Closes #90 - Implement INTC-less MSI-X architecture

* removed all interrupt controllers for VC709
* replaced by 132 MSI-X interrupts (128 PE, 4 other)
* replaced driver code

platform/vc709/module/blue_dma_ctrl.c

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file dual_dma_ctrl.c 
+ * @file dual_dma_ctrl.c
  * @brief Implementation of custom dma engine specific code
 	Strips the layout of the dma-registers and setup stuff to start a dma transfer
 	in addition calls to acknowledge the interrupt in hw is given
@@ -70,7 +70,7 @@ void blue_dma_transmit_from_device(void * device_buffer, dma_addr_t host_handle,
 	fflink_info("dev_buf %lX dma_handle %lX \nsize %d dev_base %lX\n", (unsigned long) device_buffer, (unsigned long) host_handle, btt, (unsigned long) device_base_addr);
 	//if(mutex_lock_interruptible(&dma_regs_mutex))
 	//	fflink_warn("got killed while aquiring the mutex\n");
-		
+
 	/* SA */
 	pcie_writel((unsigned long)device_buffer, device_base_addr + REG_FPGA_ADDR);
 	/* DA */
@@ -81,7 +81,7 @@ void blue_dma_transmit_from_device(void * device_buffer, dma_addr_t host_handle,
 	wmb();
 	/* start cmd */
 	pcie_writel(CMD_READ, device_base_addr + REG_CMD);
-	
+
 	//mutex_unlock(&dma_regs_mutex);
 }
 
@@ -98,7 +98,7 @@ void blue_dma_transmit_to_device(void * device_buffer, dma_addr_t host_handle, i
 	fflink_info("dev_buf %lX dma_handle %lX \nsize %d dev_base %lX\n", (unsigned long) device_buffer, (unsigned long) host_handle, btt, (unsigned long) device_base_addr);
 	//if(mutex_lock_interruptible(&dma_regs_mutex))
 	//	fflink_warn("got killed while aquiring the mutex\n");
-	
+
 	/* SA */
 	pcie_writel(host_handle, device_base_addr + REG_HOST_ADDR);
 	/* DA */
@@ -109,7 +109,7 @@ void blue_dma_transmit_to_device(void * device_buffer, dma_addr_t host_handle, i
 	wmb();
 	/* start cmd */
 	pcie_writel(CMD_WRITE, device_base_addr + REG_CMD);
-	
+
 	//mutex_unlock(&dma_regs_mutex);
 }
 

platform/vc709/module/char_device_dma.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file char_device_dma.h 
+ * @file char_device_dma.h
  * @brief Composition of everything needed to handle char-device calls for dma transfers
 	here all definitions of functions and structs are given, which are used by the char-device
 	the user can adapt the method used for transfers (bounce-/double-buffering)
@@ -98,17 +98,17 @@ struct priv_data_struct {
 
 	dma_addr_t dma_handle_h2l;
 	dma_addr_t dma_handle_l2h;
-	
+
 	void * mem_addr_l2h;
 	void * mem_addr_h2l;
-	
+
 	void * device_base_addr;
 	void * ctrl_base_addr;
-	
+
 	wait_queue_head_t rw_wait_queue;
 	bool condition_rw;
 	struct mutex rw_mutex;
-	
+
 	unsigned int cache_lsize;
 	unsigned int cache_mask;
 };

platform/vc709/module/char_device_user.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file char_device_user.h 
+ * @file char_device_user.h
  * @brief Composition of everything needed to handle char-device calls for user-functions
 	here all definitions of functions and structs are given, which are used by the char-device
 	user can choose, how much minor devices will be allocated
@@ -71,8 +71,10 @@
 
 /******************************************************************************/
 
-/* number of device, which will be created for dma engines  */
-#define FFLINK_USER_NODES 4
+/* number of device, which will be created for PEs  */
+#define FFLINK_USER_NODES 1
+
+#define PE_IRQS 128
 
 /* device name to register fops with
  * fops will append a number to name for multiple minor nodes */
@@ -114,12 +116,8 @@
 
