The EFR[HATS] bits indicate maximum host translation level supported by
IOMMU. Adding support to set the maximum host page table level as indicated
by EFR[HATS]. If the HATS=11b (reserved), the driver will attempt to use
guest page table for DMA API.
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Ankit Soni <Ankit.Soni@amd.com>
Link: https://lore.kernel.org/r/df0f8562c2a20895cc185c86f1a02c4d826fd597.1749016436.git.Ankit.Soni@amd.com
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Use iommu_alloc_pages_node_sz() instead.
AMD and Intel are both using 4K pages for these structures since those
drivers only work on 4K PAGE_SIZE.
riscv is also spec'd to use SZ_4K.
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Tested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/21-v4-c8663abbb606+3f7-iommu_pages_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Change the internal freelist to use struct iommu_pages_list.
AMD uses the freelist to batch free the entire table during domain
destruction, and to replace table levels with leafs during map.
Tested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/10-v4-c8663abbb606+3f7-iommu_pages_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
If a page is mapped starting at 0 that is equal to or larger than can fit
in the current mode (number of table levels) it results in corrupting the
mapping as the following logic assumes the mode is correct for the page
size being requested.
There are two issues here, the check if the address fits within the table
uses the start address, it should use the last address to ensure that last
byte of the mapping fits within the current table mode.
The second is if the mapping is exactly the size of the full page table it
has to add another level to instead hold a single IOPTE for the large
size.
Since both corner cases require a 0 IOVA to be hit and doesn't start until
a page size of 2^48 it is unlikely to ever hit in a real system.
Reported-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/0-v1-27ab08d646a1+29-amd_0map_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Add two new kernel command line parameters to limit the page-sizes
used for v1 page-tables:
nohugepages - Limits page-sizes to 4KiB
v2_pgsizes_only - Limits page-sizes to 4Kib/2Mib/1GiB; The
same as the sizes used with v2 page-tables
This is needed for multiple scenarios. When assigning devices to
SEV-SNP guests the IOMMU page-sizes need to match the sizes in the RMP
table, otherwise the device will not be able to access all shared
memory.
Also, some ATS devices do not work properly with arbitrary IO
page-sizes as supported by AMD-Vi, so limiting the sizes used by the
driver is a suitable workaround.
All-in-all, these parameters are only workarounds until the IOMMU core
and related APIs gather the ability to negotiate the page-sizes in a
better way.
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/20240905072240.253313-1-joro@8bytes.org
The iommu driver is supposed to provide these ops to its io_pgtable
implementation so that it can hook the invalidations and do the right
thing.
They are called by wrapper functions like io_pgtable_tlb_add_page() etc,
which the AMD code never calls.
Instead it directly calls the AMD IOMMU invalidation functions by casting
to the struct protection_domain. Remove it all.
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/12-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The AMD io_pgtable stuff doesn't implement the tlb ops callbacks, instead
it invokes the invalidation ops directly on the struct protection_domain.
Narrow the use of struct protection_domain to only those few code paths.
Make everything else properly use struct amd_io_pgtable through the call
chains, which is the correct modular type for an io-pgtable module.
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/9-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
We already have memory in the union here that is being wasted in AMD's
case, use it to store the nid.
Putting the nid here further isolates the io_pgtable code from the struct
protection_domain.
Fixup protection_domain_alloc so that the NID from the device is provided,
at this point dev is never NULL for AMD so this will now allocate the
first table pointer on the correct NUMA node.
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/8-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
There is struct protection_domain iopt and struct amd_io_pgtable iopt.
Next patches are going to want to write domain.iopt.iopt.xx which is quite
unnatural to read.
Give one of them a different name, amd_io_pgtable has fewer references so
call it pgtbl, to match pgtbl_cfg, instead.
Suggested-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/6-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Looks like many refactorings here have left this confused. There is only
one storage of the root/mode, it is in the iop struct.
increase_address_space() calls amd_iommu_domain_set_pgtable() with values
that it already stored in iop a few lines above.
amd_iommu_domain_clr_pt_root() is zero'ing memory we are about to free. It
used to protect against a double free of root, but that is gone now.
Remove amd_iommu_domain_set_pgtable(), amd_iommu_domain_set_pt_root(),
amd_iommu_domain_clr_pt_root() as they are all pointless.
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/5-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
It is a serious bug if the domain is still mapped to any DTEs when it is
freed as we immediately start freeing page table memory, so any remaining
HW touch will UAF.
If it is not mapped then dev_list is empty and amd_iommu_domain_update()
does nothing.
Remove it and add a WARN_ON() to catch this class of bug.
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/4-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
All the page table memory should be allocated/free within the io_pgtable
struct. The v2 path is already doing this, make it consistent.
It is hard to see but the free of the root in protection_domain_free() is
a NOP on the success path because v1_free_pgtable() does
amd_iommu_domain_clr_pt_root().
