The opencoded legacy Memory Disambiguation logic in init_amd() neglected
Fam19h for the Zen3 microarchitecture. Further more, all Zen2 based system
have the architectural MSR_SPEC_CTRL and the SSBD bit within it, so shouldn't
be using MSR_AMD64_LS_CFG.
Implement the algorithm given in AMD's SSBD whitepaper, and leave a
printk_once() behind in the case that no controls can be found.
This now means that a user explicitly choosing `spec-ctrl=ssbd` will properly
turn off Memory Disambiguation on Fam19h/Zen3 systems.
This still remains a single system-wide setting (for now), and is not context
switched between vCPUs. As such, it doesn't interact with Intel's use of
MSR_SPEC_CTRL and default_xen_spec_ctrl (yet).
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
There is a step change in speculation protections between the Zen1 and Zen2
microarchitectures.
Zen1 and older have no special support. Control bits in non-architectural
MSRs are used to make lfence be dispatch-serialising (Spectre v1), and to
disable Memory Disambiguation (Speculative Store Bypass). IBPB was
retrofitted in a microcode update, and software methods are required for
Spectre v2 protections.
Because the bit controlling Memory Disambiguation is model specific,
hypervisors are expected to expose a MSR_VIRT_SPEC_CTRL interface which
abstracts the model specific details.
Zen2 and later implement the MSR_SPEC_CTRL interface in hardware, and
virtualise the interface for HVM guests to use. A number of hint bits are
specified too to help guide OS software to the most efficient mitigation
strategy.
Zen3 introduced a new feature, Predictive Store Forwarding, along with a
control to disable it in sensitive code.
Add CPUID and VMCB details for all the new functionality.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Separate the read-only hints from the features requiring active actions on
Xen's behalf.
Also take the opportunity split the IBRS/IBPB and IBPB mess. More features
with overlapping enumeration are on the way, and and it is not useful to split
them like this.
No practical change.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
In order to avoid flask/Makefile.common calling `make xenversion`, we
override POLICY_FILENAME with the value we are going to use anyway.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
All Zen2 CPUs actually have this behaviour, but the CPUID bit couldn't
be introduced into Zen2 due to a lack of leaves. So, it was added in a
new leaf in Zen3. Nonetheless, hypervisors can synthesize the CPUID
bit in software.
So, Xen probes for NSCB (NullSelectorClearsBit) and
synthesizes the bit, if the behaviour is present.
Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Jane Malalane <jane.malalane@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Commit 8064488062 ("xen/domain: Fix label position in
domain_teardown()" has caused Coverity to report a _new_ supposedly
un-annotated fall-through in a switch(). I find this (once again)
puzzling; I'm having an increasingly hard time figuring what patterns
the tool is actually after. I would have expected that the tool would
either have spotted an issue also before this change, or not at all. Yet
if it had spotted one before, the statistics report should have included
an eliminated instance alongside the new one (because then the issue
would simply have moved by a few lines).
Hence the only thing I could guess is that the treatment of comments in
macro expansions might be subtly different. Therefore try whether
switching the comments to the still relatively new "fallthrough" pseudo
keyword actually helps.
Coverity-ID: 1490865
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Bertrand Marquis <bertrand.marquis@arm.com>
Acked-by: Julien Grall <jgrall@amazon.com>
Once gnttab_map_frame() drops the grant table lock, the MFN it reports
back to its caller is free to other manipulation. In particular
gnttab_unpopulate_status_frames() might free it, by a racing request on
another CPU, thus resulting in a reference to a deallocated page getting
added to a domain's P2M.
Obtain a page reference in gnttab_map_frame() to prevent freeing of the
page until xenmem_add_to_physmap_one() has actually completed its acting
on the page. Do so uniformly, even if only strictly required for v2
status pages, to avoid extra conditionals (which then would all need to
be kept in sync going forward).
This is CVE-2021-28701 / XSA-384.
