| // SPDX-License-Identifier: GPL-2.0 |
| /* Copyright (c) 2018, Intel Corporation. */ |
| |
| #include "ice_common.h" |
| #include "ice_vf_mbx.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 |
| */ |
| int |
| 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); |
| } |
| |
| static const u32 ice_legacy_aq_to_vc_speed[] = { |
| VIRTCHNL_LINK_SPEED_100MB, /* BIT(0) */ |
| VIRTCHNL_LINK_SPEED_100MB, |
| VIRTCHNL_LINK_SPEED_1GB, |
| VIRTCHNL_LINK_SPEED_1GB, |
| VIRTCHNL_LINK_SPEED_1GB, |
| VIRTCHNL_LINK_SPEED_10GB, |
| VIRTCHNL_LINK_SPEED_20GB, |
| VIRTCHNL_LINK_SPEED_25GB, |
| VIRTCHNL_LINK_SPEED_40GB, |
| VIRTCHNL_LINK_SPEED_40GB, |
| VIRTCHNL_LINK_SPEED_40GB, |
| }; |
| |
| /** |
| * ice_conv_link_speed_to_virtchnl |
| * @adv_link_support: determines the format of the returned link speed |
| * @link_speed: variable containing the link_speed to be converted |
| * |
| * Convert link speed supported by HW to link speed supported by virtchnl. |
| * If adv_link_support is true, then return link speed in Mbps. Else return |
| * link speed as a VIRTCHNL_LINK_SPEED_* casted to a u32. Note that the caller |
| * needs to cast back to an enum virtchnl_link_speed in the case where |
| * adv_link_support is false, but when adv_link_support is true the caller can |
| * expect the speed in Mbps. |
| */ |
| u32 ice_conv_link_speed_to_virtchnl(bool adv_link_support, u16 link_speed) |
| { |
| /* convert a BIT() value into an array index */ |
| u32 index = fls(link_speed) - 1; |
| |
| if (adv_link_support) |
| return ice_get_link_speed(index); |
| else if (index < ARRAY_SIZE(ice_legacy_aq_to_vc_speed)) |
| /* Virtchnl speeds are not defined for every speed supported in |
| * the hardware. To maintain compatibility with older AVF |
| * drivers, while reporting the speed the new speed values are |
| * resolved to the closest known virtchnl speeds |
| */ |
| return ice_legacy_aq_to_vc_speed[index]; |
| |
| return VIRTCHNL_LINK_SPEED_UNKNOWN; |
| } |
| |
| /* The mailbox overflow detection algorithm helps to check if there |
| * is a possibility of a malicious VF transmitting too many MBX messages to the |
| * PF. |
| * 1. The mailbox snapshot structure, ice_mbx_snapshot, is initialized during |
| * driver initialization in ice_init_hw() using ice_mbx_init_snapshot(). |
| * The struct ice_mbx_snapshot helps to track and traverse a static window of |
| * messages within the mailbox queue while looking for a malicious VF. |
| * |
| * 2. When the caller starts processing its mailbox queue in response to an |
| * interrupt, the structure ice_mbx_snapshot is expected to be cleared before |
| * the algorithm can be run for the first time for that interrupt. This |
| * requires calling ice_mbx_reset_snapshot() as well as calling |
| * ice_mbx_reset_vf_info() for each VF tracking structure. |
| * |
| * 3. For every message read by the caller from the MBX Queue, the caller must |
| * call the detection algorithm's entry function ice_mbx_vf_state_handler(). |
| * Before every call to ice_mbx_vf_state_handler() the struct ice_mbx_data is |
| * filled as it is required to be passed to the algorithm. |
| * |
| * 4. Every time a message is read from the MBX queue, a tracking structure |
| * for the VF must be passed to the state handler. The boolean output |
| * report_malvf from ice_mbx_vf_state_handler() serves as an indicator to the |
