rewrite modes for CMD_IG_VLAN_REWRITE_MODE */
#define IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK 0
#define IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN 1
@@ -390,6 +448,7 @@ enum vnic_devcmd_error {
ERR_EMAXRES = 10,
ERR_ENOTSUPPORTED = 11,
ERR_EINPROGRESS = 12,
+ ERR_MAX
};
/*
@@ -435,6 +494,115 @@ struct vnic_devcmd_provinfo {
u8 data[0];
};
+/* These are used in flags field of different filters to denote
+ * valid fields used.
+ */
+#define FILTER_FIELD_VALID(fld) (1 << (fld - 1))
+
+#define FILTER_FIELDS_USNIC ( \
+ FILTER_FIELD_VALID(1) | \
+ FILTER_FIELD_VALID(2) | \
+ FILTER_FIELD_VALID(3) | \
+ FILTER_FIELD_VALID(4))
+
+#define FILTER_FIELDS_IPV4_5TUPLE ( \
+ FILTER_FIELD_VALID(1) | \
+ FILTER_FIELD_VALID(2) | \
+ FILTER_FIELD_VALID(3) | \
+ FILTER_FIELD_VALID(4) | \
+ FILTER_FIELD_VALID(5))
+
+#define FILTER_FIELDS_MAC_VLAN ( \
+ FILTER_FIELD_VALID(1) | \
+ FILTER_FIELD_VALID(2))
+
+#define FILTER_FIELD_USNIC_VLAN FILTER_FIELD_VALID(1)
+#define FILTER_FIELD_USNIC_ETHTYPE FILTER_FIELD_VALID(2)
+#define FILTER_FIELD_USNIC_PROTO FILTER_FIELD_VALID(3)
+#define FILTER_FIELD_USNIC_ID FILTER_FIELD_VALID(4)
+
+struct filter_usnic_id {
+ u32 flags;
+ u16 vlan;
+ u16 ethtype;
+ u8 proto_version;
+ u32 usnic_id;
+} __packed;
+
+#define FILTER_FIELD_5TUP_PROTO FILTER_FIELD_VALID(1)
+#define FILTER_FIELD_5TUP_SRC_AD FILTER_FIELD_VALID(2)
+#define FILTER_FIELD_5TUP_DST_AD FILTER_FIELD_VALID(3)
+#define FILTER_FIELD_5TUP_SRC_PT FILTER_FIELD_VALID(4)
+#define FILTER_FIELD_5TUP_DST_PT FILTER_FIELD_VALID(5)
+
+/* Enums for the protocol field. */
+enum protocol_e {
+ PROTO_UDP = 0,
+ PROTO_TCP = 1,
+};
+
+struct filter_ipv4_5tuple {
+ u32 flags;
+ u32 protocol;
+ u32 src_addr;
+ u32 dst_addr;
+ u16 src_port;
+ u16 dst_port;
+} __packed;
+
+#define FILTER_FIELD_VMQ_VLAN FILTER_FIELD_VALID(1)
+#define FILTER_FIELD_VMQ_MAC FILTER_FIELD_VALID(2)
+
+struct filter_mac_vlan {
+ u32 flags;
+ u16 vlan;
+ u8 mac_addr[6];
+} __packed;
+
+/* Specifies the filter_action type. */
+enum {
+ FILTER_ACTION_RQ_STEERING = 0,
+ FILTER_ACTION_MAX
+};
+
+struct filter_action {
+ u32 type;
+ union {
+ u32 rq_idx;
+ } u;
+} __packed;
+
+/* Specifies the filter type. */
+enum filter_type {
+ FILTER_USNIC_ID = 0,
+ FILTER_IPV4_5TUPLE = 1,
+ FILTER_MAC_VLAN = 2,
+ FILTER_MAX
+};
+
+struct filter {
+ u32 type;
+ union {
+ struct filter_usnic_id usnic;
+ struct filter_ipv4_5tuple ipv4;
+ struct filter_mac_vlan mac_vlan;
+ } u;
+} __packed;
+
+enum {
+ CLSF_TLV_FILTER = 0,
+ CLSF_TLV_ACTION = 1,
+};
+
+/* Maximum size of buffer to CMD_ADD_FILTER */
+#define FILTER_MAX_BUF_SIZE 100
+
+struct filter_tlv {
+ u_int32_t type;
+ u_int32_t length;
+ u_int32_t val[0];
+};
+
/*
* Writing cmd register causes STAT_BUSY to get set in status register.
