This reverts commit eae3bd1eb7.
Reverted to avoid conflicts for geometry options revert.
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This reverts commit b008326744.
Hold off removing this for one more QEMU release (current libvirt
release still uses it.)
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This reverts commit 6266e900b8.
Some deprecated -drive options were still in use by libvirt, only
fixed with libvirt commit b340c6c614 ("qemu: format serial and geometry
on frontend disk device"), which is not yet in any released version
of libvirt.
So let's hold off removing the deprecated options for one more QEMU
release.
Reported-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Cornelia Huck <cohuck@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
These two states will be missing when doing "query-migrate" on
destination VM. Add these states so that we can get the query results
as expected.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180710091902.28780-5-peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The calculation on size of received bitmap is incorrect for postcopy
recovery. Here we wanted to let the size to cover all the valid bits in
the bitmap, we should use DIV_ROUND_UP() instead of a division.
For example, a RAMBlock with size=4K (which contains only one single 4K
page) will have nbits=1, then nbits/8=0, then the real bitmap won't be
sent to source at all.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180710091902.28780-4-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We were checking against -EIO, assuming that it will cover all IO
failures. But actually it is not. One example is that in
qemu_loadvm_section_start_full() we can have tons of places that will
return -EINVAL even if the error is caused by IO failures on the
network.
Let's loosen the recovery check logic here to cover all the error cases
happened by removing the explicit check against -EIO. After all we
won't lose anything here if any other failure happened.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180710091902.28780-3-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Firstly, renaming the old matching_page_sizes variable to
matches_target_page_size, which suites more to what it did (it only
checks against target page size rather than multiple page sizes).
Meanwhile, simplify the check logic a bit, and enhance the comments.
Should have no functional change.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180710091902.28780-2-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This is the 2nd patch to unbreak postcopy recovery.
Let's unify the migration_incoming_process() call at a single place
rather than calling it in connection setup codes. This fixes a problem
that we will go into incoming migration procedure even if we are trying
to recovery from a paused postcopy migration.
Fixes: 36c2f8be2c ("migration: Delay start of migration main routines")
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180627132246.5576-5-peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The whole postcopy recovery logic was accidentally broken. We need to
fix it in two steps.
This is the first step that we should do the recovery when needed. It
was bypassed before after commit 36c2f8be2c.
Introduce postcopy_try_recovery() helper for the postcopy recovery
logic. Call it both in migration_fd_process_incoming() and
migration_ioc_process_incoming().
Fixes: 36c2f8be2c ("migration: Delay start of migration main routines")
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180627132246.5576-4-peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Move the call to migration_incoming_process() out of multifd code. It's
a bit strange that we can migration generic calls in multifd code.
Instead, let multifd_recv_new_channel() return a boolean showing whether
it's ready to continue the incoming migration.
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180627132246.5576-3-peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Before this patch we firstly setup the postcopy-paused state then we
clean up the QEMUFile handles. That can be racy if there is a very fast
"migrate-recover" command running in parallel. Fix that up.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <20180627132246.5576-2-peterx@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The sector size needs to be large enough to accommodate the data
structures for the log super block and log write entries. This was
previously not properly checked, which made it possible to cause
QEMU to badly misbehave.
Signed-off-by: Ari Sundholm <ari@tuxera.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This was accidentally omitted. Thanks to Eric Blake for spotting this.
Signed-off-by: Ari Sundholm <ari@tuxera.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Fleecing scheme works as follows: we want a kind of temporary snapshot
of active drive A. We create temporary image B, with B->backing = A.
Then we start backup(sync=none) from A to B. From this point, B reads
as point-in-time snapshot of A (A continues to be active drive,
accepting guest IO).
This scheme needs some additional synchronization between reads from B
and backup COW operations, otherwise, the following situation is
theoretically possible:
(assume B is qcow2, client is NBD client, reading from B)
1. client starts reading and take qcow2 mutex in qcow2_co_preadv, and
goes up to l2 table loading (assume cache miss)
2) guest write => backup COW => qcow2 write =>
try to take qcow2 mutex => waiting
3. l2 table loaded, we see that cluster is UNALLOCATED, go to
"case QCOW2_CLUSTER_UNALLOCATED" and unlock mutex before
bdrv_co_preadv(bs->backing, ...)
