OSDN Git Service

qmiga/qemu.git
11 years agopixman: fix warning
Gerd Hoffmann [Wed, 9 Jan 2013 09:17:08 +0000 (10:17 +0100)]
pixman: fix warning

Cc: afaerber@suse.de
Cc: agraf@suse.de
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agogcc: rename CONFIG_PRAGMA_DISABLE_UNUSED_BUT_SET to CONFIG_PRAGMA_DIAGNOSTIC_AVAILABLE
Gerd Hoffmann [Wed, 9 Jan 2013 09:17:07 +0000 (10:17 +0100)]
gcc: rename CONFIG_PRAGMA_DISABLE_UNUSED_BUT_SET to CONFIG_PRAGMA_DIAGNOSTIC_AVAILABLE

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agooptionrom: build with discrete CPP and AS steps
Blue Swirl [Sat, 5 Jan 2013 10:10:27 +0000 (10:10 +0000)]
optionrom: build with discrete CPP and AS steps

Build option ROM .S files with separate preprocessor and
assembler steps because the C compiler could be unsuitable.

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agoslirp: remove unused field tt
Blue Swirl [Sat, 5 Jan 2013 14:48:22 +0000 (14:48 +0000)]
slirp: remove unused field tt

Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agouser: Consider symbolic links as possible directories
Richard Henderson [Sat, 5 Jan 2013 00:39:33 +0000 (16:39 -0800)]
user: Consider symbolic links as possible directories

Commit 2296f194dfde4c0a54f249d3fdb8c8ca21dc611b reduced the number
of syscalls performed during user emulation startup, but failed to
consider the use of symbolic links in creating directory structures.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agoalpha-linux-user: Fix sigaction
Richard Henderson [Sat, 5 Jan 2013 00:39:32 +0000 (16:39 -0800)]
alpha-linux-user: Fix sigaction

Unconditional bswap replaced by __get_user/__put_user.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agolinux-user: Rewrite __get_user/__put_user with __builtin_choose_expr
Richard Henderson [Sat, 5 Jan 2013 00:39:31 +0000 (16:39 -0800)]
linux-user: Rewrite __get_user/__put_user with __builtin_choose_expr

The previous formuation with multiple assignments to __typeof(*hptr) falls
down when hptr is qualified const.  E.g. with const struct S *p, p->f is
also qualified const.

With this formulation, there's no assignment to any local variable.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agobswap: Rewrite cpu_to_<endian><type>u with {ld,st}<type>_<endian>_p
Richard Henderson [Sat, 5 Jan 2013 00:39:30 +0000 (16:39 -0800)]
bswap: Rewrite cpu_to_<endian><type>u with {ld,st}<type>_<endian>_p

We've now optimized the ld/st versions; reuse that for the "legacy"
versions.  Always use inlines so that we get the type checking that
we expect.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agobswap: Rewrite all ld<type>_<endian>_p functions
Richard Henderson [Sat, 5 Jan 2013 00:39:29 +0000 (16:39 -0800)]
bswap: Rewrite all ld<type>_<endian>_p functions

Use the new host endian unaligned access functions instead of
open coding byte-by-byte references.  Remove assembly special
cases for i386 and ppc -- we've now exposed the operation to
the compiler sufficiently for these to be optimized automatically.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agobswap: Add host endian unaligned access functions
Richard Henderson [Sat, 5 Jan 2013 00:39:28 +0000 (16:39 -0800)]
bswap: Add host endian unaligned access functions

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agobswap: Tidy base definitions of bswapN
Richard Henderson [Sat, 5 Jan 2013 00:39:27 +0000 (16:39 -0800)]
bswap: Tidy base definitions of bswapN

Move the bswap_N -> bswapN wrappers inside CONFIG_BYTESWAP_H.

Change the ultimate fallback defintions from macros to inline functions.
The proper types recieved by the function arguments means we can remove
unnecessary casts, making the code more readable.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agofdt: Use bswapN instead of bswap_N
Richard Henderson [Sat, 5 Jan 2013 00:39:26 +0000 (16:39 -0800)]
fdt: Use bswapN instead of bswap_N

Fixes the libfdt enabled build for hosts that have <machine/bswap.h>.

The code at the beginning of qemu/bswap.h is attempting to standardize
on bswapN.  In the case of CONFIG_MACHINE_BSWAP_H, this is all we get.
In the case of CONFIG_BYTESWAP_H, we get bswap_N from the system header
and then wrap these with inline functions to get bswapN.

