eHEA: Introducing support vor DLPAR memory add

This patch adds support for DLPAR memory add to the eHEA driver. To detect
whether memory was added the driver uses its own memory mapping table and
checks for kernel addresses whether they're located in already known memory
sections. If not the function ehea_rereg_mrs() is triggered which performs
a rebuild of the mapping table and a re-registration of the global memory
region.
Signed-off-by: Thomas Klein <tklein@de.ibm.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>

drivers/net/ehea/ehea.h

@@ -39,13 +39,13 @@
 #include <asm/io.h>
 
 #define DRV_NAME	"ehea"
-#define DRV_VERSION	"EHEA_0067"
+#define DRV_VERSION	"EHEA_0070"
 
-/* EHEA capability flags */
+/* eHEA capability flags */
 #define DLPAR_PORT_ADD_REM 1
-#define DLPAR_MEM_ADD 2
-#define DLPAR_MEM_REM 4
-#define EHEA_CAPABILITIES (DLPAR_PORT_ADD_REM)
+#define DLPAR_MEM_ADD      2
+#define DLPAR_MEM_REM      4
+#define EHEA_CAPABILITIES  (DLPAR_PORT_ADD_REM)
 
 #define EHEA_MSG_DEFAULT (NETIF_MSG_LINK | NETIF_MSG_TIMER \
 	| NETIF_MSG_RX_ERR | NETIF_MSG_TX_ERR)
@@ -113,6 +113,8 @@
 /* Memory Regions */
 #define EHEA_MR_ACC_CTRL       0x00800000
 
+#define EHEA_BUSMAP_START      0x8000000000000000ULL
+
 #define EHEA_WATCH_DOG_TIMEOUT 10*HZ
 
 /* utility functions */
@@ -186,6 +188,12 @@ struct h_epas {
 				   set to 0 if unused */
 };
 
+struct ehea_busmap {
+	unsigned int entries;		/* total number of entries */
+	unsigned int valid_sections;	/* number of valid sections */
+	u64 *vaddr;
+};
+
 struct ehea_qp;
 struct ehea_cq;
 struct ehea_eq;
@@ -382,6 +390,8 @@ struct ehea_adapter {
 	struct ehea_mr mr;
 	u32 pd;                    /* protection domain */
 	u64 max_mc_mac;            /* max number of multicast mac addresses */
+	int active_ports;
+	struct list_head list;
 };
 
 
@@ -431,6 +441,9 @@ struct port_res_cfg {
 	int max_entries_rq3;
 };
 
+enum ehea_flag_bits {
+	__EHEA_STOP_XFER
+};
 
 void ehea_set_ethtool_ops(struct net_device *netdev);
 int ehea_sense_port_attr(struct ehea_port *port);

drivers/net/ehea/ehea_main.c

@@ -79,6 +79,11 @@ MODULE_PARM_DESC(sq_entries, " Number of entries for the Send Queue  "
 MODULE_PARM_DESC(use_mcs, " 0:NAPI, 1:Multiple receive queues, Default = 1 ");
 
 static int port_name_cnt = 0;
+static LIST_HEAD(adapter_list);
+u64 ehea_driver_flags = 0;
+struct workqueue_struct *ehea_driver_wq;
+struct work_struct ehea_rereg_mr_task;
+
 
 static int __devinit ehea_probe_adapter(struct ibmebus_dev *dev,
 					const struct of_device_id *id);
@@ -238,13 +243,17 @@ static int ehea_refill_rq_def(struct ehea_port_res *pr,
 		rwqe->wr_id = EHEA_BMASK_SET(EHEA_WR_ID_TYPE, wqe_type)
 			    | EHEA_BMASK_SET(EHEA_WR_ID_INDEX, index);
 		rwqe->sg_list[0].l_key = pr->recv_mr.lkey;
-		rwqe->sg_list[0].vaddr = (u64)skb->data;
+		rwqe->sg_list[0].vaddr = ehea_map_vaddr(skb->data);
 		rwqe->sg_list[0].len = packet_size;
 		rwqe->data_segments = 1;
 
 		index++;
 		index &= max_index_mask;
+
+		if (unlikely(test_bit(__EHEA_STOP_XFER, &ehea_driver_flags)))
+			goto out;
 	}
+
 	q_skba->index = index;
 
 	/* Ring doorbell */
@@ -253,7 +262,7 @@ static int ehea_refill_rq_def(struct ehea_port_res *pr,
 		ehea_update_rq2a(pr->qp, i);
 	else
 		ehea_update_rq3a(pr->qp, i);
-
+out:
 	return ret;
 }
 
