Let's support RAM_NORESERVE via MAP_NORESERVE on Linux. The flag has no
effect on most shared mappings - except for hugetlbfs and anonymous memory.
Linux man page:
"MAP_NORESERVE: Do not reserve swap space for this mapping. When swap
space is reserved, one has the guarantee that it is possible to modify
the mapping. When swap space is not reserved one might get SIGSEGV
upon a write if no physical memory is available. See also the discussion
of the file /proc/sys/vm/overcommit_memory in proc(5). In kernels before
2.6, this flag had effect only for private writable mappings."
Note that the "guarantee" part is wrong with memory overcommit in Linux.
Also, in Linux hugetlbfs is treated differently - we configure reservation
of huge pages from the pool, not reservation of swap space (huge pages
cannot be swapped).
The rough behavior is [1]:
a) !Hugetlbfs:
1) Without MAP_NORESERVE *or* with memory overcommit under Linux
disabled ("/proc/sys/vm/overcommit_memory == 2"), the following
accounting/reservation happens:
For a file backed map
SHARED or READ-only - 0 cost (the file is the map not swap)
PRIVATE WRITABLE - size of mapping per instance
For an anonymous or /dev/zero map
SHARED - size of mapping
PRIVATE READ-only - 0 cost (but of little use)
PRIVATE WRITABLE - size of mapping per instance
2) With MAP_NORESERVE, no accounting/reservation happens.
b) Hugetlbfs:
1) Without MAP_NORESERVE, huge pages are reserved.
2) With MAP_NORESERVE, no huge pages are reserved.
Note: With "/proc/sys/vm/overcommit_memory == 0", we were already able
to configure it for !hugetlbfs globally; this toggle now allows
configuring it more fine-grained, not for the whole system.
The target use case is virtio-mem, which dynamically exposes memory
inside a large, sparse memory area to the VM.
[1] https://www.kernel.org/doc/Documentation/vm/overcommit-accounting
Reviewed-by: Peter Xu <peterx@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com> for memory backend and machine core
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210510114328.21835-10-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Let's introduce RAM_NORESERVE, allowing mmap'ing with MAP_NORESERVE. The
new flag has the following semantics:
"
RAM is mmap-ed with MAP_NORESERVE. When set, reserving swap space (or huge
pages if applicable) is skipped: will bail out if not supported. When not
set, the OS will do the reservation, if supported for the memory type.
"
Allow passing it into:
- memory_region_init_ram_nomigrate()
- memory_region_init_resizeable_ram()
- memory_region_init_ram_from_file()
... and teach qemu_ram_mmap() and qemu_anon_ram_alloc() about the flag.
Bail out if the flag is not supported, which is the case right now for
both, POSIX and win32. We will add Linux support next and allow specifying
RAM_NORESERVE via memory backends.
The target use case is virtio-mem, which dynamically exposes memory
inside a large, sparse memory area to the VM.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com> for memory backend and machine core
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210510114328.21835-9-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Let's pass flags instead of bools to prepare for passing other flags and
update the documentation of qemu_ram_mmap(). Introduce new QEMU_MAP_
flags that abstract the mmap() PROT_ and MAP_ flag handling and simplify
it.
We expose only flags that are currently supported by qemu_ram_mmap().
Maybe, we'll see qemu_mmap() in the future as well that can implement these
flags.
Note: We don't use MAP_ flags as some flags (e.g., MAP_SYNC) are only
defined for some systems and we want to always be able to identify
these flags reliably inside qemu_ram_mmap() -- for example, to properly
warn when some future flags are not available or effective on a system.
Also, this way we can simplify PROT_ handling as well.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com> for memory backend and machine core
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210510114328.21835-8-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Let's pass ram_flags to qemu_ram_alloc() and qemu_ram_alloc_internal(),
preparing for passing additional flags.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Acked-by: Eduardo Habkost <ehabkost@redhat.com> for memory backend and machine core
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210510114328.21835-7-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
RAM_SHARED now also properly indicates shared anonymous memory. Let's check
that flag for anonymous memory as well, to restore the proper mapping.