The root memory is already freed because free_sub_pt() put it on the
freelist. The free path in protection_domain_free() is only used during
error unwind of protection_domain_alloc().
Reviewed-by: Vasant Hegde <vasant.hegde@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/1-v2-831cdc4d00f3+1a315-amd_iopgtbl_jgg@nvidia.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
AMD driver uses amd_iommu_domain_flush_complete() function to make sure
IOMMU processed invalidation commands before proceeding. Ideally this
should be called from functions which updates DTE/invalidates caches.
There is no need to call this function explicitly. This patches makes
below changes :
- Rename amd_iommu_domain_flush_complete() -> domain_flush_complete()
and make it as static function.
- Rearrage domain_flush_complete() to avoid forward declaration.
- Update amd_iommu_update_and_flush_device_table() to call
domain_flush_complete().
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/20240828111029.5429-7-vasant.hegde@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Replace a comma between expression statements by a semicolon.
Fixes: c9b258c6be ("iommu/amd: Prepare for generic IO page table framework")
Signed-off-by: Chen Ni <nichen@iscas.ac.cn>
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/20240716072545.968690-1-nichen@iscas.ac.cn
Signed-off-by: Will Deacon <will@kernel.org>
- Rename domain_flush_pages() -> amd_iommu_domain_flush_pages() and make
it as global function.
- Rename amd_iommu_domain_flush_tlb_pde() -> amd_iommu_domain_flush_all()
and make it as static.
- Convert v1 page table (io_pgtble.c) to use amd_iommu_domain_flush_pages().
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/20231122090215.6191-9-vasant.hegde@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Call amd_iommu_domain_flush_complete() from domain_flush_pages().
That way we can remove explicit call of amd_iommu_domain_flush_complete()
from various places.
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/20231122090215.6191-8-vasant.hegde@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
IOMMU advertises Access/Dirty bits if the extended feature register reports
it. Relevant AMD IOMMU SDM ref[0] "1.3.8 Enhanced Support for Access and
Dirty Bits"
To enable it set the DTE flag in bits 7 and 8 to enable access, or
access+dirty. With that, the IOMMU starts marking the D and A flags on
every Memory Request or ATS translation request. It is on the VMM side to
steer whether to enable dirty tracking or not, rather than wrongly doing in
IOMMU. Relevant AMD IOMMU SDM ref [0], "Table 7. Device Table Entry (DTE)
Field Definitions" particularly the entry "HAD".
To actually toggle on and off it's relatively simple as it's setting 2 bits
on DTE and flush the device DTE cache.
To get what's dirtied use existing AMD io-pgtable support, by walking the
pagetables over each IOVA, with fetch_pte(). The IOTLB flushing is left to
the caller (much like unmap), and iommu_dirty_bitmap_record() is the one
adding page-ranges to invalidate. This allows caller to batch the flush
over a big span of IOVA space, without the iommu wondering about when to
flush.
Worthwhile sections from AMD IOMMU SDM:
"2.2.3.1 Host Access Support"
"2.2.3.2 Host Dirty Support"
For details on how IOMMU hardware updates the dirty bit see, and expects
from its consequent clearing by CPU:
"2.2.7.4 Updating Accessed and Dirty Bits in the Guest Address Tables"
"2.2.7.5 Clearing Accessed and Dirty Bits"
Quoting the SDM:
"The setting of accessed and dirty status bits in the page tables is
visible to both the CPU and the peripheral when sharing guest page tables.
The IOMMU interlocked operations to update A and D bits must be 64-bit
operations and naturally aligned on a 64-bit boundary"
.. and for the IOMMU update sequence to Dirty bit, essentially is states:
1. Decodes the read and write intent from the memory access.
2. If P=0 in the page descriptor, fail the access.
3. Compare the A & D bits in the descriptor with the read and write
intent in the request.
4. If the A or D bits need to be updated in the descriptor:
* Start atomic operation.
* Read the descriptor as a 64-bit access.
* If the descriptor no longer appears to require an update, release the
atomic lock with
no further action and continue to step 5.
* Calculate the new A & D bits.
* Write the descriptor as a 64-bit access.
* End atomic operation.
5. Continue to the next stage of translation or to the memory access.
Access/Dirty bits readout also need to consider the non-default page-sizes
(aka replicated PTEs as mentined by manual), as AMD supports all powers of
two (except 512G) page sizes.
Select IOMMUFD_DRIVER only if IOMMUFD is enabled considering that IOMMU
dirty tracking requires IOMMUFD.
Link: https://lore.kernel.org/r/20231024135109.73787-12-joao.m.martins@oracle.com
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Introduce 'struct protection_domain->nid' variable. It will contain
IOMMU NUMA node ID. And allocate page table pages using IOMMU numa
locality info. This optimizes page table walk by IOMMU.
