The AHCI spec states that:
For NCQ, PxCI is cleared on command queued successfully.
For non-NCQ, PxCI is cleared on command completed successfully.
(A non-NCQ command that completes with error does not clear PxCI.)
The current QEMU implementation either clears PxCI in check_cmd(),
or in ahci_cmd_done().
check_cmd() will clear PxCI for a command if handle_cmd() returns 0.
handle_cmd() will return -1 if BUSY or DRQ is set.
The QEMU implementation for NCQ commands will currently not set BUSY
or DRQ, so they will always have PxCI cleared by handle_cmd().
ahci_cmd_done() will never even get called for NCQ commands.
Non-NCQ commands are executed by ide_bus_exec_cmd().
Non-NCQ commands in QEMU are implemented either in a sync or in an async
way.
For non-NCQ commands implemented in a sync way, the command handler will
return true, and when ide_bus_exec_cmd() sees that a command handler
returns true, it will call ide_cmd_done() (which will call
ahci_cmd_done()). For a command implemented in a sync way,
ahci_cmd_done() will do nothing (since busy_slot is not set). Instead,
after ide_bus_exec_cmd() has finished, check_cmd() will clear PxCI for
these commands.
For non-NCQ commands implemented in an async way (using either aiocb or
pio_aiocb), the command handler will return false, ide_bus_exec_cmd()
will not call ide_cmd_done(), instead it is expected that the async
callback function will call ide_cmd_done() once the async command is
done. handle_cmd() will set busy_slot, if and only if BUSY or DRQ is
set, and this is checked _after_ ide_bus_exec_cmd() has returned.
handle_cmd() will return -1, so check_cmd() will not clear PxCI.
When the async callback calls ide_cmd_done() (which will call
ahci_cmd_done()), it will see that busy_slot is set, and
ahci_cmd_done() will clear PxCI.
This seems racy, since busy_slot is set _after_ ide_bus_exec_cmd() has
returned. The callback might come before busy_slot gets set. And it is
quite confusing that ahci_cmd_done() will be called for all non-NCQ
commands when the command is done, but will only clear PxCI in certain
cases, even though it will always write a D2H FIS and raise an IRQ.
Even worse, in the case where ahci_cmd_done() does not clear PxCI, it
still raises an IRQ. Host software might thus read an old PxCI value,
since PxCI is cleared (by check_cmd()) after the IRQ has been raised.
Try to simplify this by always setting busy_slot for non-NCQ commands,
such that ahci_cmd_done() will always be responsible for clearing PxCI
for non-NCQ commands.
For NCQ commands, clear PxCI when we receive the D2H FIS, but before
raising the IRQ, see AHCI 1.3.1, section 5.3.8, states RegFIS:Entry and
RegFIS:ClearCI.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Message-id: 20230609140844.202795-5-nks@flawful.org
Signed-off-by: John Snow <jsnow@redhat.com>
The way that BUSY + PxCI is cleared for NCQ (FPDMA QUEUED) commands is
described in SATA 3.5a Gold:
11.15 FPDMA QUEUED command protocol
DFPDMAQ2: ClearInterfaceBsy
"Transmit Register Device to Host FIS with the BSY bit cleared to zero
and the DRQ bit cleared to zero and Interrupt bit cleared to zero to
mark interface ready for the next command."
PxCI is currently cleared by handle_cmd(), but we don't write the D2H
FIS to the FIS Receive Area that actually caused PxCI to be cleared.
Similar to how ahci_pio_transfer() calls ahci_write_fis_pio() with an
additional parameter to write a PIO Setup FIS without raising an IRQ,
add a parameter to ahci_write_fis_d2h() so that ahci_write_fis_d2h()
also can write the FIS to the FIS Receive Area without raising an IRQ.
Change process_ncq_command() to call ahci_write_fis_d2h() without
raising an IRQ (similar to ahci_pio_transfer()), such that the FIS
Receive Area is in sync with the PxTFD shadow register.
E.g. Linux reads status and error fields from the FIS Receive Area
directly, so it is wise to keep the FIS Receive Area and the PxTFD
shadow register in sync.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Message-id: 20230609140844.202795-4-nks@flawful.org
Signed-off-by: John Snow <jsnow@redhat.com>
Currently, the first time sending an unsupported command
(e.g. READ LOG DMA EXT) will not have ERR_STAT set in the completion.
Sending the unsupported command again, will correctly have ERR_STAT set.
