Attempting to cancel a dma_blk_io() request causes an abort(3):
void bdrv_aio_cancel(BlockAIOCB *acb)
{
...
while (acb->refcnt > 1) {
if (acb->aiocb_info->get_aio_context) {
aio_poll(acb->aiocb_info->get_aio_context(acb), true);
} else if (acb->bs) {
aio_poll(bdrv_get_aio_context(acb->bs), true);
} else {
abort();
}
}
...
}
This happens because DMAAIOCB->bs is NULL and
dma_aiocb_info.get_aio_context() is also NULL.
This patch trivially implements dma_aiocb_info.get_aio_context() by
fetching the DMAAIOCB->ctx field.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1466451417-27988-1-git-send-email-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Callers of dma_blk_io have no way to pass extra data to the DMAIOFunc,
because the original callback and opaque are gone by the time DMAIOFunc
is called. On the other hand, the BlockBackend is usually derived
from those extra data that you could pass to the DMAIOFunc (in the
next patch, that would be the SCSIRequest).
So change DMAIOFunc's prototype, decoupling it from blk_aio_readv
and blk_aio_writev's. The new prototype loses the BlockBackend
and gains an extra opaque value which, in the case of dma_blk_readv
and dma_blk_writev, is of course used for the BlockBackend.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Sector-based blk_aio_readv() and blk_aio_writev() should die; switch
to byte-based blk_aio_preadv() and blk_aio_pwritev() instead.
The patch had to touch multiple files at once, because dma_blk_io()
takes pointers to the functions, and ide_issue_trim() piggybacks on
the same interface (while ignoring offset under the hood).
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1454089805-5470-16-git-send-email-peter.maydell@linaro.org
- next part in the thread-safe address_space_* saga: atomic access
to the bounce buffer and the map_clients list, from Fam
- optional support for linking with tcmalloc, also from Fam
- reapplying Peter Crosthwaite's "Respect as_translate_internal
length clamp" after fixing the SPARC fallout.
- build system fix from Wei Liu
- small acpi-build and ioport cleanup by myself
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAABCAAGBQJVQJd4AAoJEL/70l94x66DYFYH/3ifhqWZsd4dfJri0CGAHI4i
SpPmNeouc8W+F/3lwf6Inrh5NnTgd5QzoUBMQaWVkQKwUiWls8g2mXkT3jo0iDqT
/B40YXnZjNm20MixNaZmk9AsOF6OqPM8EMufau874k5zTlx3tCGAW1QD+I1N7WK7
DfsFsIUD1svo2prn55fSoitMG1TIVPnpcklb4YGJRbAacQYUDhr5KAIhT1quDR2R
93BvToyQmPqRQ4YKqnJLp8HAkL4FaJumfFZVvyh2cZvyaYGN/RVdi2Dw985dJDPX
/z4enE4GCAs4RDw3lZ1RDbiZDqpT2ibFgASg/arX3SxzqHirOGvMdkOjO99r9j4=
=aLjh
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
- miscellaneous cleanups for TCG (Emilio) and NBD (Bogdan)
- next part in the thread-safe address_space_* saga: atomic access
to the bounce buffer and the map_clients list, from Fam
- optional support for linking with tcmalloc, also from Fam
- reapplying Peter Crosthwaite's "Respect as_translate_internal
length clamp" after fixing the SPARC fallout.
