[Qemu-devel] Re: [PATCH v3 1/1] Shared memory uio_pci driver

[Cam Macdonell <cam@cs.ualberta.ca>]

On Thu, Mar 25, 2010 at 10:34 AM, Michael S. Tsirkin <mst@redhat.com> wrote:
> On Thu, Mar 25, 2010 at 10:30:42AM -0600, Cam Macdonell wrote:
>> On Thu, Mar 25, 2010 at 3:05 AM, Michael S. Tsirkin <mst@redhat.com> wrote:
>> > On Thu, Mar 25, 2010 at 12:09:36AM -0600, Cam Macdonell wrote:
>> >> This patch adds a driver for my shared memory PCI device using the uio_pci
>> >> interface.  The driver has three memory regions.  The first memory region is for
>> >> device registers for sending interrupts. The second BAR is for receiving MSI-X
>> >> interrupts and the third memory region maps the shared memory.  The device only
>> >> exports the first and third memory regions to userspace.
>> >>
>> >> This driver supports MSI-X and regular pin interrupts.  Currently, the number of
>> >> MSI vectors is set to 4 which could be increased, but the driver will work with
>> >> fewer vectors.  If MSI is not available, then regular interrupts will be used.
>> >> ---
>> >>  drivers/uio/Kconfig       |    8 ++
>> >>  drivers/uio/Makefile      |    1 +
>> >>  drivers/uio/uio_ivshmem.c |  235 +++++++++++++++++++++++++++++++++++++++++++++
>> >>  3 files changed, 244 insertions(+), 0 deletions(-)
>> >>  create mode 100644 drivers/uio/uio_ivshmem.c
>> >>
>> >> diff --git a/drivers/uio/Kconfig b/drivers/uio/Kconfig
>> >> index 1da73ec..b92cded 100644
>> >> --- a/drivers/uio/Kconfig
>> >> +++ b/drivers/uio/Kconfig
>> >> @@ -74,6 +74,14 @@ config UIO_SERCOS3
>> >>
>> >>         If you compile this as a module, it will be called uio_sercos3.
>> >>
>> >> +config UIO_IVSHMEM
>> >> +     tristate "KVM shared memory PCI driver"
>> >> +     default n
>> >> +     help
>> >> +       Userspace I/O interface for the KVM shared memory device.  This
>> >> +       driver will make available two memory regions, the first is
>> >> +       registers and the second is a region for sharing between VMs.
>> >> +
>> >>  config UIO_PCI_GENERIC
>> >>       tristate "Generic driver for PCI 2.3 and PCI Express cards"
>> >>       depends on PCI
>> >> diff --git a/drivers/uio/Makefile b/drivers/uio/Makefile
>> >> index 18fd818..25c1ca5 100644
>> >> --- a/drivers/uio/Makefile
>> >> +++ b/drivers/uio/Makefile
>> >> @@ -6,3 +6,4 @@ obj-$(CONFIG_UIO_AEC) += uio_aec.o
>> >>  obj-$(CONFIG_UIO_SERCOS3)    += uio_sercos3.o
>> >>  obj-$(CONFIG_UIO_PCI_GENERIC)        += uio_pci_generic.o
>> >>  obj-$(CONFIG_UIO_NETX)       += uio_netx.o
>> >> +obj-$(CONFIG_UIO_IVSHMEM) += uio_ivshmem.o
>> >> diff --git a/drivers/uio/uio_ivshmem.c b/drivers/uio/uio_ivshmem.c
>> >> new file mode 100644
>> >> index 0000000..607435b
>> >> --- /dev/null
>> >> +++ b/drivers/uio/uio_ivshmem.c
>> >> @@ -0,0 +1,235 @@
>> >> +/*
>> >> + * UIO IVShmem Driver
>> >> + *
>> >> + * (C) 2009 Cam Macdonell
>> >> + * based on Hilscher CIF card driver (C) 2007 Hans J. Koch <hjk@linutronix.de>
>> >> + *
>> >> + * Licensed under GPL version 2 only.
>> >> + *
>> >> + */
>> >> +
>> >> +#include <linux/device.h>
>> >> +#include <linux/module.h>
>> >> +#include <linux/pci.h>
>> >> +#include <linux/uio_driver.h>
>> >> +
>> >> +#include <asm/io.h>
>> >> +
>> >> +#define IntrStatus 0x04
>> >> +#define IntrMask 0x00
>> >> +
>> >> +struct ivshmem_info {
>> >> +    struct uio_info *uio;
>> >> +    struct pci_dev *dev;
>> >> +    char (*msix_names)[256];
>> >> +    struct msix_entry *msix_entries;
>> >> +    int nvectors;
>> >> +};
>> >> +
>> >> +static irqreturn_t ivshmem_handler(int irq, struct uio_info *dev_info)
>> >> +{
>> >> +
>> >> +    void __iomem *plx_intscr = dev_info->mem[0].internal_addr
>> >> +                    + IntrStatus;
>> >> +    u32 val;
>> >> +
>> >> +    val = readl(plx_intscr);
>> >> +    if (val == 0)
>> >> +        return IRQ_NONE;
>> >> +
>> >> +    printk(KERN_INFO "Regular interrupt (val = %d)\n", val);
>> >> +    return IRQ_HANDLED;
>> >> +}
>> >> +
>> >> +static irqreturn_t ivshmem_msix_handler(int irq, void *opaque)
>> >> +{
>> >> +
>> >> +    struct uio_info * dev_info = (struct uio_info *) opaque;
>> >> +
>> >> +    /* we have to do this explicitly when using MSI-X */
>> >> +    uio_event_notify(dev_info);
>> >
>> > How does userspace know which vector was triggered?
