linux-user: Split out mmap_end

Use a subroutine instead of a goto within target_mmap__locked.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-20-richard.henderson@linaro.org>

linux-user: Fix sub-host-page mmap

We cannot skip over the_end1 to the_end, because we fail to
record the validity of the guest page with the interval tree.
Remove "the_end" and rename "the_end1" to "the_end".

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-19-richard.henderson@linaro.org>

linux-user: Move some mmap checks outside the lock

Basic validation of operands does not require the lock.
Hoist them from target_mmap__locked back into target_mmap.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-18-richard.henderson@linaro.org>

linux-user: Split out target_mmap__locked

All "goto fail" may be transformed to "return -1".

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-17-richard.henderson@linaro.org>

linux-user: Remove qemu_host_page_size from main

Use qemu_real_host_page_size() instead.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-16-richard.henderson@linaro.org>

softmmu/physmem: Remove HOST_PAGE_ALIGN

Align allocation sizes to the maximum of host and target page sizes.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-15-richard.henderson@linaro.org>

softmmu/physmem: Remove qemu_host_page_size

Use qemu_real_host_page_size() instead.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-14-richard.henderson@linaro.org>

hw/tpm: Remove HOST_PAGE_ALIGN from tpm_ppi_init

This removes a hidden use of qemu_host_page_size, hoisting
two uses of qemu_real_host_page_size to a local variable.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>

migration: Remove qemu_host_page_size

Replace with the maximum of the real host page size
and the target page size.  This is an exact replacement.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-12-richard.henderson@linaro.org>

linux-user: Remove HOST_PAGE_ALIGN from mmap.c

This removes a hidden use of qemu_host_page_size, using instead
the existing host_page_size local within each function.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-11-richard.henderson@linaro.org>

linux-user: Remove REAL_HOST_PAGE_ALIGN from mmap.c

We already have qemu_real_host_page_size() in a local variable.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-10-richard.henderson@linaro.org>

linux-user: Remove qemu_host_page_{size, mask} from mmap.c

Use qemu_real_host_page_size instead.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-9-richard.henderson@linaro.org>

linux-user: Remove qemu_host_page_size from elf_core_dump

Used only once in wmr_page_unprotect_regions.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/arm: Remove qemu_host_page_size from init_guest_commpage

Use qemu_real_host_page_size.
If the commpage is not within reserved_va, use MAP_FIXED_NOREPLACE.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-8-richard.henderson@linaro.org>

linux-user/nios2: Remove qemu_host_page_size from init_guest_commpage

Use qemu_real_host_page_size.
If !reserved_va, use MAP_FIXED_NOREPLACE.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-7-richard.henderson@linaro.org>

linux-user/hppa: Simplify init_guest_commpage

If reserved_va, then we have already reserved the entire
guest virtual address space; no need to remap page.
If !reserved_va, then use MAP_FIXED_NOREPLACE.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-6-richard.henderson@linaro.org>

linux-user: Remove qemu_host_page_size from create_elf_tables

AT_PAGESZ is supposed to advertise the guest page size.
The random adjustment made here using qemu_host_page_size
does not match anything else within linux-user.

The idea here is good, but should be done more systemically
via adjustment to TARGET_PAGE_SIZE.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-5-richard.henderson@linaro.org>

linux-user: Remove qemu_host_page_{size, mask} in probe_guest_base

The host SHMLBA is by definition a multiple of the host page size.
Thus the remaining component of qemu_host_page_size is the
target page size.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-4-richard.henderson@linaro.org>

linux-user: Adjust SVr4 NULL page mapping

Use TARGET_PAGE_SIZE and MAP_FIXED_NOREPLACE.

We really should be attending to this earlier during
probe_guest_base, as well as better detection and
emulation of various Linux personalities.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-3-richard.henderson@linaro.org>

accel/tcg: Remove qemu_host_page_size from page_protect/page_unprotect

Use qemu_real_host_page_size instead.  Except for the final mprotect
within page_protect, we already handled host < target page size.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Ilya Leoshkevich <iii@linux.ibm.com>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20240102015808.132373-2-richard.henderson@linaro.org>

tcg: Avoid double lock if page tables happen to be in mmio memory.

On i386, after fixing the page walking code to work with pages in
MMIO memory (specifically CXL emulated interleaved memory),
a crash was seen in an interrupt handling path.

