From bf1f6c15b6c6c516e8c3664a7837296fb796ba78 Mon Sep 17 00:00:00 2001 From: Cheng Fei Phung Date: Tue, 4 Dec 2018 02:19:38 +0000 Subject: [PATCH] Staging : Add RIFFA PCIe staging driver This patch adds RIFFA PCIe linux driver for https://github.com/promach/riffa/tree/full_duplex/driver/linux This staging driver is modified from this upstream driver at https://github.com/KastnerRG/riffa/tree/master/driver/linux For further details, please refer to https://github.com/KastnerRG/riffa/pull/31 Signed-off-by: Cheng Fei Phung --- drivers/staging/riffa/Kconfig | 5 + drivers/staging/riffa/Makefile | 1 + drivers/staging/riffa/TODO | 7 + drivers/staging/riffa/circ_queue.c | 188 +++ drivers/staging/riffa/circ_queue.h | 96 ++ drivers/staging/riffa/riffa.c | 152 +++ drivers/staging/riffa/riffa.h | 121 ++ drivers/staging/riffa/riffa_driver.c | 1643 ++++++++++++++++++++++++++ drivers/staging/riffa/riffa_driver.h | 131 ++ 9 files changed, 2344 insertions(+) create mode 100644 drivers/staging/riffa/Kconfig create mode 100644 drivers/staging/riffa/Makefile create mode 100644 drivers/staging/riffa/TODO create mode 100644 drivers/staging/riffa/circ_queue.c create mode 100644 drivers/staging/riffa/circ_queue.h create mode 100644 drivers/staging/riffa/riffa.c create mode 100644 drivers/staging/riffa/riffa.h create mode 100644 drivers/staging/riffa/riffa_driver.c create mode 100644 drivers/staging/riffa/riffa_driver.h diff --git a/drivers/staging/riffa/Kconfig b/drivers/staging/riffa/Kconfig new file mode 100644 index 00000000000000..b8fa2d51fbc0f4 --- /dev/null +++ b/drivers/staging/riffa/Kconfig @@ -0,0 +1,5 @@ +config RIFFA_PCIE + tristate "a simple framework for communicating data from a host CPU to a FPGA via a PCI Express bus" + depends on PCI + ---help--- + Transfers data with full duplex capability using PCIe protocol diff --git a/drivers/staging/riffa/Makefile b/drivers/staging/riffa/Makefile new file mode 100644 index 00000000000000..79ef3b9b8c8f33 --- /dev/null +++ b/drivers/staging/riffa/Makefile @@ -0,0 +1 @@ +obj-$(CONFIG_RIFFA) += riffa.o circ_queue.o riffa_driver.o riffa_mod.o diff --git a/drivers/staging/riffa/TODO b/drivers/staging/riffa/TODO new file mode 100644 index 00000000000000..5f13b32a1f912a --- /dev/null +++ b/drivers/staging/riffa/TODO @@ -0,0 +1,7 @@ +TODO: +- optimize the driver code for further speed improvement although it can now achieve defined PCIe speed grade +- solve all the coding style errors from scripts/checkpatch.pl +- add vendor and device IDs for more supported devices after actual hardware testing + +Please send any patches to Greg Kroah-Hartman +and Cheng Fei Phung diff --git a/drivers/staging/riffa/circ_queue.c b/drivers/staging/riffa/circ_queue.c new file mode 100644 index 00000000000000..fb43ca22e3c0f4 --- /dev/null +++ b/drivers/staging/riffa/circ_queue.c @@ -0,0 +1,188 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: circ_queue.c + * Version: 1.0 + * Description: A lock-free single-producer circular queue implementation + * modeled after the more elaborate C++ version from Faustino Frechilla at: + * http://www.codeproject.com/Articles/153898/Yet-another-implementation-of-a-lock-free-circular + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 1.0. + */ + +#include +#include "circ_queue.h" + +circ_queue * init_circ_queue(int len) +{ + int i; + circ_queue * q; + + q = kzalloc(sizeof(circ_queue), GFP_KERNEL); + if (q == NULL) { + printk(KERN_ERR "Not enough memory to allocate circ_queue"); + return NULL; + } + + atomic_set(&q->writeIndex, 0); + atomic_set(&q->readIndex, 0); + q->len = len; + + q->vals = (unsigned int**) kzalloc(len*sizeof(unsigned int*), GFP_KERNEL); + if (q->vals == NULL) { + printk(KERN_ERR "Not enough memory to allocate circ_queue array"); + return NULL; + } + for (i = 0; i < len; i++) { + q->vals[i] = (unsigned int*) kzalloc(2*sizeof(unsigned int), GFP_KERNEL); + if (q->vals[i] == NULL) { + printk(KERN_ERR "Not enough memory to allocate circ_queue array position"); + return NULL; + } + } + + return q; +} + +/** + * Internal function to help count. Returns the queue size normalized position. + */ +unsigned int queue_count_to_index(unsigned int count, unsigned int len) +{ + return (count % len); +} + +int push_circ_queue(circ_queue * q, unsigned int val1, unsigned int val2) +{ + unsigned int currReadIndex; + unsigned int currWriteIndex; + + currWriteIndex = atomic_read(&q->writeIndex); + currReadIndex = atomic_read(&q->readIndex); + if (queue_count_to_index(currWriteIndex+1, q->len) == queue_count_to_index(currReadIndex, q->len)) { + // The queue is full + return 1; + } + + // Save the data into the queue + q->vals[queue_count_to_index(currWriteIndex, q->len)][0] = val1; + q->vals[queue_count_to_index(currWriteIndex, q->len)][1] = val2; + // Increment atomically write index. Now a consumer thread can read + // the piece of data that was just stored. + atomic_inc(&q->writeIndex); + + return 0; +} + +int pop_circ_queue(circ_queue * q, unsigned int * val1, unsigned int * val2) +{ + unsigned int currReadIndex; + unsigned int currMaxReadIndex; + + do + { + currReadIndex = atomic_read(&q->readIndex); + currMaxReadIndex = atomic_read(&q->writeIndex); + if (queue_count_to_index(currReadIndex, q->len) == queue_count_to_index(currMaxReadIndex, q->len)) { + // The queue is empty or a producer thread has allocate space in the queue + // but is waiting to commit the data into it + return 1; + } + + // Retrieve the data from the queue + *val1 = q->vals[queue_count_to_index(currReadIndex, q->len)][0]; + *val2 = q->vals[queue_count_to_index(currReadIndex, q->len)][1]; + + // Try to perfrom now the CAS operation on the read index. If we succeed + // label & val already contain what q->readIndex pointed to before we + // increased it. + if (atomic_cmpxchg(&q->readIndex, currReadIndex, currReadIndex+1) == currReadIndex) { + // The lable & val were retrieved from the queue. Note that the + // data inside the label or value arrays are not deleted. + return 0; + } + + // Failed to retrieve the elements off the queue. Someone else must + // have read the element stored at countToIndex(currReadIndex) + // before we could perform the CAS operation. + } while(1); // keep looping to try again! + + return 1; +} + +int circ_queue_empty(circ_queue * q) +{ + unsigned int currReadIndex; + unsigned int currMaxReadIndex; + + currReadIndex = atomic_read(&q->readIndex); + currMaxReadIndex = atomic_read(&q->writeIndex); + if (queue_count_to_index(currReadIndex, q->len) == queue_count_to_index(currMaxReadIndex, q->len)) { + // The queue is empty or a producer thread has allocate space in the queue + // but is waiting to commit the data into it + return 1; + } + return 0; +} + +int circ_queue_full(circ_queue * q) +{ + unsigned int currReadIndex; + unsigned int currWriteIndex; + + currWriteIndex = atomic_read(&q->writeIndex); + currReadIndex = atomic_read(&q->readIndex); + if (queue_count_to_index(currWriteIndex+1, q->len) == queue_count_to_index(currReadIndex, q->len)) { + // The queue is full + return 1; + } + return 0; +} + +void free_circ_queue(circ_queue * q) +{ + int i; + + if (q == NULL) + return; + + for (i = 0; i < q->len; i++) { + kfree(q->vals[i]); + } + kfree(q->vals); + kfree(q); +} + diff --git a/drivers/staging/riffa/circ_queue.h b/drivers/staging/riffa/circ_queue.h new file mode 100644 index 00000000000000..e458cdb6aca837 --- /dev/null +++ b/drivers/staging/riffa/circ_queue.h @@ -0,0 +1,96 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: circ_queue.h + * Version: 1.0 + * Description: A lock-free single-producer circular queue implementation + * modeled after the more elaborate C++ version from Faustino Frechilla at: + * http://www.codeproject.com/Articles/153898/Yet-another-implementation-of-a-lock-free-circular + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 1.0. + */ +#ifndef CIRC_QUEUE_H +#define CIRC_QUEUE_H + +#include + +/* Struct for the circular queue. */ +struct circ_queue { + atomic_t writeIndex; + atomic_t readIndex; + unsigned int ** vals; + unsigned int len; +}; +typedef struct circ_queue circ_queue; + +/** + * Initializes a circ_queue with depth/length len. Returns non-NULL on success, + * NULL if there was a problem creating the queue. + */ +circ_queue * init_circ_queue(int len); + +/** + * Pushes a pair of unsigned int values into the specified queue at the head. + * Returns 0 on success, non-zero if there is no more space in the queue. + */ +int push_circ_queue(circ_queue * q, unsigned int val1, unsigned int val2); + +/** + * Pops a pair of unsigned int values out of the specified queue from the tail. + * Returns 0 on success, non-zero if the queue is empty. + */ +int pop_circ_queue(circ_queue * q, unsigned int * val1, unsigned int * val2); + +/** + * Returns 1 if the circ_queue is empty, 0 otherwise. Note, this is not a + * synchronized function. If another thread is accessing this circ_queue, the + * return value may not be valid. + */ +int circ_queue_empty(circ_queue * q); + +/** + * Returns 1 if the circ_queue is full, 0 otherwise. Note, this is not a + * synchronized function. If another thread is accessing this circ_queue, the + * return value may not be valid. + */ +int circ_queue_full(circ_queue * q); + +/** + * Frees the resources associated with the specified circ_queue. + */ +void free_circ_queue(circ_queue * q); + +#endif diff --git a/drivers/staging/riffa/riffa.c b/drivers/staging/riffa/riffa.c new file mode 100644 index 00000000000000..91b9b9af3dc804 --- /dev/null +++ b/drivers/staging/riffa/riffa.c @@ -0,0 +1,152 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: riffa.c + * Version: 2.0 + * Description: Linux PCIe communications API for RIFFA. + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 2.0. + */ + +#include +#include +#include +#include +#include +#include +#include "riffa.h" + +struct thread_info { /* Used as argument to thread_start() */ + // please refer to API of fpga_send() and fpga_recv() at http://riffa.ucsd.edu/node/10 or https://github.com/KastnerRG/riffa/blob/master/driver/linux/riffa.c#L84-L111 + fpga_t * fpga; + unsigned int chnl; + unsigned int * buffer; + unsigned int len; + unsigned int offset; + unsigned int last; + long long timeout; +}; + +struct fpga_t +{ + int fd; + int id; +}; + +fpga_t * fpga_open(int id) +{ + fpga_t * fpga; + + // Allocate space for the fpga_dev + fpga = (fpga_t *)malloc(sizeof(fpga_t)); + if (fpga == NULL) + return NULL; + fpga->id = id; + + // Open the device file. + fpga->fd = open("/dev/" DEVICE_NAME, O_RDWR | O_SYNC); + if (fpga->fd < 0) { + free(fpga); + return NULL; + } + + return fpga; +} + +void fpga_close(fpga_t * fpga) +{ + // Close the device file. + close(fpga->fd); + free(fpga); +} + +//int fpga_send(fpga_t * fpga, int chnl, void * data, int len, int destoff, int last, long long timeout) +void* fpga_send(void *arg) +{ + struct thread_info *tinfo_send = (struct thread_info *) arg; + + fpga_chnl_io io_send; + + io_send.id = tinfo_send->fpga->id; + io_send.chnl = tinfo_send->chnl; + io_send.len = tinfo_send->len; + io_send.offset = tinfo_send->offset; + io_send.last = tinfo_send->last; + io_send.timeout = tinfo_send->timeout; + io_send.data = (char *)(tinfo_send->buffer); + + int number_of_words_sent = ioctl(tinfo_send->fpga->fd, IOCTL_SEND, &io_send); + + pthread_exit((void *)(intptr_t)number_of_words_sent); +} + +//int fpga_recv(fpga_t * fpga, int chnl, void * data, int len, long long timeout) +void* fpga_recv(void *arg) +{ + struct thread_info *tinfo_recv = (struct thread_info *) arg; + + fpga_chnl_io io_recv; + + io_recv.id = tinfo_recv->fpga->id; + io_recv.chnl = tinfo_recv->chnl; + io_recv.len = tinfo_recv->len; + io_recv.timeout = tinfo_recv->timeout; + io_recv.data = (char *)(tinfo_recv->buffer); + + int number_of_words_recv = ioctl(tinfo_recv->fpga->fd, IOCTL_RECV, &io_recv); + + pthread_exit((void *)(intptr_t)number_of_words_recv); +} + +void fpga_reset(fpga_t * fpga) +{ + ioctl(fpga->fd, IOCTL_RESET, fpga->id); +} + +int fpga_list(fpga_info_list * list) { + int fd; + int rc; + + fd = open("/dev/" DEVICE_NAME, O_RDWR | O_SYNC); + if (fd < 0) + return fd; + rc = ioctl(fd, IOCTL_LIST, list); + close(fd); + return rc; +} + + + diff --git a/drivers/staging/riffa/riffa.h b/drivers/staging/riffa/riffa.h new file mode 100644 index 00000000000000..a8db183bd6b8c8 --- /dev/null +++ b/drivers/staging/riffa/riffa.h @@ -0,0 +1,121 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: riffa.h + * Version: 2.0 + * Description: Linux PCIe communications API for RIFFA. + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 2.0. + */ + +#ifndef RIFFA_H +#define RIFFA_H + +#include "riffa_driver.h" + +#ifdef __cplusplus +extern "C" { +#endif + +struct fpga_t; +typedef struct fpga_t fpga_t; + +/** + * Populates the fpga_info_list pointer with all FPGAs registered in the system. + * Returns 0 on success, a negative value on error. + */ +int fpga_list(fpga_info_list * list); + +/** + * Initializes the FPGA specified by id. On success, returns a pointer to a + * fpga_t struct. On error, returns NULL. Each FPGA must be opened before any + * channels can be accessed. Once opened, any number of threads can use the + * fpga_t struct. + */ +fpga_t * fpga_open(int id); + +/** + * Cleans up memory/resources for the FPGA specified by the fd descriptor. + */ +void fpga_close(fpga_t * fpga); + +/** + * Sends len words (4 byte words) from data to FPGA channel chnl using the + * fpga_t struct. The FPGA channel will be sent len, destoff, and last. If last + * is 1, the channel should interpret the end of this send as the end of a + * transaction. If last is 0, the channel should wait for additional sends + * before the end of the transaction. If timeout is non-zero, this call will + * send data and wait up to timeout ms for the FPGA to respond (between + * packets) before timing out. If timeout is zero, this call may block + * indefinitely. Multiple threads sending on the same channel may result in + * corrupt data or error. This function is thread safe across channels. + * On success, returns the number of words sent. On error returns a negative + * value. + */ +//int fpga_send(fpga_t * fpga, int chnl, void * data, int len, int destoff, int last, long long timeout); +void* fpga_send(void * arg); + +/** + * Receives data from the FPGA channel chnl to the data pointer, using the + * fpga_t struct. The FPGA channel can send any amount of data, so the data + * array should be large enough to accommodate. The len parameter specifies the + * actual size of the data buffer in words (4 byte words). The FPGA channel will + * specify an offset which will determine where in the data array the data will + * start being written. If the amount of data (plus offset) exceed the size of + * the data array (len), then that data will be discarded. If timeout is + * non-zero, this call will wait up to timeout ms for the FPGA to respond + * (between packets) before timing out. If timeout is zero, this call may block + * indefinitely. Multiple threads receiving on the same channel may result in + * corrupt data or error. This function is thread safe across channels. + * On success, returns the number of words written to the data array. On error + * returns a negative value. + */ +//int fpga_recv(fpga_t * fpga, int chnl, void * data, int len, long long timeout); +void* fpga_recv(void * arg); + +/** + * Resets the state of the FPGA and all transfers across all channels. This is + * meant to be used as an alternative to rebooting if an error occurs while + * sending/receiving. Calling this function while other threads are sending or + * receiving will result in unexpected behavior. + */ +void fpga_reset(fpga_t * fpga); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/drivers/staging/riffa/riffa_driver.c b/drivers/staging/riffa/riffa_driver.c new file mode 100644 index 00000000000000..a5a58d7edc1d15 --- /dev/null +++ b/drivers/staging/riffa/riffa_driver.c @@ -0,0 +1,1643 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: riffa_driver.c + * Version: 2.0 + * Description: Linux PCIe device driver for RIFFA. Uses Linux kernel APIs in + * version 2.6.27+ (tested on version 2.6.32 - 3.3.0). + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 2.0. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4,11,0) +#include +#else +#include +#endif + +#include +#include +#include +#include +#include +#include +#include "riffa_driver.h" +#include "circ_queue.h" + +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION("PCIe driver for RIFFA, Linux (2.6.27+)"); +MODULE_AUTHOR("Matt Jacobsen, Patrick Lai"); + +#ifndef __devinit +#define __devinit +#define __devexit +#define __devexit_p +#endif + +#define CHNL_REG(c, o) ((c<<4) + o) +#if !defined(__LP64__) && !defined(_LP64) +#define BUILD_32 1 +#endif + +#define CHNL_FLAG_BUSY 0 + +struct sg_mapping { + struct page ** pages; + struct scatterlist * sgl; + enum dma_data_direction direction; + int num_sg; + unsigned long num_pages; + unsigned long long length; + unsigned long long overflow; +}; + +struct chnl_dir { + volatile unsigned long flags; + wait_queue_head_t waitq; + struct circ_queue * msgs; + void * buf_addr; + dma_addr_t buf_hw_addr; + struct sg_mapping * sg_map_0; + struct sg_mapping * sg_map_1; +}; + +struct fpga_state { + struct pci_dev * dev; + unsigned long long irq; + void __iomem *bar0; + unsigned long long bar0_addr; + unsigned long long bar0_len; + unsigned long long bar0_flags; + atomic_t intr_disabled; + void * spill_buf_addr; + dma_addr_t spill_buf_hw_addr; + int num_sg; + int sg_buf_size; + int id; + char name[16]; + int vendor_id; + int device_id; + int num_chnls; + struct chnl_dir ** recv; + struct chnl_dir ** send; +}; + +// Global variables (to this file only) +static struct class * mymodule_class; +static dev_t devt; +static atomic_t used_fpgas[NUM_FPGAS]; +static struct fpga_state * fpgas[NUM_FPGAS]; + +static unsigned int tx_len; +static bool recv_sg_buf_populated; + +/////////////////////////////////////////////////////// +// MEMORY ALLOCATION & HELPER FUNCTIONS +/////////////////////////////////////////////////////// + +/** + * Returns the value at the specified address. + */ +static inline unsigned int read_reg(struct fpga_state * sc, int offset) +{ + return readl(sc->bar0 + (offset<<2)); +} + +/** + * Writes the value to the specified address. + */ +static inline void write_reg(struct fpga_state * sc, int offset, unsigned int val) +{ + writel(val, sc->bar0 + (offset<<2)); +} + +#ifdef BUILD_32 +/** + * Needed for 32 bit OS because dma_map_sg macro eventually does some 64 bit + * division. + */ +unsigned long long __udivdi3(unsigned long long num, unsigned long long den) +{ + do_div(num, den); + return num; +} +#endif + + +// These are not defined in the 2.x.y kernels, so just define them +#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,39) +#define PCI_EXP_DEVCTL2_IDO_REQ_EN 0x100 +#define PCI_EXP_DEVCTL2_IDO_CMP_EN 0x200 +#else +/** + * These are badly named in pre-3.6.11 kernel versions. We COULD do the same + * check as above, however (annoyingly) linux for tegra (based on post-3.6.11) + * picked up the header file from some pre-3.6.11 version, so we'll just make + * our code ugly and handle the check here: + */ +#ifndef PCI_EXP_DEVCTL2_IDO_REQ_EN +#define PCI_EXP_DEVCTL2_IDO_REQ_EN PCI_EXP_IDO_REQ_EN +#endif +#ifndef PCI_EXP_DEVCTL2_IDO_CMP_EN +#define PCI_EXP_DEVCTL2_IDO_CMP_EN PCI_EXP_IDO_CMP_EN +#endif +#endif + +#if LINUX_VERSION_CODE <= KERNEL_VERSION(3,6,11) +/** + * Code used to set ETB and RCB, but not available before 3.0, or incorrectly + * defined before 3.7. As it is peppered throughout the clean up code, it's just + * easier to copy the declarations (not verbatim) here than a bunch of conditionals + * everywhere else. + */ + +int pcie_capability_read_word(struct pci_dev *dev, int pos, u16 *val) +{ + int ret; + + *val = 0; + if (pos & 1) + return -EINVAL; + + ret = pci_read_config_word(dev, pci_pcie_cap(dev) + pos, val); + /* + * Reset *val to 0 if pci_read_config_word() fails, it may + * have been written as 0xFFFF if hardware error happens + * during pci_read_config_word(). + */ + if (ret) + *val = 0; + return ret; +} + +int pcie_capability_read_dword(struct pci_dev *dev, int pos, u32 *val) +{ + int ret; + + *val = 0; + if (pos & 3) + return -EINVAL; + + ret = pci_read_config_dword(dev, pci_pcie_cap(dev) + pos, val); + /* + * Reset *val to 0 if pci_read_config_dword() fails, it may + * have been written as 0xFFFFFFFF if hardware error happens + * during pci_read_config_dword(). + */ + if (ret) + *val = 0; + return ret; + +} + +int pcie_capability_write_word(struct pci_dev *dev, int pos, u16 val) +{ + if (pos & 1) + return -EINVAL; + + return pci_write_config_word(dev, pci_pcie_cap(dev) + pos, val); +} + +int pcie_capability_write_dword(struct pci_dev *dev, int pos, u32 val) +{ + if (pos & 3) + return -EINVAL; + + return pci_write_config_dword(dev, pci_pcie_cap(dev) + pos, val); +} +#endif + + + + +/////////////////////////////////////////////////////// +// INTERRUPT HANDLER +/////////////////////////////////////////////////////// + +/** + * Reads the interrupt vector and processes it. If processing VECT0, off will + * be 0. If processing VECT1, off will be 6. + */ +static inline void process_intr_vector(struct fpga_state * sc, int off, + unsigned int vect) +{ + // VECT_0/VECT_1 are organized from right to left (LSB to MSB) as: + // [ 0] TX_TXN for channel 0 in VECT_0, channel 6 in VECT_1 + // [ 1] TX_SG_BUF_RECVD for channel 0 in VECT_0, channel 6 in VECT_1 + // [ 2] TX_TXN_DONE for channel 0 in VECT_0, channel 6 in VECT_1 + // [ 3] RX_SG_BUF_RECVD for channel 0 in VECT_0, channel 6 in VECT_1 + // [ 4] RX_TXN_DONE for channel 0 in VECT_0, channel 6 in VECT_1 + // ... + // [25] TX_TXN for channel 5 in VECT_0, channel 11 in VECT_1 + // [26] TX_SG_BUF_RECVD for channel 5 in VECT_0, channel 11 in VECT_1 + // [27] TX_TXN_DONE for channel 5 in VECT_0, channel 11 in VECT_1 + // [28] RX_SG_BUF_RECVD for channel 5 in VECT_0, channel 11 in VECT_1 + // [29] RX_TXN_DONE for channel 5 in VECT_0, channel 11 in VECT_1 + // Positions 30 - 31 in both VECT_0 and VECT_1 are zero. + + unsigned int offlast; + unsigned int len; + int recv; + int send; + int chnl; + int i; + +//printk(KERN_INFO "riffa: intrpt_handler received:%08x\n", vect); + if (vect & 0xC0000000) { + printk(KERN_ERR "riffa: fpga:%d, received bad interrupt vector:%08x\n", sc->id, vect); + return; + } + + for (i = 0; i < 6 && (i+off) < sc->num_chnls; ++i) { + chnl = i + off; + recv = 0; + send = 0; + + // TX (PC receive) scatter gather buffer is read. + if (vect & (1<<((5*i)+1))) { + recv = 1; + // Keep track so the thread can handle this. + if (push_circ_queue(sc->recv[chnl]->msgs, EVENT_SG_BUF_READ, 0)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv sg buf read msg queue full\n", sc->id, chnl); + } + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, recv sg buf read\n", sc->id, chnl); + } + + // TX (PC receive) transaction done. + if (vect & (1<<((5*i)+2))) { + recv = 1; + // Read the transferred amount. + len = read_reg(sc, CHNL_REG(chnl, TX_TNFR_LEN_REG_OFF)); + // Notify the thread. + if (push_circ_queue(sc->recv[chnl]->msgs, EVENT_TXN_DONE, len)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv txn done msg queue full\n", sc->id, chnl); + } + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, recv txn done\n", sc->id, chnl); + } + + // New TX (PC receive) transaction. + if (vect & (1<<((5*i)+0))) { + recv = 1; + recv_sg_buf_populated = 0; // resets for new transaction + + // Read the offset/last and length + offlast = read_reg(sc, CHNL_REG(chnl, TX_OFFLAST_REG_OFF)); + tx_len = read_reg(sc, CHNL_REG(chnl, TX_LEN_REG_OFF)); + // Keep track of this transaction + if (push_circ_queue(sc->recv[chnl]->msgs, EVENT_TXN_OFFLAST, offlast)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv txn offlast msg queue full\n", sc->id, chnl); + } + /*if (push_circ_queue(sc->recv[chnl]->msgs, EVENT_TXN_LEN, len)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv txn len msg queue full\n", sc->id, chnl); + }*/ + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, recv txn (len:%d off:%d last:%d)\n", sc->id, chnl, tx_len, (offlast>>1), (offlast & 0x1)); + } + + // RX (PC send) scatter gather buffer is read. + if (vect & (1<<((5*i)+3))) { + send = 1; + // Keep track so the thread can handle this. + if (push_circ_queue(sc->send[chnl]->msgs, EVENT_SG_BUF_READ, 0)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, send sg buf read msg queue full\n", sc->id, chnl); + } + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, send sg buf read\n", sc->id, chnl); + } + + // RX (PC send) transaction done. + if (vect & (1<<((5*i)+4))) { + send = 1; + // Read the transferred amount. + len = read_reg(sc, CHNL_REG(chnl, RX_TNFR_LEN_REG_OFF)); + // Notify the thread. + if (push_circ_queue(sc->send[chnl]->msgs, EVENT_TXN_DONE, len)) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, send txn done msg queue full\n", sc->id, chnl); + } + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, send txn done\n", sc->id, chnl); + } + + // Wake up the thread? + if (recv) + wake_up(&sc->recv[chnl]->waitq); + if (send) + wake_up(&sc->send[chnl]->waitq); + } +} + +/** + * Interrupt handler for all interrupts on all files. Reads data/values + * from FPGA and wakes up waiting threads to process the data. Always returns + * IRQ_HANDLED. + */ +static irqreturn_t