When ide_cmd_permitted() returns false, it calls ide_abort_command().
ide_abort_command() first calls ide_transfer_stop(), which will call
ide_transfer_halt() and ide_cmd_done(), after that ide_abort_command()
sets ERR_STAT in status.
ide_cmd_done() for AHCI will call ahci_write_fis_d2h() which writes the
current status in the FIS, and raises an IRQ. (The status here will not
have ERR_STAT set!).
Thus, we cannot call ide_transfer_stop() before setting ERR_STAT, as
ide_transfer_stop() will result in the FIS being written and an IRQ
being raised.
The reason why it works the second time, is that ERR_STAT will still
be set from the previous command, so when writing the FIS, the
completion will correctly have ERR_STAT set.
Set ERR_STAT before writing the FIS (calling cmd_done), so that we will
raise an error IRQ correctly when receiving an unsupported command.
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20230609140844.202795-3-nks@flawful.org
Signed-off-by: John Snow <jsnow@redhat.com>
Write a pattern to the first cluster. Corrupt the data_off field and check
if the field was repaired on image opening and the pattern has not changed.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Images repairing in parallels_open() was added, thus parallels tests fail.
Access to an image leads to repairing the image. Further image check don't
detect any corruption. Remove reads after image creation in test 131.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
In this test cluster size is 64k, but modern tools generate images with
cluster size 1M. Calculate cluster size using track field from image header.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Replace hardcoded numbers by variables.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Fill a parallels image with a pattern and write another pattern to the
second cluster. Corrupt the image and check if the pattern changes. Repair
the image and check the patterns on guest and host sides.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Write a pattern to the last cluster, extend the image by 1 claster, repair
and check that the last cluster still has the same pattern.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Fill the image with a pattern to generate entries in the BAT, set the first
BAT entry outside the image, try to read the corrupted image. At the image
opening it should be repaired, check for zeroes in the first cluster.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Place data_start/data_end calculation after reading the image header
to s->header. Set s->data_start to the offset calculated in
parallels_test_data_off(). Call bdrv_check() if data_off is incorrect.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
data_off field of the parallels image header can be corrupted. Check if
this field greater than the header + BAT size and less than file size.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
bdrv_co_getlength() should be used in coroutine context. Replace
bdrv_getlength() by bdrv_co_getlength() in parallels_check_outside_image().
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Repair an image at opening if the image is unclean or out-of-image
corruption was detected.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Cluster offsets must be unique among all the BAT entries. Find duplicate
offsets in the BAT and fix it by copying the content of the relevant
cluster to a newly allocated cluster and set the new cluster offset to the
duplicated entry.
Add host_cluster_index() helper to deduplicate the code.
When new clusters are allocated, the file size increases by 128 Mb. Call
parallels_check_leak() to fix this leak.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
In the next patch we will need the offset of the data area for host cluster
index calculation. Add this field and setting up code.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
In the on of the next patches we need to repair leaks without changing
leaks and leaks_fixed info in res. Also we don't want to print any warning
about leaks. Add "explicit" argument to skip info changing if the argument
is false.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Initially data_end is set to the data_off image header field and must not
be greater than the file size.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
The BDRVParallelsState structure contains data_end field that is measured
in sectors. In parallels_open() initially this field is set by data_off
field from parallels image header.
According to the parallels format documentation, data_off field contains
an offset, in sectors, from the start of the file to the start of the
data area. For "WithoutFreeSpace" images: if data_off is zero, the offset
is calculated as the end of the BAT table plus some padding to ensure
sector size alignment.
The parallels_open() function has code for handling zero value in
data_off, but in the result data_end contains the offset in bytes.
Replace the alignment to sector size by division by sector size and fix
the comparision with s->header_size.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Hanna Czenczek <hreitz@redhat.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
This patch is technically necessary as git patch rendering could result
in moving some code from one place to the another and that hits
checkpatch.pl warning. This problem specifically happens within next
series.