@@ -1321,7 +1330,7 @@ static void write_swqe2_TSO(struct sk_buff *skb,
 			sg1entry->len = skb_data_size - headersize;
 
 			tmp_addr = (u64)(skb->data + headersize);
-			sg1entry->vaddr = tmp_addr;
+			sg1entry->vaddr = ehea_map_vaddr(tmp_addr);
 			swqe->descriptors++;
 		}
 	} else
@@ -1352,7 +1361,7 @@ static void write_swqe2_nonTSO(struct sk_buff *skb,
 			sg1entry->l_key = lkey;
 			sg1entry->len = skb_data_size - SWQE2_MAX_IMM;
 			tmp_addr = (u64)(skb->data + SWQE2_MAX_IMM);
-			sg1entry->vaddr = tmp_addr;
+			sg1entry->vaddr = ehea_map_vaddr(tmp_addr);
 			swqe->descriptors++;
 		}
 	} else {
@@ -1391,7 +1400,7 @@ static inline void write_swqe2_data(struct sk_buff *skb, struct net_device *dev,
 			sg1entry->len = frag->size;
 			tmp_addr =  (u64)(page_address(frag->page)
 					  + frag->page_offset);
-			sg1entry->vaddr = tmp_addr;
+			sg1entry->vaddr = ehea_map_vaddr(tmp_addr);
 			swqe->descriptors++;
 			sg1entry_contains_frag_data = 1;
 		}
@@ -1406,7 +1415,7 @@ static inline void write_swqe2_data(struct sk_buff *skb, struct net_device *dev,
 
 			tmp_addr = (u64)(page_address(frag->page)
 					 + frag->page_offset);
-			sgentry->vaddr = tmp_addr;
+			sgentry->vaddr = ehea_map_vaddr(tmp_addr);
 			swqe->descriptors++;
 		}
 	}
@@ -1878,6 +1887,9 @@ static int ehea_start_xmit(struct sk_buff *skb, struct net_device *dev)
 		ehea_dump(swqe, 512, "swqe");
 	}
 
+	if (unlikely(test_bit(__EHEA_STOP_XFER, &ehea_driver_flags)))
+		goto out;
+
 	ehea_post_swqe(pr->qp, swqe);
 	pr->tx_packets++;
 
@@ -1892,7 +1904,7 @@ static int ehea_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	}
 	dev->trans_start = jiffies;
 	spin_unlock(&pr->xmit_lock);
-
+out:
 	return NETDEV_TX_OK;
 }
 
@@ -2220,6 +2232,9 @@ static int ehea_up(struct net_device *dev)
 out_clean_pr:
 	ehea_clean_all_portres(port);
 out:
+	if (ret)
+		ehea_info("Failed starting %s. ret=%i", dev->name, ret);
+
 	return ret;
 }
 
@@ -2259,8 +2274,13 @@ static int ehea_down(struct net_device *dev)
 			msleep(1);
 
 	ehea_broadcast_reg_helper(port, H_DEREG_BCMC);
-	ret = ehea_clean_all_portres(port);
 	port->state = EHEA_PORT_DOWN;
+
+	ret = ehea_clean_all_portres(port);
+	if (ret)
+		ehea_info("Failed freeing resources for %s. ret=%i",
+			  dev->name, ret);
+
 	return ret;
 }
 
@@ -2292,15 +2312,11 @@ static void ehea_reset_port(struct work_struct *work)
 	netif_stop_queue(dev);
 	netif_poll_disable(dev);
 
-	ret = ehea_down(dev);
-	if (ret)
-		ehea_error("ehea_down failed. not all resources are freed");
+	ehea_down(dev);
 
 	ret = ehea_up(dev);
-	if (ret) {
-		ehea_error("Reset device %s failed: ret=%d", dev->name, ret);
+	if (ret)
 		goto out;
-	}
 
 	if (netif_msg_timer(port))
 		ehea_info("Device %s resetted successfully", dev->name);
@@ -2312,6 +2328,88 @@ static void ehea_reset_port(struct work_struct *work)
 	return;
 }
 