Useful part of backtrace

7  0x0000555555ab1929 in bql_lock_impl (file=0x555556049122 "../../accel/tcg/cputlb.c", line=2033) at ../../system/cpus.c:524
8  bql_lock_impl (file=file@entry=0x555556049122 "../../accel/tcg/cputlb.c", line=line@entry=2033) at ../../system/cpus.c:520
9  0x0000555555c9f7d6 in do_ld_mmio_beN (cpu=0x5555578e0cb0, full=0x7ffe88012950, ret_be=ret_be@entry=0, addr=19595792376, size=size@entry=8, mmu_idx=4, type=MMU_DATA_LOAD, ra=0) at ../../accel/tcg/cputlb.c:2033
10 0x0000555555ca0fbd in do_ld_8 (cpu=cpu@entry=0x5555578e0cb0, p=p@entry=0x7ffff4efd1d0, mmu_idx=<optimized out>, type=type@entry=MMU_DATA_LOAD, memop=<optimized out>, ra=ra@entry=0) at ../../accel/tcg/cputlb.c:2356
11 0x0000555555ca341f in do_ld8_mmu (cpu=cpu@entry=0x5555578e0cb0, addr=addr@entry=19595792376, oi=oi@entry=52, ra=0, ra@entry=52, access_type=access_type@entry=MMU_DATA_LOAD) at ../../accel/tcg/cputlb.c:2439
12 0x0000555555ca5f59 in cpu_ldq_mmu (ra=52, oi=52, addr=19595792376, env=0x5555578e3470) at ../../accel/tcg/ldst_common.c.inc:169
13 cpu_ldq_le_mmuidx_ra (env=0x5555578e3470, addr=19595792376, mmu_idx=<optimized out>, ra=ra@entry=0) at ../../accel/tcg/ldst_common.c.inc:301
14 0x0000555555b4b5fc in ptw_ldq (ra=0, in=0x7ffff4efd320) at ../../target/i386/tcg/sysemu/excp_helper.c:98
15 ptw_ldq (ra=0, in=0x7ffff4efd320) at ../../target/i386/tcg/sysemu/excp_helper.c:93
16 mmu_translate (env=env@entry=0x5555578e3470, in=0x7ffff4efd3e0, out=0x7ffff4efd3b0, err=err@entry=0x7ffff4efd3c0, ra=ra@entry=0) at ../../target/i386/tcg/sysemu/excp_helper.c:174
17 0x0000555555b4c4b3 in get_physical_address (ra=0, err=0x7ffff4efd3c0, out=0x7ffff4efd3b0, mmu_idx=0, access_type=MMU_DATA_LOAD, addr=18446741874686299840, env=0x5555578e3470) at ../../target/i386/tcg/sysemu/excp_helper.c:580
18 x86_cpu_tlb_fill (cs=0x5555578e0cb0, addr=18446741874686299840, size=<optimized out>, access_type=MMU_DATA_LOAD, mmu_idx=0, probe=<optimized out>, retaddr=0) at ../../target/i386/tcg/sysemu/excp_helper.c:606
19 0x0000555555ca0ee9 in tlb_fill (retaddr=0, mmu_idx=0, access_type=MMU_DATA_LOAD, size=<optimized out>, addr=18446741874686299840, cpu=0x7ffff4efd540) at ../../accel/tcg/cputlb.c:1315
20 mmu_lookup1 (cpu=cpu@entry=0x5555578e0cb0, data=data@entry=0x7ffff4efd540, mmu_idx=0, access_type=access_type@entry=MMU_DATA_LOAD, ra=ra@entry=0) at ../../accel/tcg/cputlb.c:1713
21 0x0000555555ca2c61 in mmu_lookup (cpu=cpu@entry=0x5555578e0cb0, addr=addr@entry=18446741874686299840, oi=oi@entry=32, ra=ra@entry=0, type=type@entry=MMU_DATA_LOAD, l=l@entry=0x7ffff4efd540) at ../../accel/tcg/cputlb.c:1803
22 0x0000555555ca3165 in do_ld4_mmu (cpu=cpu@entry=0x5555578e0cb0, addr=addr@entry=18446741874686299840, oi=oi@entry=32, ra=ra@entry=0, access_type=access_type@entry=MMU_DATA_LOAD) at ../../accel/tcg/cputlb.c:2416
23 0x0000555555ca5ef9 in cpu_ldl_mmu (ra=0, oi=32, addr=18446741874686299840, env=0x5555578e3470) at ../../accel/tcg/ldst_common.c.inc:158
24 cpu_ldl_le_mmuidx_ra (env=env@entry=0x5555578e3470, addr=addr@entry=18446741874686299840, mmu_idx=<optimized out>, ra=ra@entry=0) at ../../accel/tcg/ldst_common.c.inc:294
25 0x0000555555bb6cdd in do_interrupt64 (is_hw=1, next_eip=18446744072399775809, error_code=0, is_int=0, intno=236, env=0x5555578e3470) at ../../target/i386/tcg/seg_helper.c:889
26 do_interrupt_all (cpu=cpu@entry=0x5555578e0cb0, intno=236, is_int=is_int@entry=0, error_code=error_code@entry=0, next_eip=next_eip@entry=0, is_hw=is_hw@entry=1) at ../../target/i386/tcg/seg_helper.c:1130
27 0x0000555555bb87da in do_interrupt_x86_hardirq (env=env@entry=0x5555578e3470, intno=<optimized out>, is_hw=is_hw@entry=1) at ../../target/i386/tcg/seg_helper.c:1162
28 0x0000555555b5039c in x86_cpu_exec_interrupt (cs=0x5555578e0cb0, interrupt_request=<optimized out>) at ../../target/i386/tcg/sysemu/seg_helper.c:197
29 0x0000555555c94480 in cpu_handle_interrupt (last_tb=<synthetic pointer>, cpu=0x5555578e0cb0) at ../../accel/tcg/cpu-exec.c:844

Peter identified this as being due to the BQL already being
held when the page table walker encounters MMIO memory and attempts
to take the lock again.  There are other examples of similar paths
TCG, so this follows the approach taken in those of simply checking
if the lock is already held and if it is, don't take it again.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Message-Id: <20240219173153.12114-4-Jonathan.Cameron@huawei.com>
[rth: Use BQL_LOCK_GUARD]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

accel/tcg: Set can_do_io at at start of lookup_tb_ptr helper

If a page table is in IO memory and lookup_tb_ptr probes
the TLB it can result in a page table walk for the instruction
fetch.  If this hits IO memory and io_prepare falsely assumes
it needs to do a TLB recompile.