- build system fix from Wei Liu
- small acpi-build and ioport cleanup by myself
# gpg: Signature made Wed Apr 29 09:34:00 2015 BST using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>"
# gpg: aka "Paolo Bonzini <pbonzini@redhat.com>"
# 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: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
* remotes/bonzini/tags/for-upstream: (22 commits)
nbd/trivial: fix type cast for ioctl
translate-all: use bitmap helpers for PageDesc's bitmap
target-i386: disable LINT0 after reset
Makefile.target: prepend $libs_softmmu to $LIBS
milkymist: do not modify libs-softmmu
configure: Add support for tcmalloc
exec: Respect as_translate_internal length clamp
ioport: reserve the whole range of an I/O port in the AddressSpace
ioport: loosen assertions on emulation of 16-bit ports
ioport: remove wrong comment
ide: there is only one data port
gus: clean up MemoryRegionPortio
sb16: remove useless mixer_write_indexw
sun4m: fix slavio sysctrl and led register sizes
acpi-build: remove dependency from ram_addr.h
memory: add memory_region_ram_resize
dma-helpers: Fix race condition of continue_after_map_failure and dma_aio_cancel
exec: Notify cpu_register_map_client caller if the bounce buffer is available
exec: Protect map_client_list with mutex
linux-user, bsd-user: Remove two calls to cpu_exec_init_all
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
If DMA's owning thread cancels the IO while the bounce buffer's owning thread
is notifying the "cpu client list", a use-after-free happens:
continue_after_map_failure dma_aio_cancel
------------------------------------------------------------------
aio_bh_new
qemu_bh_delete
qemu_bh_schedule (use after free)
Also, the old code doesn't run the bh in the right AioContext.
Fix both problems by passing a QEMUBH to cpu_register_map_client.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1426496617-10702-6-git-send-email-famz@redhat.com>
[Remove unnecessary forward declaration. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Make address_space_rw take transaction attributes, rather
than always using the 'unspecified' attributes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Edgar E. Iglesias <edgar.iglesias@xilinx.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Device models should access their block backends only through the
block-backend.h API. Convert them, and drop direct includes of
inappropriate headers.
Just four uses of BlockDriverState are left:
* The Xen paravirtual block device backend (xen_disk.c) opens images
itself when set up via xenbus, bypassing blockdev.c. I figure it
should go through qmp_blockdev_add() instead.
* Device model "usb-storage" prompts for keys. No other device model
does, and this one probably shouldn't do it, either.
* ide_issue_trim_cb() uses bdrv_aio_discard() instead of
blk_aio_discard() because it fishes its backend out of a BlockAIOCB,
which has only the BlockDriverState.
* PC87312State has an unused BlockDriverState[] member.
The next two commits take care of the latter two.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I'll use it with block backends shortly, and the name is going to fit
badly there. It's a block layer thing anyway, not just a block driver
thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I'll use BlockDriverAIOCB with block backends shortly, and the name is
going to fit badly there. It's a block layer thing anyway, not just a
block driver thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is the next step for decoupling block accounting functions from
BlockDriverState.
In a future commit the BlockAcctStats structure will be moved from
BlockDriverState to the device models structures.
Note that bdrv_get_stats was introduced so device models can retrieve the
BlockAcctStats structure of a BlockDriverState without being aware of it's
layout.
This function should go away when BlockAcctStats will be embedded in the device
models structures.
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Keith Busch <keith.busch@intel.com>
CC: Anthony Liguori <aliguori@amazon.com>
CC: "Michael S. Tsirkin" <mst@redhat.com>
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Eric Blake <eblake@redhat.com>
CC: Peter Maydell <peter.maydell@linaro.org>
CC: Michael Tokarev <mjt@tls.msk.ru>
CC: John Snow <jsnow@redhat.com>
CC: Markus Armbruster <armbru@redhat.com>
CC: Alexander Graf <agraf@suse.de>
CC: Max Reitz <mreitz@redhat.com>
Signed-off-by: Benoît Canet <benoit.canet@nodalink.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the size of the scatter/gather list isn't a multiple of 512, the
number of sectors for the block layer request is rounded down, resulting
in a qiov that doesn't match the request length. Truncate the qiov to the
new length of the request.
This fixes the IDE qtest case /x86_64/ide/bmdma/short_prdt.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Calling dma_bdrv_unmap() twice is not necessary and may cause
potential problems if some code changes.