Reported-by: Julien Grall <jgrall@amazon.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
The initial if() was inverted, invalidating all output from this
function. Which in turn means the mirroring of P2M mappings into the
IOMMU didn't always work as intended: Mappings may have got updated when
there was no need to. There would not have been too few (un)mappings;
what saves us is that alongside the flags comparison MFNs also get
compared, with non-present entries always having an MFN of 0 or
INVALID_MFN while present entries always have MFNs different from these
two (0 in the table also meant to cover INVALID_MFN):
OLD NEW
P W access MFN P W access MFN
0 0 r 0 0 0 n 0
0 1 rw 0 0 1 n 0
1 0 n non-0 1 0 r non-0
1 1 n non-0 1 1 rw non-0
present <-> non-present transitions are fine because the MFNs differ.
present -> present transitions as well as non-present -> non-present
ones are potentially causing too many map/unmap operations, but never
too few, because in that case old (bogus) and new access differ.
Fixes: d1bb6c97c3 ("IOMMU: also pass p2m_access_t to p2m_get_iommu_flags())
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
AMD Zen3 adds the NullSelectorClearsBase bit to indicate that loading
a NULL segment selector zeroes the base and limit fields, as well as
just attributes.
Expose bit to all guests.
Suggested-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Jane Malalane <jane.malalane@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
One of the changes comprising the fixes for XSA-378 disallows replacing
MMIO mappings by code paths not intended for this purpose. At least in
the case of PVH Dom0 hitting an RMRR covered by an E820 ACPI region,
this is too strict. Generally short-circuit requests establishing the
same kind of mapping (mfn, type), but allow permissions to differ.
While there, also add a log message to the other domain_crash()
invocation that did prevent PVH Dom0 from coming up after the XSA-378
changes.
Fixes: 753cb68e65 ("x86/p2m: guard (in particular) identity mapping entries")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
Both comment and message string associated with GNTST_no_device_space
suggest a connection to the IOMMU. A lack of maptrack handles has
nothing to do with that; it's unclear to me why commit 6213b696ba
("Grant-table interface redone") introduced it this way. Introduce a
new error indicator.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
There's no point checking ->dev_bus_addr when GNTMAP_device_map isn't
set (and hence the field isn't going to be consumed). And if there is a
mismatch, use the so far unused GNTST_bad_dev_addr error indicator - if
not here, where else would this (so far unused) value be used?
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
Passing byte granular values will not have the intended effect. Address
the immediate issue, but I don't think what we do is actually
sufficient: At least some devices allow access to their registers via
either I/O ports or MMIO. In such aliasing cases we'd need to protect
the MMIO range even when we use I/O port accesses to drive the port.
Note that this way we may write-protect MMIO ranges of unrelated devices
as well. To deal with this, faults resulting from this would need
handling, to emulate the accesses outside of the protected range. (An
alternative would be to relocate the BAR, but I'm afraid this might end
up even more challenging.)
Fixes: c9f8e0aee5 ("ns16550: Add support for UART present in Broadcom TruManage capable NetXtreme chips")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
Like done in gnttab_setup_table(), check the handle once early in the
function and use the lighter-weight (for PV) copying function in the
loop.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
In all cases call the function just once instead of up to four times, at
the same time avoiding to store a dangling pointer in a local variable.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
This gnttab_host_mapping_get_page_type() invocation sits in the "else"
path of a conditional controlled by "map->flags & GNTMAP_readonly".
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Julien Grall <jgrall@amazon.com>
Avoid different spelling for the location of "xen-syms", and simply
use the dependency variable. This avoid the assumption about $(TARGET)
value.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
And thus avoiding checking for those variable over and over again.
Also, add "e.g." in the error messages to hint that "menuconfig"
isn't the only way.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
The make variable $(subdir-y) isn't used yet but will be in a
following patch. Anything in $(subdir-y) doesn't to have a '/' as
suffix as we already now it's a directory.
Rework the rules so that it doesn't matter whether there is a '/' or
not. It also mimic more closely to the way Linux's Kbuild descend in
subdirectories.
FORCE phony target isn't needed anymore running clean, so it can be
removed.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Replace a mix of shell script and python script by all python script.
No change to the final generated headers.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Wei Liu <wl@xen.org>
Improvement are:
- give the path to xlat.lst as argument
- include `grep -v` in compat-build-source.py script, we don't need to
write this in several scripted language.
No changes in final compat/%.h headers.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
Use $(dot-target) to have the target name prefix with a dot.