Avoid that by setting can_do_io at the start of lookup_tb_ptr.

Link: https://lore.kernel.org/qemu-devel/CAFEAcA_a_AyQ=Epz3_+CheAT8Crsk9mOu894wbNW_FywamkZiw@mail.gmail.com/#t
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Message-Id: <20240219173153.12114-2-Jonathan.Cameron@huawei.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

tcg/aarch64: Apple does not align __int128_t in even registers

From https://developer.apple.com/documentation/xcode/writing-arm64-code-for-apple-platforms

  When passing an argument with 16-byte alignment in integer registers,
  Apple platforms allow the argument to start in an odd-numbered xN
  register. The standard ABI requires it to begin in an even-numbered
  xN register.

Cc: qemu-stable@nongnu.org
Fixes: 5427a9a7604 ("tcg: Add TCG_TARGET_CALL_{RET,ARG}_I128")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2169
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <9fc0c2c7-dd57-459e-aecb-528edb74b4a7@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

linux-user/elfload: Unprotect regions before core dump

By unprotecting regions, we re-instate writability and
unify regions that have been split, which may reduce
the total number of regions.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Rely on walk_memory_regions for vmas

Rather than creating new data structures for vma,
rely on the IntervalTree used by walk_memory_regions.

Use PAGE_* constants, per the page table api, rather
than PROT_* constants, per the mmap api.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Simplify vma_dump_size

Use the flags that we've already saved in order to test
accessibility.  Use g2h_untagged and compare guest memory
directly instead of copy_from_user.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Write process memory to core file in larger chunks

We do not need to copy pages from guest memory before writing
them out.  Because vmas are contiguous in host memory, we can
write them in one go.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Write corefile elf header in one block

Fixes a bug in which write_note() wrote namesz_rounded
and datasz_rounded bytes, even though name and data
pointers contain only the unrounded number of bytes.

Instead of many small writes, allocate a block to contain all
of the elf headers and all of the notes.  Copy the data into the
block piecemeal and the write it to the file as a chunk.
This also avoids the need to lseek forward for alignment.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Size corefile before opening

Verify the size of the corefile vs the rlimit before
opening and creating the core file at all.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Lock cpu list and mmap during elf_core_dump

Do not allow changes to the set of cpus and memory regions
while we are dumping core.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Truncate core file on open

While we usually create a new corefile, truncate otherwise.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Open core file after vma_init

Swap the ordering of vma_init and open.  This will be necessary
for further changes, and adjusts the error cleanup path.  Narrow
the scope of corefile, as the variable can be freed immediately
after use in open().

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Latch errno before cleanup in elf_core_dump

On the off-chance that one of the cleanup functions changes
errno, latch the errno that we want to return beforehand.

Flush errno to 0 upon success, rather than at the beginning.
No need to avoid negation of 0.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Stack allocate struct mm_struct

Ignoring the fact that g_malloc cannot fail, the structure
is quite small and might as well be allocated locally.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Tidy fill_note_info and struct elf_note_info

In fill_note_info, there were unnecessary checks for
success of g_new/g_malloc.  But these structures do not
need to be dyamically allocated at all, and can in fact
be statically allocated within the parent structure.

This removes all error paths from fill_note_info, so
change the return type to void.

Change type of signr to match both caller (elf_core_dump)
and callee (fill_prstatus), which both use int for signr.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Merge init_note_info and fill_note_info

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

linux-user/elfload: Disable core dump if getrlimit fails

Do not dump core at all if getrlimit fails; this ensures
that dumpsize is valid throughout the function, not just
for the initial test vs rlim_cur.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Merge tag 'migration-next-pull-request' of https://gitlab.com/peterx/qemu into staging

Migration pull request

- Fabiano's fixed-ram patches (1-5 only)
- Peter's cleanups on multifd tls IOC referencing
- Steve's cpr patches for vfio (migration patches only)
- Fabiano's fix on mbps stats racing with COMPLETE state
- Fabiano's fix on return path thread hang

# -----BEGIN PGP SIGNATURE-----
#
# iIcEABYKADAWIQS5GE3CDMRX2s990ak7X8zN86vXBgUCZd7AbhIccGV0ZXJ4QHJl
# ZGhhdC5jb20ACgkQO1/MzfOr1wbg0gDyA3Vg3pIqCJ+u+hLZ+QKxY/pnu8Y5kF+E
# HK2IdslQUQD+OX4ATUnl+CGMiVX9fjs1fKx0Z0Qetq8gC1YJF13yuA0=
# =P2QF
# -----END PGP SIGNATURE-----
# gpg: Signature made Wed 28 Feb 2024 05:11:10 GMT
# gpg:                using EDDSA key B9184DC20CC457DACF7DD1A93B5FCCCDF3ABD706
# gpg:                issuer "peterx@redhat.com"
# gpg: Good signature from "Peter Xu <xzpeter@gmail.com>" [marginal]
# gpg:                 aka "Peter Xu <peterx@redhat.com>" [marginal]
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg:          It is not certain that the signature belongs to the owner.
# Primary key fingerprint: B918 4DC2 0CC4 57DA CF7D  D1A9 3B5F CCCD F3AB D706