+static void ehea_rereg_mrs(struct work_struct *work)
+{
+	int ret, i;
+	struct ehea_adapter *adapter;
+
+	ehea_info("LPAR memory enlarged - re-initializing driver");
+
+	list_for_each_entry(adapter, &adapter_list, list)
+		if (adapter->active_ports) {
+			/* Shutdown all ports */
+			for (i = 0; i < EHEA_MAX_PORTS; i++) {
+				struct ehea_port *port = adapter->port[i];
+
+				if (port) {
+					struct net_device *dev = port->netdev;
+
+					if (dev->flags & IFF_UP) {
+						ehea_info("stopping %s",
+							  dev->name);
+						down(&port->port_lock);
+						netif_stop_queue(dev);
+						netif_poll_disable(dev);
+						ehea_down(dev);
+						up(&port->port_lock);
+					}
+				}
+			}
+
+			/* Unregister old memory region */
+			ret = ehea_rem_mr(&adapter->mr);
+			if (ret) {
+				ehea_error("unregister MR failed - driver"
+					   " inoperable!");
+				goto out;
+			}
+		}
+
+	ehea_destroy_busmap();
+
+	ret = ehea_create_busmap();
+	if (ret)
+		goto out;
+
+	clear_bit(__EHEA_STOP_XFER, &ehea_driver_flags);
+
+	list_for_each_entry(adapter, &adapter_list, list)
+		if (adapter->active_ports) {
+			/* Register new memory region */
+			ret = ehea_reg_kernel_mr(adapter, &adapter->mr);
+			if (ret) {
+				ehea_error("register MR failed - driver"
+					   " inoperable!");
+				goto out;
+			}
+
+			/* Restart all ports */
+			for (i = 0; i < EHEA_MAX_PORTS; i++) {
+				struct ehea_port *port = adapter->port[i];
+
+				if (port) {
+					struct net_device *dev = port->netdev;
+
+					if (dev->flags & IFF_UP) {
+						ehea_info("restarting %s",
+							  dev->name);
+						down(&port->port_lock);
+
+						ret = ehea_up(dev);
+						if (!ret) {
+							netif_poll_enable(dev);
+							netif_wake_queue(dev);
+						}
+
+						up(&port->port_lock);
+					}
+				}
+			}
+		}
+out:
+	return;
+}
+
 static void ehea_tx_watchdog(struct net_device *dev)
 {
 	struct ehea_port *port = netdev_priv(dev);
@@ -2573,6 +2671,8 @@ struct ehea_port *ehea_setup_single_port(struct ehea_adapter *adapter,
 	ehea_info("%s: Jumbo frames are %sabled", dev->name,
 		  jumbo == 1 ? "en" : "dis");
 
+	adapter->active_ports++;
+
 	return port;
 
 out_unreg_port:
@@ -2596,6 +2696,7 @@ static void ehea_shutdown_single_port(struct ehea_port *port)
 	ehea_unregister_port(port);
 	kfree(port->mc_list);
 	free_netdev(port->netdev);
+	port->adapter->active_ports--;
 }
 
 static int ehea_setup_ports(struct ehea_adapter *adapter)
@@ -2788,6 +2889,8 @@ static int __devinit ehea_probe_adapter(struct ibmebus_dev *dev,
 		goto out;
 	}
 
+	list_add(&adapter->list, &adapter_list);
+
 	adapter->ebus_dev = dev;
 
 	adapter_handle = of_get_property(dev->ofdev.node, "ibm,hea-handle",
@@ -2891,7 +2994,10 @@ static int __devexit ehea_remove(struct ibmebus_dev *dev)
 
 	ehea_destroy_eq(adapter->neq);
 	ehea_remove_adapter_mr(adapter);
+	list_del(&adapter->list);
+
 	kfree(adapter);
+
 	return 0;
 }
 
@@ -2939,9 +3045,18 @@ int __init ehea_module_init(void)
 	printk(KERN_INFO "IBM eHEA ethernet device driver (Release %s)\n",
 	       DRV_VERSION);
 
+	ehea_driver_wq = create_workqueue("ehea_driver_wq");
+
+	INIT_WORK(&ehea_rereg_mr_task, ehea_rereg_mrs);
+
 	ret = check_module_parm();
 	if (ret)
 		goto out;
+
+	ret = ehea_create_busmap();
+	if (ret)
+		goto out;
+
 	ret = ibmebus_register_driver(&ehea_driver);
 	if (ret) {
 		ehea_error("failed registering eHEA device driver on ebus");
@@ -2965,6 +3080,7 @@ static void __exit ehea_module_exit(void)
 {
 	driver_remove_file(&ehea_driver.driver, &driver_attr_capabilities);
 	ibmebus_unregister_driver(&ehea_driver);
+	ehea_destroy_busmap();
 }
 
 module_init(ehea_module_init);

drivers/net/ehea/ehea_phyp.h

@@ -60,6 +60,9 @@ static inline u32 get_longbusy_msecs(int long_busy_ret_code)
 	}
 }
 