Now, when the CC command has run, it is recorded in .*.cmd
file, then if_changed_rules will compare it on subsequent runs.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
In the case where $(obj-y) is empty, we also replace $(c_flags) by
$(XEN_CFLAGS) to avoid generating an .%.d dependency file. This avoid
make trying to include %.h file in the ld command if $(obj-y) isn't
empty anymore on a second run.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
According to ARMv8A architecture, AArch64 registers are 64bit wide
even though in many cases the upper 32bit is reserved. Therefore there
is no need for function vreg_emulate_sysreg32 on arm64. This means
that we can have just one function vreg_emulate_sysreg using new
function pointer:
typedef bool (*vreg_reg_fn_t)(struct cpu_user_regs *regs,
register_t *r, bool read);
Modify vreg_emulate_cp32 to use the new function pointer as well.
This change allows to properly use 64bit registers in AArch64 state.
In case of AArch32 the documentation (D1.20.2, DDI 0487A.j) states
that "the upper 32 bits either become zero, or hold the value that
the same architectural register held before any AArch32 execution." As
the choice between them is IMPLEMENTATION DEFINED we cannot assume they
are zeroed. Xen should ensure that but currently it does not. This is
not a new bug and must be fixed as agreed during a discussion over this
patch.
Take the opportunity to switch CNTx_CTL_* to use UL to avoid any
surprise with the negation of any bits (as used in vtimer_cntp_ctl)
Signed-off-by: Michal Orzel <michal.orzel@arm.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
Old enough glibc has clock_gettime() in librt.so, hence the library
needs to be specified to the linker. Newer glibc has the symbol
available in both libraries, so make sure that libc.so is preferred (to
avoid an unnecessary dependency on librt.so).
Fixes: 93c9edbef5 ("tests/xenstore: Rework Makefile")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: Ian Jackson <iwj@xenproject.org>
For one xc_private.h needlessly repeats xen-tools/libs.h's definition.
And then there are two suspicious uses (resulting from the inconsistency
with the respective 2nd parameter of DIV_ROUNDUP()): While the one in
tools/console/daemon/io.c - as per the code comment - intentionally uses
8 as the second argument (meaning to align to a multiple of 256), the
one in alloc_magic_pages_hvm() pretty certainly does not: There the goal
is to align to a uint64_t boundary, for the following module struct to
end up aligned.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: Ian Jackson <iwj@xenproject.org>
For log-dirty operations a 64-bit field is being truncated to become an
"int" return value. Seeing the large number of arguments the present
function takes, reduce its set of parameters to that needed for all
operations not involving the log-dirty bitmap, while introducing a new
wrapper for the log-dirty bitmap operations. This new function in turn
doesn't need an "mb" parameter, but has a 64-bit return type. (Using the
return value in favor of a pointer-type parameter is left as is, to
disturb callers as little as possible.)
While altering xc_shadow_control() anyway, also adjust the types of the
last two of the remaining parameters.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Christian Lindig <christian.lindig@citrix.com>
Reviewed-by: Juergen Gross <jgross@suse.com>
Acked-by: Ian Jackson <iwj@xenproject.org>
Don't assume tv_sec is a unsigned long, it is 64 bits on NetBSD 32 bits.
Use %jd and cast to (intmax_t) instead
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Acked-by: Ian Jackson <iwj@xenproject.org>
One of the changes comprising the fixes for XSA-378 disallows replacing
MMIO mappings by code paths not intended for this purpose. This means we
need to be more careful about the mappings put in place in this range -
mappings should be created exactly once:
- iommu_hwdom_init() comes first; it should avoid the first Mb,
- pvh_populate_p2m() should insert identity mappings only into ranges
not populated as RAM,
- pvh_setup_acpi() should again avoid the first Mb, which was already
dealt with at that point.
Fixes: 753cb68e65 ("x86/p2m: guard (in particular) identity mapping entries")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
VT-d spec 3.2 converted this bit (back) to reserved. Since there's no
use of it anywhere in the tree, simply rename it and adjust its comment.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
We unconditionally set HTT, so merely doubling the value read from
hardware isn't going to be correct if that value is zero.
Reported-by: Julien Grall <julien@xen.org>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Tested-by: Julien Grall <julien@xen.org>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
As explained in the comments, a progress label wants to be before the function
it refers to for the higher level logic to make sense. As it happens, the
effects are benign because gnttab_mappings is immediately adjacent to teardown
in terms of co-routine exit points.