* tag 'migration-next-pull-request' of https://gitlab.com/peterx/qemu: (25 commits)
  migration: Use migrate_has_error() in close_return_path_on_source()
  migration: Join the return path thread before releasing to_dst_file
  migration: Fix qmp_query_migrate mbps value
  migration: options incompatible with cpr
  migration: update cpr-reboot description
  migration: stop vm for cpr
  migration: notifier error checking
  migration: refactor migrate_fd_connect failures
  migration: per-mode notifiers
  migration: MigrationNotifyFunc
  migration: remove postcopy_after_devices
  migration: MigrationEvent for notifiers
  migration: convert to NotifierWithReturn
  migration: remove error from notifier data
  notify: pass error to notifier with return
  migration/multifd: Drop unnecessary helper to destroy IOC
  migration/multifd: Cleanup outgoing_args in state destroy
  migration/multifd: Make multifd_channel_connect() return void
  migration/multifd: Drop registered_yank
  migration/multifd: Cleanup TLS iochannel referencing
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'for-upstream' of https://gitlab.com/bonzini/qemu into staging

* target/i386: Fix physical address truncation on 32-bit PAE
* Remove globals for options -no-fd-bootchk and -win2k-hack

# -----BEGIN PGP SIGNATURE-----
#
# iQFIBAABCAAyFiEE8TM4V0tmI4mGbHaCv/vSX3jHroMFAmXebwQUHHBib256aW5p
# QHJlZGhhdC5jb20ACgkQv/vSX3jHroPozAf/Vgc9u6C+8PcPDrol6qxjI+EOHLNy
# 7M3/OFpUkwLXuOSawb6syYxHpLS38fKRcsb2ninngUmbRWA6p+KNUizlAFMj7op5
# wJmtdamCwCwXXaw20SfWxx2Ih0JS7FQsRsU94HTOdaDB17C9+hBcYwcggsOAXCmq
# gyVenEF1mov2A4jLMhdVIRX784AAoEP+QAuhBKQBrQwRLCTTyNdHl7jXdB9w+2sh
# KafokoFLcozJHz/tN3AhRKy6zjPugJyQmJwBRuj9tstCILtXpvf/ZE/3pUq5l3ZY
# A6dCI0zWAlGNTkpKRXsMFozNIVP2htnyidy29XHptlY5acfjtQ++rMu3BQ==
# =WY4H
# -----END PGP SIGNATURE-----
# gpg: Signature made Tue 27 Feb 2024 23:23:48 GMT
# gpg:                using RSA key F13338574B662389866C7682BFFBD25F78C7AE83
# gpg:                issuer "pbonzini@redhat.com"
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>" [full]
# gpg:                 aka "Paolo Bonzini <pbonzini@redhat.com>" [full]
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4  E2F7 7E15 100C CD36 69B1
#      Subkey fingerprint: F133 3857 4B66 2389 866C  7682 BFFB D25F 78C7 AE83

* tag 'for-upstream' of https://gitlab.com/bonzini/qemu:
  ide, vl: turn -win2k-hack into a property on IDE devices
  ide: collapse parameters to ide_init_drive
  target/i386: leave the A20 bit set in the final NPT walk
  target/i386: remove unnecessary/wrong application of the A20 mask
  target/i386: Fix physical address truncation
  target/i386: use separate MMU indexes for 32-bit accesses
  target/i386: introduce function to query MMU indices
  target/i386: check validity of VMCB addresses
  target/i386: mask high bits of CR3 in 32-bit mode
  vl, pc: turn -no-fd-bootchk into a machine property

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

Merge tag 'pull-maintainer-updates-280224-1' of https://gitlab.com/stsquad/qemu into staging

Testing, gdbstub and plugin updates:

  - fix some test/tcg license headers to GPLv2+
  - bump up check-tcg timeout to 120s
  - avoid re-building VM images too often
  - update OpenBSD to 7.4
  - use GDBFeature to build gdbstub XML
  - unify plugin vcpu count under qemu_plugin_num_vcpus
  - avoid spurious idle/resume callbacks on new vCPUs
  - ensure nios2-linux-user processes async work
  - call vcpu_init plugin callback through async work
  - define plugin helpers when registers being read
  - add plugin API for reading register values
  - add support for register tracking to execlog
  - update plugin docs with assumptions
  - mention plugins can trigger tb_flush in mttcg design doc

# -----BEGIN PGP SIGNATURE-----
#
# iQEzBAABCgAdFiEEZoWumedRZ7yvyN81+9DbCVqeKkQFAmXfAv0ACgkQ+9DbCVqe
# KkQyogf/X6T5lWsdZGb22FOYzaTLf5gfCPXArIVN+GsjSae3dU6qy/qVM1VRJQPw
# mH8kvMY7QO5V9M2tL33WtZZg6hqWypXYU+Hit6sMmveKYMKS9ESEX28x3yybgt8Y
# fyDywNODX7bs8Wb6NQjVkZvTmM2llrHEtQXPffaXaPyxOAzlGTV9Mf3Sop9rk4nG
# 8IchzLmOOQ7XnVst/KRyq+29oOYsbyUtj13tNeWBZ5iXFDT6Q/nGwPQ12U2Ztn9N
# FZvyzGG707dFaEDxIr4pl7n+lHJto29LMlSXlocANwG6wFNP3nfkSw/dXw3nkZZK
# pOfrQKvnnunJKBd7495LYZxTDe505Q==
# =/k97
# -----END PGP SIGNATURE-----
# gpg: Signature made Wed 28 Feb 2024 09:55:09 GMT
# gpg:                using RSA key 6685AE99E75167BCAFC8DF35FBD0DB095A9E2A44
# gpg: Good signature from "Alex Bennée (Master Work Key) <alex.bennee@linaro.org>" [full]
# Primary key fingerprint: 6685 AE99 E751 67BC AFC8  DF35 FBD0 DB09 5A9E 2A44