 /* struct array to hold data over multiple fops-calls */
 struct priv_data_struct {
-	unsigned int minor;
-	
-	wait_queue_head_t user_wait_queue;
-	unsigned long user_condition;
-	
-	struct mutex rw_mutex;
+	wait_queue_head_t user_wait_queue[PE_IRQS];
+	int user_condition[PE_IRQS];
 };
 
 /******************************************************************************/
@@ -132,11 +130,6 @@ static ssize_t user_read(struct file *, char __user *, size_t count, loff_t *);
 static ssize_t user_write(struct file *, const char __user *, size_t count, loff_t *);
 static int user_mmap(struct file *filp, struct vm_area_struct *vma);
 
-/******************************************************************************/
-/* helper function */
-
-static inline void user_set_mask(unsigned long *condition, uint32_t mask);
-
 /******************************************************************************/
 
 #endif // __CHAR_DEVICE_USER_H

platform/vc709/module/common/debug_print.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file debug_print.h 
+ * @file debug_print.h
  * @brief Composition of everything needed for debug output
  * currently three different levels of outputs defined
  * */
@@ -53,7 +53,7 @@
 
 #define fflink_warn(msg, ...)	printk(KERN_WARNING "ffLink (%s): " msg, \
 		__func__, ##__VA_ARGS__)
-		
+
 /******************************************************************************/
 
 #endif // __DEBUG_PRINT_H

platform/vc709/module/common/device_dma.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file device_dma.h 
+ * @file device_dma.h
  * @brief Composition of everything needed to (un/)load the char_device respondsible for dma calls
 	additionally helper functions defined to ease access throughout different minor nodes
  * */

platform/vc709/module/common/device_pcie.h

@@ -17,9 +17,9 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file device_pcie.h 
+ * @file device_pcie.h
  * @brief Composition of everything needed to handle pcie device
-	Responsible for (un/)load pcie device, allows acces to pci-specific information 
+	Responsible for (un/)load pcie device, allows acces to pci-specific information
 	and supports wrapper for safe access to pcie-bar0 space
  * */
 
@@ -46,6 +46,8 @@ unsigned long pcie_readl(void * addr);
 void pcie_writel_bar2(unsigned long data, void * addr);
 unsigned long pcie_readl_bar2(void * addr);
 
+int pcie_translate_irq_number(int irq);
+
 /******************************************************************************/
 
 #endif // __DEVICE_PCIE_H

platform/vc709/module/common/device_user.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file device_user.h 
+ * @file device_user.h
  * @brief Composition of everything needed for the char-device(s) for user-calls
 	Functions to (un/)load this device and interrupt handlers
 	these handlers will be registered as msi irqs, when the pcie_device is loaded
@@ -35,10 +35,7 @@ void char_user_unregister(void);
 /******************************************************************************/
 /* interrupt handler used by user cores registered in pcie_device.c */
 
-irqreturn_t intr_handler_user_0(int irq, void * dev_id);
-irqreturn_t intr_handler_user_1(int irq, void * dev_id);
-irqreturn_t intr_handler_user_2(int irq, void * dev_id);
-irqreturn_t intr_handler_user_3(int irq, void * dev_id);
+irqreturn_t intr_handler_user(int irq, void * dev_id);
 
 /******************************************************************************/
 

platform/vc709/module/common/dma_ctrl.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file dma_ctrl.h 
+ * @file dma_ctrl.h
  * @brief Composition of everything needed to handle dma engine(s)
 	interrupt handlers will be registered as msi irqs, when the pcie_device is loaded
 	common header used for differrent hw-implementations of the dma engine (Xilinx, custom)

platform/vc709/module/dma_ctrl.c

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file dual_dma_ctrl.c 
+ * @file dual_dma_ctrl.c
  * @brief Implementation of custom dma engine specific code
 	Strips the layout of the dma-registers and setup stuff to start a dma transfer
 	in addition calls to acknowledge the interrupt in hw is given
@@ -84,10 +84,10 @@ void dma_ctrl_init(void * device_base_addr)
 irqreturn_t intr_handler_dma(int irq, void * dev_id)
 {
 	fflink_info("Interrupt called with irq %d\n", irq);
-	
+
 	fflink_dma[dma_used].intr(irq, dev_id);
 	wake_up_queue(0);
-	
+
 	return IRQ_HANDLED;
 }
 

