Merge: qed: lower the scheduling latency caused by ethtool -d

MR: https://gitlab.com/redhat/centos-stream/src/kernel/centos-stream-9/-/merge_requests/5288

JIRA: https://issues.redhat.com/browse/RHEL-6372

This fixes a customer case where ethtool -d (run by a sosreport cronjob) causes soft-lockups due to the qed driver keeping the CPU busy without scheduling for more than 26 seconds.

Signed-off-by: Michal Schmidt <mschmidt@redhat.com>

Approved-by: Kamal Heib <kheib@redhat.com>
Approved-by: José Ignacio Tornos Martínez <jtornosm@redhat.com>
Approved-by: CKI KWF Bot <cki-ci-bot+kwf-gitlab-com@redhat.com>

Merged-by: Rado Vrbovsky <rvrbovsk@redhat.com>

Author: emmil

drivers/net/ethernet/qlogic/qed/qed_debug.c

@@ -1832,7 +1832,8 @@ static enum dbg_status qed_find_nvram_image(struct qed_hwfn *p_hwfn,
 					    struct qed_ptt *p_ptt,
 					    u32 image_type,
 					    u32 *nvram_offset_bytes,
-					    u32 *nvram_size_bytes)
+					    u32 *nvram_size_bytes,
+					    bool b_can_sleep)
 {
 	u32 ret_mcp_resp, ret_mcp_param, ret_txn_size;
 	struct mcp_file_att file_att;
@@ -1846,7 +1847,8 @@ static enum dbg_status qed_find_nvram_image(struct qed_hwfn *p_hwfn,
 					&ret_mcp_resp,
 					&ret_mcp_param,
 					&ret_txn_size,
-					(u32 *)&file_att, false);
+					(u32 *)&file_att,
+					b_can_sleep);
 
 	/* Check response */
 	if (nvm_result || (ret_mcp_resp & FW_MSG_CODE_MASK) !=
@@ -1873,7 +1875,9 @@ static enum dbg_status qed_find_nvram_image(struct qed_hwfn *p_hwfn,
 static enum dbg_status qed_nvram_read(struct qed_hwfn *p_hwfn,
 				      struct qed_ptt *p_ptt,
 				      u32 nvram_offset_bytes,
-				      u32 nvram_size_bytes, u32 *ret_buf)
+				      u32 nvram_size_bytes,
+				      u32 *ret_buf,
+				      bool b_can_sleep)
 {
 	u32 ret_mcp_resp, ret_mcp_param, ret_read_size, bytes_to_copy;
 	s32 bytes_left = nvram_size_bytes;
@@ -1899,7 +1903,7 @@ static enum dbg_status qed_nvram_read(struct qed_hwfn *p_hwfn,
 				       &ret_mcp_resp,
 				       &ret_mcp_param, &ret_read_size,
 				       (u32 *)((u8 *)ret_buf + read_offset),
-				       false))
+				       b_can_sleep))
 			return DBG_STATUS_NVRAM_READ_FAILED;
 
 		/* Check response */
@@ -2868,6 +2872,7 @@ static u32 qed_grc_dump_ctx_data(struct qed_hwfn *p_hwfn,
 							  false,
 							  SPLIT_TYPE_NONE, 0);
 		}
+		cond_resched();
 	}
 
 	return offset;
@@ -3379,7 +3384,8 @@ static u32 qed_grc_dump_mcp_hw_dump(struct qed_hwfn *p_hwfn,
 				      p_ptt,
 				      NVM_TYPE_HW_DUMP_OUT,
 				      &hw_dump_offset_bytes,
-				      &hw_dump_size_bytes);
+				      &hw_dump_size_bytes,
+				      false);
 	if (status != DBG_STATUS_OK)
 		return 0;
 
@@ -3396,7 +3402,9 @@ static u32 qed_grc_dump_mcp_hw_dump(struct qed_hwfn *p_hwfn,
 		status = qed_nvram_read(p_hwfn,
 					p_ptt,
 					hw_dump_offset_bytes,
-					hw_dump_size_bytes, dump_buf + offset);
+					hw_dump_size_bytes,
+					dump_buf + offset,
+					false);
 		if (status != DBG_STATUS_OK) {
 			DP_NOTICE(p_hwfn,
 				  "Failed to read MCP HW Dump image from NVRAM\n");
@@ -4122,7 +4130,9 @@ static enum dbg_status qed_mcp_trace_get_meta_info(struct qed_hwfn *p_hwfn,
 	return qed_find_nvram_image(p_hwfn,
 				    p_ptt,
 				    nvram_image_type,
-				    trace_meta_offset, trace_meta_size);
+				    trace_meta_offset,
+				    trace_meta_size,
+				    true);
 }
 
 /* Reads the MCP Trace meta data from NVRAM into the specified buffer */
@@ -4138,7 +4148,10 @@ static enum dbg_status qed_mcp_trace_read_meta(struct qed_hwfn *p_hwfn,
 	/* Read meta data from NVRAM */
 	status = qed_nvram_read(p_hwfn,
 				p_ptt,
-				nvram_offset_in_bytes, size_in_bytes, buf);
+				nvram_offset_in_bytes,
+				size_in_bytes,
+				buf,
+				true);
 	if (status != DBG_STATUS_OK)
 		return status;
 

drivers/net/ethernet/qlogic/qed/qed_hw.c

@@ -578,6 +578,7 @@ static int qed_dmae_operation_wait(struct qed_hwfn *p_hwfn)
 	barrier();
 	while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
 		udelay(DMAE_MIN_WAIT_TIME);
+		cond_resched();
 		if (++wait_cnt > wait_cnt_limit) {
 			DP_NOTICE(p_hwfn->cdev,
 				  "Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",

drivers/net/ethernet/qlogic/qed/qed_mcp.c

@@ -459,12 +459,11 @@ static void qed_mcp_print_cpu_info(struct qed_hwfn *p_hwfn,
 static int
 _qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 		       struct qed_ptt *p_ptt,
-		       struct qed_mcp_mb_params *p_mb_params,
-		       u32 max_retries, u32 usecs)
+		       struct qed_mcp_mb_params *p_mb_params)
 {
-	u32 cnt = 0, msecs = DIV_ROUND_UP(usecs, 1000);
 	struct qed_mcp_cmd_elem *p_cmd_elem;
 	u16 seq_num;
+	u32 cnt = 0;
 	int rc = 0;
 
 	/* Wait until the mailbox is non-occupied */
@@ -488,12 +487,13 @@ _qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 		spin_unlock_bh(&p_hwfn->mcp_info->cmd_lock);
 
 		if (QED_MB_FLAGS_IS_SET(p_mb_params, CAN_SLEEP))
-			msleep(msecs);
+			usleep_range(QED_MCP_RESP_ITER_US,
+				     QED_MCP_RESP_ITER_US * 2);
 		else
-			udelay(usecs);
-	} while (++cnt < max_retries);
+			udelay(QED_MCP_RESP_ITER_US);
+	} while (++cnt < QED_DRV_MB_MAX_RETRIES);
 
-	if (cnt >= max_retries) {
+	if (cnt >= QED_DRV_MB_MAX_RETRIES) {
 		DP_NOTICE(p_hwfn,
 			  "The MFW mailbox is occupied by an uncompleted command. Failed to send command 0x%08x [param 0x%08x].\n",
 			  p_mb_params->cmd, p_mb_params->param);
@@ -520,9 +520,10 @@ _qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 		 */
 
 		if (QED_MB_FLAGS_IS_SET(p_mb_params, CAN_SLEEP))
-			msleep(msecs);
+			usleep_range(QED_MCP_RESP_ITER_US,
+				     QED_MCP_RESP_ITER_US * 2);
 		else
-			udelay(usecs);
+			udelay(QED_MCP_RESP_ITER_US);
 
 		spin_lock_bh(&p_hwfn->mcp_info->cmd_lock);
 
@@ -536,9 +537,9 @@ _qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 			goto err;
 
 		spin_unlock_bh(&p_hwfn->mcp_info->cmd_lock);
-	} while (++cnt < max_retries);
+	} while (++cnt < QED_DRV_MB_MAX_RETRIES);
 
-	if (cnt >= max_retries) {
+	if (cnt >= QED_DRV_MB_MAX_RETRIES) {
 		DP_NOTICE(p_hwfn,
 			  "The MFW failed to respond to command 0x%08x [param 0x%08x].\n",
 			  p_mb_params->cmd, p_mb_params->param);
@@ -564,7 +565,8 @@ _qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 		   "MFW mailbox: response 0x%08x param 0x%08x [after %d.%03d ms]\n",
 		   p_mb_params->mcp_resp,
 		   p_mb_params->mcp_param,
-		   (cnt * usecs) / 1000, (cnt * usecs) % 1000);
+		   (cnt * QED_MCP_RESP_ITER_US) / 1000,
+		   (cnt * QED_MCP_RESP_ITER_US) % 1000);
 
 	/* Clear the sequence number from the MFW response */
 	p_mb_params->mcp_resp &= FW_MSG_CODE_MASK;
@@ -581,8 +583,6 @@ static int qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 				 struct qed_mcp_mb_params *p_mb_params)
 {
 	size_t union_data_size = sizeof(union drv_union_data);
-	u32 max_retries = QED_DRV_MB_MAX_RETRIES;
-	u32 usecs = QED_MCP_RESP_ITER_US;
 
 	/* MCP not initialized */
 	if (!qed_mcp_is_init(p_hwfn)) {
@@ -606,13 +606,7 @@ static int qed_mcp_cmd_and_union(struct qed_hwfn *p_hwfn,
 		return -EINVAL;
 	}
 
-	if (QED_MB_FLAGS_IS_SET(p_mb_params, CAN_SLEEP)) {
-		max_retries = DIV_ROUND_UP(max_retries, 1000);
-		usecs *= 1000;
-	}
-
-	return _qed_mcp_cmd_and_union(p_hwfn, p_ptt, p_mb_params, max_retries,
-				      usecs);
+	return _qed_mcp_cmd_and_union(p_hwfn, p_ptt, p_mb_params);
 }
 
 static int _qed_mcp_cmd(struct qed_hwfn *p_hwfn,
@@ -3085,20 +3079,13 @@ int qed_mcp_nvm_read(struct qed_dev *cdev, u32 addr, u8 *p_buf, u32 len)
 					 DRV_MB_PARAM_NVM_LEN_OFFSET),
 					&resp, &resp_param,
 					&read_len,
-					(u32 *)(p_buf + offset), false);
+					(u32 *)(p_buf + offset), true);
 
 		if (rc || (resp != FW_MSG_CODE_NVM_OK)) {
 			DP_NOTICE(cdev, "MCP command rc = %d\n", rc);
 			break;
 		}
 
-		/* This can be a lengthy process, and it's possible scheduler
-		 * isn't preemptible. Sleep a bit to prevent CPU hogging.
-		 */
-		if (bytes_left % 0x1000 <
-		    (bytes_left - read_len) % 0x1000)
-			usleep_range(1000, 2000);
-
 		offset += read_len;
 		bytes_left -= read_len;
 	}