[1/2] net/qede: Fix VF re-load failure after ungraceful termination of DPDK
diff mbox series

Message ID 20200206200146.14808-1-manishc@marvell.com
State Accepted
Delegated to: Jerin Jacob
Headers show
Series
  • [1/2] net/qede: Fix VF re-load failure after ungraceful termination of DPDK
Related show

Checks

Context Check Description
ci/Intel-compilation fail apply issues
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-testing success Testing PASS
ci/iol-nxp-Performance success Performance Testing PASS
ci/checkpatch success coding style OK

Commit Message

Manish Chopra Feb. 6, 2020, 8:01 p.m. UTC
On ungraceful termination of DPDK application, PMD VF driver
fails to re-load due to PF seeing the VF in unexpected state
during VF acquisition handshake.

This patch fixes it by allowing VF to request the PF for soft
FLR during the load in such cases so that it can get cleanly
re-loaded.

Fixes: 2ea6f76aff40 ("qede: add core driver")

Signed-off-by: Manish Chopra <manishc@marvell.com>
Signed-off-by: Shahed Shaikh <shshaikh@marvell.com>
Signed-off-by: Rasesh Mody <rmody@marvell.com>
---
 drivers/net/qede/base/ecore_dev.c     |  2 +-
 drivers/net/qede/base/ecore_dev_api.h |  3 ++
 drivers/net/qede/base/ecore_iov_api.h |  1 +
 drivers/net/qede/base/ecore_sriov.c   | 33 ++++++++++++++++
 drivers/net/qede/base/ecore_vf.c      | 55 +++++++++++++++++++++++++--
 drivers/net/qede/base/ecore_vf.h      | 11 +++++-
 drivers/net/qede/base/ecore_vf_api.h  |  3 ++
 drivers/net/qede/base/ecore_vfpf_if.h | 41 ++++++++++++++++++++
 drivers/net/qede/base/mcp_public.h    |  2 +
 drivers/net/qede/qede_main.c          |  4 ++
 10 files changed, 149 insertions(+), 6 deletions(-)

Comments

Rasesh Mody Feb. 7, 2020, 6:22 a.m. UTC | #1
+Jerin

>From: dev <dev-bounces@dpdk.org> On Behalf Of Manish Chopra
>Sent: Thursday, February 06, 2020 12:02 PM
>
>On ungraceful termination of DPDK application, PMD VF driver fails to re-load
>due to PF seeing the VF in unexpected state during VF acquisition handshake.
>
>This patch fixes it by allowing VF to request the PF for soft FLR during the load
>in such cases so that it can get cleanly re-loaded.
>
>Fixes: 2ea6f76aff40 ("qede: add core driver")

Cc: stable@dpdk.org

>
>Signed-off-by: Manish Chopra <manishc@marvell.com>
>Signed-off-by: Shahed Shaikh <shshaikh@marvell.com>
>Signed-off-by: Rasesh Mody <rmody@marvell.com>
>---

Acked-by: Rasesh Mody <rmody@marvell.com>
Jerin Jacob Feb. 15, 2020, 1:33 p.m. UTC | #2
On Fri, Feb 7, 2020 at 11:52 AM Rasesh Mody <rmody@marvell.com> wrote:
>
> +Jerin
>
> >From: dev <dev-bounces@dpdk.org> On Behalf Of Manish Chopra
> >Sent: Thursday, February 06, 2020 12:02 PM
> >
> >On ungraceful termination of DPDK application, PMD VF driver fails to re-load
> >due to PF seeing the VF in unexpected state during VF acquisition handshake.
> >
> >This patch fixes it by allowing VF to request the PF for soft FLR during the load
> >in such cases so that it can get cleanly re-loaded.
> >
> >Fixes: 2ea6f76aff40 ("qede: add core driver")
>
> Cc: stable@dpdk.org
>
> >
> >Signed-off-by: Manish Chopra <manishc@marvell.com>
> >Signed-off-by: Shahed Shaikh <shshaikh@marvell.com>
> >Signed-off-by: Rasesh Mody <rmody@marvell.com>
> >---
>
> Acked-by: Rasesh Mody <rmody@marvell.com>


Fixed following check-git-log.sh warnings[1]
Series applied to dpdk-next-net-mrvl/master. Thanks.