platform/vc709/module/dual_dma_ctrl.c

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file dual_dma_ctrl.c 
+ * @file dual_dma_ctrl.c
  * @brief Implementation of custom dma engine specific code
 	Strips the layout of the dma-registers and setup stuff to start a dma transfer
 	in addition calls to acknowledge the interrupt in hw is given
@@ -75,7 +75,7 @@ void dual_dma_transmit_from_device(void * device_buffer, dma_addr_t host_handle,
 	fflink_info("dev_buf %lX dma_handle %lX \nsize %d dev_base %lX\n", (unsigned long) device_buffer, (unsigned long) host_handle, btt, (unsigned long) device_base_addr);
 	//if(mutex_lock_interruptible(&dma_regs_mutex))
 	//	fflink_warn("got killed while aquiring the mutex\n");
-		
+
 	/* SA */
 	pcie_writel((unsigned long) device_buffer, device_base_addr + REG_FPGA_ADDR_LOW);
 	/* DA */
@@ -86,7 +86,7 @@ void dual_dma_transmit_from_device(void * device_buffer, dma_addr_t host_handle,
 	wmb();
 	/* start cmd */
 	pcie_writel(CMD_READ, device_base_addr + REG_CMD);
-	
+
 	//mutex_unlock(&dma_regs_mutex);
 }
 
@@ -103,7 +103,7 @@ void dual_dma_transmit_to_device(void * device_buffer, dma_addr_t host_handle, i
 	fflink_info("dev_buf %lX dma_handle %lX \nsize %d dev_base %lX\n", (unsigned long) device_buffer, (unsigned long) host_handle, btt, (unsigned long) device_base_addr);
 	//if(mutex_lock_interruptible(&dma_regs_mutex))
 	//	fflink_warn("got killed while aquiring the mutex\n");
-	
+
 	/* SA */
 	pcie_writel(host_handle, device_base_addr + REG_HOST_ADDR_LOW);
 	/* DA */

platform/vc709/module/ffLink_driver.c

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file ffLink_driver.c 
+ * @file ffLink_driver.c
  * @brief Implementation of linux kernel driver module
 	this is the global entry point for the operating system to (un-)load the driver
 	afterwards the register routines of all submodules are called
@@ -39,24 +39,24 @@
  * @param none
  * @return Zero, if initialization was successfull
  * */
-static int __init fflink_init(void) 
+static int __init fflink_init(void)
 {
 	int err = 0;
-	
+
 	fflink_notice("Init char-dev(s), dev-entries and register pci-device\n");
-	
+
 	err = pcie_register();
 	if(err) {
 		fflink_info("Could not register pcie device\n");
 		goto error_pcie_register;
 	}
-	
+
 	err = char_dma_register();
 	if(err) {
 		fflink_info("Could not register dma char device(s)\n");
 		goto error_dma_register;
 	}
-	
+
 	err = char_user_register();
 	if(err) {
 		fflink_info("Could not register user char device(s)\n");
@@ -64,9 +64,9 @@ static int __init fflink_init(void)
 	}
 
 	fflink_warn("Successfully registered driver\n");
-	
+
 	return 0;
-	
+
 error_dma_register:
 	return -EACCES;
 error_user_register:
@@ -85,13 +85,13 @@ error_pcie_register:
 static void __exit fflink_exit(void)
 {
 	fflink_notice("Deallocate char-dev(s)/pci-device\n");
-	
+
 	char_dma_unregister();
-	
+
 	char_user_unregister();
-	
+
 	pcie_unregister();
-	
+
 	fflink_warn("Successfully unregistered driver\n");
 }
 

platform/vc709/module/ffLink_driver.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file ffLink_driver.h 
+ * @file ffLink_driver.h
  * @brief Composition of everything needed for kernel module
 	mainly consists of the init-/exit-functions
 	handles versioning, licensing and small description can be found in the implementation

platform/vc709/module/include/dma_ioctl_calls.h

@@ -17,11 +17,11 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file dma_ioctl_calls.c 
+ * @file dma_ioctl_calls.c
  * @brief Define all ioctl commands with corresponding structs for the dma char device
 	these calls will be used by the platform_api to start a dma transfer on the pci-bus
 	Two different modes can be used here: either choose mmaped memory for zero copy access
-	or let the system call handle copying the data by itself 
+	or let the system call handle copying the data by itself
  * */
 