There is and will always be a corner case with 0. Help alleviate this
visually (at least slightly) with a BUILD_BUG_ON() to ensure the property
which makes this function do anything useful.
There is also a visual corner case when changing from PROGRESS() to
PROGRESS_VCPU(). The important detail is to check that there is a "return
rc;" logically between each PROGRESS*() marker.
Fixes: b1ee10be56 ("gnttab: add preemption check to gnttab_release_mappings()")
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Jan Beulich <jbeulich@suse.com>
In some configurations, it is safe to not overwrite the RSB on entry to Xen.
Both Intel and AMD have guidelines in this area, because of the performance
difference it makes for native kernels.
A simple microperf test, measuring the amount of time a XENVER_version
hypercall takes, shows the following improvements:
KabyLake: -13.9175% +/- 6.85387%
CoffeeLake-R: -9.1183% +/- 5.04519%
Milan: -17.7803% +/- 1.29808%
This is best case improvement, because no real workloads are making
XENVER_version hypercalls in a tight loop. However, this is the hypercall
used by PV kernels to force evtchn delivery if one is pending, so it is a
common hypercall to see, especially in dom0.
The avoidance of RSB-overwriting speeds up all interrupts, exceptions and
system calls from PV or Xen context. RSB-overwriting is still required on
VMExit from HVM guests for now.
In terms of more realistic testing, LMBench in dom0 on an AMD Rome system
shows improvements across the board, with the best improvement at 8% for
simple syscall and simple write.
Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Jan Beulich <jbeulich@suse.com>
Relevant quotes from the C11 standard:
"Except where explicitly stated otherwise, for the purposes of this
subclause unnamed members of objects of structure and union type do not
participate in initialization. Unnamed members of structure objects
have indeterminate value even after initialization."
"If there are fewer initializers in a brace-enclosed list than there are
elements or members of an aggregate, [...], the remainder of the
aggregate shall be initialized implicitly the same as objects that have
static storage duration."
"If an object that has static or thread storage duration is not
initialized explicitly, then:
[...]
— if it is an aggregate, every member is initialized (recursively)
according to these rules, and any padding is initialized to zero
bits;
[...]"
"A bit-field declaration with no declarator, but only a colon and a
width, indicates an unnamed bit-field." Footnote: "An unnamed bit-field
structure member is useful for padding to conform to externally imposed
layouts."
"There may be unnamed padding within a structure object, but not at its
beginning."
Which makes me conclude:
- Whether an unnamed bit-field member is an unnamed member or padding is
unclear, and hence also whether the last quote above would render the
big endian case of the structure declaration invalid.
- Whether the number of members of an aggregate includes unnamed ones is
also not really clear.
- The initializer in map_grant_ref() initializes all fields of the "cnt"
sub-structure of the union, so assuming the second quote above applies
here (indirectly), the compiler isn't required to implicitly
initialize the rest (i.e. in particular any padding) like would happen
for static storage duration objects.
Gcc 7.4.1 can be observed (apparently in debug builds only) to translate
aforementioned initializer to a read-modify-write operation of a stack
variable, leaving unchanged the top two bits of whatever was previously
in that stack slot. Clearly if either of the two bits were set,
radix_tree_ulong_to_ptr()'s assertion would trigger.
Therefore, to be on the safe side, add an explicit padding field for the
non-big-endian-bitfields case and give a dummy name to both padding
fields.
Fixes: 9781b51efd ("gnttab: replace mapkind()")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>
There's neither documentation of what this flag is supposed to mean, nor
any implementation. Commit 4d45702cf0 ("paging: Updates to public
grant table header file") suggests there might have been plans to use it
for interaction with mem-paging, but no such functionality has ever
materialized. With this, don't even bother enclosing the #define-s in a
__XEN_INTERFACE_VERSION__ conditional, but drop them altogether.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Both callers only care about the target (level 1) MFN. I also cannot
see what we might need higher level MFNs for down the road. And even
modern gcc doesn't recognize the optimization potential.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
While for context cache entry flushing use of did 0 is indeed correct
(after all upon reading the context entry the IOMMU wouldn't know any
domain ID if the entry is not present, and hence a surrogate one needs
to be used), for IOTLB entries the normal domain ID (from the [present]
context entry) gets used. See sub-section "IOTLB" of section "Address
Translation Caches" in the VT-d spec.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Currently, both dom0less domUs and dom0 can allocate an "unlimited"
amount of memory because d->max_pages is set to ~0U.