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
11 years agoMerge remote-tracking branch 'stefanha/trivial-patches' into staging
Anthony Liguori [Fri, 11 Jan 2013 14:43:18 +0000 (08:43 -0600)]
Merge remote-tracking branch 'stefanha/trivial-patches' into staging

* stefanha/trivial-patches:
  hw/pc.c: Fix converting of ioport_register* to MemoryRegion
  Replace remaining gmtime, localtime by gmtime_r, localtime_r
  savevm: Remove MinGW specific code which is no longer needed
  qga/channel-posix.c: Explicitly include string.h
  configure: Fix comment (copy+paste bug)
  readline: avoid memcpy() of overlapping regions

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'kraxel/q35.1' into staging
Anthony Liguori [Fri, 11 Jan 2013 14:36:52 +0000 (08:36 -0600)]
Merge remote-tracking branch 'kraxel/q35.1' into staging

* kraxel/q35.1:
  Makefile: install the "acpi-dsdt.aml" and "q35-acpi-dsdt.aml" blobs too
  pc: rename machine types
  q35: document chipset devices
  q35: add ich9 intel hda controller

Conflicts:
hw/intel-hda.c

aliguori: resolve conflict with static const change from Andreas.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'afaerber-or/prep-up' into staging
Anthony Liguori [Fri, 11 Jan 2013 14:23:16 +0000 (08:23 -0600)]
Merge remote-tracking branch 'afaerber-or/prep-up' into staging

* afaerber-or/prep-up:
  prep: Use pc87312 device instead of collection of random ISA devices
  prep: Add pc87312 Super I/O emulation
  prep: Include devices for ppc64 as well

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agohw/pc.c: Fix converting of ioport_register* to MemoryRegion
Julien Grall [Wed, 9 Jan 2013 18:10:22 +0000 (18:10 +0000)]
hw/pc.c: Fix converting of ioport_register* to MemoryRegion

The commit 258711 introduced MemoryRegion to replace ioport_region*
for ioport 80h and F0h.
A MemoryRegion needs to have both read and write callback otherwise a segfault
will occur when an access is made.

The previous behaviour of this both ioport is to return 0xffffffffffffffff.
So keep this behaviour.

Reported-by: Adam Lackorzynski <adam@os.inf.tu-dresden.de>
Signed-off-by: Julien Grall <julien.grall@citrix.com>
Tested-by: Adam Lackorzynski <adam@os.inf.tu-dresden.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoReplace remaining gmtime, localtime by gmtime_r, localtime_r
Stefan Weil [Mon, 7 Jan 2013 22:08:13 +0000 (23:08 +0100)]
Replace remaining gmtime, localtime by gmtime_r, localtime_r

This allows removing of MinGW specific code and improves
reentrancy for POSIX hosts.

[Removed unused ret variable in qemu_get_timedate() to fix warning:
vl.c: In function ‘qemu_get_timedate’:
vl.c:451:16: error: variable ‘ret’ set but not used [-Werror=unused-but-set-variable]
-- Stefan Hajnoczi]

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agosavevm: Remove MinGW specific code which is no longer needed
Stefan Weil [Mon, 7 Jan 2013 21:20:27 +0000 (22:20 +0100)]
savevm: Remove MinGW specific code which is no longer needed

QEMU provides a portable function qemu_gettimeofday instead of
gettimeofday and also an implementation of localtime_r for MinGW.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoqga/channel-posix.c: Explicitly include string.h
Peter Maydell [Mon, 7 Jan 2013 17:29:55 +0000 (17:29 +0000)]
qga/channel-posix.c: Explicitly include string.h

Explicitly include string.h to avoid warnings under MacOS X/clang
about implicit declarations of strerror() and strlen().

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoconfigure: Fix comment (copy+paste bug)
Stefan Weil [Sat, 5 Jan 2013 11:17:38 +0000 (12:17 +0100)]
configure: Fix comment (copy+paste bug)

Signed-off-by: Stefan Weil <sw@weilnetz.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoMakefile: install the "acpi-dsdt.aml" and "q35-acpi-dsdt.aml" blobs too
Laszlo Ersek [Tue, 8 Jan 2013 18:52:20 +0000 (19:52 +0100)]
Makefile: install the "acpi-dsdt.aml" and "q35-acpi-dsdt.aml" blobs too

The WARNING message from commit f7e4dd6c made me notice.

Signed-off-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agopc: rename machine types
Gerd Hoffmann [Tue, 8 Jan 2013 07:53:13 +0000 (08:53 +0100)]
pc: rename machine types

Starting with release 1.4 we have a fully functional q35 machine type,
i.e. "qemu -M q35" JustWorks[tm].  Update machine type names to reflect
that:

  * pc-1.4 becomes pc-i440fx-1.4
  * q35-next becomes pc-q35-1.4

The pc-1.3 (+older) names are maintained for compatibility reasons.
For the same reason the "pc" and "q35" aliases are kept.  pc-piix-1.4
continues to be the default machine type, again for compatibility
reasons.

