实现 PCIe Linux 设备驱动程序(想从内核驱动程序访问我的卡寄存器) [英] Implementing PCIe Linux device driver (want to access my card registers from kernel driver)

问题描述

我正在编写一个设备驱动程序来访问 PCIe 卡上 FPGA 中的内存.
该卡启动并被探测/发现:-

/proc/iomem

80000000-840fffff : PCI 总线 #0380000000-83ffffff:0000:03:00.084000000-840fffff : 0000:03:00.0

所以阅读 ldd/etc 我在 80000000 编码了对 request_mem_region 的调用，并通过 ioremap_nocache 请求一个指向它的指针>

1) 我是否需要 request_mem_region 以及 ioremap_nocache，我不能只使用后者吗?

/proc/iomem 在 insmod 之后我的设备驱动程序 :-

80000000-840fffff : PCI 总线 #0380000000-83ffffff:0000:03:00.080000000-8003ffff : fp284000000-840fffff : 0000:03:00.0

2) 我觉得不太合适...?

无论如何，读取不起作用(它没有像下面这样编码，它有检查等):-

#define BAR_ADDR 0x80000000void *base = ioremap_nocache(BAR_ADDR, 0x40000);void *address = base + KNOWN_REG_LOCATION;int 数据 = ioread32(地址)；printk("fp2: address:0x%08x, data:0x%08x
", address, data);

输出:-

地址:0xfd500000，数据:0xffffffff

我可以从 mmap 用户空间读取 x80000000+KNOWN_REG_LOCATION.

3) 我试过 __raw_readl/readl 也没有运气.

4) 我可以只读取当前映射的地址 x80000000 吗?

解决方案

Ian,

我为设备编写了一个 PCI 驱动程序(完整来源).不过寄存器空间的映射应该是一样的.这是我的做法.

dm7820_device->pci[region].virt_addr = ioremap_nocache(address, length);如果 (dm7820_device->pci[region].virt_addr == NULL) {printk(KERN_ERR "%s: ERROR: BAR%u 重新映射失败
",&((dm7820_device->device_name)[0]), region);dm7820_release_resources();返回-ENOMEM;}if (request_mem_region(address, length, &((dm7820_device->device_name)[0])) == NULL) {printk(KERN_ERR "%s: ERROR: I/O memory range %#lx-%#lx allocation FAILED
",&((dm7820_device->device_name)[0]), address, (address + length - 1));dm7820_release_resources();返回-EBUSY;}

地址和长度值从 pci_resource_start() 和 pci_resource_length() 调用返回.

然后你可以使用 ioread32() 访问它，使用 dm7820_device->pci[region].virt_addr +

如果您有任何问题，请告诉我.

I'm writing a device driver to access the memory in a FPGA on a PCIe card.
The card boots and is probed/found :-

/proc/iomem

80000000-840fffff : PCI Bus #03
  80000000-83ffffff : 0000:03:00.0
  84000000-840fffff : 0000:03:00.0

So reading ldd/etc I coded up a call to request_mem_region at the 80000000, and requested a pointer to it via ioremap_nocache

1) Do I need to request_mem_region as well as a ioremap_nocache, cant I use just the latter?

/proc/iomem After insmod my device driver :-

80000000-840fffff : PCI Bus #03
  80000000-83ffffff : 0000:03:00.0
    80000000-8003ffff : fp2
  84000000-840fffff : 0000:03:00.0

2) Doesnt look quite right to me...?

Anyway, reads don't work (its not coded like below, it has checks etc):-

#define BAR_ADDR 0x80000000
void *base = ioremap_nocache(BAR_ADDR, 0x40000);
void *address = base + KNOWN_REG_LOCATION;
int data = ioread32(address);
printk("fp2: address:0x%08x, data:0x%08x
", address, data);

Outputs :-

address:0xfd500000, data:0xffffffff

I can read the x80000000+KNOWN_REG_LOCATION from mmap userspace.

3) I've tried __raw_readl/readl with no luck as well.

4) Can I just read at the currently mapped address x80000000?

解决方案

Ian,

I wrote a PCI driver for a device (full source). The mapping of the register space should be the same though. Here is how I do it.

dm7820_device->pci[region].virt_addr = ioremap_nocache(address, length);
if (dm7820_device->pci[region].virt_addr == NULL) {
    printk(KERN_ERR "%s: ERROR: BAR%u remapping FAILED
",
        &((dm7820_device->device_name)[0]), region);
    dm7820_release_resources();
    return -ENOMEM;
}

if (request_mem_region(address, length, &((dm7820_device->device_name)[0])) == NULL) {
    printk(KERN_ERR "%s: ERROR: I/O memory range %#lx-%#lx allocation FAILED
",
        &((dm7820_device->device_name)[0]), address, (address + length - 1));
    dm7820_release_resources();
    return -EBUSY;
}

The address and length values are returned from pci_resource_start() and pci_resource_length() calls.

Then you can access it using ioread32() using dm7820_device->pci[region].virt_addr + <register offset>

Let me know if you have any questions.

这篇关于实现 PCIe Linux 设备驱动程序(想从内核驱动程序访问我的卡寄存器)的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！