Signed-off-by: Alexander Ivanov <alexander.ivanov@virtuozzo.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Driver changes are driving by me for now. At least we need to get
functionally complete check and repair procedure for now.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
CC: Stefan Hajnoczi <stefanha@redhat.com>
* Fixes for the Aspeed I2C model
* New SDK image for avocado tests
* blockdev support for flash device definition
* SD refactoring preparing ground for eMMC support
-----BEGIN PGP SIGNATURE-----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=ymmX
-----END PGP SIGNATURE-----
Merge tag 'pull-aspeed-20230901' of https://github.com/legoater/qemu into staging
aspeed queue:
* Fixes for the Aspeed I2C model
* New SDK image for avocado tests
* blockdev support for flash device definition
* SD refactoring preparing ground for eMMC support
# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCAAdFiEEoPZlSPBIlev+awtgUaNDx8/77KEFAmTxsaQACgkQUaNDx8/7
# 7KGXmg//XJNisscl/VWSBaGmH5MbQUAg/QCRalXx1V/lJ8rhE/JqwnWKuoPFd4EN
# iDlh3ufpzxPhHFc9boechuM5ytlrJxpLJoCIJ4sw/4qnO3Dy3Q6BCy1t8Ma62D1u
# oE7cAMHsriJ1uTJNHUTFo72VapTaH2XwFN9lFDuQW45d+WWAXtVJsqvRgFETNmw6
# YYnTTpH2gLTZZFEgOixhWpGLh4Ibc/l8U1VzL0ctQmC11xng0bqk3PAqU9NGzcM5
# MJmEGAxg43CnFu9NJI1nMqC/coi/8PFtrM7HprSwE3H8Jkwncs4ePVT+kZQC+VNQ
# 7EaVkksfEGHlN8XP5+eQDrQ5yT6ve+fbHTLQhwULfeyt0GlQ8h1yewvHCDWo/zw3
# XI1ZyOcNZ2yiaenSUrTPzu0LiqZEJQnzRjPCpgTi1fU08ryEMEaPtr176YDLCguQ
# cpRj4QSZHCrGl/Eo9NlkFP/2rQDKTvCcedKPkYLQtsurSiH/36Oj9YvZycNtZ574
# ortKAtru4YV/rglNX4L8JDhdI+nqvy1liifpJsiS/2KBZDpVFaP8PzGIV40HNy3G
# 8/LVTnaggZaScF3ftHhkg84uQumELS9l2dhsNCL9HqdlrNXLQrVAIR6iuQlpOKBa
# 5S/6h7ZXGOb1qNVQjYp4HCrB7X1KIJYksZ3GdUREf8ot5Ds1FhE=
# =ymmX
# -----END PGP SIGNATURE-----
# gpg: Signature made Fri 01 Sep 2023 05:40:52 EDT
# gpg: using RSA key A0F66548F04895EBFE6B0B6051A343C7CFFBECA1
# gpg: Good signature from "Cédric Le Goater <clg@redhat.com>" [unknown]
# gpg: aka "Cédric Le Goater <clg@kaod.org>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg: There is no indication that the signature belongs to the owner.
# Primary key fingerprint: A0F6 6548 F048 95EB FE6B 0B60 51A3 43C7 CFFB ECA1
* tag 'pull-aspeed-20230901' of https://github.com/legoater/qemu: (26 commits)
hw/sd: Introduce a "sd-card" SPI variant model
hw/sd: Add sd_cmd_SET_BLOCK_COUNT() handler
hw/sd: Add sd_cmd_SEND_TUNING_BLOCK() handler
hw/sd: Add sd_cmd_SEND_RELATIVE_ADDR() handler
hw/sd: Add sd_cmd_ALL_SEND_CID() handler
hw/sd: Add sd_cmd_SEND_OP_CMD() handler
hw/sd: Add sd_cmd_GO_IDLE_STATE() handler
hw/sd: Add sd_cmd_unimplemented() handler
hw/sd: Add sd_cmd_illegal() handler
hw/sd: Introduce sd_cmd_handler type
hw/sd: Move proto_name to SDProto structure
hw/sd: When card is in wrong state, log which spec version is used
hw/sd: When card is in wrong state, log which state it is
hw/sd/sdcard: Return ILLEGAL for CMD19/CMD23 prior SD spec v3.01
aspeed: Get the BlockBackend of FMC0 from the flash device
m25p80: Introduce an helper to retrieve the BlockBackend of a device
aspeed: Create flash devices only when defaults are enabled
hw/ssi: Check for duplicate CS indexes
aspeed/smc: Wire CS lines at reset
hw/ssi: Introduce a ssi_get_cs() helper
...
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The XIVE interrupt contoller maintains various fields on interrupt
targets in a structure called NVT. Each unit has a NVT cache, backed
by RAM.
When the NVT structure is not local (in RAM) to the chip, the XIVE
interrupt controller forwards the memory operation to the owning chip
using the PC MMIO region configured for this purpose. QEMU does not
need to be so precise since software shouldn't perform any of these
operations. The model implementation is simplified to return the RAM
address of the NVT structure which is then used by pnv_xive_vst_write
or read to perform the operation in RAM.
Remove the last use of pnv_xive_get_remote().
Reviewed-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The notify page of the interrupt controller can either be used to
receive trigger events from the HW controllers (PHB, PSI) or to
reroute interrupts between Interrupt Controllers. In which case, the
VSD table is used to determine the address of the notify page of the
remote IC and the store data is forwarded.