 #ifndef __DMA_IOCTL_CALLS_H

platform/vc709/module/include/user_ioctl_calls.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file user_ioctl_calls.c 
+ * @file user_ioctl_calls.c
  * @brief Define all ioctl commands with corresponding structs for user calls
 	this allows simplified setup of kernel functions in the bitstream composition
 	Two mechanisms are implemented: make use of read/write system calls to access registers,

platform/vc709/module/pcie_device.h

@@ -17,7 +17,7 @@
 // along with Tapasco.  If not, see <http://www.gnu.org/licenses/>.
 //
 /**
- * @file pcie_device.h 
+ * @file pcie_device.h
  * @brief Composition of everything needed to handle pcie device
  * */
 
@@ -67,6 +67,8 @@
 
 #define FFLINK_PCI_NAME "FFLINK_PCI_DRIVER"
 
+#define REQUIRED_INTERRUPTS 132
+
 // pci id to find device on the bus
 #define XILINX_VENDOR_ID   0x10EE
 #define XILINX_DEVICE_ID   0x7038
@@ -95,16 +97,14 @@ static struct pci_driver fflink_pci_driver = {
 
 /* struct to hold data related to the pcie device */
 struct pci_data_struct{
-	struct pci_dev* pdev;
+	struct pci_dev *pdev;
 	unsigned long long phy_addr_bar0;
 	unsigned long long phy_len_bar0;
 	unsigned long long phy_flags_bar0;
 	unsigned long long phy_addr_bar2;
 	unsigned long long phy_len_bar2;
 	unsigned long long phy_flags_bar2;
-	unsigned int irq_first;	
-	unsigned int irq_length;
-	unsigned int irq_assigned;
+	int irq_mapping[REQUIRED_INTERRUPTS];
 	void * kvirt_addr_bar0;
 	void * kvirt_addr_bar2;
 };
@@ -116,6 +116,7 @@ static int register_intr_handler(struct pci_dev *pdev, int c);
 static int claim_device(struct pci_dev *pdev);
 static int configure_device(struct pci_dev *pdev);
 static int claim_msi(struct pci_dev *pdev);
+static int free_irqs(struct pci_dev *pdev);
 static void report_link_status(struct pci_dev *pdev);
 
 /******************************************************************************/

platform/vc709/vc709.tcl

@@ -71,12 +71,22 @@ namespace eval platform {
     set msix_int [create_bd_pin -dir "O" "msix_int"]
     set m_axi [create_bd_intf_pin -mode Master -vlnv xilinx.com:interface:aximm_rtl:1.0 "M_AXI"]
 
-    set irq_concat [tapasco::createConcat "interrupt_concat" 8]
+    set num_irqs 132
+    set num_irqs_threadpools 128
+
+    set irq_concat_ss [tapasco::createConcat "interrupt_concat" 8]
 
     # create MSIX interrupt controller
     set msix_intr_ctrl [tapasco::createMSIXIntrCtrl "msix_intr_ctrl"]
-    connect_bd_net [get_bd_pin -of_objects $irq_concat -filter {NAME == "dout"}] [get_bd_pin -of_objects $msix_intr_ctrl -filter {NAME == "interrupt"}]
+    connect_bd_net [get_bd_pin -of_objects $irq_concat_ss -filter {NAME == "dout"}] [get_bd_pin -of_objects $msix_intr_ctrl -filter {NAME == "interrupt"}]
+
 