Also updated the description (shown by "qemu -M ?") with host bridge
name, south bridge name and chipset release year.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoq35: document chipset devices
Gerd Hoffmann [Tue, 8 Jan 2013 08:38:19 +0000 (09:38 +0100)]
q35: document chipset devices

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoq35: add ich9 intel hda controller
Gerd Hoffmann [Tue, 8 Jan 2013 08:25:16 +0000 (09:25 +0100)]
q35: add ich9 intel hda controller

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agovirtio-scsi: abort in-flight I/O when the device is reset
Paolo Bonzini [Thu, 10 Jan 2013 14:49:08 +0000 (15:49 +0100)]
virtio-scsi: abort in-flight I/O when the device is reset

When the device is reset, the SCSI bus should also be reset so
that in-flight I/O is cancelled.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoqdev: add qbus_reset_all
Paolo Bonzini [Thu, 10 Jan 2013 14:49:07 +0000 (15:49 +0100)]
qdev: add qbus_reset_all

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMake all static TypeInfos const
Andreas Färber [Thu, 10 Jan 2013 15:19:07 +0000 (16:19 +0100)]
Make all static TypeInfos const

Since 39bffca2030950ef6efe57c2fac8327a45ae1015 (qdev: register all
types natively through QEMU Object Model), TypeInfo as used in
the common, non-iterative pattern is no longer amended with information
and should therefore be const.

Fix the documented QOM examples:

 sed -i 's/static TypeInfo/static const TypeInfo/g' include/qom/object.h

Since frequently the wrong examples are being copied by contributors of
new devices, fix all types in the tree:

 sed -i 's/^static TypeInfo/static const TypeInfo/g' */*.c
 sed -i 's/^static TypeInfo/static const TypeInfo/g' */*/*.c

This also avoids to piggy-back these changes onto real functional
changes or other refactorings.

Signed-off-by: Andreas Färber <afaerber@suse.de>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge branch 'master' of git://git.qemu.org/qemu into prep-up
Andreas Färber [Thu, 10 Jan 2013 20:52:28 +0000 (21:52 +0100)]
Merge branch 'master' of git://git.qemu.org/qemu into prep-up

Conflicts:
hw/Makefile.objs
hw/ppc_prep.c

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
11 years agoMerge remote-tracking branch 'kraxel/build.1' into staging
Anthony Liguori [Thu, 10 Jan 2013 19:26:31 +0000 (13:26 -0600)]
Merge remote-tracking branch 'kraxel/build.1' into staging

* kraxel/build.1:
  m48t59-test: don't touch watchdog
  rtc-test: skip year-2038 overflow check in case time_t is 32bit only

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'awilliam/tags/qemu-1.4-vfio-20130109.0' into staging
Anthony Liguori [Thu, 10 Jan 2013 19:26:12 +0000 (13:26 -0600)]
Merge remote-tracking branch 'awilliam/tags/qemu-1.4-vfio-20130109.0' into staging

vfio-pci: Fixes for qemu 1.4 & stable

* awilliam/tags/qemu-1.4-vfio-20130109.0:
  vfio-pci: Loosen sanity checks to allow future features
  vfio-pci: Make host MSI-X enable track guest

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agom48t59-test: don't touch watchdog
Gerd Hoffmann [Fri, 4 Jan 2013 16:12:18 +0000 (17:12 +0100)]
m48t59-test: don't touch watchdog

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agortc-test: skip year-2038 overflow check in case time_t is 32bit only
Gerd Hoffmann [Fri, 4 Jan 2013 09:37:50 +0000 (10:37 +0100)]
rtc-test: skip year-2038 overflow check in case time_t is 32bit only

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoCheck return values from g_poll and select
Fabien Chouteau [Tue, 8 Jan 2013 15:30:56 +0000 (16:30 +0100)]
Check return values from g_poll and select

The current implementation of os_host_main_loop_wait() on Windows,
returns 1 only when a g_poll() event occurs because the return value of
select() is overridden. This is wrong as we may skip a socket event, as
shown in this example:

1. select() returns 0
2. g_poll() returns 1  (socket event occurs)
3. os_host_main_loop_wait() returns 1
4. qemu_iohandler_poll() sees no socket event because select() has
   return before the event occurs
5. select() returns 1
6. g_poll() returns 0 (g_poll overrides select's return value)
7. os_host_main_loop_wait() returns 0
8. qemu_iohandler_poll() doesn't check for socket events because the
   return value of os_host_main_loop_wait() is zero.
9. goto 5

This patch use one variable for each of these return values, so we don't
miss a select() event anymore.