| * caller whether it must report this VF as malicious or not. |
| * |
| * 5. When a VF is identified to be malicious, the caller can send a message |
| * to the system administrator. |
| * |
| * 6. The PF is responsible for maintaining the struct ice_mbx_vf_info |
| * structure for each VF. The PF should clear the VF tracking structure if the |
| * VF is reset. When a VF is shut down and brought back up, we will then |
| * assume that the new VF is not malicious and may report it again if we |
| * detect it again. |
| * |
| * 7. The function ice_mbx_reset_snapshot() is called to reset the information |
| * in ice_mbx_snapshot for every new mailbox interrupt handled. |
| */ |
| #define ICE_RQ_DATA_MASK(rq_data) ((rq_data) & PF_MBX_ARQH_ARQH_M) |
| /* Using the highest value for an unsigned 16-bit value 0xFFFF to indicate that |
| * the max messages check must be ignored in the algorithm |
| */ |
| #define ICE_IGNORE_MAX_MSG_CNT 0xFFFF |
| |
| /** |
| * ice_mbx_reset_snapshot - Reset mailbox snapshot structure |
| * @snap: pointer to the mailbox snapshot |
| */ |
| static void ice_mbx_reset_snapshot(struct ice_mbx_snapshot *snap) |
| { |
| struct ice_mbx_vf_info *vf_info; |
| |
| /* Clear mbx_buf in the mailbox snaphot structure and setting the |
| * mailbox snapshot state to a new capture. |
| */ |
| memset(&snap->mbx_buf, 0, sizeof(snap->mbx_buf)); |
| snap->mbx_buf.state = ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT; |
| |
| /* Reset message counts for all VFs to zero */ |
| list_for_each_entry(vf_info, &snap->mbx_vf, list_entry) |
| vf_info->msg_count = 0; |
| } |
| |
| /** |
| * ice_mbx_traverse - Pass through mailbox snapshot |
| * @hw: pointer to the HW struct |
| * @new_state: new algorithm state |
| * |
| * Traversing the mailbox static snapshot without checking |
| * for malicious VFs. |
| */ |
| static void |
| ice_mbx_traverse(struct ice_hw *hw, |
| enum ice_mbx_snapshot_state *new_state) |
| { |
| struct ice_mbx_snap_buffer_data *snap_buf; |
| u32 num_iterations; |
| |
| snap_buf = &hw->mbx_snapshot.mbx_buf; |
| |
| /* As mailbox buffer is circular, applying a mask |
| * on the incremented iteration count. |
| */ |
| num_iterations = ICE_RQ_DATA_MASK(++snap_buf->num_iterations); |
| |
| /* Checking either of the below conditions to exit snapshot traversal: |
| * Condition-1: If the number of iterations in the mailbox is equal to |
| * the mailbox head which would indicate that we have reached the end |
| * of the static snapshot. |
| * Condition-2: If the maximum messages serviced in the mailbox for a |
| * given interrupt is the highest possible value then there is no need |
| * to check if the number of messages processed is equal to it. If not |
| * check if the number of messages processed is greater than or equal |
| * to the maximum number of mailbox entries serviced in current work item. |
| */ |
| if (num_iterations == snap_buf->head || |
| (snap_buf->max_num_msgs_mbx < ICE_IGNORE_MAX_MSG_CNT && |
| ++snap_buf->num_msg_proc >= snap_buf->max_num_msgs_mbx)) |
| *new_state = ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT; |
| } |
| |
| /** |
| * ice_mbx_detect_malvf - Detect malicious VF in snapshot |
| * @hw: pointer to the HW struct |
| * @vf_info: mailbox tracking structure for a VF |
| * @new_state: new algorithm state |
| * @is_malvf: boolean output to indicate if VF is malicious |
| * |
| * This function tracks the number of asynchronous messages |