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/20230321092348.6127-2-vasant.hegde@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Implement the io_pgtable_ops->unmap_pages() callback for AMD driver
and deprecate io_pgtable_ops->unmap callback.
Also if fetch_pte() returns NULL then return from unmap_mapages()
instead of trying to continue to unmap remaining pages.
Suggested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Vasant Hegde <vasant.hegde@amd.com>
Link: https://lore.kernel.org/r/20220825063939.8360-3-vasant.hegde@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Use try_cmpxchg64 instead of cmpxchg64 (*ptr, old, new) != old in
alloc_pte and free_clear_pte. cmpxchg returns success in ZF flag, so this
change saves a compare after cmpxchg (and related move instruction
in front of cmpxchg). Also, remove racy explicit assignment to pteval
when cmpxchg fails, this is what try_cmpxchg does implicitly from
*pte in an atomic way.
Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20220525145416.10816-1-ubizjak@gmail.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
The current logic updates the I/O page table mode for the domain
before calling the logic to free memory used for the page table.
This results in IOMMU page table memory leak, and can be observed
when launching VM w/ pass-through devices.
Fix by freeing the memory used for page table before updating the mode.
Cc: Joerg Roedel <joro@8bytes.org>
Reported-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Tested-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Fixes: e42ba06330 ("iommu/amd: Restructure code for freeing page table")
Link: https://lore.kernel.org/all/20220118194720.urjgi73b7c3tq2o6@oracle.com/
Link: https://lore.kernel.org/r/20220210154745.11524-1-suravee.suthikulpanit@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
page->freelist is for the use of slab. We already have the ability
to free a list of pages in the core mm, but it requires the use of a
list_head and for the pages to be chained together through page->lru.
Switch the AMD IOMMU code over to using free_pages_list().
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
[rm: split from original patch, cosmetic tweaks]
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/73af128f651aaa1f38f69e586c66765a88ad2de0.1639753638.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
For reasons unclear, pagetable freeing is an effectively recursive
method implemented via an elaborate system of templated functions that
turns out to account for 25% of the object file size. Implementing it
using regular straightforward recursion makes the code simpler, and
seems like a good thing to do before we work on it further. As part of
that, also fix the types to avoid all the needless casting back and
forth which just gets in the way.
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/d3d00c9f3fa0df4756b867072c201e6e82f9ce39.1639753638.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
If people are going to insist on calling iommu_iova_to_phys()
pointlessly and expecting it to work, we can at least do ourselves a
favour by handling those cases in the core code, rather than repeatedly
across an inconsistent handful of drivers.
Since all the existing drivers implement the internal callback, and any
future ones are likely to want to work with iommu-dma which relies on
iova_to_phys a fair bit, we may as well remove that currently-redundant
check as well and consider it mandatory.
Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Link: https://lore.kernel.org/r/f564f3f6ff731b898ff7a898919bf871c2c7745a.1626354264.git.robin.murphy@arm.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
increase_address_space() calls get_zeroed_page(gfp) under spin_lock with
disabled interrupts. gfp flags passed to increase_address_space() may allow
sleeping, so it comes to this:
BUG: sleeping function called from invalid context at mm/page_alloc.c:4342
in_atomic(): 1, irqs_disabled(): 1, pid: 21555, name: epdcbbf1qnhbsd8
Call Trace:
dump_stack+0x66/0x8b
___might_sleep+0xec/0x110
__alloc_pages_nodemask+0x104/0x300
get_zeroed_page+0x15/0x40
iommu_map_page+0xdd/0x3e0
amd_iommu_map+0x50/0x70
iommu_map+0x106/0x220
vfio_iommu_type1_ioctl+0x76e/0x950 [vfio_iommu_type1]
do_vfs_ioctl+0xa3/0x6f0
ksys_ioctl+0x66/0x70
__x64_sys_ioctl+0x16/0x20
do_syscall_64+0x4e/0x100
entry_SYSCALL_64_after_hwframe+0x44/0xa9
Fix this by moving get_zeroed_page() out of spin_lock/unlock section.
Fixes: 754265bcab ("iommu/amd: Fix race in increase_address_space()")
Signed-off-by: Andrey Ryabinin <arbn@yandex-team.com>
Acked-by: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20210217143004.19165-1-arbn@yandex-team.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
These implement map and unmap for AMD IOMMU v1 pagetable, which
will be used by the IO pagetable framework.
Also clean up unused extern function declarations.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/20201215073705.123786-13-suravee.suthikulpanit@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Since the IO page table root and mode parameters have been moved into
the struct amd_io_pg, the function is no longer needed. Therefore,
remove it along with the struct domain_pgtable.
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
Link: https://lore.kernel.org/r/20201215073705.123786-9-suravee.suthikulpanit@amd.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