Today, our model grabs the remote VSD (EAS, END, NVT) address using
pnv_xive_get_remote() helper. Be more precise and implement remote END
triggers using a store on the remote IC notify page.
We still have a shortcut in the model for the NVT accesses which we
will address later.
Reviewed-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
It will help us model the END triggers on the PowerNV machine, which
can be rerouted to another interrupt controller.
Reviewed-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
to log an error in case of bad configuration of the XIVE tables by the FW.
Reviewed-by: Frederic Barrat <fbarrat@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
It's unnecessary for non-KVM accelerators(TCG, for example),
to call this function, so change the order of kvm_enable() judgment.
The static inline function that returns -1 directly does not work
in TCG's situation.
Signed-off-by: jianchunfu <chunfu.jian@shingroup.cn>
Tested-by: Gautam Menghani <gautam@linux.ibm.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
ppce500_reset_device_tree is registered for system reset, but after
c4b075318e this function rerandomizes rng-seed via
qemu_guest_getrandom_nofail. And when loading a snapshot, it tries to read
EVENT_RANDOM that doesn't exist, so we have an error:
qemu-system-ppc: Missing random event in the replay log
To fix this, use qemu_register_reset_nosnapshotload instead of
qemu_register_reset.
Reported-by: Vitaly Cheptsov <cheptsov@ispras.ru>
Fixes: c4b075318e ("hw/ppc: pass random seed to fdt ")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1634
Signed-off-by: Maksim Kostin <maksim.kostin@ispras.ru>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1779
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
LQ, STQ have the same register-pair ordering as LQARX/STQARX., which is
the even (lower) register contains the most significant bits. This is
not implemented correctly for big-endian.
do_ldst_quad() has variables low_addr_gpr and high_addr_gpr which is
confusing because they are low and high addresses, whereas LQARX/STQARX.
and most such things use the low and high values for lo/hi variables.
The conversion to native 128-bit memory access functions missed this
strangeness.
Fix this by changing the if condition, and change the variable names to
hi/lo to match convention.
Cc: qemu-stable@nongnu.org
Reported-by: Ivan Warren <ivan@vmfacility.fr>
Fixes: 57b38ffd0c ("target/ppc: Use tcg_gen_qemu_{ld,st}_i128 for LQARX, LQ, STQ")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1836
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
These machines run reverse-debugging well enough to pass basic tests.
Wire them up.
Reviewed-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The reverse-debugging test creates a trace, then replays it and:
1. Steps the first 10 instructions and records their addresses.
2. Steps backward and verifies their addresses match.
3. Runs to (near) the end of the trace.
4. Sets breakpoints on the first 10 instructions.
5. Continues backward and verifies execution stops at the last
breakpoint.
Step 5 breaks if any of the other 9 breakpoints are re-executed in the
trace after the 10th instruction is run, because those will be
unexpectedly hit when reverse continuing. This situation does arise
with the ppc pseries machine, the SLOF bios branches to its own entry
point.
Deal with this by switching steps 3 and 4, so the trace will be run to
the end *or* one of the breakpoints being re-executed. Step 5 then
reverses from there to the 10th instruction will not hit a breakpoint in
between, by definition.
Another step is added between steps 2 and 3, which steps forward over
the first 10 instructions and verifies their addresses, to support this.
Reviewed-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
This the ppc64 record-replay test is able to replay the full kernel boot
so try enabling it.
Acked-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
spapr_machine_reset gets a random number to populate the device-tree
rng seed with. When loading a snapshot for record-replay, the machine
is reset again, and that tries to consume the random event record
again, crashing due to inconsistent record
Fix this by saving the seed to populate the device tree with, and
skipping the rng on snapshot load.
Acked-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
When the machine is reset to load a new snapshot while being debugged
with replay-record, it is done from another thread, so the CPU does
not run the register setting operations. Set CPU registers directly in
machine reset.
Cc: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Timebase save uses a random number for a legacy vmstate field, which
makes rr snapshot loading unbalanced. The easiest way to deal with this
is just to skip the rng if record-replay is active.
Reviewed-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
ppc only migrates reserve_addr, so the destination machine can get a
valid reservation with an incorrect reservation value of 0. Prior to
commit 392d328abe ("target/ppc: Ensure stcx size matches larx"),
this could permit a stcx. to incorrectly succeed. That commit
inadvertently fixed that bug because the target machine starts with an
impossible reservation size of 0, so any stcx. will fail.
