NVDIMMs can belong to their own proximity domains, as described by the
NFIT. In such cases, the SRAT needs to have Memory Affinity structures
in the SRAT for these NVDIMMs, otherwise Linux doesn't populate node
data structures properly during NUMA initialization. See the following
for an example failure case.
https://lore.kernel.org/linux-nvdimm/20200416225438.15208-1-vishal.l.verma@intel.com/
Introduce a new helper, nvdimm_build_srat(), and call it for both the
i386 and arm versions of 'build_srat()' to augment the SRAT with
memory affinity information for NVDIMMs.
The relevant command line options to exercise this are below. Nodes 0-1
contain CPUs and regular memory, and nodes 2-3 are the NVDIMM address
space.
-object memory-backend-ram,id=mem0,size=2048M
-numa node,nodeid=0,memdev=mem0,
-numa cpu,node-id=0,socket-id=0
-object memory-backend-ram,id=mem1,size=2048M
-numa node,nodeid=1,memdev=mem1,
-numa cpu,node-id=1,socket-id=1
-numa node,nodeid=2,
-object memory-backend-file,id=nvmem0,share,mem-path=nvdimm-0,size=16384M,align=1G
-device nvdimm,memdev=nvmem0,id=nv0,label-size=2M,node=2
-numa node,nodeid=3,
-object memory-backend-file,id=nvmem1,share,mem-path=nvdimm-1,size=16384M,align=1G
-device nvdimm,memdev=nvmem1,id=nv1,label-size=2M,node=3
Cc: Jingqi Liu <jingqi.liu@intel.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Jingqi Liu <jingqi.liu@intel.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Message-Id: <20200606000911.9896-3-vishal.l.verma@intel.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
PCI spec says:
For all accesses to MSI-X Table and MSI-X PBA fields, software must use
aligned full DWORD or aligned full QWORD transactions; otherwise, the
result is undefined.
However, since MSI-X was converted to use memory API, QEMU
started blocking qword transactions, only allowing DWORD
ones. Guests do not seem to use QWORD accesses, but let's
be spec compliant.
Fixes: 95524ae8dc ("msix: convert to memory API")
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
The Plug & Play region of the AHB/APB bridge can be accessed
by various word size, however the implementation is clearly
restricted to 32-bit:
static uint64_t grlib_ahb_pnp_read(void *opaque, hwaddr offset, unsigned size)
{
AHBPnp *ahb_pnp = GRLIB_AHB_PNP(opaque);
return ahb_pnp->regs[offset >> 2];
}
Similarly to commit 0fbe394a64 with the APB PnP registers,
set the MemoryRegionOps::impl min/max fields to 32-bit, so
memory.c::access_with_adjusted_size() can adjust when the
access is not 32-bit.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: KONRAD Frederic <frederic.konrad@adacore.com>
Message-Id: <20200331105048.27989-4-f4bug@amsat.org>
Empty slots model RAZ/WI access on a bus. Since we can still
(hot) plug devices on the bus, lower the slot priority, so
device added later is accessed first.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Artyom Tarasenko <atar4qemu@gmail.com>
Message-Id: <20200510152840.13558-3-f4bug@amsat.org>
These devices are not slots on a bus, but real I/O devices
that we do not implement. As the ISDN ROM would be a ROMD
device, also model it as UnimplementedDevice.
Reviewed-by: Artyom Tarasenko <atar4qemu@gmail.com>
Message-Id: <20200510152840.13558-2-f4bug@amsat.org>
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Create ExcMultiPattern to hold an set of non-overlapping patterns.
The body of build_tree, prop_format become member functions on this
class. Add minimal member functions to Pattern and MultiPattern
to allow recusion through the tree.
Move the bulk of build_incmulti_pattern to prop_masks and prop_width
in MultiPattern, since we will need this for both kinds of containers.
Only perform prop_width for variablewidth.
Remove global patterns variable, and pass down container object into
parse_file from main.
No functional change in all of this.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This is an edge case for sure, but the logic that disallowed
this case was faulty. Further, a few fixes scattered about
can allow this to work.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Name the current node for "inclusive" multi-pattern, in
preparation for adding a node for "exclusive" multi-pattern.
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Relaxing the restrictions on 64 bit guests leads to the user being
able to attempt to map right at the edge of addressable memory. This
in turn lead to address overflow tripping the assert in page_set_flags
when the end address wrapped around.
Detect the wrap earlier and correctly -ENOMEM the guest (in the
reported case LTP mmap15).
Fixes: 7d8cbbabcb
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reported-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20200605154929.26910-15-alex.bennee@linaro.org>
We rely on the pointer to wrap when accessing the high address of the
COMMPAGE so it lands somewhere reasonable. However on 32 bit hosts we
cannot afford just to map the entire 4gb address range. The old mmap
trial and error code handled this by just checking we could map both
the guest_base and the computed COMMPAGE address.
We can't just manipulate loadaddr to get what we want so we introduce
an offset which pgb_find_hole can apply when looking for a gap for
guest_base that ensures there is space left to map the COMMPAGE
afterwards.
This is arguably a little inefficient for the one 32 bit
value (kuser_helper_version) we need to keep there given all the
actual code entries are picked up during the translation phase.
Fixes: ee94743034
Bug: https://bugs.launchpad.net/qemu/+bug/1880225
Cc: Bug 1880225 <1880225@bugs.launchpad.net>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>
Cc: Richard Henderson <richard.henderson@linaro.org>
Cc: Peter Maydell <peter.maydell@linaro.org>
Message-Id: <20200605154929.26910-13-alex.bennee@linaro.org>
When running QEMU out of a chroot environment we may not have access
to /proc/self/maps. As there is no other "official" way to introspect
our memory map we need to fall back to the original technique of
repeatedly trying to mmap an address range until we find one that
works.
Fortunately it's not quite as ugly as the original code given we
already re-factored the complications of dealing with the
ARM_COMMPAGE. We do make an attempt to skip over brk() which is about
the only concrete piece of information we have about the address map
at this moment.
Fixes: ee9474303
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20200605154929.26910-12-alex.bennee@linaro.org>