In particular, the former are meant to be unprivileged. Therefore the
memory they could allocate should be bounded. As the domain are not yet
officially aware of Xen (we don't expose advertise it in the DT, yet
the hypercalls are accessible), they should not need to allocate more
than the initial amount. So cap set d->max_pages directly the amount of
memory we are meant to allocate.
Take the opportunity to also restrict the memory for dom0 as the
domain is direct mapped (e.g. MFN == GFN) and therefore cannot
allocate outside of the pre-allocated region.
This is CVE-2021-28700 / XSA-383.
Reported-by: Julien Grall <jgrall@amazon.com>
Signed-off-by: Julien Grall <jgrall@amazon.com>
Reviewed-by: Stefano Stabellini <sstabellini@kernel.org>
Tested-by: Stefano Stabellini <sstabellini@kernel.org>
The number of grant frames is of no interest here; converting the passed
in op.nr_frames this way means we allow for 8 times as many GFNs to be
written as actually fit in the array. We would corrupt xlat areas of
higher vCPU-s (after having faulted many times while trying to write to
the guard pages between any two areas) for 32-bit PV guests. For HVM
guests we'd simply crash as soon as we hit the first guard page, as
accesses to the xlat area are simply memcpy() there.
This is CVE-2021-28699 / XSA-382.
Fixes: 18b1be5e32 ("gnttab: make resource limits per domain")
Signed-off-by: Jan Beulich <jbeulich@suse.com>
mapkind() doesn't scale very well with larger maptrack entry counts,
using a brute force linear search through all entries, with the only
option of an early loop exit if a matching writable entry was found.
Introduce a radix tree alongside the main maptrack table, thus
allowing much faster MFN-based lookup. To avoid the need to actually
allocate space for the individual nodes, encode the two counters in the
node pointers themselves, thus limiting the number of permitted
simultaneous r/o and r/w mappings of the same MFN to 2³¹-1 (64-bit) /
2¹⁵-1 (32-bit) each.
To avoid enforcing an unnecessarily low bound on the number of
simultaneous mappings of a single MFN, introduce
radix_tree_{ulong_to_ptr,ptr_to_ulong} paralleling
radix_tree_{int_to_ptr,ptr_to_int}.
As a consequence locking changes are also applicable: With there no
longer being any inspection of the remote domain's active entries,
there's also no need anymore to hold the remote domain's grant table
lock. And since we're no longer iterating over the local domain's map
track table, the lock in map_grant_ref() can also be dropped before the
new maptrack entry actually gets populated.
As a nice side effect this also reduces the number of IOMMU operations
in unmap_common(): Previously we would have "established" a readable
mapping whenever we didn't find a writable entry anymore (yet, of
course, at least one readable one). But we only need to do this if we
actually dropped the last writable entry, not if there were none already
before.
This is part of CVE-2021-28698 / XSA-380.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
A guest may die with many grant mappings still in place, or simply with
a large maptrack table. Iterating through this may take more time than
is reasonable without intermediate preemption (to run softirqs and
perhaps the scheduler).
Move the invocation of the function to the section where other
restartable functions get invoked, and have the function itself check
for preemption every once in a while. Have it iterate the table
backwards, such that decreasing the maptrack limit is all it takes to
convey restart information.
In domain_teardown() introduce PROG_none such that inserting at the
front will be easier going forward.
This is part of CVE-2021-28698 / XSA-380.
Reported-by: Andrew Cooper <andrew.cooper3@citrix.com>
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
For pages which can be made part of the P2M by the guest, but which can
also later be de-allocated (grant table v2 status pages being the
present example), it is imperative that they be mapped at no more than a
single GFN. We therefore need to make sure that of two parallel
XENMAPSPACE_grant_table requests for the same status page one completes
before the second checks at which other GFN the underlying MFN is
presently mapped.