[1]

check-git-log failed
Wrong headline format:
        net/qede: Do not stop vport if not started.
Wrong headline uppercase:
        net/qede: Fix VF re-load failure after ungraceful termination of DPDK
        net/qede: Do not stop vport if not started.
Headline too long:
        net/qede: Fix VF re-load failure after ungraceful termination of DPDK
Is it candidate for Cc: stable@dpdk.org backport?
        net/qede: Fix VF re-load failure after ungraceful termination of DPDK]


>

Patch
diff mbox series

diff --git a/drivers/net/qede/base/ecore_dev.c b/drivers/net/qede/base/ecore_dev.c
index f33b9910c..86ecfb269 100644
--- a/drivers/net/qede/base/ecore_dev.c
+++ b/drivers/net/qede/base/ecore_dev.c
@@ -5617,7 +5617,7 @@  ecore_hw_prepare_single(struct ecore_hwfn *p_hwfn, void OSAL_IOMEM *p_regview,
 	p_hwfn->db_phys_addr = db_phys_addr;
 
 	if (IS_VF(p_dev))
-		return ecore_vf_hw_prepare(p_hwfn);
+		return ecore_vf_hw_prepare(p_hwfn, p_params);
 
 	/* Validate that chip access is feasible */
 	if (REG_RD(p_hwfn, PXP_PF_ME_OPAQUE_ADDR) == 0xffffffff) {
diff --git a/drivers/net/qede/base/ecore_dev_api.h b/drivers/net/qede/base/ecore_dev_api.h
index 4d5cc1a0f..5ea8427a0 100644
--- a/drivers/net/qede/base/ecore_dev_api.h
+++ b/drivers/net/qede/base/ecore_dev_api.h
@@ -277,6 +277,9 @@  struct ecore_hw_prepare_params {
 
 	/* Indicates whether this PF serves a storage target */
 	bool b_is_target;
+
+	/* retry count for VF acquire on channel timeout */
+	u8 acquire_retry_cnt;
 };
 
 /**
diff --git a/drivers/net/qede/base/ecore_iov_api.h b/drivers/net/qede/base/ecore_iov_api.h
index c998dbf8d..545001812 100644
--- a/drivers/net/qede/base/ecore_iov_api.h
+++ b/drivers/net/qede/base/ecore_iov_api.h
@@ -51,6 +51,7 @@  enum ecore_iov_pf_to_vf_status {
 	PFVF_STATUS_NO_RESOURCE,
 	PFVF_STATUS_FORCED,
 	PFVF_STATUS_MALICIOUS,
+	PFVF_STATUS_ACQUIRED,
 };
 