Fixes: 06329ccecf ("mem: add share parameter to memory-backend-ram")
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210406080126.24010-4-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We can create shared anonymous memory via
"-object memory-backend-ram,share=on,..."
which is, for example, required by PVRDMA for mremap() to work.
Shared anonymous memory is weird, though. Instead of MADV_DONTNEED, we
have to use MADV_REMOVE: MADV_DONTNEED will only remove / zap all
relevant page table entries of the current process, the backend storage
will not get removed, resulting in no reduced memory consumption and
a repopulation of previous content on next access.
Shared anonymous memory is internally really just shmem, but without a
fd exposed. As we cannot use fallocate() without the fd to discard the
backing storage, MADV_REMOVE gets the same job done without a fd as
documented in "man 2 madvise". Removing backing storage implicitly
invalidates all page table entries with relevant mappings - an additional
MADV_DONTNEED is not required.
Fixes: 06329ccecf ("mem: add share parameter to memory-backend-ram")
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210406080126.24010-3-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Let's drop the "shared" parameter from ram_block_add() and properly
store it in the flags of the ram block instead, such that
qemu_ram_is_shared() properly succeeds on all ram blocks that were mapped
MAP_SHARED.
We'll use this information next to fix some cases with shared anonymous
memory.
Reviewed-by: Igor Kotrasinski <i.kotrasinsk@partner.samsung.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210406080126.24010-2-david@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch enables reverse debugging with watchpoints.
Reverse continue scans the execution to find the breakpoints
and watchpoints that should fire. It uses helper function
replay_breakpoint() for that. But this function needs to access
icount, which can't be correct in the middle of TB.
Therefore, in case of watchpoint, we have to retranslate the block
to allow this access.
Signed-off-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Message-Id: <162072430303.827403.7379783546934958566.stgit@pasha-ThinkPad-X280>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We want to make use of ram_block_discard_range() in the RAM block resize
callback when growing a RAM block, *before* used_length is changed.
Let's relax the check. As RAM blocks always mmap the whole max_length area,
we cannot corrupt unrelated data.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210429112708.12291-6-david@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Resizing while migrating is dangerous and does not work as expected.
The whole migration code works on the usable_length of ram blocks and does
not expect this to change at random points in time.
In the case of precopy, the ram block size must not change on the source,
after syncing the RAM block list in ram_save_setup(), so as long as the
guest is still running on the source.
Resizing can be trigger *after* (but not during) a reset in
ACPI code by the guest
- hw/arm/virt-acpi-build.c:acpi_ram_update()
- hw/i386/acpi-build.c:acpi_ram_update()
Use the ram block notifier to get notified about resizes. Let's simply
cancel migration and indicate the reason. We'll continue running on the
source. No harm done.
Update the documentation. Postcopy will be handled separately.
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210429112708.12291-5-david@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Manual merge
Ram block notifiers are currently not aware of resizes. To properly
handle resizes during migration, we want to teach ram block notifiers about
resizeable ram.
Introduce the basic infrastructure but keep using max_size in the
existing notifiers. Supply the max_size when adding and removing ram
blocks. Also, notify on resizes.
Acked-by: Paul Durrant <paul@xen.org>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: xen-devel@lists.xenproject.org
Cc: haxm-team@intel.com
Cc: Paul Durrant <paul@xen.org>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Anthony Perard <anthony.perard@citrix.com>
Cc: Wenchao Wang <wenchao.wang@intel.com>
Cc: Colin Xu <colin.xu@intel.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210429112708.12291-3-david@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Factor it out into common code when a new notifier is registered, just
as done with the memory region notifier. This keeps logic about how to
process existing ram blocks at a central place.
Just like when adding a new ram block, we have to register the max_length.
Ram blocks are only "fake resized". All memory (max_length) is mapped.
Print the warning from inside qemu_vfio_ram_block_added().
