Register our DRHD IOMMUs, cross link devices, and provide a base set
of attributes for the IOMMU.  Note that IRQ remapping support parses
the DMAR table very early in boot, well before the iommu_class can
reasonably be setup, so our registration is split between
intel_iommu_init(), which occurs later, and alloc_iommu(), which
typically occurs much earlier, but may happen at any time later
with IOMMU hot-add support.

On a typical desktop system, this provides the following (pruned):

$ find /sys | grep dmar
/sys/devices/virtual/iommu/dmar0
/sys/devices/virtual/iommu/dmar0/devices
/sys/devices/virtual/iommu/dmar0/devices/0000:00:02.0
/sys/devices/virtual/iommu/dmar0/intel-iommu
/sys/devices/virtual/iommu/dmar0/intel-iommu/cap
/sys/devices/virtual/iommu/dmar0/intel-iommu/ecap
/sys/devices/virtual/iommu/dmar0/intel-iommu/address
/sys/devices/virtual/iommu/dmar0/intel-iommu/version
/sys/devices/virtual/iommu/dmar1
/sys/devices/virtual/iommu/dmar1/devices
/sys/devices/virtual/iommu/dmar1/devices/0000:00:00.0
/sys/devices/virtual/iommu/dmar1/devices/0000:00:01.0
/sys/devices/virtual/iommu/dmar1/devices/0000:00:16.0
/sys/devices/virtual/iommu/dmar1/devices/0000:00:1a.0
/sys/devices/virtual/iommu/dmar1/devices/0000:00:1b.0
/sys/devices/virtual/iommu/dmar1/devices/0000:00:1c.0
...
/sys/devices/virtual/iommu/dmar1/intel-iommu
/sys/devices/virtual/iommu/dmar1/intel-iommu/cap
/sys/devices/virtual/iommu/dmar1/intel-iommu/ecap
/sys/devices/virtual/iommu/dmar1/intel-iommu/address
/sys/devices/virtual/iommu/dmar1/intel-iommu/version
/sys/class/iommu/dmar0
/sys/class/iommu/dmar1

(devices also link back to the dmar units)

This makes address, version, capabilities, and extended capabilities
available, just like printed on boot.  I've tried not to duplicate
data that can be found in the DMAR table, with the exception of the
address, which provides an easy way to associate the sysfs device with
a DRHD entry in the DMAR.  It's tempting to add scopes and RMRR data
here, but the full DMAR table is already exposed under /sys/firmware/
and therefore already provides a way for userspace to learn such
details.
Signed-off-by: NAlex Williamson <alex.williamson@redhat.com>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

VT-d code currently makes use of pci_find_upstream_pcie_bridge() in
order to find the topology based alias of a device.  This function has
a few problems.  First, it doesn't check the entire alias path of the
device to the root bus, therefore if a PCIe device is masked upstream,
the wrong result is produced.  Also, it's known to get confused and
give up when it crosses a bridge from a conventional PCI bus to a PCIe
bus that lacks a PCIe capability.  The PCI-core provided DMA alias
support solves both of these problems and additionally adds support
for DMA function quirks allowing VT-d to work with devices like
Marvell and Ricoh with known broken requester IDs.
Signed-off-by: NAlex Williamson <alex.williamson@redhat.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

The IOMMU code now provides a common interface for finding or
creating an IOMMU group for a device on PCI buses.  Make use of it
and remove piles of code.
Signed-off-by: NAlex Williamson <alex.williamson@redhat.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

Function dmar_iommu_notify_scope_dev() makes a wrong assumption that
there's one RMRR for each PCI device at most, which causes DMA failure
on some HP platforms. So enhance dmar_iommu_notify_scope_dev() to
handle multiple RMRRs for the same PCI device.

Fixbug: https://bugzilla.novell.com/show_bug.cgi?id=879482

Cc: <stable@vger.kernel.org> # 3.15
Reported-by: NTom Mingarelli <thomas.mingarelli@hp.com>
Tested-by: NLinda Knippers <linda.knippers@hp.com>
Signed-off-by: NJiang Liu <jiang.liu@linux.intel.com>
Signed-off-by: NJoerg Roedel <jroedel@suse.de>

This adds support for the DMA Contiguous Memory Allocator for
intel-iommu.  This change enables dma_alloc_coherent() to allocate big
contiguous memory.