+    set curr_pcie_line 4
+    # connect interrupts to interrupt controller
+    foreach irq $irqs {
+      connect_bd_net -boundary_type upper $irq [get_bd_pins -of $irq_concat_ss -filter "NAME == [format "In%d" $curr_pcie_line]"]
+      incr curr_pcie_line 1
+    }
 
     connect_bd_intf_net [get_bd_intf_pins -of_objects $msix_intr_ctrl -filter {NAME == "M_AXI"}] $m_axi
     connect_bd_net [get_bd_pin -of_objects $msix_intr_ctrl -filter {NAME == "cfg_interrupt_msix_address"}] $msix_addr
@@ -87,29 +97,7 @@ namespace eval platform {
     connect_bd_net $msix_enable [get_bd_pin -of_objects $msix_intr_ctrl -filter {NAME == "cfg_interrupt_msix_enable"}]
     connect_bd_net $msix_mask [get_bd_pin -of_objects $msix_intr_ctrl -filter {NAME == "cfg_interrupt_msix_mask"}]
 
-    connect_bd_net $dma_irq [get_bd_pin -of_objects $irq_concat -filter {NAME == "In0"}]
-
-    set curr_pcie_line 4
-    # create interrupt controllers and connect them to GP1
-    set intcs [list]
-    foreach irq $irqs {
-      set intc [tapasco::createIntCtrl [format "axi_intc_%02d" [llength $intcs]]]
-      lappend intcs $intc
-      connect_bd_net -boundary_type upper $irq [get_bd_pins -of $intc -filter {NAME=="intr"}]
-      connect_bd_net -boundary_type upper [get_bd_pins -of $intc -filter {NAME=="irq"}] [get_bd_pins -of $irq_concat -filter "NAME == [format "In%d" $curr_pcie_line]"]
-      incr curr_pcie_line 1
-    }
-
-    set intcic [tapasco::createInterconnect "axi_intc_ic" 1 [expr {1 + [llength $intcs]}]]
-    connect_bd_intf_net -boundary_type upper [get_bd_intf_pins -of $intcic -filter "NAME == M00_AXI"] [get_bd_intf_pins -of $msix_intr_ctrl -filter { MODE == "Slave" }]
-    set i 0
-    foreach intc $intcs {
-      set slave [get_bd_intf_pins -of $intc -filter { MODE == "Slave" }]
-      set master [get_bd_intf_pins -of $intcic -filter "NAME == [format "M%02d_AXI" [expr {$i + 1}]]"]
-      puts "Connecting $master to $slave ..."
-      connect_bd_intf_net -boundary_type upper $master $slave
-      incr i
-    }
+    connect_bd_net $dma_irq [get_bd_pin -of_objects $irq_concat_ss -filter {NAME == "In0"}]
 
     # connect internal clocks
     connect_bd_net -net intc_clock_net $aclk [get_bd_pins -of_objects [get_bd_cells] -filter {TYPE == "clk" && DIR == "I"}]
@@ -121,7 +109,7 @@ namespace eval platform {
     connect_bd_net -net intc_p_reset_net $p_aresetn $p_resets
 
     # connect S_AXI
-    connect_bd_intf_net $s_axi [get_bd_intf_pins -of_objects $intcic -filter {NAME == "S00_AXI"}]
+    connect_bd_intf_net $s_axi [get_bd_intf_pins -of_objects $msix_intr_ctrl -filter {NAME == "S_AXI"}]
 
     current_bd_instance $instance
     return $group
@@ -402,9 +390,9 @@ namespace eval platform {
       CONFIG.axi_addr_width {64} \
       CONFIG.pf0_msi_enabled {false} \
       CONFIG.pf0_msix_enabled {true} \
-      CONFIG.pf0_msix_cap_table_size {007} \
+      CONFIG.pf0_msix_cap_table_size {83} \
       CONFIG.pf0_msix_cap_table_offset {500000} \
-      CONFIG.pf0_msix_cap_pba_offset {502000} \
+      CONFIG.pf0_msix_cap_pba_offset {508000} \
       CONFIG.comp_timeout {50ms} \
       CONFIG.pf0_interrupt_pin {NONE} \
       CONFIG.c_s_axi_supports_narrow_burst {false} \