>>
>> Right now, it doesn't.  If a user had a particular need they would
>> need to write their own uio driver.  I guess this leads to a
>> discussion of MSI support in UIO and how they would work with the
>> userspace.
>
> So why request more than one vector then?

No strong reason, to some extent an artifact of my playing with uio
and MSI together.  But, since interrupts on vector 0 are raised on
guest joins/exits, a user could modify the driver to ignore those if
they wanted.  Also, this way the device doesn't need to know if the
guest is using 1 or 2 vectors.

>
>> >
>> >> +    printk(KERN_INFO "MSI-X interrupt (%d)\n", irq);
>> >> +    return IRQ_HANDLED;
>> >> +
>> >
>> > extra empty line
>> >
>> >> +}
>> >> +
>> >> +static int request_msix_vectors(struct ivshmem_info *ivs_info, int nvectors)
>> >> +{
>> >> +    int i, err;
>> >> +    const char *name = "ivshmem";
>> >> +
>> >> +    printk(KERN_INFO "devname is %s\n", name);
>> >
>> > These KERN_INFO messages need to be cleaned up, they would be
>> > look confusing in the log.
>>
>> Agreed.  I will clean most of these out.
>>
>> >
>> >> +    ivs_info->nvectors = nvectors;
>> >> +
>> >> +
>> >> +    ivs_info->msix_entries = kmalloc(nvectors * sizeof *ivs_info->msix_entries,
>> >> +            GFP_KERNEL);
>> >> +    ivs_info->msix_names = kmalloc(nvectors * sizeof *ivs_info->msix_names,
>> >> +            GFP_KERNEL);
>> >> +
>> >
>> > need to handle errors
>> >
>> >> +    for (i = 0; i < nvectors; ++i)
>> >> +        ivs_info->msix_entries[i].entry = i;
>> >> +
>> >> +    err = pci_enable_msix(ivs_info->dev, ivs_info->msix_entries,
>> >> +                    ivs_info->nvectors);
>> >> +    if (err > 0) {
>> >> +        ivs_info->nvectors = err; /* msi-x positive error code
>> >> +                                     returns the number available*/
>> >> +        err = pci_enable_msix(ivs_info->dev, ivs_info->msix_entries,
>> >> +                    ivs_info->nvectors);
>> >> +        if (err > 0) {
>> >> +            printk(KERN_INFO "no MSI (%d). Back to INTx.\n", err);
>> >> +            return -ENOSPC;
>> >> +        }
>> >> +    }
>> >
>> > we can also get err < 0.
>> >
>> >> +
>> >> +    printk(KERN_INFO "err is %d\n", err);
>> >> +    if (err) return err;
>> >
>> > coding style rule violation
>> >
>> >> +
>> >> +    for (i = 0; i < ivs_info->nvectors; i++) {
>> >> +
>> >> +        snprintf(ivs_info->msix_names[i], sizeof *ivs_info->msix_names,
>> >> +            "%s-config", name);
>> >> +
>> >> +        ivs_info->msix_entries[i].entry = i;
>> >> +        err = request_irq(ivs_info->msix_entries[i].vector,
>> >> +            ivshmem_msix_handler, 0,
>> >> +            ivs_info->msix_names[i], ivs_info->uio);
>> >> +
>> >> +        if (err) {
>> >> +            return -ENOSPC;
>> >
>> > coding style rule violation
>> > no undo on error handling
>> > why override error code with -ENOSPC?
>>
>> Ah, I think I've confused linux and qemu coding styles perhaps.  I'll
>> fix these and others above.
