Peter Maydell [Tue, 20 Sep 2011 15:18:46 +0000 (16:18 +0100)]
MAINTAINERS: update maintainer for target-arm and ARM devboards
Add myself as co-maintainer alongside Paul Brook for the TCG ARM
guest implementation (target-arm) and the ARM dev boards (integratorcp,
realview, stellaris, versatilepb).
Peter Maydell [Fri, 16 Sep 2011 14:40:16 +0000 (15:40 +0100)]
Makefile: Remove 'tarbin' target
Remove the 'tarbin' target -- it isn't used as part of the official
QEMU release process, and it's out of date (various new bios files
were never added to its list of files). It's better not to provide
it at all than to have a broken makefile target we never use or test.
(Creating a tarball by just pulling in binaries that have been installed
directly to the system you're running the build on is a bad idea anyway:
the better way to create a binary tarball would be just to install to
a temporary DESTDIR and then tar up that.)
Now that qemu_set_fd_handler and qemu_set_fd_handler2 have different
implementations, one using qemu iohandlers and the other glib, it is not
safe to mix the two when inserting/deleting handlers.
Avi Kivity [Sun, 18 Sep 2011 12:58:26 +0000 (15:58 +0300)]
irq: introduce qemu_irq_proxy()
In some cases we have a circular dependency involving irqs - the irq
controller depends on a bus, which in turn depends on the irq controller.
Add qemu_irq_proxy() which acts as a passthrough, except that the target
irq may be set later on.
David Gibson [Tue, 20 Sep 2011 02:05:21 +0000 (12:05 +1000)]
Barriers in qemu-barrier.h should not be x86 specific
qemu-barrier.h contains a few macros implementing memory barrier
primitives used in several places throughout qemu. However, apart
from the compiler-only barrier, the defined wmb() is correct only for
x86, or platforms which are similarly strongly ordered.
This patch addresses the FIXME about this by making the wmb() macro
arch dependent. On x86, it remains a compiler barrier only, but with
a comment explaining in more detail the conditions under which this is
correct. On weakly-ordered powerpc, an "eieio" instruction is used,
again with explanation of the conditions under which it is sufficient.
On other platforms, we use the __sync_synchronize() primitive,
available in sufficiently recent gcc (4.2 and after?). This should
implement a full barrier which will be sufficient on all platforms,
although it may be overkill in some cases. Other platforms can add
optimized versions in future if it's worth it for them.
Without proper memory barriers, it is easy to reproduce ordering
problems with virtio on powerpc; specifically, the QEMU puts new
element into the "used" ring and then updates the ring free-running
counter. Without a barrier between these under the right
circumstances, the guest linux driver can receive an interrupt, read
the counter change but find the ring element to be handled still has
an old value, leading to an "id %u is not a head!\n" error message.
Similar problems are likely to be possible with kvm on other weakly
ordered platforms.
David Gibson [Tue, 20 Sep 2011 02:05:20 +0000 (12:05 +1000)]
virtio: Use global memory barrier macros
The virtio code uses wmb() macros in several places, as required by the
SMP-aware virtio protocol. However the wmb() macro is locally defined
to be a compiler barrier only. This is probably sufficient on x86
due to its strong storage ordering model, but it certainly isn't on other
platforms, such as ppc.
In any case, qemu already has some globally defined memory barrier macros
in qemu-barrier.h. This patch, therefore converts virtio.c to use those
barrier macros. The macros in qemu-barrier.h are also wrong (or at least,
safe for x86 only) but this way at least there's only one place to fix
them.
Stefan Weil [Fri, 16 Sep 2011 20:03:08 +0000 (22:03 +0200)]
Fix and clean code which tests the gcc version
The code which tests whether gcc supports warn_unused_result was wrong.
Remove the wrong test from configure and replace it by code using
macro QEMU_GNUC_PREREQ in compiler.h.
Donald Dutile [Wed, 21 Sep 2011 19:25:11 +0000 (15:25 -0400)]
pci-devfn: check that device/slot number is within range
Need to check that guest slot/device number is not > 31 or walk off
the devfn table when checking if a devfn is available or not in a guest.
before this fix, passing in an addr=abc or addr=34,
can crash qemu, sometimes fail gracefully if data past end
of devfn table fails the availability test.
with this fix, get clean error:
Property 'pci-assign.addr' doesn't take value '34'
also tested when no addr= param passed for guest (pcicfg) address,
and that worked as well.