| * sent per VF and marks the VF as malicious if it exceeds |
| * the permissible number of messages to send. |
| */ |
| static int |
| ice_mbx_detect_malvf(struct ice_hw *hw, struct ice_mbx_vf_info *vf_info, |
| enum ice_mbx_snapshot_state *new_state, |
| bool *is_malvf) |
| { |
| /* increment the message count for this VF */ |
| vf_info->msg_count++; |
| |
| if (vf_info->msg_count >= ICE_ASYNC_VF_MSG_THRESHOLD) |
| *is_malvf = true; |
| |
| /* continue to iterate through the mailbox snapshot */ |
| ice_mbx_traverse(hw, new_state); |
| |
| return 0; |
| } |
| |
| /** |
| * ice_mbx_vf_state_handler - Handle states of the overflow algorithm |
| * @hw: pointer to the HW struct |
| * @mbx_data: pointer to structure containing mailbox data |
| * @vf_info: mailbox tracking structure for the VF in question |
| * @report_malvf: boolean output to indicate whether VF should be reported |
| * |
| * The function serves as an entry point for the malicious VF |
| * detection algorithm by handling the different states and state |
| * transitions of the algorithm: |
| * New snapshot: This state is entered when creating a new static |
| * snapshot. The data from any previous mailbox snapshot is |
| * cleared and a new capture of the mailbox head and tail is |
| * logged. This will be the new static snapshot to detect |
| * asynchronous messages sent by VFs. On capturing the snapshot |
| * and depending on whether the number of pending messages in that |
| * snapshot exceed the watermark value, the state machine enters |
| * traverse or detect states. |
| * Traverse: If pending message count is below watermark then iterate |
| * through the snapshot without any action on VF. |
| * Detect: If pending message count exceeds watermark traverse |
| * the static snapshot and look for a malicious VF. |
| */ |
| int |
| ice_mbx_vf_state_handler(struct ice_hw *hw, struct ice_mbx_data *mbx_data, |
| struct ice_mbx_vf_info *vf_info, bool *report_malvf) |
| { |
| struct ice_mbx_snapshot *snap = &hw->mbx_snapshot; |
| struct ice_mbx_snap_buffer_data *snap_buf; |
| struct ice_ctl_q_info *cq = &hw->mailboxq; |
| enum ice_mbx_snapshot_state new_state; |
| bool is_malvf = false; |
| int status = 0; |
| |
| if (!report_malvf || !mbx_data || !vf_info) |
| return -EINVAL; |
| |
| *report_malvf = false; |
| |
| /* When entering the mailbox state machine assume that the VF |
| * is not malicious until detected. |
| */ |
| /* Checking if max messages allowed to be processed while servicing current |
| * interrupt is not less than the defined AVF message threshold. |
| */ |
| if (mbx_data->max_num_msgs_mbx <= ICE_ASYNC_VF_MSG_THRESHOLD) |
| return -EINVAL; |
| |
| /* The watermark value should not be lesser than the threshold limit |
| * set for the number of asynchronous messages a VF can send to mailbox |
| * nor should it be greater than the maximum number of messages in the |
| * mailbox serviced in current interrupt. |
| */ |
| if (mbx_data->async_watermark_val < ICE_ASYNC_VF_MSG_THRESHOLD || |
| mbx_data->async_watermark_val > mbx_data->max_num_msgs_mbx) |
| return -EINVAL; |
| |
| new_state = ICE_MAL_VF_DETECT_STATE_INVALID; |
| snap_buf = &snap->mbx_buf; |
| |
| switch (snap_buf->state) { |
| case ICE_MAL_VF_DETECT_STATE_NEW_SNAPSHOT: |
| /* Clear any previously held data in mailbox snapshot structure. */ |
| ice_mbx_reset_snapshot(snap); |
| |