>>
>> >
>> >> +        }
>> >> +    }
>> >> +
>> >> +    return 0;
>> >> +}
>> >> +
>> >> +static int __devinit ivshmem_pci_probe(struct pci_dev *dev,
>> >> +                    const struct pci_device_id *id)
>> >> +{
>> >> +    struct uio_info *info;
>> >> +    struct ivshmem_info * ivshmem_info;
>> >> +    int nvectors = 4;
>> >> +
>> >> +    info = kzalloc(sizeof(struct uio_info), GFP_KERNEL);
>> >> +    if (!info)
>> >> +        return -ENOMEM;
>> >> +
>> >> +    ivshmem_info = kzalloc(sizeof(struct ivshmem_info), GFP_KERNEL);
>> >> +    if (!ivshmem_info) {
>> >> +        kfree(info);
>> >> +        return -ENOMEM;
>> >> +    }
>> >> +
>> >> +    if (pci_enable_device(dev))
>> >> +        goto out_free;
>> >> +
>> >> +    if (pci_request_regions(dev, "ivshmem"))
>> >> +        goto out_disable;
>> >> +
>> >> +    info->mem[0].addr = pci_resource_start(dev, 0);
>> >> +    if (!info->mem[0].addr)
>> >> +        goto out_release;
>> >> +
>> >> +    info->mem[0].size = pci_resource_len(dev, 0);
>> >> +    info->mem[0].internal_addr = pci_ioremap_bar(dev, 0);
>> >> +    if (!info->mem[0].internal_addr) {
>> >> +        printk(KERN_INFO "got a null");
>> >> +        goto out_release;
>> >> +    }
>> >> +
>> >> +    info->mem[0].memtype = UIO_MEM_PHYS;
>> >> +
>> >> +    info->mem[1].addr = pci_resource_start(dev, 2);
>> >> +    if (!info->mem[1].addr)
>> >> +        goto out_unmap;
>> >> +    info->mem[1].internal_addr = pci_ioremap_bar(dev, 2);
>> >> +    if (!info->mem[1].internal_addr)
>> >> +        goto out_unmap;
>> >> +
>> >> +    info->mem[1].size = pci_resource_len(dev, 2);
>> >> +    info->mem[1].memtype = UIO_MEM_PHYS;
>> >> +
>> >> +    ivshmem_info->uio = info;
>> >> +    ivshmem_info->dev = dev;
>> >> +
>> >> +    if (request_msix_vectors(ivshmem_info, nvectors) != 0) {
>> >> +        printk(KERN_INFO "regular IRQs\n");
>> >> +        info->irq = dev->irq;
>> >> +        info->irq_flags = IRQF_SHARED;
>> >> +        info->handler = ivshmem_handler;
>> >> +        writel(0xffffffff, info->mem[0].internal_addr + IntrMask);
>> >> +    } else {
>> >> +        printk(KERN_INFO "MSI-X enabled\n");
>> >> +        info->irq = -1;
>> >> +    }
>> >> +
>> >> +    info->name = "ivshmem";
>> >> +    info->version = "0.0.1";
>> >> +
>> >> +    if (uio_register_device(&dev->dev, info))
>> >> +        goto out_unmap2;
>> >> +
>> >> +    pci_set_drvdata(dev, info);
>> >> +
>> >> +
>> >> +    return 0;
>> >> +out_unmap2:
>> >> +    iounmap(info->mem[2].internal_addr);
>> >> +out_unmap:
>> >> +    iounmap(info->mem[0].internal_addr);
>> >> +out_release:
>> >> +    pci_release_regions(dev);
>> >> +out_disable:
>> >> +    pci_disable_device(dev);
>> >> +out_free:
>> >> +    kfree (info);
>> >> +    return -ENODEV;
>> >> +}
>> >> +
>> >> +static void ivshmem_pci_remove(struct pci_dev *dev)
>> >> +{
>> >> +    struct uio_info *info = pci_get_drvdata(dev);
>> >> +
>> >> +    uio_unregister_device(info);
>> >> +    pci_release_regions(dev);
>> >> +    pci_disable_device(dev);
>> >> +    pci_set_drvdata(dev, NULL);
>> >> +    iounmap(info->mem[0].internal_addr);
>> >> +
>> >> +    kfree (info);
>> >> +}
>> >> +
>> >> +static struct pci_device_id ivshmem_pci_ids[] __devinitdata = {
>> >> +    {
>> >> +        .vendor =    0x1af4,
>> >> +        .device =    0x1110,
>> >
>> > vendor ids must be registered with PCI SIG.
>> > this one does not seem to be registered.
>> >
>> >> +        .subvendor =    PCI_ANY_ID,
>> >> +        .subdevice =    PCI_ANY_ID,
>> >> +    },
>> >> +    { 0, }
>> >> +};
>> >> +
>> >> +static struct pci_driver ivshmem_pci_driver = {
>> >> +    .name = "uio_ivshmem",
>> >> +    .id_table = ivshmem_pci_ids,
>> >> +    .probe = ivshmem_pci_probe,
>> >> +    .remove = ivshmem_pci_remove,
>> >> +};
>> >> +
>> >> +static int __init ivshmem_init_module(void)
>> >> +{
>> >> +    return pci_register_driver(&ivshmem_pci_driver);
>> >> +}
>> >> +
>> >> +static void __exit ivshmem_exit_module(void)
>> >> +{
>> >> +    pci_unregister_driver(&ivshmem_pci_driver);
>> >> +}
>> >> +
>> >> +module_init(ivshmem_init_module);
>> >> +module_exit(ivshmem_exit_module);
>> >> +
>> >> +MODULE_DEVICE_TABLE(pci, ivshmem_pci_ids);
>> >> +MODULE_LICENSE("GPL v2");
>> >> +MODULE_AUTHOR("Cam Macdonell");
>> >> --
>> >> 1.6.6.1
>> >>
>> >> --
>> >> To unsubscribe from this list: send the line "unsubscribe kvm" in
>> >> the body of a message to majordomo@vger.kernel.org
>> >> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> >
>> >
>
>