 struct ecore_mcp_link_params;
diff --git a/drivers/net/qede/base/ecore_sriov.c b/drivers/net/qede/base/ecore_sriov.c
index deee04ac4..e60257e19 100644
--- a/drivers/net/qede/base/ecore_sriov.c
+++ b/drivers/net/qede/base/ecore_sriov.c
@@ -61,6 +61,39 @@  const char *qede_ecore_channel_tlvs_string[] = {
 	"CHANNEL_TLV_COALESCE_READ",
 	"CHANNEL_TLV_BULLETIN_UPDATE_MAC",
 	"CHANNEL_TLV_UPDATE_MTU",
+	"CHANNEL_TLV_RDMA_ACQUIRE",
+	"CHANNEL_TLV_RDMA_START",
+	"CHANNEL_TLV_RDMA_STOP",
+	"CHANNEL_TLV_RDMA_ADD_USER",
+	"CHANNEL_TLV_RDMA_REMOVE_USER",
+	"CHANNEL_TLV_RDMA_QUERY_COUNTERS",
+	"CHANNEL_TLV_RDMA_ALLOC_TID",
+	"CHANNEL_TLV_RDMA_REGISTER_TID",
+	"CHANNEL_TLV_RDMA_DEREGISTER_TID",
+	"CHANNEL_TLV_RDMA_FREE_TID",
+	"CHANNEL_TLV_RDMA_CREATE_CQ",
+	"CHANNEL_TLV_RDMA_RESIZE_CQ",
+	"CHANNEL_TLV_RDMA_DESTROY_CQ",
+	"CHANNEL_TLV_RDMA_CREATE_QP",
+	"CHANNEL_TLV_RDMA_MODIFY_QP",
+	"CHANNEL_TLV_RDMA_QUERY_QP",
+	"CHANNEL_TLV_RDMA_DESTROY_QP",
+	"CHANNEL_TLV_RDMA_CREATE_SRQ",
+	"CHANNEL_TLV_RDMA_MODIFY_SRQ",
+	"CHANNEL_TLV_RDMA_DESTROY_SRQ",
+	"CHANNEL_TLV_RDMA_QUERY_PORT",
+	"CHANNEL_TLV_RDMA_QUERY_DEVICE",
+	"CHANNEL_TLV_RDMA_IWARP_CONNECT",
+	"CHANNEL_TLV_RDMA_IWARP_ACCEPT",
+	"CHANNEL_TLV_RDMA_IWARP_CREATE_LISTEN",
+	"CHANNEL_TLV_RDMA_IWARP_DESTROY_LISTEN",
+	"CHANNEL_TLV_RDMA_IWARP_PAUSE_LISTEN",
+	"CHANNEL_TLV_RDMA_IWARP_REJECT",
+	"CHANNEL_TLV_RDMA_IWARP_SEND_RTR",
+	"CHANNEL_TLV_ESTABLISH_LL2_CONN",
+	"CHANNEL_TLV_TERMINATE_LL2_CONN",
+	"CHANNEL_TLV_ASYNC_EVENT",
+	"CHANNEL_TLV_SOFT_FLR",
 	"CHANNEL_TLV_MAX"
 };
 
diff --git a/drivers/net/qede/base/ecore_vf.c b/drivers/net/qede/base/ecore_vf.c
index c5c081426..db03bc494 100644
--- a/drivers/net/qede/base/ecore_vf.c
+++ b/drivers/net/qede/base/ecore_vf.c
@@ -226,7 +226,6 @@  enum _ecore_status_t ecore_vf_pf_release(struct ecore_hwfn *p_hwfn)
 	return _ecore_vf_pf_release(p_hwfn, true);
 }
 
-#define VF_ACQUIRE_THRESH 3
 static void ecore_vf_pf_acquire_reduce_resc(struct ecore_hwfn *p_hwfn,
 					    struct vf_pf_resc_request *p_req,
 					    struct pf_vf_resc *p_resp)
@@ -251,13 +250,47 @@  static void ecore_vf_pf_acquire_reduce_resc(struct ecore_hwfn *p_hwfn,
 	p_req->num_cids = p_resp->num_cids;
 }
 
-static enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn)
+static enum _ecore_status_t
+ecore_vf_pf_soft_flr_acquire(struct ecore_hwfn *p_hwfn)
+{
+	struct ecore_vf_iov *p_iov = p_hwfn->vf_iov_info;
+	struct pfvf_def_resp_tlv *resp;
+	struct vfpf_soft_flr_tlv *req;
+	enum _ecore_status_t rc;
+
+	req = ecore_vf_pf_prep(p_hwfn, CHANNEL_TLV_SOFT_FLR, sizeof(*req));
+
+	/* add list termination tlv */
+	ecore_add_tlv(&p_iov->offset,
+		      CHANNEL_TLV_LIST_END,
+		      sizeof(struct channel_list_end_tlv));
+
+	resp = &p_iov->pf2vf_reply->default_resp;
+	rc = ecore_send_msg2pf(p_hwfn, &resp->hdr.status, sizeof(*resp));
+
+	DP_VERBOSE(p_hwfn, ECORE_MSG_IOV, "rc=0x%x\n", rc);
+
+	/* to release the mutex as ecore_vf_pf_acquire() take the mutex */
+	ecore_vf_pf_req_end(p_hwfn, ECORE_AGAIN);
+
+	/* As of today, there is no mechanism in place for VF to know the FLR
+	 * status, so sufficiently (worst case time) wait for FLR to complete,
+	 * as mailbox request to MFW by the PF for initiating VF flr and PF
+	 * processing VF FLR could take time.
+	 */
+	OSAL_MSLEEP(3000);
+
+	return ecore_vf_pf_acquire(p_hwfn);
+}
+
+enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn)
 {
 	struct ecore_vf_iov *p_iov = p_hwfn->vf_iov_info;
 	struct pfvf_acquire_resp_tlv *resp = &p_iov->pf2vf_reply->acquire_resp;
 	struct pf_vf_pfdev_info *pfdev_info = &resp->pfdev_info;
 	struct ecore_vf_acquire_sw_info vf_sw_info;
 	struct ecore_dev *p_dev = p_hwfn->p_dev;
+	u8 retry_cnt = p_iov->acquire_retry_cnt;
 	struct vf_pf_resc_request *p_resc;
 	bool resources_acquired = false;
 	struct vfpf_acquire_tlv *req;
@@ -318,6 +351,14 @@  static enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn)
 		/* send acquire request */
 		rc = ecore_send_msg2pf(p_hwfn,
 				       &resp->hdr.status, sizeof(*resp));
+
+		if (retry_cnt && rc == ECORE_TIMEOUT) {
+			DP_VERBOSE(p_hwfn, ECORE_MSG_IOV,
+				   "VF retrying to acquire due to VPC timeout\n");
+			retry_cnt--;
+			continue;
+		}
+
 		if (rc != ECORE_SUCCESS)
 			goto exit;
 