* When cmd completes, STAT_BUSY will be cleared.
diff --git a/drivers/net/ethernet/cisco/enic/vnic_rq.c b/drivers/net/ethernet/cisco/enic/vnic_rq.c
index 7e1488f..36a2ed6 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_rq.c
+++ b/drivers/net/ethernet/cisco/enic/vnic_rq.c
@@ -30,12 +30,9 @@
static int vnic_rq_alloc_bufs(struct vnic_rq *rq)
{
struct vnic_rq_buf *buf;
- struct vnic_dev *vdev;
unsigned int i, j, count = rq->ring.desc_count;
unsigned int blks = VNIC_RQ_BUF_BLKS_NEEDED(count);
- vdev = rq->vdev;
-
for (i = 0; i < blks; i++) {
rq->bufs[i] = kzalloc(VNIC_RQ_BUF_BLK_SZ(count), GFP_ATOMIC);
if (!rq->bufs[i])
@@ -141,7 +138,7 @@ void vnic_rq_init(struct vnic_rq *rq, unsigned int cq_index,
unsigned int error_interrupt_enable,
unsigned int error_interrupt_offset)
{
- u32 fetch_index;
+ u32 fetch_index = 0;
/* Use current fetch_index as the ring starting point */
fetch_index = ioread32(&rq->ctrl->fetch_index);
diff --git a/drivers/net/ethernet/cisco/enic/vnic_rq.h b/drivers/net/ethernet/cisco/enic/vnic_rq.h
index 2056586..ee7bc95 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_rq.h
+++ b/drivers/net/ethernet/cisco/enic/vnic_rq.h
@@ -72,6 +72,7 @@ struct vnic_rq_buf {
unsigned int len;
unsigned int index;
void *desc;
+ uint64_t wr_id;
};
struct vnic_rq {
@@ -110,7 +111,8 @@ static inline unsigned int vnic_rq_next_index(struct vnic_rq *rq)
static inline void vnic_rq_post(struct vnic_rq *rq,
void *os_buf, unsigned int os_buf_index,
- dma_addr_t dma_addr, unsigned int len)
+ dma_addr_t dma_addr, unsigned int len,
+ uint64_t wrid)
{
struct vnic_rq_buf *buf = rq->to_use;
@@ -118,6 +120,7 @@ static inline void vnic_rq_post(struct vnic_rq *rq,
buf->os_buf_index = os_buf_index;
buf->dma_addr = dma_addr;
buf->len = len;
+ buf->wr_id = wrid;
buf = buf->next;
rq->to_use = buf;
diff --git a/drivers/net/ethernet/cisco/enic/vnic_wq.c b/drivers/net/ethernet/cisco/enic/vnic_wq.c
index 5e0d7a2..3e6b8d5 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_wq.c
+++ b/drivers/net/ethernet/cisco/enic/vnic_wq.c
@@ -30,12 +30,9 @@
static int vnic_wq_alloc_bufs(struct vnic_wq *wq)
{
struct vnic_wq_buf *buf;
- struct vnic_dev *vdev;
unsigned int i, j, count = wq->ring.desc_count;
unsigned int blks = VNIC_WQ_BUF_BLKS_NEEDED(count);
- vdev = wq->vdev;
-
for (i = 0; i < blks; i++) {
wq->bufs[i] = kzalloc(VNIC_WQ_BUF_BLK_SZ(count), GFP_ATOMIC);
if (!wq->bufs[i])
diff --git a/drivers/net/ethernet/cisco/enic/vnic_wq.h b/drivers/net/ethernet/cisco/enic/vnic_wq.h
index 7dd937a..b655667 100644
--- a/drivers/net/ethernet/cisco/enic/vnic_wq.h
+++ b/drivers/net/ethernet/cisco/enic/vnic_wq.h
@@ -58,6 +58,10 @@ struct vnic_wq_buf {
unsigned int index;
int sop;
void *desc;
+ uint64_t wr_id; /* Cookie */
+ uint8_t cq_entry; /* Gets completion event from hw */
+ uint8_t desc_skip_cnt; /* Num descs to occupy */
+ uint8_t compressed_send; /* Both hdr and payload in one desc */
};
/* Break the vnic_wq_buf allocations into blocks of 32/64 entries */
@@ -100,16 +104,94 @@ static inline void *vnic_wq_next_desc(struct vnic_wq *wq)
return wq->to_use->desc;
}
+#define PI_LOG2_CACHE_LINE_SIZE 5
+#define PI_INDEX_BITS 12
+#define PI_INDEX_MASK ((1U << PI_INDEX_BITS) - 1)
+#define PI_PREFETCH_LEN_MASK ((1U << PI_LOG2_CACHE_LINE_SIZE) - 1)
+#define PI_PREFETCH_LEN_OFF 16
+#define PI_PREFETCH_ADDR_BITS 43
+#define PI_PREFETCH_ADDR_MASK ((1ULL << PI_PREFETCH_ADDR_BITS) - 1)
+#define PI_PREFETCH_ADDR_OFF 21
+
+/* How many cache lines are touched by buffer (addr, len). */
+static inline unsigned int num_cache_lines_touched(dma_addr_t addr,
+ unsigned int len)
+{
+ const unsigned long mask = PI_PREFETCH_LEN_MASK;
+ const unsigned long laddr = (unsigned long)addr;
+ unsigned long lines, equiv_len;
+ /* A. If addr is aligned, our solution is just to round up len to the
+ * next boundary.