This behaviour is permitted by the ISA because reservation loss may
have implementation-dependent cause. What's more, with KVM machines it
is impossible save or reasonably restore reservation state. However if
the vmstate is being used for record-replay, the reservation must be
saved and restored exactly in order for execution from snapshot to
match the record.
This patch deprecates the existing incomplete reserve_addr vmstate,
and adds a new vmstate subsection with complete reservation state.
The new vmstate is needed only when record-replay mode is active.
Acked-by: Pavel Dovgalyuk <Pavel.Dovgalyuk@ispras.ru>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reading the time more than once to perform an operation always increases
complexity and fragility due to introduced deltas. Simplify the
decrementer write by reading the clock once for the operation.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Lower interrupts, delete timers, and set time facility registers
back to initial state on machine reset.
This is not so important for record-replay since timebase and
decrementer are migrated, but it gives a cleaner reset state.
Cc: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Cc: BALATON Zoltan <balaton@eik.bme.hu>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[ clg: checkpatch.pl fixes ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
TCG does not maintain the DEC reigster in the SPR array, so it does get
migrated. TCG also needs to re-start the decrementer timer on the
destination machine.
Load and store the decrementer into the SPR when migrating. This works
for the level-triggered (book3s) decrementer, and should be compatible
with existing KVM machines that do keep the DEC value there.
This fixes lost decrementer interrupt on migration that can cause
hangs, as well as other problems including record-replay bugs.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
When writing a value to the decrementer that raises an exception, the
irq is raised, but the value is not stored so the store doesn't appear
to have changed the register when it is read again.
Always store the write value to the register.
Fixes: e81a982aa5 ("PPC: Clean up DECR implementation")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
When storing a large decrementer value with the most significant
implemented bit set, it is to be treated as a negative and sign
extended.
This isn't hit for book3s DEC because of another bug, fixing it
in the next patch exposes this one and can cause additional
problems, so fix this first. It can be hit with HDECR and other
edge triggered types.
Fixes: a8dafa5251 ("target/ppc: Implement large decrementer support for TCG")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[ clg: removed extra cpu and pcc variables shadowing local variables ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The decrementer register contains a relative time in timebase units.
When writing to DECR this is converted and stored as an absolute value
in nanosecond units, reading DECR converts back to relative timebase.
The tb<->ns conversion of the relative part can cause rounding such that
a value writen to the decrementer can read back a different, with time
held constant. This is a particular problem for a deterministic icount
and record-replay trace.
Fix this by storing the absolute value in timebase units rather than
nanoseconds. The math before:
store: decr_next = now_ns + decr * ns_per_sec / tb_per_sec
load: decr = (decr_next - now_ns) * tb_per_sec / ns_per_sec
load(store): decr = decr * ns_per_sec / tb_per_sec * tb_per_sec /
ns_per_sec
After:
store: decr_next = now_ns * tb_per_sec / ns_per_sec + decr
load: decr = decr_next - now_ns * tb_per_sec / ns_per_sec
load(store): decr = decr
Fixes: 9fddaa0c0c ("PowerPC merge: real time TB and decrementer - faster and simpler exception handling (Jocelyn Mayer)")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
The rule of timers is typically that they should never expire before the
timeout, but some time afterward. Rounding timer intervals up when doing
conversion is the right thing to do.
Under most circumstances it is impossible observe the decrementer
interrupt before the dec register has triggered. However with icount
timing, problems can arise. For example setting DEC to 0 can schedule
the timer for now, causing it to fire before any more instructions
have been executed and DEC is still 0.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
This will be used for converting time intervals in different base units
to host units, for the purpose of scheduling timers to emulate target
timers. Timers typically must not fire before their requested expiry
time but may fire some time afterward, so rounding up is the right way
to implement these.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
[ clg: renamed __muldiv64() to muldiv64_rounding() ]
Signed-off-by: Cédric Le Goater <clg@kaod.org>
These calculations are repeated several times, and they will become
a little more complicated with subsequent changes.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Failing to reset the of_instance_last makes ihandle allocation continue
to increase, which causes record-replay replay fail to match the
recorded trace.
Not resetting claimed_base makes VOF eventually run out of memory after
some resets.
Cc: Alexey Kardashevskiy <aik@ozlabs.ru>
Fixes: fc8c745d50 ("spapr: Implement Open Firmware client interface")
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Cédric Le Goater <clg@kaod.org>
Niklas Cassel
Alexander Ivanov
Denis V. Lunev
Stefan Hajnoczi
Cédric Le Goater
jianchunfu
Maksim Kostin
Richard Henderson
Nicholas Piggin