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20210429112708.12291-2-david@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Stop including exec/address-spaces.h in files that don't need it.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210416171314.2074665-5-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Stop including cpu.h in files that don't need it.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210416171314.2074665-4-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Stop including sysemu/sysemu.h in files that don't need it.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-Id: <20210416171314.2074665-2-thuth@redhat.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
For the sparse-mem device, we want the fuzzer to populate entire DMA
reads from sparse-mem, rather than hooking into the individual MMIO
memory_region_dispatch_read operations. Otherwise, the fuzzer will treat
each sequential read separately (and populate it with a separate
pattern). Work around this by rearranging some DMA hooks. Since the
fuzzer has it's own logic to skip accidentally writing to MMIO regions,
we can call the DMA cb, outside the flatview_translate loop.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As the last user is gone, we can get rid of phys_mem_set_alloc() and
simplify.
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Halil Pasic <pasic@linux.ibm.com>
Cc: Cornelia Huck <cohuck@redhat.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Thomas Huth <thuth@redhat.com>
Cc: Igor Mammedov <imammedo@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Message-Id: <20210303130916.22553-3-david@redhat.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
There is nothing special about this compile flag that doesn't mean we
can't just compute it with curr_cflags() which we should be using when
building a new set.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20210224165811.11567-3-alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Allow RAM MemoryRegion to be created from an offset in a file, instead
of allocating at offset of 0 by default. This is needed to synchronize
RAM between QEMU & remote process.
Signed-off-by: Jagannathan Raman <jag.raman@oracle.com>
Signed-off-by: John G Johnson <john.g.johnson@oracle.com>
Signed-off-by: Elena Ufimtseva <elena.ufimtseva@oracle.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 609996697ad8617e3b01df38accc5c208c24d74e.1611938319.git.jag.raman@oracle.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We passed an is_write flag to the fuzz_dma_read_cb function to
differentiate between the mapped DMA regions that need to be populated
with fuzzed data, and those that don't. We simply passed through the
address_space_map is_write parameter. The goal was to cut down on
unnecessarily populating mapped DMA regions, when they are not read
from.
Unfortunately, nothing precludes code from reading from regions mapped
with is_write=true. For example, see:
https://lists.gnu.org/archive/html/qemu-devel/2021-01/msg04729.html
This patch removes the is_write parameter to fuzz_dma_read_cb. As a
result, we will fill all mapped DMA regions with fuzzed data, ignoring
the specified transfer direction.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20210120060255.558535-1-alxndr@bu.edu>
we cannot in principle make the TCG Operations field definitions
conditional on CONFIG_TCG in code that is included by both common_ss
and specific_ss modules.
Therefore, what we can do safely to restrict the TCG fields to TCG-only
builds, is to move all tcg cpu operations into a separate header file,
which is only included by TCG, target-specific code.
This leaves just a NULL pointer in the cpu.h for the non-TCG builds.
This also tidies up the code in all targets a bit, having all TCG cpu
operations neatly contained by a dedicated data struct.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Message-Id: <20210204163931.7358-16-cfontana@suse.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
commit 568496c0c0 ("cpu: Add callback to check architectural") and
commit 3826121d92 ("target-arm: Implement checking of fired")
introduced an ARM-specific hack for cpu_check_watchpoint.
Make debug_check_watchpoint optional, and move it to tcg_ops.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20210204163931.7358-15-cfontana@suse.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
commit 4061200059 ("arm: Correctly handle watchpoints for BE32 CPUs")
introduced this ARM-specific, TCG-specific hack to adjust the address,
before checking it with cpu_check_watchpoint.
Make adjust_watchpoint_address optional and move it to tcg_ops.
Signed-off-by: Claudio Fontana <cfontana@suse.de>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20210204163931.7358-14-cfontana@suse.de>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
There is currently no way to open(O_RDONLY) and mmap(PROT_READ) when
creating a memory region from a file. This functionality is needed since
the underlying host file may not allow writing.
Add a bool readonly argument to memory_region_init_ram_from_file() and
the APIs it calls.
Extend memory_region_init_ram_from_file() rather than introducing a
memory_region_init_rom_from_file() API so that callers can easily make a
choice between read/write and read-only at runtime without calling
different APIs.
No new RAMBlock flag is introduced for read-only because it's unclear
whether RAMBlocks need to know that they are read-only. Pass a bool
readonly argument instead.