+ *
+ * e.g. addr = 0, len = 48
+ * +--------------------+
+ * |XXXXXXXXXXXXXXXXXXXX| 32-byte cacheline a
+ * +--------------------+
+ * |XXXXXXXXXX | cacheline b
+ * +--------------------+
+ *
+ * B. If addr is not aligned, however, we may use an extra
+ * cacheline. e.g. addr = 12, len = 22
+ *
+ * +--------------------+
+ * | XXXXXXXXXXXXX|
+ * +--------------------+
+ * |XX |
+ * +--------------------+
+ *
+ * Our solution is to make the problem equivalent to case A
+ * above by adding the empty space in the first cacheline to the length:
+ * unsigned long len;
+ *
+ * +--------------------+
+ * |eeeeeeeXXXXXXXXXXXXX| "e" is empty space, which we add to len
+ * +--------------------+
+ * |XX |
+ * +--------------------+
+ */
+ equiv_len = len + (laddr & mask);
+
+ /* Now we can just round up this len to the next 32-byte boundary. */
+ lines = (equiv_len + mask) & (~mask);
+
+ /* Scale bytes -> cachelines. */
+ return lines >> PI_LOG2_CACHE_LINE_SIZE;
+}
+
+static inline u64 vnic_cached_posted_index(dma_addr_t addr,
+ unsigned int len, unsigned int index)
+{
+ unsigned int num_cache_lines = num_cache_lines_touched(addr, len);
+ /* Wish we could avoid a branch here. We could have separate
+ * vnic_wq_post() and vinc_wq_post_inline(), the latter
+ * only supporting < 1k (2^5 * 2^5) sends, I suppose. This would
+ * eliminate the if (eop) branch as well.
+ */
+ if (num_cache_lines > PI_PREFETCH_LEN_MASK)
+ num_cache_lines = 0;
+ return (index & PI_INDEX_MASK) |
+ ((num_cache_lines & PI_PREFETCH_LEN_MASK)
+ << PI_PREFETCH_LEN_OFF) |
+ (((addr >> PI_LOG2_CACHE_LINE_SIZE) &
+ PI_PREFETCH_ADDR_MASK) << PI_PREFETCH_ADDR_OFF);
+}
+
static inline void vnic_wq_post(struct vnic_wq *wq,
void *os_buf, dma_addr_t dma_addr,
- unsigned int len, int sop, int eop)
+ unsigned int len, int sop, int eop,
+ uint8_t desc_skip_cnt, uint8_t cq_entry,
+ uint8_t compressed_send, uint64_t wrid)
{
struct vnic_wq_buf *buf = wq->to_use;
buf->sop = sop;
+ buf->cq_entry = cq_entry;
+ buf->compressed_send = compressed_send;
+ buf->desc_skip_cnt = desc_skip_cnt;
buf->os_buf = eop ? os_buf : NULL;
buf->dma_addr = dma_addr;
buf->len = len;
+ buf->wr_id = wrid;
buf = buf->next;
if (eop) {
@@ -123,7 +205,7 @@ static inline void vnic_wq_post(struct vnic_wq *wq,
}
wq->to_use = buf;
- wq->ring.desc_avail--;
+ wq->ring.desc_avail -= desc_skip_cnt;
}
static inline void vnic_wq_service(struct vnic_wq *wq,
--
1.8.4-rc0
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists