ice: Add support for VF reset events

Post VF initialization, there are a couple of different ways in which a
VF reset can be triggered. One is when the underlying PF itself goes
through a reset and other is via a VFLR interrupt. ice_reset_vf introduced
in this patch handles both these cases.

Also introduced in this patch is a helper function ice_aq_send_msg_to_vf
to send messages to VF over the mailbox queue. The PF uses this to send
reset notifications to VFs.

Signed-off-by: Anirudh Venkataramanan <[email protected]>
Tested-by: Andrew Bowers <[email protected]>
Signed-off-by: Jeff Kirsher <[email protected]>

Author: refactorman

drivers/net/ethernet/intel/ice/Makefile

@@ -16,4 +16,4 @@ ice-y := ice_main.o	\
 	 ice_lib.o	\
 	 ice_txrx.o	\
 	 ice_ethtool.o
-ice-$(CONFIG_PCI_IOV) += ice_virtchnl_pf.o
+ice-$(CONFIG_PCI_IOV) += ice_virtchnl_pf.o ice_sriov.o

drivers/net/ethernet/intel/ice/ice.h

@@ -37,6 +37,7 @@
 #include "ice_common.h"
 #include "ice_sched.h"
 #include "ice_virtchnl_pf.h"
+#include "ice_sriov.h"
 
 extern const char ice_drv_ver[];
 #define ICE_BAR0		0
@@ -155,6 +156,7 @@ enum ice_state {
 	__ICE_ADMINQ_EVENT_PENDING,
 	__ICE_MAILBOXQ_EVENT_PENDING,
 	__ICE_MDD_EVENT_PENDING,
+	__ICE_VFLR_EVENT_PENDING,
 	__ICE_FLTR_OVERFLOW_PROMISC,
 	__ICE_VF_DIS,
 	__ICE_CFG_BUSY,

drivers/net/ethernet/intel/ice/ice_adminq_cmd.h

@@ -1077,6 +1077,19 @@ struct ice_aqc_nvm {
 	__le32 addr_low;
 };
 
+/**
+ * Send to PF command (indirect 0x0801) id is only used by PF
+ *
+ * Send to VF command (indirect 0x0802) id is only used by PF
+ *
+ */
+struct ice_aqc_pf_vf_msg {
+	__le32 id;
+	u32 reserved;
+	__le32 addr_high;
+	__le32 addr_low;
+};
+
 /* Get/Set RSS key (indirect 0x0B04/0x0B02) */
 struct ice_aqc_get_set_rss_key {
 #define ICE_AQC_GSET_RSS_KEY_VSI_VALID	BIT(15)
@@ -1334,6 +1347,7 @@ struct ice_aq_desc {
 		struct ice_aqc_query_txsched_res query_sched_res;
 		struct ice_aqc_add_move_delete_elem add_move_delete_elem;
 		struct ice_aqc_nvm nvm;
+		struct ice_aqc_pf_vf_msg virt;
 		struct ice_aqc_get_set_rss_lut get_set_rss_lut;
 		struct ice_aqc_get_set_rss_key get_set_rss_key;
 		struct ice_aqc_add_txqs add_txqs;
@@ -1431,6 +1445,9 @@ enum ice_adminq_opc {
 	/* NVM commands */
 	ice_aqc_opc_nvm_read				= 0x0701,
 
+	/* PF/VF mailbox commands */
+	ice_mbx_opc_send_msg_to_vf			= 0x0802,
+
 	/* RSS commands */
 	ice_aqc_opc_set_rss_key				= 0x0B02,
 	ice_aqc_opc_set_rss_lut				= 0x0B03,

drivers/net/ethernet/intel/ice/ice_hw_autogen.h

@@ -136,6 +136,7 @@
 #define PFINT_OICR_PCI_EXCEPTION_M		BIT(21)
 #define PFINT_OICR_HMC_ERR_M			BIT(26)
 #define PFINT_OICR_PE_CRITERR_M			BIT(28)
+#define PFINT_OICR_VFLR_M			BIT(29)
 #define PFINT_OICR_CTL				0x0016CA80
 #define PFINT_OICR_CTL_MSIX_INDX_M		ICE_M(0x7FF, 0)
 #define PFINT_OICR_CTL_ITR_INDX_S		11

drivers/net/ethernet/intel/ice/ice_main.c

@@ -342,6 +342,10 @@ ice_prepare_for_reset(struct ice_pf *pf)
 {
 	struct ice_hw *hw = &pf->hw;
 
+	/* Notify VFs of impending reset */
+	if (ice_check_sq_alive(hw, &hw->mailboxq))
+		ice_vc_notify_reset(pf);
+
 	/* disable the VSIs and their queues that are not already DOWN */
 	ice_pf_dis_all_vsi(pf);
 
@@ -1064,6 +1068,7 @@ static void ice_service_task(struct work_struct *work)
 	ice_check_for_hang_subtask(pf);
 	ice_sync_fltr_subtask(pf);
 	ice_handle_mdd_event(pf);
+	ice_process_vflr_event(pf);
 	ice_watchdog_subtask(pf);
 	ice_clean_adminq_subtask(pf);
 	ice_clean_mailboxq_subtask(pf);
@@ -1077,6 +1082,7 @@ static void ice_service_task(struct work_struct *work)
 	 */
 	if (time_after(jiffies, (start_time + pf->serv_tmr_period)) ||
 	    test_bit(__ICE_MDD_EVENT_PENDING, pf->state) ||
+	    test_bit(__ICE_VFLR_EVENT_PENDING, pf->state) ||
 	    test_bit(__ICE_MAILBOXQ_EVENT_PENDING, pf->state) ||
 	    test_bit(__ICE_ADMINQ_EVENT_PENDING, pf->state))
 		mod_timer(&pf->serv_tmr, jiffies);
@@ -1229,6 +1235,7 @@ static void ice_ena_misc_vector(struct ice_pf *pf)
 	       PFINT_OICR_MAL_DETECT_M |
 	       PFINT_OICR_GRST_M |
 	       PFINT_OICR_PCI_EXCEPTION_M |
+	       PFINT_OICR_VFLR_M |
 	       PFINT_OICR_HMC_ERR_M |
 	       PFINT_OICR_PE_CRITERR_M);
 
@@ -1261,6 +1268,10 @@ static irqreturn_t ice_misc_intr(int __always_unused irq, void *data)
 		ena_mask &= ~PFINT_OICR_MAL_DETECT_M;
 		set_bit(__ICE_MDD_EVENT_PENDING, pf->state);
 	}
+	if (oicr & PFINT_OICR_VFLR_M) {
+		ena_mask &= ~PFINT_OICR_VFLR_M;
+		set_bit(__ICE_VFLR_EVENT_PENDING, pf->state);
+	}
 
 	if (oicr & PFINT_OICR_GRST_M) {
 		u32 reset;
@@ -3224,6 +3235,10 @@ static int ice_vsi_rebuild_all(struct ice_pf *pf)
 		if (!pf->vsi[i])
 			continue;
 
+		/* VF VSI rebuild isn't supported yet */
+		if (pf->vsi[i]->type == ICE_VSI_VF)
+			continue;
+
 		err = ice_vsi_rebuild(pf->vsi[i]);
 		if (err) {
 			dev_err(&pf->pdev->dev,

drivers/net/ethernet/intel/ice/ice_sriov.c

@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2018, Intel Corporation. */
+
+#include "ice_common.h"
+#include "ice_adminq_cmd.h"
+#include "ice_sriov.h"
+
+/**
+ * ice_aq_send_msg_to_vf
+ * @hw: pointer to the hardware structure
+ * @vfid: VF ID to send msg
+ * @v_opcode: opcodes for VF-PF communication
+ * @v_retval: return error code
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ * @cd: pointer to command details
+ *
+ * Send message to VF driver (0x0802) using mailbox
+ * queue and asynchronously sending message via
+ * ice_sq_send_cmd() function
+ */
+enum ice_status
+ice_aq_send_msg_to_vf(struct ice_hw *hw, u16 vfid, u32 v_opcode, u32 v_retval,
+		      u8 *msg, u16 msglen, struct ice_sq_cd *cd)
+{
+	struct ice_aqc_pf_vf_msg *cmd;
+	struct ice_aq_desc desc;
+
+	ice_fill_dflt_direct_cmd_desc(&desc, ice_mbx_opc_send_msg_to_vf);
+
+	cmd = &desc.params.virt;
+	cmd->id = cpu_to_le32(vfid);
+
+	desc.cookie_high = cpu_to_le32(v_opcode);
+	desc.cookie_low = cpu_to_le32(v_retval);
+
+	if (msglen)
+		desc.flags |= cpu_to_le16(ICE_AQ_FLAG_RD);
+
+	return ice_sq_send_cmd(hw, &hw->mailboxq, &desc, msg, msglen, cd);
+}

drivers/net/ethernet/intel/ice/ice_sriov.h

@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright (c) 2018, Intel Corporation. */
+
+#ifndef _ICE_SRIOV_H_
+#define _ICE_SRIOV_H_
+
+#include "ice_common.h"
+
+#ifdef CONFIG_PCI_IOV
+enum ice_status
+ice_aq_send_msg_to_vf(struct ice_hw *hw, u16 vfid, u32 v_opcode, u32 v_retval,
+		      u8 *msg, u16 msglen, struct ice_sq_cd *cd);
+
+#else /* CONFIG_PCI_IOV */
+static inline enum ice_status
+ice_aq_send_msg_to_vf(struct ice_hw __always_unused *hw,
+		      u16 __always_unused vfid, u32 __always_unused v_opcode,
+		      u32 __always_unused v_retval, u8 __always_unused *msg,
+		      u16 __always_unused msglen,
+		      struct ice_sq_cd __always_unused *cd)
+{
+	return 0;
+}
+#endif /* CONFIG_PCI_IOV */
+#endif /* _ICE_SRIOV_H_ */

drivers/net/ethernet/intel/ice/ice_status.h

@@ -6,6 +6,9 @@
 
 /* Error Codes */
 enum ice_status {
+	ICE_SUCCESS				= 0,
+
+	/* Generic codes : Range -1..-49 */
 	ICE_ERR_PARAM				= -1,
 	ICE_ERR_NOT_IMPL			= -2,
 	ICE_ERR_NOT_READY			= -3,

drivers/net/ethernet/intel/ice/ice_virtchnl_pf.c

@@ -4,6 +4,36 @@
 #include "ice.h"
 #include "ice_lib.h"
 
+/**
+ * ice_vc_vf_broadcast - Broadcast a message to all VFs on PF
+ * @pf: pointer to the PF structure
+ * @v_opcode: operation code
+ * @v_retval: return value
+ * @msg: pointer to the msg buffer
+ * @msglen: msg length
+ */
+static void
+ice_vc_vf_broadcast(struct ice_pf *pf, enum virtchnl_ops v_opcode,
+		    enum ice_status v_retval, u8 *msg, u16 msglen)
+{
+	struct ice_hw *hw = &pf->hw;
+	struct ice_vf *vf = pf->vf;
+	int i;
+
+	for (i = 0; i < pf->num_alloc_vfs; i++, vf++) {
+		/* Not all vfs are enabled so skip the ones that are not */
+		if (!test_bit(ICE_VF_STATE_INIT, vf->vf_states) &&
+		    !test_bit(ICE_VF_STATE_ACTIVE, vf->vf_states))
+			continue;
+
+		/* Ignore return value on purpose - a given VF may fail, but
+		 * we need to keep going and send to all of them
+		 */
+		ice_aq_send_msg_to_vf(hw, vf->vf_id, v_opcode, v_retval, msg,
+				      msglen, NULL);
+	}
+}
+
 /**
  * ice_get_vf_vector - get VF interrupt vector register offset
  * @vf_msix: number of MSIx vector per VF on a PF
@@ -693,6 +723,97 @@ bool ice_reset_all_vfs(struct ice_pf *pf, bool is_vflr)
 	return true;
 }
 
+/**
+ * ice_reset_vf - Reset a particular VF
+ * @vf: pointer to the VF structure
+ * @is_vflr: true if VFLR was issued, false if not
+ *
+ * Returns true if the VF is reset, false otherwise.
+ */
+static bool ice_reset_vf(struct ice_vf *vf, bool is_vflr)
+{
+	struct ice_pf *pf = vf->pf;
+	struct ice_hw *hw = &pf->hw;
+	bool rsd = false;
+	u32 reg;
+	int i;
+
+	/* If the VFs have been disabled, this means something else is
+	 * resetting the VF, so we shouldn't continue.
+	 */
+	if (test_and_set_bit(__ICE_VF_DIS, pf->state))
+		return false;
+
+	ice_trigger_vf_reset(vf, is_vflr);
+
+	if (test_bit(ICE_VF_STATE_ENA, vf->vf_states)) {
+		ice_vsi_stop_tx_rings(pf->vsi[vf->lan_vsi_idx], ICE_VF_RESET,
+				      vf->vf_id);
+		ice_vsi_stop_rx_rings(pf->vsi[vf->lan_vsi_idx]);
+		clear_bit(ICE_VF_STATE_ENA, vf->vf_states);
+	} else {
+		/* Call Disable LAN Tx queue AQ call even when queues are not
+		 * enabled. This is needed for successful completiom of VFR
+		 */
+		ice_dis_vsi_txq(pf->vsi[vf->lan_vsi_idx]->port_info, 0,
+				NULL, NULL, ICE_VF_RESET, vf->vf_id, NULL);
+	}
+
+	/* poll VPGEN_VFRSTAT reg to make sure
+	 * that reset is complete
+	 */
+	for (i = 0; i < 10; i++) {
+		/* VF reset requires driver to first reset the VF and then
+		 * poll the status register to make sure that the reset
+		 * completed successfully.
+		 */
+		usleep_range(10000, 20000);
+		reg = rd32(hw, VPGEN_VFRSTAT(vf->vf_id));
+		if (reg & VPGEN_VFRSTAT_VFRD_M) {
+			rsd = true;
+			break;
+		}
+	}
+
+	/* Display a warning if VF didn't manage to reset in time, but need to
+	 * continue on with the operation.
+	 */
+	if (!rsd)
+		dev_warn(&pf->pdev->dev, "VF reset check timeout on VF %d\n",
+			 vf->vf_id);
+
+	usleep_range(10000, 20000);
+
+	/* free VF resources to begin resetting the VSI state */
+	ice_free_vf_res(vf);
+
+	ice_cleanup_and_realloc_vf(vf);
+
+	ice_flush(hw);
+	clear_bit(__ICE_VF_DIS, pf->state);
+
+	return true;
+}
+
+/**
+ * ice_vc_notify_reset - Send pending reset message to all VFs
+ * @pf: pointer to the PF structure
+ *
+ * indicate a pending reset to all VFs on a given PF
+ */
+void ice_vc_notify_reset(struct ice_pf *pf)
+{
+	struct virtchnl_pf_event pfe;
+
+	if (!pf->num_alloc_vfs)
+		return;
+
+	pfe.event = VIRTCHNL_EVENT_RESET_IMPENDING;
+	pfe.severity = PF_EVENT_SEVERITY_CERTAIN_DOOM;
+	ice_vc_vf_broadcast(pf, VIRTCHNL_OP_EVENT, ICE_SUCCESS,
+			    (u8 *)&pfe, sizeof(struct virtchnl_pf_event));
+}
+
 /**
  * ice_alloc_vfs - Allocate and set up VFs resources
  * @pf: pointer to the PF structure
@@ -845,3 +966,45 @@ int ice_sriov_configure(struct pci_dev *pdev, int num_vfs)
 
 	return 0;
 }
+
+/**
+ * ice_process_vflr_event - Free VF resources via IRQ calls
+ * @pf: pointer to the PF structure
+ *
+ * called from the VLFR IRQ handler to
+ * free up VF resources and state variables
+ */
+void ice_process_vflr_event(struct ice_pf *pf)
+{
+	struct ice_hw *hw = &pf->hw;
+	int vf_id;
+	u32 reg;
+
+	if (!test_bit(__ICE_VFLR_EVENT_PENDING, pf->state) ||
+	    !pf->num_alloc_vfs)
+		return;
+
+	/* Re-enable the VFLR interrupt cause here, before looking for which
+	 * VF got reset. Otherwise, if another VF gets a reset while the
+	 * first one is being processed, that interrupt will be lost, and
+	 * that VF will be stuck in reset forever.
+	 */
+	reg = rd32(hw, PFINT_OICR_ENA);
+	reg |= PFINT_OICR_VFLR_M;
+	wr32(hw, PFINT_OICR_ENA, reg);
+	ice_flush(hw);
+
+	clear_bit(__ICE_VFLR_EVENT_PENDING, pf->state);
+	for (vf_id = 0; vf_id < pf->num_alloc_vfs; vf_id++) {
+		struct ice_vf *vf = &pf->vf[vf_id];
+		u32 reg_idx, bit_idx;
+
+		reg_idx = (hw->func_caps.vf_base_id + vf_id) / 32;
+		bit_idx = (hw->func_caps.vf_base_id + vf_id) % 32;
+		/* read GLGEN_VFLRSTAT register to find out the flr VFs */
+		reg = rd32(hw, GLGEN_VFLRSTAT(reg_idx));
+		if (reg & BIT(bit_idx))
+			/* GLGEN_VFLRSTAT bit will be cleared in ice_reset_vf */
+			ice_reset_vf(vf, true);
+	}
+}

drivers/net/ethernet/intel/ice/ice_virtchnl_pf.h

@@ -51,11 +51,15 @@ struct ice_vf {
 };
 
 #ifdef CONFIG_PCI_IOV
+void ice_process_vflr_event(struct ice_pf *pf);
 int ice_sriov_configure(struct pci_dev *pdev, int num_vfs);
 void ice_free_vfs(struct ice_pf *pf);
+void ice_vc_notify_reset(struct ice_pf *pf);
 bool ice_reset_all_vfs(struct ice_pf *pf, bool is_vflr);
 #else /* CONFIG_PCI_IOV */
+#define ice_process_vflr_event(pf) do {} while (0)
 #define ice_free_vfs(pf) do {} while (0)
+#define ice_vc_notify_reset(pf) do {} while (0)
 
 static inline bool
 ice_reset_all_vfs(struct ice_pf __always_unused *pf,