Michael Roth [Wed, 7 Sep 2011 23:40:52 +0000 (18:40 -0500)]
build: fix race with creating qapi-generated
Since qapi-generated/ is a global QEMU include path, we need to make
sure it is created before anything is compiled, so do this in the
configure phase rather than via the Makefile.
Jan Kiszka [Wed, 21 Sep 2011 18:49:29 +0000 (20:49 +0200)]
pc: Unbreak ROM mapping for ISA machine
This is based on the original fix by Hervé Poussineau: pc_memory_init
actually takes a memory region for mapping BIOS and extension ROMs. That
equals the PCI memory region if PCI is available, but must be system
memory in the ISA case.
dann frazier [Wed, 21 Sep 2011 20:06:25 +0000 (14:06 -0600)]
e1000: Don't set the Capabilities List bit
[Originally sent to qemu-kvm list, but I was redirected here]
The Capabilities Pointer is NULL, so this bit shouldn't be set. The state of
this bit doesn't appear to change any behavior on Linux/Windows versions we've
tested, but it does cause Windows' PCI/PCI Express Compliance Test to balk.
I happen to have a physical 82540EM controller, and it also sets the
Capabilities Bit, but it actually has items on the capabilities list to go
with it :)
Nick Thomas [Wed, 21 Sep 2011 10:55:50 +0000 (11:55 +0100)]
block/curl: Don't finish AIOCBs too early
The previous behaviour was to finish AIOCBs inside curl_aio_readv()
if the data was cached. This caused the following failed assertion
at hw/ide/pci.c:314: bmdma_cmd_writeb
Daniel Verkamp [Sat, 27 Aug 2011 09:12:28 +0000 (02:12 -0700)]
ahci: add port I/O index-data pair
Implement an I/O space index-data register pair as defined by the AHCI
spec, including the corresponding SATA PCI capability and BAR.
This allows real-mode code to access the AHCI registers; real-mode
code cannot address the memory-mapped register space because it is
beyond the first megabyte.
Stefan Hajnoczi [Mon, 5 Sep 2011 17:31:21 +0000 (18:31 +0100)]
trace: use binary file open mode in simpletrace
For Windows portability the simple trace backend must use the 'b' file
open mode. This prevents the stdio library from mangling 0x0a/0x0d
newline characters.
Stefan Hajnoczi [Mon, 5 Sep 2011 07:30:17 +0000 (08:30 +0100)]
trace: portable simple trace backend using glib
Convert the simple trace backend to glib so that it works under Windows.
We cannot use pthread directly but glib provides portable abstractions.
Also use glib atomics instead of newish gcc builtins which may not be
supported on Windows toolchains.
Stefan Hajnoczi [Mon, 5 Sep 2011 15:31:45 +0000 (16:31 +0100)]
trace: allow trace events with string arguments
String arguments are useful for producing human-readable traces without
post-processing (e.g. stderr backend). Although the simple backend
cannot handles strings all others can. Strings should be allowed and
the simple backend can be extended to support them.
Peter Maydell [Thu, 15 Sep 2011 11:15:56 +0000 (12:15 +0100)]
configure: Make missing pkg-config an error rather than a warning
If pkg-config doesn't exist then make configure fail immediately
with a useful error message. Now that glib is a required dependency,
proceeding despite the missing pkg-config will just cause us to
fail later with a misleading message about glib not being present.
Now that iothread is always compiled sending a signal seems only an
additional step. This patch also avoid writing to two pipe (one from signal
and one in qemu_service_io).
Work with kvm enabled or disabled. strace output is more readable (less syscalls).
Paolo Bonzini [Fri, 9 Sep 2011 14:47:26 +0000 (16:47 +0200)]
scsi: fix sign extension problems
When assigning a 32-bit value to cmd->xfer (which is 64-bits)
it can be erroneously sign extended because the intermediate
32-bit computation is signed. Fix this by standardizing on
the ld*_be_p functions.
Sage Weil [Thu, 15 Sep 2011 21:11:08 +0000 (14:11 -0700)]
rbd: ignore failures when reading from default conf location
If we are reading from the default config location, ignore any failures.