@@ -343,7 +384,7 @@  static enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn)
 			resources_acquired = true;
 		} /* PF refuses to allocate our resources */
 		else if (resp->hdr.status == PFVF_STATUS_NO_RESOURCE &&
-			 attempts < VF_ACQUIRE_THRESH) {
+			 attempts < ECORE_VF_ACQUIRE_THRESH) {
 			ecore_vf_pf_acquire_reduce_resc(p_hwfn, p_resc,
 							&resp->resc);
 
@@ -391,6 +432,9 @@  static enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn)
 				  "PF rejected acquisition by VF\n");
 			rc = ECORE_INVAL;
 			goto exit;
+		} else if (resp->hdr.status == PFVF_STATUS_ACQUIRED) {
+			ecore_vf_pf_req_end(p_hwfn, ECORE_AGAIN);
+			return ecore_vf_pf_soft_flr_acquire(p_hwfn);
 		} else {
 			DP_ERR(p_hwfn,
 			       "PF returned err %d to VF acquisition request\n",
@@ -477,7 +521,9 @@  u32 ecore_vf_hw_bar_size(struct ecore_hwfn *p_hwfn,
 	return 0;
 }
 
-enum _ecore_status_t ecore_vf_hw_prepare(struct ecore_hwfn *p_hwfn)
+enum _ecore_status_t
+ecore_vf_hw_prepare(struct ecore_hwfn *p_hwfn,
+		    struct ecore_hw_prepare_params *p_params)
 {
 	struct ecore_hwfn *p_lead = ECORE_LEADING_HWFN(p_hwfn->p_dev);
 	struct ecore_vf_iov *p_iov;
@@ -583,6 +629,7 @@  enum _ecore_status_t ecore_vf_hw_prepare(struct ecore_hwfn *p_hwfn)
 #endif
 	OSAL_MUTEX_INIT(&p_iov->mutex);
 
+	p_iov->acquire_retry_cnt = p_params->acquire_retry_cnt;
 	p_hwfn->vf_iov_info = p_iov;
 
 	p_hwfn->hw_info.personality = ECORE_PCI_ETH;
diff --git a/drivers/net/qede/base/ecore_vf.h b/drivers/net/qede/base/ecore_vf.h
index a07f82ebd..f027eba3e 100644
--- a/drivers/net/qede/base/ecore_vf.h
+++ b/drivers/net/qede/base/ecore_vf.h
@@ -11,6 +11,7 @@ 
 #include "ecore_vf_api.h"
 #include "ecore_l2_api.h"
 #include "ecore_vfpf_if.h"
+#include "ecore_dev_api.h"
 
 /* Default number of CIDs [total of both Rx and Tx] to be requested
  * by default.
@@ -59,6 +60,9 @@  struct ecore_vf_iov {
 	 * bar or via the doorbell bar.
 	 */
 	bool b_doorbell_bar;
+
+	/* retry count for VF acquire on channel timeout */
+	u8 acquire_retry_cnt;
 };
 