* tag 'pull-maintainer-updates-280224-1' of https://gitlab.com/stsquad/qemu: (29 commits)
  docs/devel: plugins can trigger a tb flush
  docs/devel: document some plugin assumptions
  docs/devel: lift example and plugin API sections up
  contrib/plugins: extend execlog to track register changes
  contrib/plugins: fix imatch
  tests/tcg: expand insn test case to exercise register API
  plugins: add an API to read registers
  plugins: create CPUPluginState and migrate plugin_mask
  gdbstub: expose api to find registers
  plugins: Use different helpers when reading registers
  cpu: call plugin init hook asynchronously
  linux-user: ensure nios2 processes queued work
  plugins: fix order of init/idle/resume callback
  plugins: add qemu_plugin_num_vcpus function
  plugins: remove previous n_vcpus functions from API
  gdbstub: Add members to identify registers to GDBFeature
  hw/core/cpu: Remove gdb_get_dynamic_xml member
  gdbstub: Infer number of core registers from XML
  gdbstub: Simplify XML lookup
  gdbstub: Change gdb_get_reg_cb and gdb_set_reg_cb
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

docs/devel: plugins can trigger a tb flush

When scoreboards need to be reallocated.

Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20240213094009.150349-8-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-30-alex.bennee@linaro.org>

docs/devel: document some plugin assumptions

While we attempt to hide implementation details from the plugin we
shouldn't be totally obtuse. Let the user know what they can and can't
expect with the various instrumentation options.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-29-alex.bennee@linaro.org>

docs/devel: lift example and plugin API sections up

This makes them a bit more visible in the TCG emulation menu rather
than hiding them away bellow the ToC limit.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-28-alex.bennee@linaro.org>

contrib/plugins: extend execlog to track register changes

With the new plugin register API we can now track changes to register
values. Currently the implementation is fairly dumb which will slow
down if a large number of register values are being tracked. This
could be improved by only instrumenting instructions which mention
registers we are interested in tracking.

Example usage:

  ./qemu-aarch64 -D plugin.log -d plugin \
     -cpu max,sve256=on \
     -plugin contrib/plugins/libexeclog.so,reg=sp,reg=z\* \
     ./tests/tcg/aarch64-linux-user/sha512-sve

will display in the execlog any changes to the stack pointer (sp) and
the SVE Z registers.

As testing registers every instruction will be quite a heavy operation
there is an additional flag which attempts to optimise the register
tracking by only instrumenting instructions which are likely to change
its value. This relies on the QEMU disassembler showing up the register
names in disassembly so is an explicit opt-in.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Cc: Akihiko Odaki <akihiko.odaki@daynix.com>
Based-On:  <20231025093128.33116-19-akihiko.odaki@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-27-alex.bennee@linaro.org>

contrib/plugins: fix imatch

We can't directly save the ephemeral imatch from argv as that memory
will get recycled.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-26-alex.bennee@linaro.org>

tests/tcg: expand insn test case to exercise register API

This ensure we at least read every register the plugin API reports at
least once during the check-tcg checks.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-25-alex.bennee@linaro.org>

plugins: add an API to read registers

We can only request a list of registers once the vCPU has been
initialised so the user needs to use either call the get function on
vCPU initialisation or during the translation phase.

We don't expose the reg number to the plugin instead hiding it behind
an opaque handle. For now this is just the gdb_regnum encapsulated in
an anonymous GPOINTER but in future as we add more state for plugins
to track we can expand it.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1706
Based-on:  <20231025093128.33116-18-akihiko.odaki@daynix.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-24-alex.bennee@linaro.org>

plugins: create CPUPluginState and migrate plugin_mask

As we expand the per-vCPU data for plugins we don't want to pollute
CPUState. For now this just moves the plugin_mask (renamed to
event_mask) as the memory callbacks are accessed directly by TCG
generated code.

Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-23-alex.bennee@linaro.org>

gdbstub: expose api to find registers

Expose an internal API to QEMU to return all the registers for a vCPU.
The list containing the details required to called gdb_read_register().

Based-on:  <20231025093128.33116-15-akihiko.odaki@daynix.com>
Cc: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-22-alex.bennee@linaro.org>

plugins: Use different helpers when reading registers

This avoids optimizations incompatible when reading registers.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-Id: <20231213-gdb-v17-12-777047380591@daynix.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-21-alex.bennee@linaro.org>

cpu: call plugin init hook asynchronously

This ensures we run during a cpu_exec, which allows to call start/end
exclusive from this init hook (needed for new scoreboard API introduced
later).

async work is run before any tb is translated/executed, so we can
guarantee plugin init will be called before any other hook.

The previous change made sure that any idle/resume cb call will not be
done before initializing plugin for a given vcpu.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-Id: <20240213094009.150349-5-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-20-alex.bennee@linaro.org>

linux-user: ensure nios2 processes queued work

While async processes are rare for linux-user we do use them from time
to time. The most obvious one is tb_flush when we run out of
translation space. We will also need this when we move plugin
vcpu_init to an async task.

Fix nios2 to follow its older, wiser and more stable siblings.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-19-alex.bennee@linaro.org>

plugins: fix order of init/idle/resume callback

We found that vcpu_init_hook was called *after* idle callback.
vcpu_init is called from cpu_realize_fn, while idle/resume cb are called
from qemu_wait_io_event (in vcpu thread).