Both of these design decisions can be changed in the future. It just
seemed like the simplest approach to me.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Liam Merwick <liam.merwick@oracle.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Message-Id: <20210104171320.575838-2-stefanha@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
When building with GCC 10.2 configured with --extra-cflags=-Os, we get:
softmmu/physmem.c: In function 'address_space_translate_for_iotlb':
softmmu/physmem.c:643:26: error: 'notifier' may be used uninitialized in this function [-Werror=maybe-uninitialized]
643 | notifier->active = true;
| ^
softmmu/physmem.c:608:23: note: 'notifier' was declared here
608 | TCGIOMMUNotifier *notifier;
| ^~~~~~~~
Initialize 'notifier' to silence the warning.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20210117170411.4106949-1-f4bug@amsat.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We are shortly going to have a split rw/rx jit buffer. Depending
on the host, we need to flush the dcache at the rw data pointer and
flush the icache at the rx code pointer.
For now, the two passed pointers are identical, so there is no
effective change in behaviour.
Reviewed-by: Joelle van Dyne <j@getutm.app>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This has been a tcg-specific function, but is also in use
by hardware accelerators via physmem.c. This can cause
link errors when tcg is disabled.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Joelle van Dyne <j@getutm.app>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20201214140314.18544-3-richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Enable removing tcg/$tcg_arch from the include path when TCG is disabled.
Move translate-all.h to include/exec, since stubs exist for the functions
defined therein.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
In using the address_space_translate_internal API, address_space_cache_init
forgot one piece of advice that can be found in the code for
address_space_translate_internal:
/* MMIO registers can be expected to perform full-width accesses based only
* on their address, without considering adjacent registers that could
* decode to completely different MemoryRegions. When such registers
* exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO
* regions overlap wildly. For this reason we cannot clamp the accesses
* here.
*
* If the length is small (as is the case for address_space_ldl/stl),
* everything works fine. If the incoming length is large, however,
* the caller really has to do the clamping through memory_access_size.
*/
address_space_cache_init is exactly one such case where "the incoming length
is large", therefore we need to clamp the resulting length---not to
memory_access_size though, since we are not doing an access yet, but to
the size of the resulting section. This ensures that subsequent accesses
to the cached MemoryRegionSection will be in range.
With this patch, the enclosed testcase notices that the used ring does
not fit into the MSI-X table and prints a "qemu-system-x86_64: Cannot map used"
error.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
There is no "version 2" of the "Lesser" General Public License.
It is either "GPL version 2.0" or "Lesser GPL version 2.1".
This patch replaces all occurrences of "Lesser GPL version 2" with
"Lesser GPL version 2.1" in comment section.
This patch contains all the files, whose maintainer I could not get
from ‘get_maintainer.pl’ script.
Signed-off-by: Chetan Pant <chetan4windows@gmail.com>
Message-Id: <20201023124424.20177-1-chetan4windows@gmail.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
[thuth: Adapted exec.c and qdev-monitor.c to new location]
Signed-off-by: Thomas Huth <thuth@redhat.com>
We removed the global_locking field in commit 4174495408,
leaving dead code around the 'unlocked' variable. Remove it
to fix the DEADCODE issue reported by Coverity (CID 1432876).
Fixes: 4174495408 ("exec: Remove MemoryRegion::global_locking field")
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20201030153752.1557776-1-philmd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We should be careful to not call any functions besides fuzz_dma_read_cb.
Without --enable-fuzzing, fuzz_dma_read_cb is an empty inlined function.
Signed-off-by: Alexander Bulekov <alxndr@bu.edu>
Reviewed-by: Darren Kenny <darren.kenny@oracle.com>
Message-Id: <20201023150746.107063-7-alxndr@bu.edu>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Over the years, most parts of exec.c that were not specific to softmmu
have been moved to accel/tcg; what's left is mostly the low-level part
of the memory API, which includes RAMBlock and AddressSpaceDispatch.
However exec.c also hosts 4-500 lines of code for the target specific
parts of the CPU QOM object, plus a few functions for user-mode
emulation that do not have a better place (they are not TCG-specific so
accel/tcg/user-exec.c is not a good place either).
Move these parts to a new file, so that exec.c can be moved to
softmmu/physmem.c.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
