This is the second of 2 related, but independent, patches. This is for
uio.c, the previous is for uio_pci_generic.c.
The 2 patches were previously one large patch. Changes for this version:
- uio_pci_generic.c just gets extensions so that a single fd can be used
by non-privileged processes for interrupt control and mmaps
- All of the DMA and IOMMU related stuff move to uio.c; no longer a need
to pass ioctls to individual uio drivers. It turns out that the code
is not PCI specific anyways.
- A new ioctl to pin DMA buffers to certain IO virtual addresses for KVM.
- New eventfd based interrupt notifications, including support for PCI
specific MSI and MSI-X interrupts.
- PCI specific code to reset PCI functions before and after use
diff -ruNP linux-2.6.33/drivers/uio/uio.c uio-2.6.33/drivers/uio/uio.c
--- linux-2.6.33/drivers/uio/uio.c 2010-02-24 10:52:17.000000000 -0800
+++ uio-2.6.33/drivers/uio/uio.c 2010-04-15 12:39:02.000000000 -0700
@@ -23,6 +23,11 @@
#include <linux/string.h>
#include <linux/kobject.h>
#include <linux/uio_driver.h>
+#include <linux/mmu_notifier.h>
+#include <linux/pci.h>
+#include <linux/iommu.h>
+#include <linux/eventfd.h>
+#include <linux/semaphore.h>
#define UIO_MAX_DEVICES 255
@@ -37,8 +42,37 @@
struct uio_info *info;
struct kobject *map_dir;
struct kobject *portio_dir;
+ struct pci_dev *pdev;
+ int pmaster;
+ struct semaphore gate;
+ int listeners;
+ atomic_t mapcount;
+ struct msix_entry *msix;
+ int nvec;
+ struct iommu_domain *domain;
+ int cachec;
+ struct eventfd_ctx *ev_irq;
+ struct eventfd_ctx *ev_msi;
+ struct eventfd_ctx **ev_msix;
};
+/*
+ * Structure for keeping track of memory nailed down by the
+ * user for DMA
+ */
+struct dma_map_page {
+ struct list_head list;
+ struct page **pages;
+ struct scatterlist *sg;
+ dma_addr_t daddr;
+ unsigned long vaddr;
+ int npage;
+ int rdwr;
+};
+
+static void uio_disable_msi(struct uio_device ...Why do you want to allow multiple opens for a device? Is it not
If userspace calls this path this will make all the addresses mapped
with DMA-API paths unusable by the device. This doesn't look like a sane
userspace interface.
For better and more in-depth review I suggest that you split up this
large patch into a series of smaler which implement specific aspects of
your work.
Joerg
P.S.: I got these warning when applying your patches ...
Applying: drivers/uio/uio_pci_generic.c: allow access for non-privileged processes
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:86: trailing whitespace.
info->mem[j].name = name;
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:87: trailing whitespace.
info->mem[j].addr = pci_resource_start(pdev, i);
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:89: trailing whitespace.
info->mem[j].memtype = UIO_MEM_PHYS;
warning: 3 lines add whitespace errors.
Applying: drivers/uio/uio.c: DMA mapping, interrupt extensions, etc.
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:315: trailing whitespace.
ret = iommu_map_range(idev->domain, iova,
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:325: trailing whitespace.
}
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:366: trailing whitespace.
* adjacent pages, but noone seems to really do that. So we squash
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:368: trailing whitespace.
* This works if (a) there is an iommu, or (b) the user allocates
/home/joro/src/linux.trees.git/.git/rebase-apply/patch:578: trailing whitespace.
unsigned int cmd, unsigned long arg)
warning: squelched 1 whitespace error
warning: 6 lines add whitespace errors.
And there are also some coding-style issues.
--
The current uio and uio_pci_generic allow multiple opens; I was just preserving that behavior. --
Use __u64 for both, otherwise you need to rewrite the structures in the What units? Size is probably too small. Suggest unsigned type to avoid These three need some documentation. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic. --