This change ensures we only call idle and resume cb only once a plugin
was init for a given vcpu.

Next change in the series will run vcpu_init asynchronously, which will
make it run *after* resume callback as well. So we fix this now.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Message-Id: <20240213094009.150349-4-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-18-alex.bennee@linaro.org>

plugins: add qemu_plugin_num_vcpus function

We now keep track of how many vcpus were started. This way, a plugin can
easily query number of any vcpus at any point of execution, which
unifies user and system mode workflows.

Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20240213094009.150349-3-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-17-alex.bennee@linaro.org>

plugins: remove previous n_vcpus functions from API

This information is already accessible using qemu_info_t during plugin
install.

We will introduce another function (qemu_plugin_num_vcpus) which
represent how many cpus were enabled, by tracking new cpu indexes.

It's a breaking change, so we bump API version.

Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20240213094009.150349-2-pierrick.bouvier@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-16-alex.bennee@linaro.org>

gdbstub: Add members to identify registers to GDBFeature

These members will be used to help plugins to identify registers.
The added members in instances of GDBFeature dynamically generated by
CPUs will be filled in later changes.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-Id: <20231213-gdb-v17-10-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-15-alex.bennee@linaro.org>

hw/core/cpu: Remove gdb_get_dynamic_xml member

This function is no longer used.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20231213-gdb-v17-9-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-14-alex.bennee@linaro.org>

gdbstub: Infer number of core registers from XML

GDBFeature has the num_regs member so use it where applicable to
remove magic numbers.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-Id: <20231213-gdb-v17-8-777047380591@daynix.com>
[AJB: remove core reg check from microblaze read reg]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-13-alex.bennee@linaro.org>

gdbstub: Simplify XML lookup

Now we know all instances of GDBFeature that is used in CPU so we can
traverse them to find XML. This removes the need for a CPU-specific
lookup function for dynamic XMLs.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20231213-gdb-v17-7-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-12-alex.bennee@linaro.org>

gdbstub: Change gdb_get_reg_cb and gdb_set_reg_cb

Align the parameters of gdb_get_reg_cb and gdb_set_reg_cb with the
gdb_read_register and gdb_write_register members of CPUClass to allow
to unify the logic to access registers of the core and coprocessors
in the future.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20231213-gdb-v17-6-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-11-alex.bennee@linaro.org>

gdbstub: Use GDBFeature for GDBRegisterState

Simplify GDBRegisterState by replacing num_regs and xml members with
one member that points to GDBFeature.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-Id: <20231213-gdb-v17-5-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-10-alex.bennee@linaro.org>

gdbstub: Use GDBFeature for gdb_register_coprocessor

This is a tree-wide change to introduce GDBFeature parameter to
gdb_register_coprocessor(). The new parameter just replaces num_regs
and xml parameters for now. GDBFeature will be utilized to simplify XML
lookup in a following change.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Acked-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20231213-gdb-v17-4-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-9-alex.bennee@linaro.org>

target/riscv: Use GDBFeature for dynamic XML

In preparation for a change to use GDBFeature as a parameter of
gdb_register_coprocessor(), convert the internal representation of
dynamic feature from plain XML to GDBFeature.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Message-Id: <20231213-gdb-v17-3-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-8-alex.bennee@linaro.org>

target/ppc: Use GDBFeature for dynamic XML

In preparation for a change to use GDBFeature as a parameter of
gdb_register_coprocessor(), convert the internal representation of
dynamic feature from plain XML to GDBFeature.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20231213-gdb-v17-2-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-7-alex.bennee@linaro.org>

target/arm: Use GDBFeature for dynamic XML

In preparation for a change to use GDBFeature as a parameter of
gdb_register_coprocessor(), convert the internal representation of
dynamic feature from plain XML to GDBFeature.

Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
Acked-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20231213-gdb-v17-1-777047380591@daynix.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-6-alex.bennee@linaro.org>

tests/vm: update openbsd image to 7.4

The old links are dead so even if we have the ISO cached we can't
finish the install. Update to the current stable and tweak the install
strings.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2192
Tested-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-5-alex.bennee@linaro.org>

tests/vm: avoid re-building the VM images all the time

The main problem is that "check-venv" is a .PHONY target will always
evaluate and trigger a full re-build of the VM images. While its
tempting to drop it from the dependencies that does introduce a
breakage on freshly configured builds.

Fortunately we do have the otherwise redundant --force flag for the
script which up until now was always on. If we make the usage of
--force conditional on dependencies other than check-venv triggering
the update we can avoid the costly rebuild and still run cleanly on a
fresh checkout.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2118
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-4-alex.bennee@linaro.org>

tests/tcg: bump TCG test timeout to 120s

This is less than ideal but easier than making sure we get all the
iterations of the memory test. Update the comment accordingly.

Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-3-alex.bennee@linaro.org>

tests/tcg: update licenses to GPLv2 as intended

My default header template is GPLv3 but for QEMU code we really should
stick to GPLv2-or-later (allowing others to up-license it if they
wish). While this is test code we should still be consistent on the
source distribution.

I wrote all of this code so its not a problem. However there remains
one GPLv3 file left which is the crt0-tc2x.S for TriCore.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20240227144335.1196131-2-alex.bennee@linaro.org>

migration: Use migrate_has_error() in close_return_path_on_source()

close_return_path_on_source() retrieves the migration error from the
the QEMUFile '->to_dst_file' to know if a shutdown is required. This
shutdown is required to exit the return-path thread.