 /**
@@ -72,6 +76,8 @@  struct ecore_vf_iov {
 enum _ecore_status_t ecore_vf_pf_get_coalesce(struct ecore_hwfn *p_hwfn,
 					      u16 *p_coal,
 					      struct ecore_queue_cid *p_cid);
+
+enum _ecore_status_t ecore_vf_pf_acquire(struct ecore_hwfn *p_hwfn);
 /**
  * @brief VF - Set Rx/Tx coalesce per VF's relative queue.
  *             Coalesce value '0' will omit the configuration.
@@ -92,10 +98,13 @@  enum _ecore_status_t ecore_vf_pf_set_coalesce(struct ecore_hwfn *p_hwfn,
  *	sends ACQUIRE message
  *
  * @param p_hwfn
+ * @param p_params
  *
  * @return enum _ecore_status_t
  */
-enum _ecore_status_t ecore_vf_hw_prepare(struct ecore_hwfn *p_hwfn);
+enum _ecore_status_t
+ecore_vf_hw_prepare(struct ecore_hwfn *p_hwfn,
+		    struct ecore_hw_prepare_params *p_params);
 
 /**
  * @brief VF - start the RX Queue by sending a message to the PF
diff --git a/drivers/net/qede/base/ecore_vf_api.h b/drivers/net/qede/base/ecore_vf_api.h
index 1a9fb3b1f..43951a9a3 100644
--- a/drivers/net/qede/base/ecore_vf_api.h
+++ b/drivers/net/qede/base/ecore_vf_api.h
@@ -11,6 +11,9 @@ 
 #include "ecore_mcp_api.h"
 
 #ifdef CONFIG_ECORE_SRIOV
+
+#define ECORE_VF_ACQUIRE_THRESH 3
+
 /**
  * @brief Read the VF bulletin and act on it if needed
  *
diff --git a/drivers/net/qede/base/ecore_vfpf_if.h b/drivers/net/qede/base/ecore_vfpf_if.h
index c7ecb01c2..f92dc428a 100644
--- a/drivers/net/qede/base/ecore_vfpf_if.h
+++ b/drivers/net/qede/base/ecore_vfpf_if.h
@@ -251,6 +251,13 @@  struct vfpf_qid_tlv {
 	u8			padding[3];
 };
 
+/* Soft FLR req */
+struct vfpf_soft_flr_tlv {
+	struct vfpf_first_tlv first_tlv;
+	u32 reserved1;
+	u32 reserved2;
+};
+
 /* Setup Queue */
 struct vfpf_start_rxq_tlv {
 	struct vfpf_first_tlv	first_tlv;
@@ -557,6 +564,7 @@  union vfpf_tlvs {
 	struct vfpf_read_coal_req_tlv		read_coal_req;
 	struct vfpf_bulletin_update_mac_tlv	bulletin_update_mac;
 	struct vfpf_update_mtu_tlv		update_mtu;
+	struct vfpf_soft_flr_tlv		soft_flr;
 	struct tlv_buffer_size			tlv_buf_size;
 };
 