Signed-off-by: Jules Wang <junqing.wang@cs2c.com.cn>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Initialize the dbs->in_cancel flag in dma_bdrv_io(), since qemu_aio_get()
does not return zero-initialized memory. Spotted by the clang sanitizer
(which complained when the value loaded in dma_complete() was not valid
for a bool type); this might have resulted in leaking the AIO block.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
include/qemu/timer.h has no need to include main-loop.h and
doing so causes an issue for the next patch. Unfortunately
various files assume including timers.h will pull in main-loop.h.
Untangle this mess.
Signed-off-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The DMAContext is a simple pointer to an AddressSpace that is now always
already available. Make everyone hold the address space directly,
and clean up the DMA API to use the AddressSpace directly.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The translate function in the DMAContext is now always NULL.
Remove every reference to it.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The old-style IOMMU lets you check whether an access is valid in a
given DMAContext. There is no equivalent for AddressSpace in the
memory API, implement it with a lookup of the dispatch tree.
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Now that AIOPool no longer keeps a freelist, it isn't really a "pool"
anymore. Rename it to AIOCBInfo and make it const since it no longer
needs to be modified.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
target_phys_addr_t is unwieldly, violates the C standard (_t suffixes are
reserved) and its purpose doesn't match the name (most target_phys_addr_t
addresses are not target specific). Replace it with a finger-friendly,
standards conformant hwaddr.
Outstanding patchsets can be fixed up with the command
git rebase -i --exec 'find -name "*.[ch]"
| xargs s/target_phys_addr_t/hwaddr/g' origin
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Instead of accessing the cpu address space, use an address space
configured by the caller.
Eventually all dma functionality will be folded into AddressSpace,
but we have to start from something.
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Hi hard a brain fart when coding that function, it will
fail to "set" the memory beyond the first 512 bytes. This
is in turn causing guest crashes in ibmveth (spapr_llan.c
on the qemu side) due to the receive queue not being
properly initialized.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
I noticed that in hw/ide/ahci:ahci_dma_rw_buf() we do not free the sglist. Thus,
I've added a call to qemu_sglist_destroy() to fix this memory leak.
In addition, I've adeed a call in qemu_sglist_destroy() to 0 all of the sglist
fields, in case there is some other codepath that tries to free the sglist.
Signed-off-by: Jason Baron <jbaron@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The emulated devices can run simultaneously with the guest, so
we need to be careful with ordering of load and stores done by
them to the guest system memory, which need to be observed in
the right order by the guest operating system.
This adds a barrier call to the basic DMA read/write ops which
is currently implemented as a smp_mb(), but could be later
improved for more fine grained control of barriers.
Additionally, a _relaxed() variant of the accessors is provided
to easily convert devices who would be performance sensitive
and negatively impacted by the change.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This patch adds the basic infrastructure necessary to emulate an IOMMU
visible to the guest. The DMAContext structure is extended with
information and a callback describing the translation, and the various
DMA functions used by devices will now perform IOMMU translation using
this callback.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
dma-helpers.c contains a number of helper functions for doing
scatter/gather DMA, and various block device related DMA. Currently,
these directly access guest memory using cpu_physical_memory_*(),
assuming no IOMMU translation.
This patch updates this code to use the new universal DMA helper
functions. qemu_sglist_init() now takes a DMAContext * to describe
the DMA address space in which the scatter/gather will take place.
We minimally update the callers qemu_sglist_init() to pass NULL
(i.e. no translation, same as current behaviour). Some of those
callers should pass something else in some cases to allow proper IOMMU
translation in future, but that will be fixed in later patches.
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Not that long ago, every device implementation using DMA directly
accessed guest memory using cpu_physical_memory_*(). This meant that
adding support for a guest visible IOMMU would require changing every
one of these devices to go through IOMMU translation.
Shortly before qemu 1.0, I made a start on fixing this by providing
helper functions for PCI DMA. These are currently just stubs which
call the direct access functions, but mean that an IOMMU can be
implemented in one place, rather than for every PCI device.