4) aha, mutex unlocked, backup COW continues, and we finally finish
guest write and change cluster in our active disk A
5. actually, do bdrv_co_preadv(bs->backing, ...) and read
_new updated_ data.
To avoid this, let's make backup writes serializing, to not intersect
with reads from B.
Note: we expand range of handled cases from (sync=none and
B->backing = A) to just (A in backing chain of B), to finally allow
safe reading from B during backup for all cases when A in backing chain
of B, i.e. B formally looks like point-in-time snapshot of A.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Serialized writes should be used in copy-on-write of backup(sync=none)
for image fleecing scheme.
We need to change an assert in bdrv_aligned_pwritev, added in
28de2dcd88. The assert may fail now, because call to
wait_serialising_requests here may become first call to it for this
request with serializing flag set. It occurs if the request is aligned
(otherwise, we should already set serializing flag before calling
bdrv_aligned_pwritev and correspondingly waited for all intersecting
requests). However, for aligned requests, we should not care about
outdating of previously read data, as there no such data. Therefore,
let's just update an assert to not care about aligned requests.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Pass read flags and write flags separately. This is needed to handle
coming BDRV_REQ_NO_SERIALISING clearly in following patches.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Here two things are fixed:
1. Architecture
On each recursion step, we go to the child of src or dst, only for one
of them. So, it's wrong to create tracked requests for both on each
step. It leads to tracked requests duplication.
2. Wait for serializing requests on write path independently of
BDRV_REQ_NO_SERIALISING
Before commit 9ded4a0114 "backup: Use copy offloading",
BDRV_REQ_NO_SERIALISING was used for only one case: read in
copy-on-write operation during backup. Also, the flag was handled only
on read path (in bdrv_co_preadv and bdrv_aligned_preadv).
After 9ded4a0114, flag is used for not waiting serializing operations
on backup target (in same case of copy-on-write operation). This
behavior change is unsubstantiated and potentially dangerous, let's
drop it and add additional asserts and documentation.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Luks needs special parameters to operate the image. Since this test is
focusing on image fleecing, skip skip that format.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For M-profile exception returns, the mmu index to use for exception
return unstacking is supposed to be that of wherever we are returning to:
* if returning to handler mode, privileged
* if returning to thread mode, privileged or unprivileged depending on
CONTROL.nPRIV for the destination security state
We were passing the wrong thing as the 'priv' argument to
arm_v7m_mmu_idx_for_secstate_and_priv(). The effect was that guests
which programmed the MPU to behave differently for privileged and
unprivileged code could get spurious MemManage Unstack exceptions.
Reported-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20180709124535.1116-1-peter.maydell@linaro.org
If the virtual disk size isn't aligned to full clusters,
bdrv_co_do_copy_on_readv() may get pnum == 0 before having the full
cluster completed, which will let it run into an assertion failure:
qemu-io: block/io.c:1203: bdrv_co_do_copy_on_readv: Assertion `skip_bytes < pnum' failed.
Check for EOF, assert that we read at least as much as the read request
originally wanted to have (which is true at EOF because otherwise
bdrv_check_byte_request() would already have returned an error) and
return success early even though we couldn't copy the full cluster.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit dcf94a23b1 ('block: Don't poll in parent drain callbacks')
removed polling in bdrv_child_cb_drained_begin() on the grounds that the
original bdrv_drain() already will poll and BdrvChildRole.drained_begin
calls must not cause graph changes (and therefore must not call
aio_poll() or the recursion through the graph will break.
This reasoning is correct for calls through bdrv_do_drained_begin().
However, BdrvChildRole.drained_begin is also called when a node that is
already in a drained section (i.e. bdrv_do_drained_begin() has already
returned and therefore can't poll any more) is attached to a new parent.
In this case, we must explicitly poll to have all requests completed
before the drained new child can be attached to the parent.
In bdrv_replace_child_noperm(), we know that we're not inside the
recursion of bdrv_do_drained_begin() because graph changes are not
allowed there, and bdrv_replace_child_noperm() is a graph change. The
call of BdrvChildRole.drained_begin() must therefore be followed by a
BDRV_POLL_WHILE() that waits for the completion of requests.
Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This new test verifies that VMDK backing file reads fail when the
backing file has a non-matching CID. This includes non-VMDK backing
files.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180702210721.4847-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
VMDK performs a probing check in vmdk_co_create_opts() to prevent the
user from assigning non-VMDK files as a backing file, because it only
supports VMDK backing files. However, with the @backing runtime option,
it is possible to assign arbitrary nodes as backing nodes, regardless of
what the image header says. Therefore, VMDK may not just access backing
nodes assuming they are VMDK nodes -- which it does, because it needs to
compare the backing file's CID with the overlay's parentCID value, and
naturally the backing file only has a CID when it's a VMDK file.
Instead, it should report the CID of non-VMDK backing files not to match
the overlay because clearly a non-present CID does not match.
Without this change, vmdk_read_cid() reads from the backing file's
bs->file, which may be NULL (in which case we get a segfault). Also, it
interprets bs->opaque as a BDRVVmdkState and then reads from the
.desc_offset field, which usually will just return some arbitrary value
which then results in either garbage to be read, or bdrv_pread() to
return an error, both of which result in a non-matching CID to be
reported.
(In a very unlikely case, we could read something that looks like a
VMDK descriptor, and then get a CID which might actually match. But
that is highly unlikely, and the only result would be that VMDK accepts
the backing file which is not too bad (albeit unintentional).)
((And in theory, the seek to .desc_offset might leak data from another
block driver's opaque object. But then again, the user should realize
very quickly that a non-VMDK backing file does not work (because the
read will very likely fail, due to the reasons given above), so this
should not be exploitable.))
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180702210721.4847-2-mreitz@redhat.com
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
One of them is a typo. But update both to be more readable.
Reported-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 20180702025836.20957-3-famz@redhat.com
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Commit 7a3099fc9c5c("numa: postpone options post-processing till machine_run_board_init()")
broke the commit 7b8be49d36fc("NUMA: Enable adding NUMA node implicitly").
The machine_run_board_init() doesn't do NUMA setup if nb_numa_nodes=0,
but the numa_complete_configuration need add a new node if memory hotplug
is enabled (slots > 0) even nb_numa_nodes=0.
So, Remove the check for numa_complete_configuration() to fix this.
Fixes 7a3099fc9c5c("numa: postpone options post-processing till machine_run_board_init()")
Signed-off-by: Dou Liyang <douly.fnst@cn.fujitsu.com>
Message-Id: <20180704132239.6506-1-douly.fnst@cn.fujitsu.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Machines might have inititalized device_memory if they support memory
devices, so let's properly free it.
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20180702094152.7882-1-david@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
The translator loop does not allow the tb_start hook to set
dc->base.is_jmp; the only hook allowed to do that is translate_insn.
Split the work between init_disas_context where we validate
the gUSA parameters, and translate_insn where we emit code.
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This fixes a record-replay regression introduced by 95590e2
("translate-all: discard TB when tb_link_page returns an existing
matching TB", 2018-06-15). The problem is that code using CF_NOCACHE
assumes that the TB returned from tb_gen_code is always a
newly-generated one. This assumption, however, was broken in
the aforementioned commit.
Fix it by honouring CF_NOCACHE, so that tb_gen_code always
returns a newly-generated TB when CF_NOCACHE is passed to it.
Do this by avoiding the TB hash table if CF_NOCACHE is set.
Reported-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Tested-by: Pavel Dovgalyuk <Pavel.Dovgaluk@ispras.ru>
Signed-off-by: Emilio G. Cota <cota@braap.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Alistair Francis <alistair.francis@wdc.com>
Message-id: 1530806837-5416-1-git-send-email-cota@braap.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Currently we use memory_region_init_rom_nomigrate() to create
the "dp3893x-prom" memory region, and we don't manually register
it with vmstate_register_ram(). This currently means that its
contents are migrated but as a ram block whose name is the empty
string; in future it may mean they are not migrated at all. Use
memory_region_init_ram() instead.
Note that this is a a cross-version migration compatibility break
for the MIPS "magnum" and "pica61" machines.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Aleksandar Markovic <aleksandar.markovic@wavecomp.com>
Message-id: 20180706174309.27110-1-peter.maydell@linaro.org
commit b08199c6fb accidentally added a reference to a doc
comment to a nonexistent memory_region_allocate_aux_memory().