It is perfectly legal for the user to specify exactly the options they need
and to not rely on any config file.
Alexander Motin [Mon, 12 Sep 2011 08:19:25 +0000 (11:19 +0300)]
AHCI Port Interrupt Enable register cleaning on soft reset
I've found that FreeBSD AHCI driver doesn't work with AHCI hardware
emulation of QEMU 0.15.0. I believe the problem is on QEMU's side. As I
see, it clears port's Interrupt Enable register each time when reset of
any level happens. Is is reasonable for the global controller reset. It
is probably not good, but acceptable for FreeBSD driver for the port
hard reset. But it is IMO wrong for the device soft reset. None of real
hardware I know behaves that way.
Paolo Bonzini [Fri, 16 Sep 2011 14:40:04 +0000 (16:40 +0200)]
scsi-disk: lazily allocate bounce buffer
It will not be needed for reads and writes if the HBA provides a sglist.
In addition, this lets scsi-disk refuse commands with an excessive
allocation length, as well as limit memory on usual well-behaved guests.
Paolo Bonzini [Fri, 16 Sep 2011 14:40:02 +0000 (16:40 +0200)]
dma-helpers: rewrite completion/cancellation
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...
Support bridge filtering on top of the memory
API as suggested by Avi Kivity:
Create a memory region for the bridge's address space. This region is
not directly added to system_memory or its descendants. Devices under
the bridge see this region as its pci_address_space(). The region is
as large as the entire address space - it does not take into account
any windows.
For each of the three windows (pref, non-pref, vga), create an alias
with the appropriate start and size. Map the alias into the bridge's
parent's pci_address_space(), as subregions.
Wen Congyang [Fri, 26 Aug 2011 10:00:33 +0000 (13:00 +0300)]
pci_bridge: use parent bus's address space
The switch to the new memory API caused the following problem:
The pci device may call pci_register_bar() to use PCI bus's address
space. But we don't init PCI bus's address space if it is not bus
0. A crash was reported:
http://lists.gnu.org/archive/html/qemu-devel/2011-08/msg02243.html
More work will be needed to make bridge filtering work correctly
with the memory API.
Paolo Bonzini [Thu, 8 Sep 2011 15:24:56 +0000 (17:24 +0200)]
nbd: support NBD_SET_FLAGS ioctl
The nbd kernel module cannot enable DISCARD requests unless it is
informed about it. The flags field in the header is used for this,
and this patch adds support for it.
Blue Swirl [Sat, 17 Sep 2011 19:16:57 +0000 (19:16 +0000)]
sun4u: don't set up isa_mem_base
Since we use memory API in sun4u.c, after 71579cae30b53c910cd6c47ab4e683f647d36519, setting up isa_mem_base
puts vga.chain4 outside of the physical address space.
Max Reitz [Fri, 16 Sep 2011 15:29:04 +0000 (17:29 +0200)]
target-i386: Fix several SSE3 instructions.
haddp[sd], hsubp[sd] and addsubp[sd] operate on floats, thus it is
necessary to use the appropriate floating point calculation functions.
If this is not done, those functions operate merely on integers, which
is not correct.
Stefan Hajnoczi [Tue, 13 Sep 2011 12:34:37 +0000 (13:34 +0100)]
trace: add virtio_set_status() trace event
The virtio device lifecycle can be observed by looking at the sequence
of set status operations. This is especially important for catching the
reset operation (status value 0), which resets the device and all
virtqueues.
Stefan Hajnoczi [Tue, 13 Sep 2011 12:34:35 +0000 (13:34 +0100)]
trace: allow PRI*64 at beginning and ending of format string
The tracetool parser only picks up PRI*64 and other format string macros
when enclosed between double quoted strings. Lift this restriction by
extracting everything after the closing ')' as the format string:
One trick here: it turns out that backslashes in the format string like
"\n" were being interpreted by echo(1). Fix this by using the POSIX
printf(1) command instead. Although it normally does not make sense to
include backslashes in trace event format strings, an injected newline
causes tracetool to emit a broken header file and I want to eliminate
cases where broken output is emitted, even if the input was bad.
projects / qemu.git / log