@@ -689,6 +697,39 @@  enum {
 	CHANNEL_TLV_COALESCE_READ,
 	CHANNEL_TLV_BULLETIN_UPDATE_MAC,
 	CHANNEL_TLV_UPDATE_MTU,
+	CHANNEL_TLV_RDMA_ACQUIRE,
+	CHANNEL_TLV_RDMA_START,
+	CHANNEL_TLV_RDMA_STOP,
+	CHANNEL_TLV_RDMA_ADD_USER,
+	CHANNEL_TLV_RDMA_REMOVE_USER,
+	CHANNEL_TLV_RDMA_QUERY_COUNTERS,
+	CHANNEL_TLV_RDMA_ALLOC_TID,
+	CHANNEL_TLV_RDMA_REGISTER_TID,
+	CHANNEL_TLV_RDMA_DEREGISTER_TID,
+	CHANNEL_TLV_RDMA_FREE_TID,
+	CHANNEL_TLV_RDMA_CREATE_CQ,
+	CHANNEL_TLV_RDMA_RESIZE_CQ,
+	CHANNEL_TLV_RDMA_DESTROY_CQ,
+	CHANNEL_TLV_RDMA_CREATE_QP,
+	CHANNEL_TLV_RDMA_MODIFY_QP,
+	CHANNEL_TLV_RDMA_QUERY_QP,
+	CHANNEL_TLV_RDMA_DESTROY_QP,
+	CHANNEL_TLV_RDMA_QUERY_PORT,
+	CHANNEL_TLV_RDMA_QUERY_DEVICE,
+	CHANNEL_TLV_RDMA_IWARP_CONNECT,
+	CHANNEL_TLV_RDMA_IWARP_ACCEPT,
+	CHANNEL_TLV_RDMA_IWARP_CREATE_LISTEN,
+	CHANNEL_TLV_RDMA_IWARP_DESTROY_LISTEN,
+	CHANNEL_TLV_RDMA_IWARP_PAUSE_LISTEN,
+	CHANNEL_TLV_RDMA_IWARP_REJECT,
+	CHANNEL_TLV_RDMA_IWARP_SEND_RTR,
+	CHANNEL_TLV_ESTABLISH_LL2_CONN,
+	CHANNEL_TLV_TERMINATE_LL2_CONN,
+	CHANNEL_TLV_ASYNC_EVENT,
+	CHANNEL_TLV_RDMA_CREATE_SRQ,
+	CHANNEL_TLV_RDMA_MODIFY_SRQ,
+	CHANNEL_TLV_RDMA_DESTROY_SRQ,
+	CHANNEL_TLV_SOFT_FLR,
 	CHANNEL_TLV_MAX,
 
 	/* Required for iterating over vport-update tlvs.
diff --git a/drivers/net/qede/base/mcp_public.h b/drivers/net/qede/base/mcp_public.h
index 98b9723dd..6667c2d7a 100644
--- a/drivers/net/qede/base/mcp_public.h
+++ b/drivers/net/qede/base/mcp_public.h
@@ -1290,6 +1290,7 @@  struct public_drv_mb {
 /*deprecated don't use*/
 #define DRV_MSG_CODE_INITIATE_FLR_DEPRECATED    0x02000000
 #define DRV_MSG_CODE_INITIATE_PF_FLR            0x02010000
+#define DRV_MSG_CODE_INITIATE_VF_FLR		0x02020000
 #define DRV_MSG_CODE_VF_DISABLED_DONE           0xc0000000
 #define DRV_MSG_CODE_CFG_VF_MSIX                0xc0010000
 #define DRV_MSG_CODE_CFG_PF_VFS_MSIX            0xc0020000
@@ -1749,6 +1750,7 @@  struct public_drv_mb {
 #define FW_MSG_CODE_NIG_DRAIN_DONE              0x30000000
 #define FW_MSG_CODE_VF_DISABLED_DONE            0xb0000000
 #define FW_MSG_CODE_DRV_CFG_VF_MSIX_DONE        0xb0010000
+#define FW_MSG_CODE_INITIATE_VF_FLR_OK		0xb0030000
 #define FW_MSG_CODE_ERR_RESOURCE_TEMPORARY_UNAVAILABLE	0x008b0000
 #define FW_MSG_CODE_ERR_RESOURCE_ALREADY_ALLOCATED	0x008c0000
 #define FW_MSG_CODE_ERR_RESOURCE_NOT_ALLOCATED		0x008d0000
diff --git a/drivers/net/qede/qede_main.c b/drivers/net/qede/qede_main.c
index 4eb79d0fb..8580cbcd7 100644
--- a/drivers/net/qede/qede_main.c
+++ b/drivers/net/qede/qede_main.c
@@ -56,6 +56,10 @@  qed_probe(struct ecore_dev *edev, struct rte_pci_device *pci_dev,
 	qed_init_pci(edev, pci_dev);
 
 	memset(&hw_prepare_params, 0, sizeof(hw_prepare_params));
+
+	if (is_vf)
+		hw_prepare_params.acquire_retry_cnt = ECORE_VF_ACQUIRE_THRESH;
+
 	hw_prepare_params.personality = ECORE_PCI_ETH;
 	hw_prepare_params.drv_resc_alloc = false;
 	hw_prepare_params.chk_reg_fifo = false;