This was a leftover from a previous version of the patchset
which defined memory_region_allocate_aux_memory() for
"allocate RAM MemoryRegion and register it for migration"
and left "memory_region_init_ram()" with its original semantics
of "allocate RAM MR but do not register for migration". In
the end we decided on the approach of "memory_region_init_ram()
registers the MR for migration, and memory_region_init_ram_nomigrate()
is a new function which does not", but this comment change
got left in by mistake. Revert that part of the commit.
Reported-by: Thomas Huth <huth@tuxfamily.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 20180702130605.13611-1-peter.maydell@linaro.org
DeviceClass::reset models a "cold power-on" reset which can
also be used to powercycle a device; but there is no "hot reset"
(a.k.a. soft-reset) method available.
The OMAP MMC Power-Up Control bit is not designed to powercycle
a card, but to disable it without powering it off (pseudo-reset):
Multimedia Card (MMC/SD/SDIO) Interface [SPRU765A]
MMC_CON[11] Power-Up Control (POW)
This bit must be set to 1 before any valid transaction to either
MMC/SD or SPI memory cards.
When 1, the card is considered powered-up and the controller core
is enabled.
When 0, the card is considered powered-down (system dependent),
and the controller core logic is in pseudo-reset state. This is,
the MMC_STAT flags and the FIFO pointers are reset, any access to
MMC_DATA[DATA] has no effect, a write into the MMC.CMD register
is ignored, and a setting of MMC_SPI[STR] to 1 is ignored.
By splitting the 'pseudo-reset' code out of the 'power-on' reset
function, this patch fixes a latent bug in omap_mmc_write(MMC_CON)i
recently exposed by ecd219f7ab.
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180706162155.8432-2-f4bug@amsat.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Use MAKE_64BIT_MASK instead of open-coding. Remove an odd
vector size check that is unlikely to be more profitable
than 3 64-bit integer stores. Correct the iteration for WORD
to avoid writing too much data.
Fixes RISU tests of PTRUE for VL 256.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20180705191929.30773-3-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Normally this is automatic in the size restrictions that are placed
on vector sizes coming from the implementation. However, for the
legitimate size tuple [oprsz=8, maxsz=32], we need to clear the final
24 bytes of the vector register. Without this check, do_dup selects
TCG_TYPE_V128 and clears only 16 bytes.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Tested-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20180705191929.30773-2-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
These instructions must perform the sve_access_check, but
since they are implemented as NOPs there is no generated
code to elide when the access check fails.
Fixes: Coverity issues 1393780 & 1393779.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
If the CMSDK APB timer is set up with a zero RELOAD value
then it will count down to zero, fire once and then stay
at zero. From the point of view of the ptimer system, the
timer is disabled; but the enable bit in the CTRL register
is still set and if the guest subsequently writes to the
RELOAD or VALUE registers this should cause the timer to
start counting down again.
Add code to the write paths for RELOAD and VALUE so that
we correctly restart the timer in this situation.
Conversely, if the new RELOAD and VALUE are both zero,
we should stop the ptimer.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Message-id: 20180703171044.9503-5-peter.maydell@linaro.org
The CMSDK APB timer is currently always configured as periodic timer.
This results in the following messages when trying to boot Linux.
Timer with delta zero, disabling
If the timer limit set with the RELOAD command is 0, the timer
needs to be enabled as one-shot timer.
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The CMSDK timer interrupt triggers when the counter goes from 1 to 0,
so we want to trigger immediately, rather than waiting for a
clock cycle. Drop the incorrect NO_IMMEDIATE_TRIGGER setting.
We also do not want to get an interrupt if the guest sets the
counter directly to zero, so use the new TRIGGER_ONLY_ON_DECREMENT
policy.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Message-id: 20180703171044.9503-3-peter.maydell@linaro.org
The CMSDK timer behaviour is that an interrupt is triggered when the
counter counts down from 1 to 0; however one is not triggered if the
counter is manually set to 0 by a guest write to the counter register.
Currently ptimer can't handle this; add a policy option to allow
a ptimer user to request this behaviour.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Message-id: 20180703171044.9503-2-peter.maydell@linaro.org
Cornelia Huck