From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- drivers/firewire/nosy.c | 719 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 719 insertions(+) create mode 100644 drivers/firewire/nosy.c (limited to 'drivers/firewire/nosy.c') diff --git a/drivers/firewire/nosy.c b/drivers/firewire/nosy.c new file mode 100644 index 000000000..900b7d904 --- /dev/null +++ b/drivers/firewire/nosy.c @@ -0,0 +1,719 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * nosy - Snoop mode driver for TI PCILynx 1394 controllers + * Copyright (C) 2002-2007 Kristian Høgsberg + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include /* required for linux/wait.h */ +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "nosy.h" +#include "nosy-user.h" + +#define TCODE_PHY_PACKET 0x10 +#define PCI_DEVICE_ID_TI_PCILYNX 0x8000 + +static char driver_name[] = KBUILD_MODNAME; + +#define RCV_BUFFER_SIZE (16 * 1024) + +/* this is the physical layout of a PCL, its size is 128 bytes */ +struct pcl { + __le32 next; + __le32 async_error_next; + u32 user_data; + __le32 pcl_status; + __le32 remaining_transfer_count; + __le32 next_data_buffer; + struct { + __le32 control; + __le32 pointer; + } buffer[13]; +}; + +struct packet { + unsigned int length; + char data[]; +}; + +struct packet_buffer { + char *data; + size_t capacity; + long total_packet_count, lost_packet_count; + atomic_t size; + struct packet *head, *tail; + wait_queue_head_t wait; +}; + +struct pcilynx { + struct pci_dev *pci_device; + __iomem char *registers; + + struct pcl *rcv_start_pcl, *rcv_pcl; + __le32 *rcv_buffer; + + dma_addr_t rcv_start_pcl_bus, rcv_pcl_bus, rcv_buffer_bus; + + spinlock_t client_list_lock; + struct list_head client_list; + + struct miscdevice misc; + struct list_head link; + struct kref kref; +}; + +static inline struct pcilynx * +lynx_get(struct pcilynx *lynx) +{ + kref_get(&lynx->kref); + + return lynx; +} + +static void +lynx_release(struct kref *kref) +{ + kfree(container_of(kref, struct pcilynx, kref)); +} + +static inline void +lynx_put(struct pcilynx *lynx) +{ + kref_put(&lynx->kref, lynx_release); +} + +struct client { + struct pcilynx *lynx; + u32 tcode_mask; + struct packet_buffer buffer; + struct list_head link; +}; + +static DEFINE_MUTEX(card_mutex); +static LIST_HEAD(card_list); + +static int +packet_buffer_init(struct packet_buffer *buffer, size_t capacity) +{ + buffer->data = kmalloc(capacity, GFP_KERNEL); + if (buffer->data == NULL) + return -ENOMEM; + buffer->head = (struct packet *) buffer->data; + buffer->tail = (struct packet *) buffer->data; + buffer->capacity = capacity; + buffer->lost_packet_count = 0; + atomic_set(&buffer->size, 0); + init_waitqueue_head(&buffer->wait); + + return 0; +} + +static void +packet_buffer_destroy(struct packet_buffer *buffer) +{ + kfree(buffer->data); +} + +static int +packet_buffer_get(struct client *client, char __user *data, size_t user_length) +{ + struct packet_buffer *buffer = &client->buffer; + size_t length; + char *end; + + if (wait_event_interruptible(buffer->wait, + atomic_read(&buffer->size) > 0) || + list_empty(&client->lynx->link)) + return -ERESTARTSYS; + + if (atomic_read(&buffer->size) == 0) + return -ENODEV; + + length = buffer->head->length; + + if (length > user_length) + return 0; + + end = buffer->data + buffer->capacity; + + if (&buffer->head->data[length] < end) { + if (copy_to_user(data, buffer->head->data, length)) + return -EFAULT; + buffer->head = (struct packet *) &buffer->head->data[length]; + } else { + size_t split = end - buffer->head->data; + + if (copy_to_user(data, buffer->head->data, split)) + return -EFAULT; + if (copy_to_user(data + split, buffer->data, length - split)) + return -EFAULT; + buffer->head = (struct packet *) &buffer->data[length - split]; + } + + /* + * Decrease buffer->size as the last thing, since this is what + * keeps the interrupt from overwriting the packet we are + * retrieving from the buffer. + */ + atomic_sub(sizeof(struct packet) + length, &buffer->size); + + return length; +} + +static void +packet_buffer_put(struct packet_buffer *buffer, void *data, size_t length) +{ + char *end; + + buffer->total_packet_count++; + + if (buffer->capacity < + atomic_read(&buffer->size) + sizeof(struct packet) + length) { + buffer->lost_packet_count++; + return; + } + + end = buffer->data + buffer->capacity; + buffer->tail->length = length; + + if (&buffer->tail->data[length] < end) { + memcpy(buffer->tail->data, data, length); + buffer->tail = (struct packet *) &buffer->tail->data[length]; + } else { + size_t split = end - buffer->tail->data; + + memcpy(buffer->tail->data, data, split); + memcpy(buffer->data, data + split, length - split); + buffer->tail = (struct packet *) &buffer->data[length - split]; + } + + /* Finally, adjust buffer size and wake up userspace reader. */ + + atomic_add(sizeof(struct packet) + length, &buffer->size); + wake_up_interruptible(&buffer->wait); +} + +static inline void +reg_write(struct pcilynx *lynx, int offset, u32 data) +{ + writel(data, lynx->registers + offset); +} + +static inline u32 +reg_read(struct pcilynx *lynx, int offset) +{ + return readl(lynx->registers + offset); +} + +static inline void +reg_set_bits(struct pcilynx *lynx, int offset, u32 mask) +{ + reg_write(lynx, offset, (reg_read(lynx, offset) | mask)); +} + +/* + * Maybe the pcl programs could be set up to just append data instead + * of using a whole packet. + */ +static inline void +run_pcl(struct pcilynx *lynx, dma_addr_t pcl_bus, + int dmachan) +{ + reg_write(lynx, DMA0_CURRENT_PCL + dmachan * 0x20, pcl_bus); + reg_write(lynx, DMA0_CHAN_CTRL + dmachan * 0x20, + DMA_CHAN_CTRL_ENABLE | DMA_CHAN_CTRL_LINK); +} + +static int +set_phy_reg(struct pcilynx *lynx, int addr, int val) +{ + if (addr > 15) { + dev_err(&lynx->pci_device->dev, + "PHY register address %d out of range\n", addr); + return -1; + } + if (val > 0xff) { + dev_err(&lynx->pci_device->dev, + "PHY register value %d out of range\n", val); + return -1; + } + reg_write(lynx, LINK_PHY, LINK_PHY_WRITE | + LINK_PHY_ADDR(addr) | LINK_PHY_WDATA(val)); + + return 0; +} + +static int +nosy_open(struct inode *inode, struct file *file) +{ + int minor = iminor(inode); + struct client *client; + struct pcilynx *tmp, *lynx = NULL; + + mutex_lock(&card_mutex); + list_for_each_entry(tmp, &card_list, link) + if (tmp->misc.minor == minor) { + lynx = lynx_get(tmp); + break; + } + mutex_unlock(&card_mutex); + if (lynx == NULL) + return -ENODEV; + + client = kmalloc_obj(*client); + if (client == NULL) + goto fail; + + client->tcode_mask = ~0; + client->lynx = lynx; + INIT_LIST_HEAD(&client->link); + + if (packet_buffer_init(&client->buffer, 128 * 1024) < 0) + goto fail; + + file->private_data = client; + + return stream_open(inode, file); +fail: + kfree(client); + lynx_put(lynx); + + return -ENOMEM; +} + +static int +nosy_release(struct inode *inode, struct file *file) +{ + struct client *client = file->private_data; + struct pcilynx *lynx = client->lynx; + + spin_lock_irq(&lynx->client_list_lock); + list_del_init(&client->link); + spin_unlock_irq(&lynx->client_list_lock); + + packet_buffer_destroy(&client->buffer); + kfree(client); + lynx_put(lynx); + + return 