Avoid relying on '->to_dst_file' and use migrate_has_error() instead.

(using to_dst_file is a heuristic to infer whether
rp_state.from_dst_file might be stuck on a recvmsg(). Using a generic
method for detecting errors is more reliable. We also want to reduce
dependency on QEMUFile::last_error)

Suggested-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Cédric Le Goater <clg@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
[added some words about the motivation for this patch]
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240226203122.22894-3-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: Join the return path thread before releasing to_dst_file

The return path thread might hang at a blocking system call. Before
joining the thread we might need to issue a shutdown() on the socket
file descriptor to release it. To determine whether the shutdown() is
necessary we look at the QEMUFile error.

Make sure we only clean up the QEMUFile after the return path has been
waited for.

This fixes a hang when qemu_savevm_state_setup() produced an error
that was detected by migration_detect_error(). That skips
migration_completion() so close_return_path_on_source() would get
stuck waiting for the RP thread to terminate.

Reported-by: Cédric Le Goater <clg@redhat.com>
Tested-by: Cédric Le Goater <clg@redhat.com>
Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240226203122.22894-2-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: Fix qmp_query_migrate mbps value

The QMP command query_migrate might see incorrect throughput numbers
if it runs after we've set the migration completion status but before
migration_calculate_complete() has updated s->total_time and s->mbps.

The migration status would show COMPLETED, but the throughput value
would be the one from the last iteration and not the one from the
whole migration. This will usually be a larger value due to the time
period being smaller (one iteration).

Move migration_calculate_complete() earlier so that the status
MIGRATION_STATUS_COMPLETED is only emitted after the final counters
update. Keep everything under the BQL so the QMP thread sees the
updates as atomic.

Rename migration_calculate_complete to migration_completion_end to
reflect its new purpose of also updating s->state.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240226143335.14282-1-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: options incompatible with cpr

Fail the migration request if options are set that are incompatible
with cpr.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-15-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: update cpr-reboot description

Clarify qapi for cpr-reboot migration mode, and add vfio support.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-14-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: stop vm for cpr

When migration for cpr is initiated, stop the vm and set state
RUN_STATE_FINISH_MIGRATE before ram is saved.  This eliminates the
possibility of ram and device state being out of sync, and guarantees
that a guest in the suspended state remains suspended, because qmp_cont
rejects a cont command in the RUN_STATE_FINISH_MIGRATE state.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-11-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: notifier error checking

Check the status returned by migration notifiers for event type
MIG_EVENT_PRECOPY_SETUP, and report errors.  None of the notifiers
return an error status at this time.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-10-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: refactor migrate_fd_connect failures

Move common code for the error path in migrate_fd_connect to a shared
fail label.  No functional change.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-9-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: per-mode notifiers

Keep a separate list of migration notifiers for each migration mode.

Suggested-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-8-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: MigrationNotifyFunc

Define MigrationNotifyFunc to improve type safety and simplify migration
notifiers.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-7-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: remove postcopy_after_devices

postcopy_after_devices and migration_in_postcopy_after_devices are no
longer used, so delete them.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-6-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: MigrationEvent for notifiers

Passing MigrationState to notifiers is unsound because they could access
unstable migration state internals or even modify the state.  Instead, pass
the minimal info needed in a new MigrationEvent struct, which could be
extended in the future if needed.

Suggested-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-5-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: convert to NotifierWithReturn

Change all migration notifiers to type NotifierWithReturn, so notifiers
can return an error status in a future patch.  For now, pass NULL for the
notifier error parameter, and do not check the return value.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-4-git-send-email-steven.sistare@oracle.com
[peterx: dropped unexpected update to roms/seabios-hppa]
Signed-off-by: Peter Xu <peterx@redhat.com>

migration: remove error from notifier data

Remove the error object from opaque data passed to notifiers.
Use the new error parameter passed to the notifier instead.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-3-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

notify: pass error to notifier with return

Pass an error object as the third parameter to "notifier with return"
notifiers, so clients no longer need to bundle an error object in the
opaque data.  The new parameter is used in a later patch.

Signed-off-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/1708622920-68779-2-git-send-email-steven.sistare@oracle.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Drop unnecessary helper to destroy IOC

Both socket_send_channel_destroy() and multifd_send_channel_destroy() are
unnecessary wrappers to destroy an IOC, as the only thing to do is to
release the final IOC reference.  We have plenty of code that destroys an
IOC using direct unref() already; keep that style.

Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240222095301.171137-6-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Cleanup outgoing_args in state destroy

outgoing_args is a global cache of socket address to be reused in multifd.
Freeing the cache in per-channel destructor is more or less a hack.  Move
it to multifd_send_cleanup_state() so it only get checked once.  Use a
small helper to do so because it's internal of socket.c.

Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240222095301.171137-5-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Make multifd_channel_connect() return void

It never fails, drop the retval and also the Error**.

Suggested-by: Avihai Horon <avihaih@nvidia.com>
Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240222095301.171137-4-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Drop registered_yank

With a clear definition of p->c protocol, where we only set it up if the
channel is fully established (TLS or non-TLS), registered_yank boolean will
have equal meaning of "p->c != NULL".

Drop registered_yank by checking p->c instead.

Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240222095301.171137-3-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Cleanup TLS iochannel referencing

Commit a1af605bd5 ("migration/multifd: fix hangup with TLS-Multifd due to
blocking handshake") introduced a thread for TLS channels, which will
resolve the issue on blocking the main thread.  However in the same commit
p->c is slightly abused just to be able to pass over the pointer "p" into
the thread.