intrpt_handler(int irq, void *dev_id) +{ + unsigned int vect0; + unsigned int vect1; + struct fpga_state * sc; + + sc = (struct fpga_state *)dev_id; + vect0 = 0; + vect1 = 0; + + if (sc == NULL) { + printk(KERN_ERR "riffa: invalid fpga_state pointer\n"); + return IRQ_HANDLED; + } + + if (!atomic_read(&sc->intr_disabled)) { + // Read the interrupt vector(s): + vect0 = read_reg(sc, IRQ_REG0_OFF); + if (sc->num_chnls > 6) + vect1 = read_reg(sc, IRQ_REG1_OFF); + + // Process the vector(s) + process_intr_vector(sc, 0, vect0); + if (sc->num_chnls > 6) + process_intr_vector(sc, 6, vect1); + } + + return IRQ_HANDLED; +} + + +/////////////////////////////////////////////////////// +// FILE OPERATION HANDLERS +/////////////////////////////////////////////////////// + +/** + * Creates and returns a struct_sg_mapping that holds all the data for the user + * pages that have been mapped into a scatterlist array. Assumes the user data + * is 32 bit word aligned. Up to length bytes from the udata pointer will be + * mapped. After all length bytes are mapped, up to overflow bytes will be + * mapped using the common spill buffer for the channel. The overflow is used + * if we run out of space in the supplied udata pointer. + */ +static inline struct sg_mapping * fill_sg_buf(struct fpga_state * sc, int chnl, + void * sg_buf, unsigned long udata, unsigned long long length, + unsigned long long overflow, enum dma_data_direction direction) { + const char * dir = (direction == DMA_TO_DEVICE ? "send" : "recv"); + struct sg_mapping * sg_map; + struct page ** pages = NULL; + struct scatterlist * sgl = NULL; + struct scatterlist * sg; + unsigned long num_pages_reqd = 0; + long num_pages = 0; + unsigned int fp_offset; + unsigned int len; + unsigned int hw_len; + dma_addr_t hw_addr; + unsigned long long len_rem = length; + unsigned long long overflow_rem = overflow; + unsigned int * sg_buf_ptr = (unsigned int *)sg_buf; + int num_sg = 0; + int i; + + // Create the sg_mapping struct. + if ((sg_map = (struct sg_mapping *)kmalloc(sizeof(*sg_map), GFP_KERNEL)) == NULL) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, %s could not allocate memory for sg_mapping struct\n", sc->id, chnl, dir); + return NULL; + } + + if (length > 0) { + // Create the pages array. + num_pages_reqd = ((udata + length - 1)>>PAGE_SHIFT) - (udata>>PAGE_SHIFT) + 1; + num_pages_reqd = (num_pages_reqd > sc->num_sg ? sc->num_sg : num_pages_reqd); + if ((pages = kmalloc(num_pages_reqd * sizeof(*pages), GFP_KERNEL)) == NULL) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, %s could not allocate memory for pages array\n", sc->id, chnl, dir); + kfree(sg_map); + return NULL; + } + + // Page in the user pages. + down_read(¤t->mm->mmap_sem); + #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0) + num_pages = get_user_pages(current, current->mm, udata, num_pages_reqd, 1, 0, pages, NULL); + #elsif LINUX_VERSION_CODE < KERNEL_VERSION(4,9,0) + num_pages = get_user_pages(udata, num_pages_reqd, 1, 0, pages, NULL); + #else + num_pages = get_user_pages(udata, num_pages_reqd, FOLL_WRITE, pages, NULL); + #endif + up_read(¤t->mm->mmap_sem); + if (num_pages <= 0) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, %s unable to pin any pages in memory\n", sc->id, chnl, dir); + kfree(pages); + kfree(sg_map); + return NULL; + } + + // Create the scatterlist array. + if ((sgl = kcalloc(num_pages, sizeof(*sgl), GFP_KERNEL)) == NULL) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, %s could not allocate memory for scatterlist array\n", sc->id, chnl, dir); + for (i = 0; i < num_pages; ++i) + #if LINUX_VERSION_CODE < KERNEL_VERSION(4, 6, 0) + page_cache_release(pages[i]); + #else + put_page(pages[i]); + #endif + kfree(pages); + kfree(sg_map); + return NULL; + } + + // Set the scatterlist values + fp_offset = (udata & (~PAGE_MASK)); + sg_init_table(sgl, num_pages); + for (i = 0; i < num_pages; ++i) { + len = ((fp_offset + len_rem) > PAGE_SIZE ? (PAGE_SIZE - fp_offset) : len_rem); + sg_set_page(&sgl[i], pages[i], len, fp_offset); + len_rem -= len; + fp_offset = 0; + } + + // Map the scatterlist values and write to the common buffer area + num_sg = dma_map_sg(&sc->dev->dev, sgl, num_pages, direction); + for_each_sg(sgl, sg, num_sg, i) { + hw_addr = sg_dma_address(sg); + hw_len = sg_dma_len(sg); + sg_buf_ptr[(i*4)+0] = (hw_addr & 0xFFFFFFFF); + sg_buf_ptr[(i*4)+1] = ((hw_addr>>32) & 0xFFFFFFFF); + sg_buf_ptr[(i*4)+2] = hw_len>>2; // Words! + } + } + + // Provide scatter gather mappings for overflow data (all in spill common buffer) + while (len_rem == 0 && overflow_rem > 0 && num_sg < sc->num_sg) { + sg_buf_ptr[(num_sg*4)+0] = (sc->spill_buf_hw_addr & 0xFFFFFFFF); + sg_buf_ptr[(num_sg*4)+1] = ((sc->spill_buf_hw_addr>>32) & 0xFFFFFFFF); + sg_buf_ptr[(num_sg*4)+2] = SPILL_BUF_SIZE>>2; // Words! + num_sg++; + overflow_rem -= (SPILL_BUF_SIZE > overflow ? overflow : SPILL_BUF_SIZE); + } + + // Populate the number of bytes mapped and other sg data + sg_map->direction = direction; + sg_map->num_pages = num_pages; + sg_map->num_sg = num_sg; + sg_map->length = (length - len_rem); + sg_map->overflow = (overflow - overflow_rem); + sg_map->pages = pages; + sg_map->sgl = sgl; + + return sg_map; +} + +/** + * Frees the scatterlist mappings in the struct sg_mapping pointer and frees all + * corresponding structs. + */ +static inline void free_sg_buf(struct fpga_state * sc, struct sg_mapping * sg_map) { + int i; + + if (sg_map == NULL) + return; + + // Unmap the pages. + if (sg_map->sgl != NULL) + dma_unmap_sg(&sc->dev->dev, sg_map->sgl, sg_map->num_pages, sg_map->direction); + + // Free the pages (mark dirty if necessary). + if (sg_map->pages != NULL) { + if (sg_map->direction == DMA_FROM_DEVICE) { + for (i = 0; i < sg_map->num_pages; ++i) { + if (!PageReserved(sg_map->pages[i])) + SetPageDirty(sg_map->pages[i]); + #if LINUX_VERSION_CODE < KERNEL_VERSION(4, 6, 0) + page_cache_release(sg_map->pages[i]); + #else + put_page(sg_map->pages[i]); + #endif + } + } + else { + for (i = 0; i < sg_map->num_pages; ++i) { + #if LINUX_VERSION_CODE < KERNEL_VERSION(4, 6, 0) + page_cache_release(sg_map->pages[i]); + #else + put_page(sg_map->pages[i]); + #endif + } + } + } + + // Free the structures. + if (sg_map->pages != NULL) + kfree(sg_map->pages); + if (sg_map->sgl != NULL) + kfree(sg_map->sgl); + kfree(sg_map); +} + +/** + * Reads data from the FPGA. Will block until all the data is received from the + * FPGA unless timeout is non-zero. If timeout is non-zero, the function will + * block until all the data is received or until the timeout expires. Received + * data will be written directly into the user buffer, bufp, by the DMA process + * (using scatter gather). Up to len words (each word == 32 bits) will be + * written. On success, the number of words received are returned. On error, + * returns a negative value. + */ +static inline unsigned int chnl_recv(struct fpga_state * sc, int chnl, + char __user * bufp, unsigned int len, unsigned long long timeout) +{ + struct sg_mapping * sg_map; + long tymeouto; + long tymeout; + int nomsg; + unsigned int msg_type; + unsigned int msg; + int last = -1; + unsigned long long offset = 0; + unsigned long long length = 0; + unsigned long long overflow = 0; + unsigned long long capacity = (((unsigned long long)len)<<2); + unsigned long long recvd = 0; + unsigned long udata = (unsigned long)bufp; + unsigned long max_ptr; + + DEFINE_WAIT(wait); + + // Convert timeout to jiffies. + tymeout = (timeout == 0 ? MAX_SCHEDULE_TIMEOUT : (timeout * HZ/1000 > LONG_MAX ? LONG_MAX : timeout * HZ/1000)); + tymeouto = tymeout; + + // Initialize the sg_maps + sc->recv[chnl]->sg_map_0 = NULL; + sc->recv[chnl]->sg_map_1 = NULL; + + // Continue until we get a message or timeout. + while (1) { + while ((nomsg = pop_circ_queue(sc->recv[chnl]->msgs, &msg_type, &msg))) { + prepare_to_wait(&sc->recv[chnl]->waitq, &wait, TASK_INTERRUPTIBLE); + // Another check before we schedule. + if ((nomsg = pop_circ_queue(sc->recv[chnl]->msgs, &msg_type, &msg))) + tymeout = schedule_timeout(tymeout); + finish_wait(&sc->recv[chnl]->waitq, &wait); + if (signal_pending(current)) { + free_sg_buf(sc, sc->recv[chnl]->sg_map_0); + free_sg_buf(sc, sc->recv[chnl]->sg_map_1); + return -ERESTARTSYS; + } + if (!nomsg) + break; + if (tymeout == 0) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv timed out\n", sc->id, chnl); + /*free_sg_buf(sc, sc->recv[chnl]->sg_map_0); + free_sg_buf(sc, sc->recv[chnl]->sg_map_1); + return (unsigned int)(recvd>>2);*/ + } + } + tymeout = tymeouto; + DEBUG_MSG(KERN_INFO "msg_type: %d\n", msg_type); // added by cheng fei + // Process the message. + switch (msg_type) { + case EVENT_TXN_OFFLAST: + // Read the offset and last flags (always before reading length) + offset = (((unsigned long long)(msg>>1))<<2); + last = (msg & 0x1); + //break; + + //case EVENT_TXN_LEN: + // Read the length + //length = (((unsigned long long)msg)<<2); + length = tx_len << 2; + recvd = 0; + overflow = 0; + // Check for address overflow + max_ptr = (unsigned long)(udata + offset + length - 1); + if (max_ptr < udata) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv pointer address overflow\n", sc->id, chnl); + overflow = length; + length = 0; + } + // Check for capacity overflow + if ((offset + length) > capacity) { + if (offset > capacity) { + overflow = length; + length = 0; + } + else { + overflow = length + offset - capacity; + length = capacity - offset; + } + } + // Use the recv common buffer to share the scatter gather elements. + if (length > 0 || overflow > 0) { + udata = udata + offset; + sg_map = fill_sg_buf(sc, chnl, sc->recv[chnl]->buf_addr, udata, length, overflow, DMA_FROM_DEVICE); + if (sg_map == NULL || sg_map->num_sg == 0) + return (unsigned int)(recvd>>2); + // Update based on the sg_mapping + udata += sg_map->length; + length -= sg_map->length; + overflow -= sg_map->overflow; + sc->recv[chnl]->sg_map_1 = sg_map; + // Let FPGA know about the scatter gather buffer. + write_reg(sc, CHNL_REG(chnl, TX_SG_ADDR_LO_REG_OFF), (sc->recv[chnl]->buf_hw_addr & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, TX_SG_ADDR_HI_REG_OFF), ((sc->recv[chnl]->buf_hw_addr>>32) & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, TX_SG_LEN_REG_OFF), 4 * sg_map->num_sg); + + recv_sg_buf_populated = 1; + + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, recv sg buf populated, %d sent\n", sc->id, chnl, sg_map->num_sg); + + wake_up(&sc->send[chnl]->waitq); // https://elixir.bootlin.com/linux/v4.19-rc7/source/include/linux/wait.h#L476 + // The @condition is checked each time the waitqueue @wq_head is woken up. wake_up() has to be called after changing any variable that could change the result of the wait condition. + } + break; + + case EVENT_SG_BUF_READ: + // Ignore if we haven't received offlast/len. + if (last == -1) + break; + // Release the previous scatter gather data. + if (sc->recv[chnl]->sg_map_0 != NULL) + recvd += sc->recv[chnl]->sg_map_0->length; + free_sg_buf(sc, sc->recv[chnl]->sg_map_0); + sc->recv[chnl]->sg_map_0 = NULL; + // Populate the common buffer with more scatter gather data? + if (length > 0 || overflow > 0) { + sg_map = fill_sg_buf(sc, chnl, sc->recv[chnl]->buf_addr, udata, length, overflow, DMA_FROM_DEVICE); + if (sg_map == NULL || sg_map->num_sg == 0) { + free_sg_buf(sc, sc->recv[chnl]->sg_map_0); + free_sg_buf(sc, sc->recv[chnl]->sg_map_1); + return (unsigned int)(recvd>>2); + } + // Update based on the sg_mapping + udata += sg_map->length; + length -= sg_map->length; + overflow -= sg_map->overflow; + sc->recv[chnl]->sg_map_0 = sc->recv[chnl]->sg_map_1; + sc->recv[chnl]->sg_map_1 = sg_map; + write_reg(sc, CHNL_REG(chnl, TX_SG_ADDR_LO_REG_OFF), (sc->recv[chnl]->buf_hw_addr & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, TX_SG_ADDR_HI_REG_OFF), ((sc->recv[chnl]->buf_hw_addr>>32) & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, TX_SG_LEN_REG_OFF), 4 * sg_map->num_sg); + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, recv sg buf populated, %d sent\n", sc->id, chnl, sg_map->num_sg); + } + break; + + case EVENT_TXN_DONE: + recv_sg_buf_populated = 0; // resets recv sg buf parameters for next transaction. + + // Ignore if we haven't received offlast/len. + if (last == -1) + break; + // Update with the true value of words transferred. + recvd = (((unsigned long long)msg)<<2); + // Return if this was the last transaction. + free_sg_buf(sc, sc->recv[chnl]->sg_map_0); + free_sg_buf(sc, sc->recv[chnl]->sg_map_1); + sc->recv[chnl]->sg_map_0 = NULL; + sc->recv[chnl]->sg_map_1 = NULL; + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, received %d words\n", sc->id, chnl, (unsigned int)(recvd>>2)); + if (last) + return (unsigned int)(recvd>>2); + break; + + default: + printk(KERN_ERR "riffa: fpga:%d chnl:%d, received unknown msg: %08x\n", sc->id, chnl, msg); + break; + } + } + + return 0; +} + +static inline unsigned int chnl_recv_wrapcheck(struct fpga_state * sc, int chnl, + char __user * bufp, unsigned int len, unsigned long long timeout) +{ + unsigned long udata = (unsigned long)bufp; + unsigned int ret; + + // Validate the parameters. + if (chnl >= sc->num_chnls || chnl < 0) { + printk(KERN_INFO "riffa: fpga:%d chnl:%d, recv channel invalid!\n", sc->id, chnl); + return 0; + } + if (udata & 0x3) { + printk(KERN_INFO "riffa: fpga:%d chnl:%d, recv user buffer must be 32 bit word aligned!\n", sc->id, chnl); + return -EINVAL; + } + + // Ensure no simultaneous operations from several threads + if (test_and_set_bit(CHNL_FLAG_BUSY, &sc->recv[chnl]->flags) != 0) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, recv conflict between threads!\n", sc->id, chnl); + return -EBUSY; + } + + ret = chnl_recv(sc, chnl, bufp, len, timeout); + + // Clear the busy flag + clear_bit(CHNL_FLAG_BUSY, &sc->recv[chnl]->flags); + + return ret; +} + +/** + * Writes data to the FPGA channel specified. Will block until all the data is + * sent to the FPGA unless a non-zero timeout is configured. If timeout is non- + * zero, then the function will block until all data is sent or when the timeout + * ms elapses. User data from the bufp pointer will be sent, up to len words + * (each word == 32 bits). The channel will be told how much data to expect and + * at what offset. If last == 1, the FPGA channel will recognize this + * transaction as complete after sending. If last == 0, the FPGA channel will + * expect additional transactions. On success, returns the number of words sent. + * On error, returns a negative value. + */ +static inline unsigned int chnl_send(struct fpga_state * sc, int chnl, + const char __user * bufp, unsigned int len, unsigned int offset, + unsigned int last, unsigned long long timeout) +{ + struct sg_mapping * sg_map; + long tymeouto; + long tymeout; + int nomsg; + unsigned int msg_type; + unsigned int msg; + unsigned long long sent = 0; + unsigned long long length = (((unsigned long long)len)<<2); + unsigned long udata = (unsigned long)bufp; + + DEFINE_WAIT(wait); + + // Convert timeout to jiffies. + tymeout = (timeout == 0 ? MAX_SCHEDULE_TIMEOUT : (timeout * HZ/1000 > LONG_MAX ? LONG_MAX : timeout * HZ/1000)); + tymeouto = tymeout; + + // Clear the message queue. + while (!pop_circ_queue(sc->send[chnl]->msgs, &msg_type, &msg)); + + // Initialize the sg_maps + sc->send[chnl]->sg_map_0 = NULL; + sc->send[chnl]->sg_map_1 = NULL; + + // Let FPGA know about transfer. + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, send (len:%d off:%d last:%d)\n", sc->id, chnl, len, offset, last); + write_reg(sc, CHNL_REG(chnl, RX_OFFLAST_REG_OFF), ((offset<<1) | last)); + write_reg(sc, CHNL_REG(chnl, RX_LEN_REG_OFF), len); + if (len == 0) + return 0; + + // Use the send common buffer to share the scatter gather data + sg_map = fill_sg_buf(sc, chnl, sc->send[chnl]->buf_addr, udata, length, 0, DMA_TO_DEVICE); + if (sg_map == NULL || sg_map->num_sg == 0) + return (unsigned int)(sent>>2); + + // Update based on the sg_mapping + udata += sg_map->length; + length -= sg_map->length; + sc->send[chnl]->sg_map_1 = sg_map; + + if(tx_len > 0) { // FPGA initiates new Tx transaction, so "yield" to software chnl_recv() thread + + // gives time for software chnl_recv() thread to populate recv sg buf parameter + wait_event_interruptible_timeout(sc->send[chnl]->waitq, (recv_sg_buf_populated == 1), timeout); + } + + // Let FPGA know about the scatter gather buffer. + write_reg(sc, CHNL_REG(chnl, RX_SG_ADDR_LO_REG_OFF), (sc->send[chnl]->buf_hw_addr & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, RX_SG_ADDR_HI_REG_OFF), ((sc->send[chnl]->buf_hw_addr>>32) & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, RX_SG_LEN_REG_OFF), 4 * sg_map->num_sg); + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, send sg buf populated, %d sent\n", sc->id, chnl, sg_map->num_sg); + + // Continue until we get a message or timeout. + while (1) { + while ((nomsg = pop_circ_queue(sc->send[chnl]->msgs, &msg_type, &msg))) { + prepare_to_wait(&sc->send[chnl]->waitq, &wait, TASK_INTERRUPTIBLE); + // Another