0; +} + +static __poll_t +nosy_poll(struct file *file, poll_table *pt) +{ + struct client *client = file->private_data; + __poll_t ret = 0; + + poll_wait(file, &client->buffer.wait, pt); + + if (atomic_read(&client->buffer.size) > 0) + ret = EPOLLIN | EPOLLRDNORM; + + if (list_empty(&client->lynx->link)) + ret |= EPOLLHUP; + + return ret; +} + +static ssize_t +nosy_read(struct file *file, char __user *buffer, size_t count, loff_t *offset) +{ + struct client *client = file->private_data; + + return packet_buffer_get(client, buffer, count); +} + +static long +nosy_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + struct client *client = file->private_data; + spinlock_t *client_list_lock = &client->lynx->client_list_lock; + struct nosy_stats stats; + int ret; + + switch (cmd) { + case NOSY_IOC_GET_STATS: + spin_lock_irq(client_list_lock); + stats.total_packet_count = client->buffer.total_packet_count; + stats.lost_packet_count = client->buffer.lost_packet_count; + spin_unlock_irq(client_list_lock); + + if (copy_to_user((void __user *) arg, &stats, sizeof stats)) + return -EFAULT; + else + return 0; + + case NOSY_IOC_START: + ret = -EBUSY; + spin_lock_irq(client_list_lock); + if (list_empty(&client->link)) { + list_add_tail(&client->link, &client->lynx->client_list); + ret = 0; + } + spin_unlock_irq(client_list_lock); + + return ret; + + case NOSY_IOC_STOP: + spin_lock_irq(client_list_lock); + list_del_init(&client->link); + spin_unlock_irq(client_list_lock); + + return 0; + + case NOSY_IOC_FILTER: + spin_lock_irq(client_list_lock); + client->tcode_mask = arg; + spin_unlock_irq(client_list_lock); + + return 0; + + default: + return -EINVAL; + /* Flush buffer, configure filter. */ + } +} + +static const struct file_operations nosy_ops = { + .owner = THIS_MODULE, + .read = nosy_read, + .unlocked_ioctl = nosy_ioctl, + .poll = nosy_poll, + .open = nosy_open, + .release = nosy_release, +}; + +#define PHY_PACKET_SIZE 12 /* 1 payload, 1 inverse, 1 ack = 3 quadlets */ + +static void +packet_irq_handler(struct pcilynx *lynx) +{ + struct client *client; + u32 tcode_mask, tcode, timestamp; + size_t length; + struct timespec64 ts64; + + /* FIXME: Also report rcv_speed. */ + + length = __le32_to_cpu(lynx->rcv_pcl->pcl_status) & 0x00001fff; + tcode = __le32_to_cpu(lynx->rcv_buffer[1]) >> 4 & 0xf; + + ktime_get_real_ts64(&ts64); + timestamp = ts64.tv_nsec / NSEC_PER_USEC; + lynx->rcv_buffer[0] = (__force __le32)timestamp; + + if (length == PHY_PACKET_SIZE) + tcode_mask = 1 << TCODE_PHY_PACKET; + else + tcode_mask = 1 << tcode; + + spin_lock(&lynx->client_list_lock); + + list_for_each_entry(client, &lynx->client_list, link) + if (client->tcode_mask & tcode_mask) + packet_buffer_put(&client->buffer, + lynx->rcv_buffer, length + 4); + + spin_unlock(&lynx->client_list_lock); +} + +static void +bus_reset_irq_handler(struct pcilynx *lynx) +{ + struct client *client; + struct timespec64 ts64; + u32 timestamp; + + ktime_get_real_ts64(&ts64); + timestamp = ts64.tv_nsec / NSEC_PER_USEC; + + spin_lock(&lynx->client_list_lock); + + list_for_each_entry(client, &lynx->client_list, link) + packet_buffer_put(&client->buffer, ×tamp, 4); + + spin_unlock(&lynx->client_list_lock); +} + +static irqreturn_t +irq_handler(int irq, void *device) +{ + struct pcilynx *lynx = device; + u32 pci_int_status; + + pci_int_status = reg_read(lynx, PCI_INT_STATUS); + + if (pci_int_status == ~0) + /* Card was ejected. */ + return IRQ_NONE; + + if ((pci_int_status & PCI_INT_INT_PEND) == 0) + /* Not our interrupt, bail out quickly. */ + return IRQ_NONE; + + if ((pci_int_status & PCI_INT_P1394_INT) != 0) { + u32 link_int_status; + + link_int_status = reg_read(lynx, LINK_INT_STATUS); + reg_write(lynx, LINK_INT_STATUS, link_int_status); + + if ((link_int_status & LINK_INT_PHY_BUSRESET) > 0) + bus_reset_irq_handler(lynx); + } + + /* Clear the PCI_INT_STATUS register only after clearing the + * LINK_INT_STATUS register; otherwise the PCI_INT_P1394 will + * be set again immediately. */ + + reg_write(lynx, PCI_INT_STATUS, pci_int_status); + + if ((pci_int_status & PCI_INT_DMA0_HLT) > 0) { + packet_irq_handler(lynx); + run_pcl(lynx, lynx->rcv_start_pcl_bus, 0); + } + + return IRQ_HANDLED; +} + +static void +remove_card(struct pci_dev *dev) +{ + struct pcilynx *lynx = pci_get_drvdata(dev); + struct client *client; + + mutex_lock(&card_mutex); + list_del_init(&lynx->link); + misc_deregister(&lynx->misc); + mutex_unlock(&card_mutex); + + reg_write(lynx, PCI_INT_ENABLE, 0); + free_irq(lynx->pci_device->irq, lynx); + + spin_lock_irq(&lynx->client_list_lock); + list_for_each_entry(client, &lynx->client_list, link) + wake_up_interruptible(&client->buffer.wait); + spin_unlock_irq(&lynx->client_list_lock); + + dma_free_coherent(&lynx->pci_device->dev, sizeof(struct pcl), + lynx->rcv_start_pcl, lynx->rcv_start_pcl_bus); + dma_free_coherent(&lynx->pci_device->dev, sizeof(struct pcl), + lynx->rcv_pcl, lynx->rcv_pcl_bus); + dma_free_coherent(&lynx->pci_device->dev, RCV_BUFFER_SIZE, + lynx->rcv_buffer, lynx->rcv_buffer_bus); + + iounmap(lynx->registers); + pci_disable_device(dev); + lynx_put(lynx); +} + +static int +add_card(struct pci_dev *dev, const struct pci_device_id *unused) +{ + struct pcilynx *lynx; + u32 p, end; + int ret, i; + + if (dma_set_mask(&dev->dev, DMA_BIT_MASK(32))) { + dev_err(&dev->dev, + "DMA address limits not supported for PCILynx hardware\n"); + return -ENXIO; + } + if (pci_enable_device(dev)) { + dev_err(&dev->dev, "Failed to enable PCILynx hardware\n"); + return -ENXIO; + } + pci_set_master(dev); + + lynx = kzalloc_obj(*lynx); + if (lynx == NULL) { + dev_err(&dev->dev, "Failed to allocate control structure\n"); + ret = -ENOMEM; + goto fail_disable; + } + lynx->pci_device = dev; + pci_set_drvdata(dev, lynx); + + spin_lock_init(&lynx->client_list_lock); + INIT_LIST_HEAD(&lynx->client_list); + kref_init(&lynx->kref); + + lynx->registers = ioremap(pci_resource_start(dev, 0), + PCILYNX_MAX_REGISTER); + if (lynx->registers == NULL) { + dev_err(&dev->dev, "Failed to map registers\n"); + ret = -ENOMEM; + goto fail_deallocate_lynx; + } + + lynx->rcv_start_pcl = dma_alloc_coherent(&lynx->pci_device->dev, + sizeof(struct pcl), + &lynx->rcv_start_pcl_bus, + GFP_KERNEL); + lynx->rcv_pcl = dma_alloc_coherent(&lynx->pci_device->dev, + sizeof(struct pcl), + &lynx->rcv_pcl_bus, GFP_KERNEL); + lynx->rcv_buffer = dma_alloc_coherent(&lynx->pci_device->dev, + RCV_BUFFER_SIZE, + &lynx->rcv_buffer_bus, GFP_KERNEL); + if (lynx->rcv_start_pcl == NULL || + lynx->rcv_pcl == NULL || + lynx->rcv_buffer == NULL) { + dev_err(&dev->dev, "Failed to allocate receive buffer\n"); + ret = -ENOMEM; + goto fail_deallocate_buffers; + } + lynx->rcv_start_pcl->next = cpu_to_le32(lynx->rcv_pcl_bus); + lynx->rcv_pcl->next = cpu_to_le32(PCL_NEXT_INVALID); + lynx->rcv_pcl->async_error_next = cpu_to_le32(PCL_NEXT_INVALID); + + lynx->rcv_pcl->buffer[0].control = + cpu_to_le32(PCL_CMD_RCV | PCL_BIGENDIAN | 2044); + lynx->rcv_pcl->buffer[0].pointer = + cpu_to_le32(lynx->rcv_buffer_bus + 4); + p = lynx->rcv_buffer_bus + 2048; + end = lynx->rcv_buffer_bus + RCV_BUFFER_SIZE; + for (i = 1; p < end; i++, p += 2048) { + lynx->rcv_pcl->buffer[i].control = + cpu_to_le32(PCL_CMD_RCV | PCL_BIGENDIAN | 2048); + lynx->rcv_pcl->buffer[i].pointer = cpu_to_le32(p); + } + lynx->rcv_pcl->buffer[i - 1].control |= cpu_to_le32(PCL_LAST_BUFF); + + reg_set_bits(lynx, MISC_CONTROL, MISC_CONTROL_SWRESET); + /* Fix buggy cards with autoboot pin not tied low: */ + reg_write(lynx, DMA0_CHAN_CTRL, 0); + reg_write(lynx, DMA_GLOBAL_REGISTER, 0x00 << 24); + +#if 0 + /* now, looking for PHY register set */ + if ((get_phy_reg(lynx, 2) & 0xe0) == 0xe0) { + lynx->phyic.reg_1394a = 1; + PRINT(KERN_INFO, lynx->id, + "found 1394a conform PHY (using extended register set)"); + lynx->phyic.vendor = get_phy_vendorid(lynx); + lynx->phyic.product = get_phy_productid(lynx); + } else { + lynx->phyic.reg_1394a = 0; + PRINT(KERN_INFO, lynx->id, "found old 1394 PHY"); + } +#endif + + /* Setup the general receive FIFO max size. */ + reg_write(lynx, FIFO_SIZES, 255); + + reg_set_bits(lynx, PCI_INT_ENABLE, PCI_INT_DMA_ALL); + + reg_write(lynx, LINK_INT_ENABLE, + LINK_INT_PHY_TIME_OUT | LINK_INT_PHY_REG_RCVD | + LINK_INT_PHY_BUSRESET | LINK_INT_IT_STUCK | + LINK_INT_AT_STUCK | LINK_INT_SNTRJ | + LINK_INT_TC_ERR | LINK_INT_GRF_OVER_FLOW | + LINK_INT_ITF_UNDER_FLOW | LINK_INT_ATF_UNDER_FLOW); + + /* Disable the L flag in self ID packets. */ + set_phy_reg(lynx, 4, 0); + + /* Put this baby into snoop mode */ + reg_set_bits(lynx, LINK_CONTROL, LINK_CONTROL_SNOOP_ENABLE); + + run_pcl(lynx, lynx->rcv_start_pcl_bus, 0); + + if (request_irq(dev->irq, irq_handler, IRQF_SHARED, + driver_name, lynx)) { + dev_err(&dev->dev, + "Failed to allocate shared interrupt %d\n", dev->irq); + ret = -EIO; + goto fail_deallocate_buffers; + } + + lynx->misc.parent = &dev->dev; + lynx->misc.minor = MISC_DYNAMIC_MINOR; + lynx->misc.name = "nosy"; + lynx->misc.fops = &nosy_ops; + + mutex_lock(&card_mutex); + ret = misc_register(&lynx->misc); + if (ret) { + dev_err(&dev->dev, "Failed to register misc char device\n"); + mutex_unlock(&card_mutex); + goto fail_free_irq; + } + list_add_tail(&lynx->link, &card_list); + mutex_unlock(&card_mutex); + + dev_info(&dev->dev, + "Initialized PCILynx IEEE1394 card, irq=%d\n", dev->irq); + + return 0; + +fail_free_irq: + reg_write(lynx, PCI_INT_ENABLE, 0); + free_irq(lynx->pci_device->irq, lynx); + +fail_deallocate_buffers: + if (lynx->rcv_start_pcl) + dma_free_coherent(&lynx->pci_device->dev, sizeof(struct pcl), + lynx->rcv_start_pcl, + lynx->rcv_start_pcl_bus); + if (lynx->rcv_pcl) + dma_free_coherent(&lynx->pci_device->dev, sizeof(struct pcl), + lynx->rcv_pcl, lynx->rcv_pcl_bus); + if (lynx->rcv_buffer) + dma_free_coherent(&lynx->pci_device->dev, RCV_BUFFER_SIZE, + lynx->rcv_buffer, lynx->rcv_buffer_bus); + iounmap(lynx->registers); + +fail_deallocate_lynx: + kfree(lynx); + +fail_disable: + pci_disable_device(dev); + + return ret; +} + +static struct pci_device_id pci_table[] = { + { + .vendor = PCI_VENDOR_ID_TI, + .device = PCI_DEVICE_ID_TI_PCILYNX, + .subvendor = PCI_ANY_ID, + .subdevice = PCI_ANY_ID, + }, + { } /* Terminating entry */ +}; + +MODULE_DEVICE_TABLE(pci, pci_table); + +static struct pci_driver lynx_pci_driver = { + .name = driver_name, + .id_table = pci_table, + .probe = add_card, + .remove = remove_card, +}; + +module_pci_driver(lynx_pci_driver); + +MODULE_AUTHOR("Kristian Hoegsberg"); +MODULE_DESCRIPTION("Snoop mode driver for TI pcilynx 1394 controllers"); +MODULE_LICENSE("GPL"); -- cgit v1.3.1