That's the major reason we'll need to conditionally free the io channel in
the fault paths.

To clean it up, using a separate structure to pass over both "p" and "tioc"
in the tls handshake thread.  Then we can make it a rule that p->c will
never be set until the channel is completely setup.  With that, we can drop
the tricky conditional unref of the io channel in the error path.

Reviewed-by: Fabiano Rosas <farosas@suse.de>
Link: https://lore.kernel.org/r/20240222095301.171137-2-peterx@redhat.com
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Release recv sem_sync earlier

Now that multifd_recv_terminate_threads() is called only once, release
the recv side sem_sync earlier like we do for the send side.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/20240220224138.24759-6-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

migration/multifd: Remove p->quit from recv side

Like we did on the sending side, replace the p->quit per-channel flag
with a global atomic 'exiting' flag.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/20240220224138.24759-5-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

tests/qtest/migration: Add a fd + file test

The fd URI supports an fd that is backed by a file. The code should
select between QIOChannelFile and QIOChannelSocket, depending on the
type of the fd. Add a test for that.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/20240220224138.24759-4-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

tests/qtest/migration: Rename fd_proto test

Next patch adds another fd test. Rename the existing one closer to
what's used on other tests, with the 'precopy' prefix.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/20240220224138.24759-3-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

docs/devel/migration.rst: Document the file transport

When adding the support for file migration with the file: transport,
we missed adding documentation for it.

Signed-off-by: Fabiano Rosas <farosas@suse.de>
Reviewed-by: Peter Xu <peterx@redhat.com>
Link: https://lore.kernel.org/r/20240220224138.24759-2-farosas@suse.de
Signed-off-by: Peter Xu <peterx@redhat.com>

ide, vl: turn -win2k-hack into a property on IDE devices

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

ide: collapse parameters to ide_init_drive

All calls to ide_init_drive comes from ide_dev_initfn.  Just pass down the
IDEDevice (IDEState is kinda obsolete and should be merged into IDEDevice).

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: leave the A20 bit set in the final NPT walk

The A20 mask is only applied to the final memory access.  Nested
page tables are always walked with the raw guest-physical address.

Unlike the previous patch, in this one the masking must be kept, but
it was done too early.

Cc: qemu-stable@nongnu.org
Fixes: 4a1e9d4d11c ("target/i386: Use atomic operations for pte updates", 2022-10-18)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: remove unnecessary/wrong application of the A20 mask

If ptw_translate() does a MMU_PHYS_IDX access, the A20 mask is already
applied in get_physical_address(), which is called via probe_access_full()
and x86_cpu_tlb_fill().

If ptw_translate() on the other hand does a MMU_NESTED_IDX access,
the A20 mask must not be applied to the address that is looked up in
the nested page tables; it must be applied only to the addresses that
hold the NPT entries (which is achieved via MMU_PHYS_IDX, per the
previous paragraph).

Therefore, we can remove A20 masking from the computation of the page
table entry's address, and let get_physical_address() or mmu_translate()
apply it when they know they are returning a host-physical address.

Cc: qemu-stable@nongnu.org
Fixes: 4a1e9d4d11c ("target/i386: Use atomic operations for pte updates", 2022-10-18)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: Fix physical address truncation

The address translation logic in get_physical_address() will currently
truncate physical addresses to 32 bits unless long mode is enabled.
This is incorrect when using physical address extensions (PAE) outside
of long mode, with the result that a 32-bit operating system using PAE
to access memory above 4G will experience undefined behaviour.

The truncation code was originally introduced in commit 33dfdb5 ("x86:
only allow real mode to access 32bit without LMA"), where it applied
only to translations performed while paging is disabled (and so cannot
affect guests using PAE).

Commit 9828198 ("target/i386: Add MMU_PHYS_IDX and MMU_NESTED_IDX")
rearranged the code such that the truncation also applied to the use
of MMU_PHYS_IDX and MMU_NESTED_IDX.  Commit 4a1e9d4 ("target/i386: Use
atomic operations for pte updates") brought this truncation into scope
for page table entry accesses, and is the first commit for which a
Windows 10 32-bit guest will reliably fail to boot if memory above 4G
is present.

The truncation code however is not completely redundant.  Even though the
maximum address size for any executed instruction is 32 bits, helpers for
operations such as BOUND, FSAVE or XSAVE may ask get_physical_address()
to translate an address outside of the 32-bit range, if invoked with an
argument that is close to the 4G boundary.  Likewise for processor
accesses, for example TSS or IDT accesses, when EFER.LMA==0.

So, move the address truncation in get_physical_address() so that it
applies to 32-bit MMU indexes, but not to MMU_PHYS_IDX and MMU_NESTED_IDX.

Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2040
Fixes: 4a1e9d4d11c ("target/i386: Use atomic operations for pte updates", 2022-10-18)
Cc: qemu-stable@nongnu.org
Co-developed-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Michael Brown <mcb30@ipxe.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

target/i386: use separate MMU indexes for 32-bit accesses

Accesses from a 32-bit environment (32-bit code segment for instruction
accesses, EFER.LMA==0 for processor accesses) have to mask away the
upper 32 bits of the address.  While a bit wasteful, the easiest way
to do so is to use separate MMU indexes.  These days, QEMU anyway is
compiled with a fixed value for NB_MMU_MODES.  Split MMU_USER_IDX,
MMU_KSMAP_IDX and MMU_KNOSMAP_IDX in two.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>