check before we schedule. + if ((nomsg = pop_circ_queue(sc->send[chnl]->msgs, &msg_type, &msg))) + tymeout = schedule_timeout(tymeout); + finish_wait(&sc->send[chnl]->waitq, &wait); + if (signal_pending(current)) { + free_sg_buf(sc, sc->send[chnl]->sg_map_0); + free_sg_buf(sc, sc->send[chnl]->sg_map_1); + return -ERESTARTSYS; + } + if (!nomsg) + break; + if (tymeout == 0) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, send timed out\n", sc->id, chnl); + /*free_sg_buf(sc, sc->send[chnl]->sg_map_0); + free_sg_buf(sc, sc->send[chnl]->sg_map_1); + return (unsigned int)(sent>>2);*/ + } + } + tymeout = tymeouto; + + // Process the message. + switch (msg_type) { + case EVENT_SG_BUF_READ: + // Release the previous scatter gather data? + if (sc->send[chnl]->sg_map_0 != NULL) + sent += sc->send[chnl]->sg_map_0->length; + free_sg_buf(sc, sc->send[chnl]->sg_map_0); + sc->send[chnl]->sg_map_0 = NULL; + // Populate the common buffer with more scatter gather data? + if (length > 0) { + sg_map = fill_sg_buf(sc, chnl, sc->send[chnl]->buf_addr, udata, length, 0, DMA_TO_DEVICE); + if (sg_map == NULL || sg_map->num_sg == 0) { + free_sg_buf(sc, sc->send[chnl]->sg_map_0); + free_sg_buf(sc, sc->send[chnl]->sg_map_1); + return (unsigned int)(sent>>2); + } + // Update based on the sg_mapping + udata += sg_map->length; + length -= sg_map->length; + sc->send[chnl]->sg_map_0 = sc->send[chnl]->sg_map_1; + sc->send[chnl]->sg_map_1 = sg_map; + write_reg(sc, CHNL_REG(chnl, RX_SG_ADDR_LO_REG_OFF), (sc->send[chnl]->buf_hw_addr & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, RX_SG_ADDR_HI_REG_OFF), ((sc->send[chnl]->buf_hw_addr>>32) & 0xFFFFFFFF)); + write_reg(sc, CHNL_REG(chnl, RX_SG_LEN_REG_OFF), 4 * sg_map->num_sg); + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, send sg buf populated, %d sent\n", sc->id, chnl, sg_map->num_sg); + } + break; + + case EVENT_TXN_DONE: + // Update with the true value of words transferred. + sent = (((unsigned long long)msg)<<2); + // Return as this is the end of the transaction. + free_sg_buf(sc, sc->send[chnl]->sg_map_0); + free_sg_buf(sc, sc->send[chnl]->sg_map_1); + DEBUG_MSG(KERN_INFO "riffa: fpga:%d chnl:%d, sent %d words\n", sc->id, chnl, (unsigned int)(sent>>2)); + return (unsigned int)(sent>>2); + break; + + default: + printk(KERN_ERR "riffa: fpga:%d chnl:%d, received unknown msg: %08x\n", sc->id, chnl, msg); + break; + } + } + + return 0; +} + +static inline unsigned int chnl_send_wrapcheck(struct fpga_state * sc, int chnl, + const char __user * bufp, unsigned int len, unsigned int offset, + unsigned int last, unsigned long long timeout) +{ + unsigned long long length = (((unsigned long long)len)<<2); + unsigned long udata = (unsigned long)bufp; + unsigned long max_ptr; + unsigned int ret; + + // Validate the parameters. + if (chnl >= sc->num_chnls || chnl < 0) { + printk(KERN_INFO "riffa: fpga:%d chnl:%d, send channel invalid!\n", sc->id, chnl); + return 0; + } + max_ptr = (unsigned long)(udata + length - 1); + if (max_ptr < udata) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, send pointer address overflow\n", sc->id, chnl); + return -EINVAL; + } + if (udata & 0x3) { + printk(KERN_INFO "riffa: fpga:%d chnl:%d, send user buffer must be 32 bit word aligned!\n", sc->id, chnl); + return -EINVAL; + } + + // Ensure no simultaneous operations from several threads + if (test_and_set_bit(CHNL_FLAG_BUSY, &sc->send[chnl]->flags) != 0) { + printk(KERN_ERR "riffa: fpga:%d chnl:%d, send conflict between threads!\n", sc->id, chnl); + return -EBUSY; + } + + ret = chnl_send(sc, chnl, bufp, len, offset, last, timeout); + + // Clear the busy flag + clear_bit(CHNL_FLAG_BUSY, &sc->send[chnl]->flags); + + return ret; +} + +/** + * Populates the fpga_info struct with the current FPGA state information. On + * success, returns 0. On error, returns a negative value. + */ +static inline int list_fpgas(fpga_info_list * list) +{ + int i; + int num_fpgas = 0; + struct fpga_state * sc; + + for (i = 0; i < NUM_FPGAS; ++i) { + if (atomic_read(&used_fpgas[i])) { + sc = fpgas[i]; + list->id[num_fpgas] = sc->id; + list->num_chnls[num_fpgas] = sc->num_chnls; + list->vendor_id[num_fpgas] = sc->vendor_id; + list->device_id[num_fpgas] = sc->device_id; + memcpy(list->name[num_fpgas], sc->name, 16*sizeof(char)); + num_fpgas++; + } + } + // Zero out the rest + for (i = num_fpgas; i < NUM_FPGAS; ++i) { + list->id[i] = -1; + list->num_chnls[i] = 0; + list->vendor_id[i] = 0; + list->device_id[i] = 0; + memset(list->name[i], 0, 16*sizeof(char)); + } + list->num_fpgas = num_fpgas; + + return 0; +} + +/** + * Resets the driver for the specified FPGA. The fpga_state struct for all + * channels will be reset as will the FPGA itself. + */ +static inline void reset(int id) +{ + int i; + unsigned int dummy0; + unsigned int dummy1; + struct fpga_state * sc; + + if (atomic_read(&used_fpgas[id])) { + sc = fpgas[id]; + // Disable interrupts + atomic_set(&sc->intr_disabled, 1); + // Reset the FPGA + read_reg(sc, INFO_REG_OFF); + + // Reset all the channels + for (i = 0; i < sc->num_chnls; ++i) { + while (!pop_circ_queue(sc->send[i]->msgs, &dummy0, &dummy1)); + while (!pop_circ_queue(sc->recv[i]->msgs, &dummy0, &dummy1)); + + // resets read and write pointers of the circular queue + atomic_set(&sc->recv[i]->msgs->writeIndex, 0); + atomic_set(&sc->recv[i]->msgs->readIndex, 0); + atomic_set(&sc->send[i]->msgs->writeIndex, 0); + atomic_set(&sc->send[i]->msgs->readIndex, 0); + + wake_up(&sc->send[i]->waitq); + wake_up(&sc->recv[i]->waitq); + clear_bit(CHNL_FLAG_BUSY, &sc->send[i]->flags); + clear_bit(CHNL_FLAG_BUSY, &sc->recv[i]->flags); + } + // Enable interrupts + atomic_set(&sc->intr_disabled, 0); + } +} + +/** + * Main entry point for reading and writing on the device. Return value depends + * on ioctlnum and expected behavior. See code for details. + */ +static long fpga_ioctl(struct file *filp, unsigned int ioctlnum, + unsigned long ioctlparam) +{ + int rc; + fpga_chnl_io io; + fpga_info_list list; + + switch (ioctlnum) { + case IOCTL_SEND: + if ((rc = copy_from_user(&io, (void *)ioctlparam, sizeof(fpga_chnl_io)))) { + printk(KERN_ERR "riffa: cannot read ioctl user parameter.\n"); + return rc; + } + if (io.id < 0 || io.id >= NUM_FPGAS || !atomic_read(&used_fpgas[io.id])) + return 0; + return chnl_send_wrapcheck(fpgas[io.id], io.chnl, io.data, io.len, io.offset, + io.last, io.timeout); + case IOCTL_RECV: + if ((rc = copy_from_user(&io, (void *)ioctlparam, sizeof(fpga_chnl_io)))) { + printk(KERN_ERR "riffa: cannot read ioctl user parameter.\n"); + return rc; + } + if (io.id < 0 || io.id >= NUM_FPGAS || !atomic_read(&used_fpgas[io.id])) + return 0; + return chnl_recv_wrapcheck(fpgas[io.id], io.chnl, io.data, io.len, io.timeout); + case IOCTL_LIST: + list_fpgas(&list); + if ((rc = copy_to_user((void *)ioctlparam, &list, sizeof(fpga_info_list)))) + printk(KERN_ERR "riffa: cannot write ioctl user parameter.\n"); + return rc; + case IOCTL_RESET: + reset((int)ioctlparam); + break; + default: + return -ENOTTY; + break; + } + return 0; +} + + +/////////////////////////////////////////////////////// +// PCI DRIVER HANDLERS +/////////////////////////////////////////////////////// + +/** + * Allocates and initializes chnl_dir structs for each channel. Returns the + * number of chnl_dir structs allocated. + */ +static inline int __devinit allocate_chnls(struct pci_dev *dev, struct fpga_state * sc) +{ + int i; + dma_addr_t hw_addr; + + for (i = 0; i < sc->num_chnls; ++i) { + // Allocate the recv struct + sc->recv[i] = (struct chnl_dir *) kzalloc(sizeof(struct chnl_dir), GFP_KERNEL); + if (sc->recv[i] == NULL) + return i; + sc->recv[i]->flags = 0; + init_waitqueue_head(&sc->recv[i]->waitq); + if ((sc->recv[i]->msgs = init_circ_queue(5)) == NULL) { + kfree(sc->recv[i]); + return i; + } + sc->recv[i]->buf_addr = pci_alloc_consistent(dev, sc->sg_buf_size, &hw_addr); + sc->recv[i]->buf_hw_addr = hw_addr; + if (sc->recv[i]->buf_addr == NULL) { + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->recv[i]); + return i; + } + + // Allocate the send struct + sc->send[i] = (struct chnl_dir *) kzalloc(sizeof(struct chnl_dir), GFP_KERNEL); + if (sc->send[i] == NULL) { + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->recv[i]); + return i; + } + sc->send[i]->flags = 0; + init_waitqueue_head(&sc->send[i]->waitq); + if ((sc->send[i]->msgs = init_circ_queue(4)) == NULL) { + kfree(sc->send[i]); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->recv[i]); + return