Clearly, this doesn't help for non PCI devices, which could also be
IOMMU translated on some platforms. It is also problematic for the
devices which have both PCI and non-PCI version (e.g. OHCI, AHCI) - we
cannot use the the pci_dma_*() functions, because they assume the
presence of a PCIDevice, but we don't want to have to check between
pci_dma_*() and cpu_physical_memory_*() every time we do a DMA in the
device code.
This patch makes the first step on addressing both these problems, by
introducing new (stub) dma helper functions which can be used for any
DMA capable device.
These dma functions take a DMAContext *, a new (currently empty)
variable describing the DMA address space in which the operation is to
take place. NULL indicates untranslated DMA directly into guest
physical address space. The intention is that in future non-NULL
values will given information about any necessary IOMMU translation.
DMA using devices must obtain a DMAContext (or, potentially, contexts)
from their bus or platform. For now this patch just converts the PCI
wrappers to be implemented in terms of the universal wrappers,
converting other drivers can take place over time.
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Eduard - Gabriel Munteanu <eduard.munteanu@linux360.ro>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Currently dma_bdrv_io() takes a 'to_dev' boolean parameter to
determine the direction of DMA it is emulating. We already have a
DMADirection enum designed specifically to encode DMA directions.
This patch uses it for dma_bdrv_io() as well. This involves removing
the DMADirection definition from the #ifdef it was inside, but since that
only existed to protect the definition of dma_addr_t from places where
config.h is not included, there wasn't any reason for it to be there in
the first place.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
And remove several block_int.h inclusions that should not be there.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
These helpers do a full transfer from an in-memory buffer to target
memory, with support for scatter/gather lists. It will be used to
store the reply of an emulated command into a QEMUSGList provided by
the adapter.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch uses the newly created dma_addr_t type throughout the
scatter/gather handling code in dma-helpers.c whenever we need to
represent a dma bus address. This makes a better distinction as to
what is a bus address and what is a cpu physical address. Since we
don't support IOMMUs yet, they can't be very different for now, but
that will change in future, and this preliminary helps clarify what's
going on.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This fixes various problems with completion/cancellation:
* if the io_func fails to get an AIOCB, the callback wasn't called
* If DMA encounters a bounce buffer conflict, and the DMA operation is
canceled before the bottom half fires, bad things happen.
* memory is not unmapped after cancellation, again causing problems
when doing DMA to I/O areas
* cancellation could leak the iovec
* the callback was missed if the I/O operation failed without returning
an AIOCB
and probably more that I've missed. The patch fixes them by sharing
the cleanup code between completion and cancellation. The dma_bdrv_cb
now returns a boolean completed/not completed flag, and the wrapper
dma_continue takes care of tasks to do upon completion.
Most of these are basically impossible in practice, but it is better
to be tidy...
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Make dma_bdrv_io available for drivers, and pass an explicit I/O function
instead of hardcoding bdrv_aio_readv/bdrv_aio_writev. This is required
to implement non-READ/WRITE dma commands in the ide driver, e.g. the
upcoming TRIM support.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the very least, a change like this requires discussion on the list.
The naming convention is goofy and it causes a massive merge problem. Something
like this _must_ be presented on the list first so people can provide input
and cope with it.
This reverts commit 99a0949b72.
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Now that we have a separate aio pool structure we can remove those
aio pool details from BlockDriver.
Every driver supporting AIO now needs to declare a static AIOPool
with the aiocb size and the cancellation method. This cleans up the
current code considerably and will make it cleaner and more obvious
to support two different aio implementations behind a single
BlockDriver.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
The block layer may signal an immediate error on an asynchronous request
by returning NULL. The DMA API did not handle this correctly, returning
an AIO request which would never complete (and which would crash if
cancelled).
Fix by detecting the failure and propagating it.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@6893 c046a42c-6fe2-441c-8c8c-71466251a162
Stefan Hajnoczi