+/* Number of pages which can be registered at once by H_REGISTER_HEA_RPAGES */
+#define EHEA_MAX_RPAGE 512
+
 /* Notification Event Queue (NEQ) Entry bit masks */
 #define NEQE_EVENT_CODE		EHEA_BMASK_IBM(2, 7)
 #define NEQE_PORTNUM  		EHEA_BMASK_IBM(32, 47)

drivers/net/ehea/ehea_qmr.c

@@ -31,6 +31,13 @@
 #include "ehea_phyp.h"
 #include "ehea_qmr.h"
 
+
+struct ehea_busmap ehea_bmap = { 0, 0, NULL };
+extern u64 ehea_driver_flags;
+extern struct workqueue_struct *ehea_driver_wq;
+extern struct work_struct ehea_rereg_mr_task;
+
+
 static void *hw_qpageit_get_inc(struct hw_queue *queue)
 {
 	void *retvalue = hw_qeit_get(queue);
@@ -547,18 +554,84 @@ int ehea_destroy_qp(struct ehea_qp *qp)
 	return 0;
 }
 
+int ehea_create_busmap( void )
+{
+	u64 vaddr = EHEA_BUSMAP_START;
+	unsigned long abs_max_pfn = 0;
+	unsigned long sec_max_pfn;
+	int i;
+
+	/*
+	 * Sections are not in ascending order -> Loop over all sections and
+	 * find the highest PFN to compute the required map size.
+	*/
+	ehea_bmap.valid_sections = 0;
+
+	for (i = 0; i < NR_MEM_SECTIONS; i++)
+		if (valid_section_nr(i)) {
+			sec_max_pfn = section_nr_to_pfn(i);
+			if (sec_max_pfn > abs_max_pfn)
+				abs_max_pfn = sec_max_pfn;
+			ehea_bmap.valid_sections++;
+		}
+
+	ehea_bmap.entries = abs_max_pfn / EHEA_PAGES_PER_SECTION + 1;
+	ehea_bmap.vaddr = vmalloc(ehea_bmap.entries * sizeof(*ehea_bmap.vaddr));
+
+	if (!ehea_bmap.vaddr)
+		return -ENOMEM;
+
+	for (i = 0 ; i < ehea_bmap.entries; i++) {
+		unsigned long pfn = section_nr_to_pfn(i);
+
+		if (pfn_valid(pfn)) {
+			ehea_bmap.vaddr[i] = vaddr;
+			vaddr += EHEA_SECTSIZE;
+		} else
+			ehea_bmap.vaddr[i] = 0;
+	}
+
+	return 0;
+}
+
+void ehea_destroy_busmap( void )
+{
+	vfree(ehea_bmap.vaddr);
+}
+
+u64 ehea_map_vaddr(void *caddr)
+{
+	u64 mapped_addr;
+	unsigned long index = __pa(caddr) >> SECTION_SIZE_BITS;
+
+	if (likely(index < ehea_bmap.entries)) {
+		mapped_addr = ehea_bmap.vaddr[index];
+		if (likely(mapped_addr))
+			mapped_addr |= (((unsigned long)caddr)
+					& (EHEA_SECTSIZE - 1));
+		else
+			mapped_addr = -1;
+	} else
+		mapped_addr = -1;
+
+	if (unlikely(mapped_addr == -1))
+		if (!test_and_set_bit(__EHEA_STOP_XFER, &ehea_driver_flags))
+			queue_work(ehea_driver_wq, &ehea_rereg_mr_task);
+
+	return mapped_addr;
+}
+
 int ehea_reg_kernel_mr(struct ehea_adapter *adapter, struct ehea_mr *mr)
 {
-	int i, k, ret;
-	u64 hret, pt_abs, start, end, nr_pages;
-	u32 acc_ctrl = EHEA_MR_ACC_CTRL;
+	int ret;
 	u64 *pt;
+	void *pg;
+	u64 hret, pt_abs, i, j, m, mr_len;
+	u32 acc_ctrl = EHEA_MR_ACC_CTRL;
 
-	start = KERNELBASE;
-	end = (u64)high_memory;
-	nr_pages = (end - start) / EHEA_PAGESIZE;
+	mr_len = ehea_bmap.valid_sections * EHEA_SECTSIZE;
 
-	pt =  kzalloc(PAGE_SIZE, GFP_KERNEL);
+	pt =  kzalloc(EHEA_MAX_RPAGE * sizeof(u64), GFP_KERNEL);
 	if (!pt) {
 		ehea_error("no mem");
 		ret = -ENOMEM;
@@ -566,7 +639,8 @@ int ehea_reg_kernel_mr(struct ehea_adapter *adapter, struct ehea_mr *mr)
 	}
 	pt_abs = virt_to_abs(pt);
 