| /* Collect the pending ARQ count, number of messages processed and |
| * the maximum number of messages allowed to be processed from the |
| * Mailbox for current interrupt. |
| */ |
| snap_buf->num_pending_arq = mbx_data->num_pending_arq; |
| snap_buf->num_msg_proc = mbx_data->num_msg_proc; |
| snap_buf->max_num_msgs_mbx = mbx_data->max_num_msgs_mbx; |
| |
| /* Capture a new static snapshot of the mailbox by logging the |
| * head and tail of snapshot and set num_iterations to the tail |
| * value to mark the start of the iteration through the snapshot. |
| */ |
| snap_buf->head = ICE_RQ_DATA_MASK(cq->rq.next_to_clean + |
| mbx_data->num_pending_arq); |
| snap_buf->tail = ICE_RQ_DATA_MASK(cq->rq.next_to_clean - 1); |
| snap_buf->num_iterations = snap_buf->tail; |
| |
| /* Pending ARQ messages returned by ice_clean_rq_elem |
| * is the difference between the head and tail of the |
| * mailbox queue. Comparing this value against the watermark |
| * helps to check if we potentially have malicious VFs. |
| */ |
| if (snap_buf->num_pending_arq >= |
| mbx_data->async_watermark_val) { |
| new_state = ICE_MAL_VF_DETECT_STATE_DETECT; |
| status = ice_mbx_detect_malvf(hw, vf_info, &new_state, &is_malvf); |
| } else { |
| new_state = ICE_MAL_VF_DETECT_STATE_TRAVERSE; |
| ice_mbx_traverse(hw, &new_state); |
| } |
| break; |
| |
| case ICE_MAL_VF_DETECT_STATE_TRAVERSE: |
| new_state = ICE_MAL_VF_DETECT_STATE_TRAVERSE; |
| ice_mbx_traverse(hw, &new_state); |
| break; |
| |
| case ICE_MAL_VF_DETECT_STATE_DETECT: |
| new_state = ICE_MAL_VF_DETECT_STATE_DETECT; |
| status = ice_mbx_detect_malvf(hw, vf_info, &new_state, &is_malvf); |
| break; |
| |
| default: |
| new_state = ICE_MAL_VF_DETECT_STATE_INVALID; |
| status = -EIO; |
| } |
| |
| snap_buf->state = new_state; |
| |
| /* Only report VFs as malicious the first time we detect it */ |
| if (is_malvf && !vf_info->malicious) { |
| vf_info->malicious = 1; |
| *report_malvf = true; |
| } |
| |
| return status; |
| } |
| |
| /** |
| * ice_mbx_clear_malvf - Clear VF mailbox info |
| * @vf_info: the mailbox tracking structure for a VF |
| * |
| * In case of a VF reset, this function shall be called to clear the VF's |
| * current mailbox tracking state. |
| */ |
| void ice_mbx_clear_malvf(struct ice_mbx_vf_info *vf_info) |
| { |
| vf_info->malicious = 0; |
| vf_info->msg_count = 0; |
| } |
| |
| /** |
| * ice_mbx_init_vf_info - Initialize a new VF mailbox tracking info |
| * @hw: pointer to the hardware structure |
| * @vf_info: the mailbox tracking info structure for a VF |
| * |
| * Initialize a VF mailbox tracking info structure and insert it into the |
| * snapshot list. |
| * |
| * If you remove the VF, you must also delete the associated VF info structure |
| * from the linked list. |
| */ |
| void ice_mbx_init_vf_info(struct ice_hw *hw, struct ice_mbx_vf_info *vf_info) |
| { |
| struct ice_mbx_snapshot *snap = &hw->mbx_snapshot; |
| |
| ice_mbx_clear_malvf(vf_info); |
| list_add(&vf_info->list_entry, &snap->mbx_vf); |
| } |
| |
| /** |
| * ice_mbx_init_snapshot - Initialize mailbox snapshot data |
| * @hw: pointer to the hardware structure |
| * |
| * Clear the mailbox snapshot structure and initialize the VF mailbox list. |
| */ |
| void ice_mbx_init_snapshot(struct ice_hw *hw) |
| { |
| struct ice_mbx_snapshot *snap = &hw->mbx_snapshot; |
| |
| INIT_LIST_HEAD(&snap->mbx_vf); |
| ice_mbx_reset_snapshot(snap); |
| } |