i; + } + sc->send[i]->buf_addr = pci_alloc_consistent(dev, sc->sg_buf_size, &hw_addr); + sc->send[i]->buf_hw_addr = hw_addr; + if (sc->send[i]->buf_addr == NULL) { + free_circ_queue(sc->send[i]->msgs); + kfree(sc->send[i]); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->recv[i]); + return i; + } + } + + return i; +} + +/** + * Called by the OS when the device is ready for access. Returns 0 on success, + * negative value on failure. + */ +static int __devinit fpga_probe(struct pci_dev *dev, const struct pci_device_id *id) +{ + int i; + int j; + int error; + struct fpga_state * sc; + dma_addr_t hw_addr; + unsigned int reg; + u32 lnkctl_result = 0; + u32 devctl_result = 0; + u32 devctl2_result = 0; + + // Setup the PCIe device. + error = pci_enable_device(dev); + if (error < 0) { + printk(KERN_ERR "riffa: pci_enable_device returned %d\n", error); + return (-ENODEV); + } + + // Enable bus master + pci_set_master(dev); + + // Set the mask size + error = pci_set_dma_mask(dev, DMA_BIT_MASK(64)); + if (!error) + error = pci_set_consistent_dma_mask(dev, DMA_BIT_MASK(64)); + if (error) { + printk(KERN_ERR "riffa: cannot set 64 bit DMA mode\n"); + pci_disable_device(dev); + return error; + } + + // Allocate device structure. + sc = kzalloc(sizeof(*sc), GFP_KERNEL); + if (sc == NULL) { + printk(KERN_ERR "riffa: not enough memory to allocate sc\n"); + pci_disable_device(dev); + return (-ENOMEM); + } + atomic_set(&sc->intr_disabled, 0); + snprintf(sc->name, sizeof(sc->name), "%s%d", pci_name(dev), 0); + sc->vendor_id = dev->vendor; + sc->device_id = dev->device; + printk(KERN_INFO "riffa: found FPGA with name: %s\n", sc->name); + printk(KERN_INFO "riffa: vendor id: 0x%04X\n", sc->vendor_id); + printk(KERN_INFO "riffa: device id: 0x%04X\n", sc->device_id); + + // Setup the BAR memory regions + error = pci_request_regions(dev, sc->name); + if (error < 0) { + printk(KERN_ERR "riffa: pci_request_regions returned error: %d\n", error); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + // PCI BAR 0 + sc->bar0_addr = pci_resource_start(dev, 0); + sc->bar0_len = pci_resource_len(dev, 0); + sc->bar0_flags = pci_resource_flags(dev, 0); + printk(KERN_INFO "riffa: BAR 0 address: %llx\n", sc->bar0_addr); + printk(KERN_INFO "riffa: BAR 0 length: %lld bytes\n", sc->bar0_len); + if (sc->bar0_len != 1024) { + printk(KERN_ERR "riffa: BAR 0 incorrect length\n"); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + sc->bar0 = ioremap(sc->bar0_addr, sc->bar0_len); + if (!sc->bar0) { + printk(KERN_ERR "riffa: could not ioremap BAR 0\n"); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + // Setup MSI interrupts + error = pci_enable_msi(dev); + if (error != 0) { + printk(KERN_ERR "riffa: pci_enable_msi returned error: %d\n", error); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + + // Request an interrupt + error = request_irq(dev->irq, intrpt_handler, IRQF_SHARED, sc->name, sc); + if (error != 0) { + printk(KERN_ERR "riffa: request_irq(%d) returned error: %d\n", dev->irq, error); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + sc->irq = dev->irq; + printk(KERN_INFO "riffa: MSI setup on irq %d\n", dev->irq); + + // Set extended tag bit + error = pcie_capability_read_dword(dev,PCI_EXP_DEVCTL,&devctl_result); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_read_dword returned error: %d\n", error); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + printk(KERN_INFO "riffa: PCIE_EXP_DEVCTL register: %x\n",devctl_result); + + error = pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,(devctl_result|PCI_EXP_DEVCTL_EXT_TAG)); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_write_dword returned error: %d\n", error); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + + // Set IDO bits + error = pcie_capability_read_dword(dev,PCI_EXP_DEVCTL2,&devctl2_result); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_read_dword returned error: %d\n", error); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + printk(KERN_INFO "riffa: PCIE_EXP_DEVCTL2 register: %x\n",devctl2_result); + + error = pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,(devctl2_result | PCI_EXP_DEVCTL2_IDO_REQ_EN | PCI_EXP_DEVCTL2_IDO_CMP_EN)); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_write_dword returned error: %d\n", error); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + + // Set RCB to 128 + error = pcie_capability_read_dword(dev,PCI_EXP_LNKCTL,&lnkctl_result); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_read_dword returned error: %d\n", error); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + printk(KERN_INFO "riffa: PCIE_EXP_LNKCTL register: %x\n",lnkctl_result); + + error = pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,(lnkctl_result|PCI_EXP_LNKCTL_RCB)); + if (error != 0) { + printk(KERN_ERR "riffa: pcie_capability_write_dword returned error: %d\n", error); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return error; + } + // Read device configuration + reg = read_reg(sc, INFO_REG_OFF); + sc->num_chnls = ((reg>>0) & 0xF); + sc->num_sg = SG_ELEMS*((reg>>19) & 0xF); + sc->sg_buf_size = SG_BUF_SIZE*((reg>>19) & 0xF); + printk(KERN_INFO "riffa: number of channels: %d\n", ((reg>>0) & 0xF)); + printk(KERN_INFO "riffa: bus interface width: %d\n", ((reg>>19) & 0xF)<<5); + printk(KERN_INFO "riffa: bus master enabled: %d\n", ((reg>>4) & 0x1)); + printk(KERN_INFO "riffa: negotiated link width: %d\n", ((reg>>5) & 0x3F)); + printk(KERN_INFO "riffa: negotiated link rate: %d MTs\n", ((reg>>11) & 0x3)*2500); + printk(KERN_INFO "riffa: max downstream payload: %d bytes\n", 128<<((reg>>13) & 0x7) ); + printk(KERN_INFO "riffa: max upstream payload: %d bytes\n", 128<<((reg>>16) & 0x7) ); + + if (((reg>>4) & 0x1) != 1) { + printk(KERN_ERR "riffa: bus master not enabled!\n"); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + if (((reg>>5) & 0x3F) == 0 || ((reg>>11) & 0x3) == 0) { + printk(KERN_ERR "riffa: bad link parameters!\n"); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + if ((reg & 0xF) == 0 || (reg & 0xF) > MAX_CHNLS) { + printk(KERN_ERR "riffa: bad number of channels!\n"); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + if (((reg>>19) & 0xF) == 0 || ((reg>>19) & 0xF) > MAX_BUS_WIDTH_PARAM) { + printk(KERN_ERR "riffa: bad bus width!\n"); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENODEV); + } + + // Create chnl_dir structs. + sc->recv = (struct chnl_dir **) kzalloc(sc->num_chnls*sizeof(struct chnl_dir*), GFP_KERNEL); + sc->send = (struct chnl_dir **) kzalloc(sc->num_chnls*sizeof(struct chnl_dir*), GFP_KERNEL); + if (sc->recv == NULL || sc->send == NULL) { + printk(KERN_ERR "riffa: not enough memory to allocate chnl_dir arrays\n"); + if (sc->recv != NULL) + kfree(sc->recv); + if (sc->send != NULL) + kfree(sc->send); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENOMEM); + } + j = allocate_chnls(dev, sc); + if (j < sc->num_chnls) { + sc->num_chnls = j; + printk(KERN_ERR "riffa: not enough memory to allocate chnl_dir structs\n"); + for (i = 0; i < sc->num_chnls; ++i) { + pci_free_consistent(dev, sc->sg_buf_size, sc->send[i]->buf_addr, + (dma_addr_t)sc->send[i]->buf_hw_addr); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->send[i]->msgs); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->send[i]); + kfree(sc->recv[i]); + } + kfree(sc->recv); + kfree(sc->send); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENOMEM); + } + + // Create spill buffer (for overflow on receive). + sc->spill_buf_addr = pci_alloc_consistent(dev, SPILL_BUF_SIZE, &hw_addr); + sc->spill_buf_hw_addr = hw_addr; + if (sc->spill_buf_addr == NULL) { + printk(KERN_ERR "riffa: not enough memory to allocate spill buffer\n"); + for (i = 0; i < sc->num_chnls; ++i) { + pci_free_consistent(dev, sc->sg_buf_size, sc->send[i]->buf_addr, + (dma_addr_t)sc->send[i]->buf_hw_addr); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->send[i]->msgs); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->send[i]); + kfree(sc->recv[i]); + } + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + kfree(sc->recv); + kfree(sc->send); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + } + + // Save pointer to structure + pci_set_drvdata(dev, sc); + sc->dev = dev; + sc->id = -1; + for (i = 0; i < NUM_FPGAS; i++) { + if (!atomic_xchg(&used_fpgas[i], 1)) { + sc->id = i; + fpgas[i] = sc; + break; + } + } + if (sc->id == -1) { + printk(KERN_ERR "riffa: could not save FPGA information, %d is limit.