-	hret = ehea_h_alloc_resource_mr(adapter->handle, start, end - start,
+	hret = ehea_h_alloc_resource_mr(adapter->handle,
+					EHEA_BUSMAP_START, mr_len,
 					acc_ctrl, adapter->pd,
 					&mr->handle, &mr->lkey);
 	if (hret != H_SUCCESS) {
@@ -575,49 +649,43 @@ int ehea_reg_kernel_mr(struct ehea_adapter *adapter, struct ehea_mr *mr)
 		goto out;
 	}
 
-	mr->vaddr = KERNELBASE;
-	k = 0;
-
-	while (nr_pages > 0) {
-		if (nr_pages > 1) {
-			u64 num_pages = min(nr_pages, (u64)512);
-			for (i = 0; i < num_pages; i++)
-				pt[i] = virt_to_abs((void*)(((u64)start) +
-							    ((k++) *
-							     EHEA_PAGESIZE)));
-
-			hret = ehea_h_register_rpage_mr(adapter->handle,
-							mr->handle, 0,
-							0, (u64)pt_abs,
-							num_pages);
-			nr_pages -= num_pages;
-		} else {
-			u64 abs_adr = virt_to_abs((void*)(((u64)start) +
-							  (k * EHEA_PAGESIZE)));
-
-			hret = ehea_h_register_rpage_mr(adapter->handle,
-							mr->handle, 0,
-							0, abs_adr,1);
-			nr_pages--;
-		}
-
-		if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED)) {
-			ehea_h_free_resource(adapter->handle,
-					     mr->handle, FORCE_FREE);
-			ehea_error("register_rpage_mr failed");
-			ret = -EIO;
-			goto out;
+	for (i = 0 ; i < ehea_bmap.entries; i++)
+		if (ehea_bmap.vaddr[i]) {
+			void *sectbase = __va(i << SECTION_SIZE_BITS);
+			unsigned long k = 0;
+
+			for (j = 0; j < (PAGES_PER_SECTION / EHEA_MAX_RPAGE);
+			      j++) {
+
+				for (m = 0; m < EHEA_MAX_RPAGE; m++) {
+					pg = sectbase + ((k++) * EHEA_PAGESIZE);
+					pt[m] = virt_to_abs(pg);
+				}
+
+				hret = ehea_h_register_rpage_mr(adapter->handle,
+								mr->handle,
+								0, 0, pt_abs,
+								EHEA_MAX_RPAGE);
+				if ((hret != H_SUCCESS)
+				    && (hret != H_PAGE_REGISTERED)) {
+					ehea_h_free_resource(adapter->handle,
+							     mr->handle,
+							     FORCE_FREE);
+					ehea_error("register_rpage_mr failed");
+					ret = -EIO;
+					goto out;
+				}
+			}
 		}
-	}
 
 	if (hret != H_SUCCESS) {
-		ehea_h_free_resource(adapter->handle, mr->handle,
-				     FORCE_FREE);
-		ehea_error("register_rpage failed for last page");
+		ehea_h_free_resource(adapter->handle, mr->handle, FORCE_FREE);
+		ehea_error("registering mr failed");
 		ret = -EIO;
 		goto out;
 	}
 
+	mr->vaddr = EHEA_BUSMAP_START;
 	mr->adapter = adapter;
 	ret = 0;
 out:

drivers/net/ehea/ehea_qmr.h

@@ -36,8 +36,14 @@
  * page size of ehea hardware queues
  */
 
-#define EHEA_PAGESHIFT  12
-#define EHEA_PAGESIZE   4096UL
+#define EHEA_PAGESHIFT         12
+#define EHEA_PAGESIZE          (1UL << EHEA_PAGESHIFT)
+#define EHEA_SECTSIZE          (1UL << 24)
+#define EHEA_PAGES_PER_SECTION (EHEA_SECTSIZE >> PAGE_SHIFT)
+
+#if (1UL << SECTION_SIZE_BITS) < EHEA_SECTSIZE
+#error eHEA module can't work if kernel sectionsize < ehea sectionsize
+#endif
 
 /* Some abbreviations used here:
  *
@@ -372,4 +378,8 @@ int ehea_rem_mr(struct ehea_mr *mr);
 
 void ehea_error_data(struct ehea_adapter *adapter, u64 res_handle);
 
+int ehea_create_busmap( void );
+void ehea_destroy_busmap( void );
+u64 ehea_map_vaddr(void *caddr);
+
 #endif	/* __EHEA_QMR_H__ */