It is achieved in the same way as nommu_dma_ops currently does, i.e.
trying to allocate memory by dma_alloc_from_contiguous() and
alloc_pages() is used as a fallback.
Signed-off-by: NAkinobu Mita <akinobu.mita@gmail.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 146922ec ("iommu/vt-d: Make get_domain_for_dev() take struct
device") introduced new variables bridge_bus and bridge_devfn to
identify the upstream PCIe to PCI bridge responsible for the given
target device. Leaving the original bus/devfn variables to identify
the target device itself, now that it is no longer assumed to be PCI
and we can no longer trivially find that information.

However, the patch failed to correctly use the new variables in all
cases; instead using the as-yet-uninitialised 'bus' and 'devfn'
variables.
Reported-by: NAlex Williamson <alex.williamson@redhat.com>
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Commit ea8ea460 "iommu/vt-d: Clean up and fix page table clear/free
behaviour" introduces possible leakage of DMA page tables due to:
        for (pte = page_address(pg); !first_pte_in_page(pte); pte++) {
                if (dma_pte_present(pte) && !dma_pte_superpage(pte))
                        freelist = dma_pte_list_pagetables(domain, level - 1,
                                                           pte, freelist);
        }

For the first pte in a page, first_pte_in_page(pte) will always be true,
thus dma_pte_list_pagetables() will never be called and leak DMA page
tables if level is bigger than 1.
Signed-off-by: NJiang Liu <jiang.liu@linux.intel.com>
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

If we hit this error condition then we want to return a NULL pointer and
not a freed variable.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

It might not be...
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Mostly made redundant by using dev_name() instead of pci_name(), and one
instance of using *dev->dma_mask instead of pdev->dma_mask.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Should hopefully never happen (RMRRs are an abomination) but while we're
busy eliminating all the PCI assumptions, we might as well do it.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Pass the struct device to it, and also make it return the bus/devfn to use,
since that is also stored in the DMAR table.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

It's accessible via info->iommu->segment so this is redundant.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

This was problematic because it works by domain/bus/devfn and we want
to make device_to_iommu() use only a struct device * (for handling non-PCI
devices). Now that the iommu pointer is reliably stored in the
device_domain_info, we don't need to look it up.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Now we store the iommu in the device_domain_info, we don't need to do a
lookup.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

By moving this into get_domain_for_dev() we can make dmar_insert_dev_info()
suitable for use with "special" domains such as the si_domain, which
currently use domain_add_dev_info().
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

It's not only for PCI devices any more, and the scope information for an
ACPI device provides the bus and devfn so that has to be stored here too.

It is the device pointer itself which needs to be protected with RCU,
so the __rcu annotation follows it into the definition of struct
dmar_dev_scope, since we're no longer just passing arrays of device
pointers around.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

In commit 2e12bc29 ("intel-iommu: Default to non-coherent for domains
unattached to iommus") we decided to err on the side of caution and
always assume that it's possible that a device will be attached which is
behind a non-coherent IOMMU.

In some cases, however, that just *cannot* happen. If there *are* no
IOMMUs in the system which are non-coherent, then we don't need to do
it. And flushing the dcache is a *significant* performance hit.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

There is a race condition between the existing clear/free code and the
hardware. The IOMMU is actually permitted to cache the intermediate
levels of the page tables, and doesn't need to walk the table from the
very top of the PGD each time. So the existing back-to-back calls to
dma_pte_clear_range() and dma_pte_free_pagetable() can lead to a
use-after-free where the IOMMU reads from a freed page table.

When freeing page tables we actually need to do the IOTLB flush, with
the 'invalidation hint' bit clear to indicate that it's not just a
leaf-node flush, after unlinking each page table page from the next level
up but before actually freeing it.

So in the rewritten domain_unmap() we just return a list of pages (using
pg->freelist to make a list of them), and then the caller is expected to
do the appropriate IOTLB flush (or tear down the domain completely,
whatever), before finally calling dma_free_pagelist() to free the pages.

As an added bonus, we no longer need to flush the CPU's data cache for
pages which are about to be *removed* from the page table hierarchy anyway,
in the non-cache-coherent case. This drastically improves the performance
of large unmaps.

As a side-effect of all these changes, this also fixes the fact that
intel_iommu_unmap() was neglecting to free the page tables for the range
in question after clearing them.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>

We have this horrid API where iommu_unmap() can unmap more than it's asked
to, if the IOVA in question happens to be mapped with a large page.

Instead of propagating this nonsense to the point where we end up returning
the page order from dma_pte_clear_range(), let's just do it once and adjust
the 'size' parameter accordingly.

Augment pfn_to_dma_pte() to return the level at which the PTE was found,
which will also be useful later if we end up changing the API for
iommu_iova_to_phys() to behave the same way as is being discussed upstream.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>