\n", NUM_FPGAS); + for (i = 0; i < sc->num_chnls; ++i) { + pci_free_consistent(dev, sc->sg_buf_size, sc->send[i]->buf_addr, + (dma_addr_t)sc->send[i]->buf_hw_addr); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->send[i]->msgs); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->send[i]); + kfree(sc->recv[i]); + } + kfree(sc->recv); + kfree(sc->send); + pci_free_consistent(dev, SPILL_BUF_SIZE, sc->spill_buf_addr, + (dma_addr_t)sc->spill_buf_hw_addr); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,lnkctl_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,devctl2_result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,devctl_result); + free_irq(dev->irq, sc); + pci_disable_msi(dev); + iounmap(sc->bar0); + pci_release_regions(dev); + pci_disable_device(dev); + kfree(sc); + return (-ENOMEM); + } + else { + printk(KERN_INFO "riffa: saved FPGA with id: %d\n", sc->id); + } + + return 0; +} + +/** + * Called when the device is unloaded. + */ +static void __devexit fpga_remove(struct pci_dev *dev) +{ + int i; + u32 result; + struct fpga_state * sc; + + pcie_capability_read_dword(dev,PCI_EXP_DEVCTL,&result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL,result & ~(PCI_EXP_DEVCTL_EXT_TAG | PCI_EXP_DEVCTL_RELAX_EN)); + + pcie_capability_read_dword(dev,PCI_EXP_DEVCTL2,&result); + pcie_capability_write_dword(dev,PCI_EXP_DEVCTL2,result & ~(PCI_EXP_DEVCTL2_IDO_REQ_EN | PCI_EXP_DEVCTL2_IDO_CMP_EN)); + + pcie_capability_read_dword(dev,PCI_EXP_DEVCTL,&result); + pcie_capability_write_dword(dev,PCI_EXP_LNKCTL,result & (~PCI_EXP_LNKCTL_RCB)); + + if ((sc = (struct fpga_state *)pci_get_drvdata(dev)) != NULL) { + // Free structs, memory regions, etc. + atomic_set(&used_fpgas[sc->id], 0); + for (i = 0; i < sc->num_chnls; ++i) { + pci_free_consistent(dev, sc->sg_buf_size, sc->send[i]->buf_addr, + (dma_addr_t)sc->send[i]->buf_hw_addr); + pci_free_consistent(dev, sc->sg_buf_size, sc->recv[i]->buf_addr, + (dma_addr_t)sc->recv[i]->buf_hw_addr); + free_circ_queue(sc->send[i]->msgs); + free_circ_queue(sc->recv[i]->msgs); + kfree(sc->send[i]); + kfree(sc->recv[i]); + } + kfree(sc->recv); + kfree(sc->send); + pci_free_consistent(dev, SPILL_BUF_SIZE, sc->spill_buf_addr, + (dma_addr_t)sc->spill_buf_hw_addr); + free_irq(dev->irq, sc); + iounmap(sc->bar0); + kfree(sc); + } + pci_disable_msi(dev); + pci_release_regions(dev); + pci_disable_device(dev); + pci_set_drvdata(dev, NULL); +} + + +/////////////////////////////////////////////////////// +// MODULE INIT/EXIT FUNCTIONS +/////////////////////////////////////////////////////// + +#define ALTERA_VENDOR_ID 1172 +#define ALTERA_DE4_DEVICE_ID 0004 + +#if LINUX_VERSION_CODE < KERNEL_VERSION(4,8,0) +static DEFINE_PCI_DEVICE_TABLE(fpga_ids) = +#else +static const struct pci_device_id fpga_ids[] = +#endif +{ + {PCI_DEVICE(ALTERA_VENDOR_ID, ALTERA_DE4_DEVICE_ID)}, //{PCI_DEVICE(VENDOR_ID0, PCI_ANY_ID)}, + {PCI_DEVICE(VENDOR_ID1, PCI_ANY_ID)}, + {0}, +}; + +MODULE_DEVICE_TABLE(pci, fpga_ids); +static struct pci_driver fpga_driver = { + .name = DEVICE_NAME, + .id_table = fpga_ids, + .probe = fpga_probe, + .remove = __devexit_p(fpga_remove), +}; + +static const struct file_operations fpga_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = fpga_ioctl, +}; + +/** + * Called to initialize the PCI device. + */ +static int __init fpga_init(void) +{ + int i; + int error; + + for (i = 0; i < NUM_FPGAS; i++) + atomic_set(&used_fpgas[i], 0); + + error = pci_register_driver(&fpga_driver); + if (error != 0) { + printk(KERN_ERR "riffa: pci_module_register returned %d\n", error); + return (error); + } + + error = register_chrdev(MAJOR_NUM, DEVICE_NAME, &fpga_fops); + if (error < 0) { + printk(KERN_ERR "riffa: register_chrdev returned %d\n", error); + return (error); + } + + mymodule_class = class_create(THIS_MODULE, DEVICE_NAME); + if (IS_ERR(mymodule_class)) { + error = PTR_ERR(mymodule_class); + printk(KERN_ERR "riffa: class_create() returned %d\n", error); + return (error); + } + + devt = MKDEV(MAJOR_NUM, 0); + device_create(mymodule_class, NULL, devt, "%s", DEVICE_NAME); + + return 0; +} + +/** + * Called to destroy the PCI device. + */ +static void __exit fpga_exit(void) +{ + device_destroy(mymodule_class, devt); + class_destroy(mymodule_class); + pci_unregister_driver(&fpga_driver); + unregister_chrdev(MAJOR_NUM, DEVICE_NAME); +} + +module_init(fpga_init); +module_exit(fpga_exit); diff --git a/drivers/staging/riffa/riffa_driver.h b/drivers/staging/riffa/riffa_driver.h new file mode 100644 index 00000000000000..cf366b269c18d8 --- /dev/null +++ b/drivers/staging/riffa/riffa_driver.h @@ -0,0 +1,131 @@ +// ---------------------------------------------------------------------- +// Copyright (c) 2016, The Regents of the University of California All +// rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// +// * Neither the name of The Regents of the University of California +// nor the names of its contributors may be used to endorse or +// promote products derived from this software without specific +// prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL REGENTS OF THE +// UNIVERSITY OF CALIFORNIA BE LIABLE FOR ANY DIRECT, INDIRECT, +// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, +// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS +// OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR +// TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE +// USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH +// DAMAGE. +// ---------------------------------------------------------------------- + +/* + * Filename: riffa_driver.h + * Version: 2.0 + * Description: Linux PCIe device driver for RIFFA. Uses Linux kernel APIs in + * version 2.6.27+ (tested on version 2.6.32 - 3.3.0). + * Author: Matthew Jacobsen + * History: @mattj: Initial release. Version 2.0. + */ + +#ifndef RIFFA_DRIVER_H +#define RIFFA_DRIVER_H + +#include + +#define DEBUG 1 + +#ifdef DEBUG +#define DEBUG_MSG(...) printk(__VA_ARGS__) +#else +#define DEBUG_MSG(...) +#endif + + +// The major device number. We can't rely on dynamic registration because ioctls +// need to know it. +#define MAJOR_NUM 100 +#define DEVICE_NAME "riffa" +#define VENDOR_ID0 0x10EE +#define VENDOR_ID1 0x1172 + +// Message events for readmsgs/writemsgs queues. +#define EVENT_TXN_LEN 1 +#define EVENT_TXN_OFFLAST 2 +#define EVENT_TXN_DONE 3 +#define EVENT_SG_BUF_READ 4 + +// Constants and device offsets +#define NUM_FPGAS 5 // max # of FPGAs to support in a single PC +#define MAX_CHNLS 12 // max # of channels per FPGA +#define MAX_BUS_WIDTH_PARAM 4 // max bus width parameter +#define SG_BUF_SIZE (4*1024) // size of shared SG buffer +#define SG_ELEMS 200 // # of SG elements to transfer at a time +#define SPILL_BUF_SIZE (4*1024) // size of shared spill common buffer + +#define RX_SG_LEN_REG_OFF 0x0 // config offset for RX SG buf length +#define RX_SG_ADDR_LO_REG_OFF 0x1 // config offset for RX SG buf low addr +#define RX_SG_ADDR_HI_REG_OFF 0x2 // config offset for RX SG buf high addr +#define RX_LEN_REG_OFF 0x3 // config offset for RX txn length +#define RX_OFFLAST_REG_OFF 0x4 // config offset for RX txn last/offset +#define RX_TNFR_LEN_REG_OFF 0xD // config offset for RX transfer length +#define TX_SG_LEN_REG_OFF 0x5 // config offset for TX SG buf length +#define TX_SG_ADDR_LO_REG_OFF 0x6 // config offset for TX SG buf low addr +#define TX_SG_ADDR_HI_REG_OFF 0x7 // config offset for TX SG buf high addr +#define TX_LEN_REG_OFF 0x8 // config offset for TX txn length +#define TX_OFFLAST_REG_OFF 0x9 // config offset for TX txn last/offset +#define TX_TNFR_LEN_REG_OFF 0xE // config offset for TX transfer length + +#define INFO_REG_OFF 0xA // config offset for link info + +#define IRQ_REG0_OFF 0xB // config offset for interrupt reg 0 +#define IRQ_REG1_OFF 0xC // config offset for interrupt reg 1 + + +// Structs +struct fpga_chnl_io +{ + int id; + int chnl; + unsigned int len; + unsigned int offset; + unsigned int last; + unsigned long long timeout; + char * data; +}; +typedef struct fpga_chnl_io fpga_chnl_io; + +struct fpga_info_list +{ + int num_fpgas; + int id[NUM_FPGAS]; + int num_chnls[NUM_FPGAS]; + char name[NUM_FPGAS][16]; + int vendor_id[NUM_FPGAS]; + int device_id[NUM_FPGAS]; +}; +typedef struct fpga_info_list fpga_info_list; + +// IOCTLs +#define IOCTL_SEND _IOW(MAJOR_NUM, 1, fpga_chnl_io *) +#define IOCTL_RECV _IOR(MAJOR_NUM, 2, fpga_chnl_io *) +#define IOCTL_LIST _IOR(MAJOR_NUM, 3, fpga_info_list *) +#define IOCTL_RESET _IOW(MAJOR_NUM, 4, int) + + + +#endif