Pull ahead the respective get_gfn() and push down the respective
put_gfn(). This leverages that gfn_lock() really aliases p2m_lock(), but
the function makes this assumption already anyway: In the
XENMAPSPACE_gmfn case lock nesting constraints for both involved GFNs
would otherwise need to be enforced to avoid ABBA deadlocks.
This is CVE-2021-28697 / XSA-379.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
Such entries, created by set_identity_p2m_entry(), should only be
destroyed by clear_identity_p2m_entry(). However, similarly, entries
created by set_mmio_p2m_entry() should only be torn down by
clear_mmio_p2m_entry(), so the logic gets based upon p2m_mmio_direct as
the entry type (separation between "ordinary" and 1:1 mappings would
require a further indicator to tell apart the two).
As to the guest_remove_page() change, commit 48dfb297a2 ("x86/PVH:
allow guest_remove_page to remove p2m_mmio_direct pages"), which
introduced the call to clear_mmio_p2m_entry(), claimed this was done for
hwdom only without this actually having been the case. However, this
code shouldn't be there in the first place, as MMIO entries shouldn't be
dropped this way. Avoid triggering the warning again that 48dfb297a2
silenced by an adjustment to xenmem_add_to_physmap_one() instead.
Note that guest_physmap_mark_populate_on_demand() gets tightened beyond
the immediate purpose of this change.
Note also that I didn't inspect code which isn't security supported,
e.g. sharing, paging, or altp2m.
This is CVE-2021-28694 / part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Seeing the similarity of grant, foreign, and (subsequently) direct-MMIO
handling, introduce a new P2M type group named "special" (as in "needing
special accessors to create/destroy").
Also use -EPERM instead of other error codes on the two domain_crash()
paths touched.
This is part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
The spec makes no provisions for OS behavior here to depend on the
amount of RAM found on the system. While the spec may not sufficiently
clearly distinguish both kinds of regions, they are surely meant to be
separate things: Only regions with ACPI_IVMD_EXCLUSION_RANGE set should
be candidates for putting in the exclusion range registers. (As there's
only a single such pair of registers per IOMMU, secondary non-adjacent
regions with the flag set already get converted to unity mapped
regions.)
First of all, drop the dependency on max_page. With commit b4f042236a
("AMD/IOMMU: Cease using a dynamic height for the IOMMU pagetables") the
use of it here was stale anyway; it was bogus already before, as it
didn't account for max_page getting increased later on. Simply try an
exclusion range registration first, and if it fails (for being
unsuitable or non-mergeable), register a unity mapping range.
With this various local variables become unnecessary and hence get
dropped at the same time.
With the max_page boundary dropped for using unity maps, the minimum
page table tree height now needs both recording and enforcing in
amd_iommu_domain_init(). Since we can't predict which devices may get
assigned to a domain, our only option is to uniformly force at least
that height for all domains, now that the height isn't dynamic anymore.
Further don't make use of the exclusion range unless ACPI data says so.
Note that exclusion range registration in
register_range_for_all_devices() is on a best effort basis. Hence unity
map entries also registered are redundant when the former succeeded, but
they also do no harm. Improvements in this area can be done later imo.
Also adjust types where suitable without touching extra lines.
This is part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Prior to the assignment step having completed successfully, devices
should not get associated with their new owner. Hand the device to DomIO
(perhaps temporarily), until after the de-assignment step has completed.
De-assignment of a device (from other than Dom0) as well as failure of
reassign_device() during assignment should result in unity mappings
getting torn down. This in turn requires switching to a refcounted
mapping approach, as was already used by VT-d for its RMRRs, to prevent
unmapping a region used by multiple devices.
This is CVE-2021-28696 / part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
In order to re-use it elsewhere, move the logic to vendor independent
code and strip it of RMRR specifics.
Note that the prior "map" parameter gets folded into the new "p2ma" one
(which AMD IOMMU code will want to make use of), assigning alternative
meaning ("unmap") to p2m_access_x. Prepare set_identity_p2m_entry() and
p2m_get_iommu_flags() for getting passed access types other than
p2m_access_rw (in the latter case just for p2m_mmio_direct requests).
Note also that, to be on the safe side, an overlap check gets added to
the main loop of iommu_identity_mapping().
This is part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
A subsequent change will want to customize the IOMMU permissions based
on this.
This is part of XSA-378.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Andrew Cooper