Also move the call to select() after g_poll(), this will improve latency
as we don't have to go through two os_host_main_loop_wait() calls to
detect a socket event.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Fabien Chouteau <chouteau@adacore.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'mdroth/qga-pull-1-8-2013' into staging
Anthony Liguori [Wed, 9 Jan 2013 15:55:51 +0000 (09:55 -0600)]
Merge remote-tracking branch 'mdroth/qga-pull-1-8-2013' into staging

* mdroth/qga-pull-1-8-2013:
  qemu-ga: sample fsfreeze hooks
  qemu-ga: execute hook to quiesce the guest on fsfreeze-freeze/thaw
  qemu-ga: guest_suspend(): improve error reporting
  qemu-ga: bios_supports_mode(): improve error reporting
  qemu-ga: qmp_guest_network_get_interfaces(): get rid of snprintf() + error_set()
  qemu-ga: qmp_guest_fstrim(): get rid of sprintf() + error_set()
  qemu-ga: qmp_guest_fsfreeze_*(): get rid of sprintf() + error_set()
  qemu-ga: build_fs_mount_list(): take an Error argument
  qemu-ga: qmp_guest_shutdown(): improve error reporting
  qemu-ga: qmp_guest_file_*: improve error reporting
  qemu-ga: qmp_guest_file_close(): fix fclose() error check
  qemu-ga: guest_file_handle_find(): take an Error argument

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoqemu-ga: sample fsfreeze hooks
Tomoki Sekiyama [Wed, 12 Dec 2012 03:55:57 +0000 (12:55 +0900)]
qemu-ga: sample fsfreeze hooks

Adds sample hook scripts for --fsfreeze-hook option of qemu-ga.
  - fsfreeze-hook : execute scripts in fsfreeze-hook.d/
  - fsfreeze-hook.d/mysql-flush.sh.sample : quiesce MySQL before snapshot

Signed-off-by: Tomoki Sekiyama <tomoki.sekiyama.qu@hitachi.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: execute hook to quiesce the guest on fsfreeze-freeze/thaw
Tomoki Sekiyama [Wed, 12 Dec 2012 03:55:55 +0000 (12:55 +0900)]
qemu-ga: execute hook to quiesce the guest on fsfreeze-freeze/thaw

To use the online disk snapshot for online-backup, application-level
consistency of the snapshot image is required. However, currently the
guest agent can provide only filesystem-level consistency, and the
snapshot may contain dirty data, for example, incomplete transactions.
This patch provides the opportunity to quiesce applications before
snapshot is taken.

If --fsfreeze-hook option is specified, the hook is executed with
"freeze" argument before the filesystem is frozen by fsfreeze-freeze
command. As for fsfreeze-thaw command, the hook is executed with "thaw"
argument after the filesystem is thawed.

This patch depends on patchset to improve error reporting by Luiz Capitulino:
  http://lists.gnu.org/archive/html/qemu-devel/2012-11/msg03016.html

Signed-off-by: Tomoki Sekiyama <tomoki.sekiyama.qu@hitachi.com>
Reviewed-by: Luiz Capitulino <lcapitulino@redhat.com>
*clarified usage in help output

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: guest_suspend(): improve error reporting
Luiz Capitulino [Tue, 27 Nov 2012 13:02:04 +0000 (11:02 -0200)]
qemu-ga: guest_suspend(): improve error reporting

Most errors are QERR_UNDEFINED_ERROR today.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: bios_supports_mode(): improve error reporting
Luiz Capitulino [Tue, 27 Nov 2012 13:02:03 +0000 (11:02 -0200)]
qemu-ga: bios_supports_mode(): improve error reporting

Most errors are QERR_UNDEFINED_ERROR today.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_network_get_interfaces(): get rid of snprintf() + error_set()
Luiz Capitulino [Tue, 27 Nov 2012 13:02:02 +0000 (11:02 -0200)]
qemu-ga: qmp_guest_network_get_interfaces(): get rid of snprintf() + error_set()

Convert them to error_setg_errno().

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_fstrim(): get rid of sprintf() + error_set()
Luiz Capitulino [Tue, 27 Nov 2012 13:02:01 +0000 (11:02 -0200)]
qemu-ga: qmp_guest_fstrim(): get rid of sprintf() + error_set()

Convert them to error_setg_errno().

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_fsfreeze_*(): get rid of sprintf() + error_set()
Luiz Capitulino [Tue, 27 Nov 2012 13:02:00 +0000 (11:02 -0200)]
qemu-ga: qmp_guest_fsfreeze_*(): get rid of sprintf() + error_set()

Convert them to error_setg_errno().

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: build_fs_mount_list(): take an Error argument
Luiz Capitulino [Thu, 29 Nov 2012 17:29:11 +0000 (15:29 -0200)]
qemu-ga: build_fs_mount_list(): take an Error argument

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_shutdown(): improve error reporting
Luiz Capitulino [Tue, 27 Nov 2012 13:01:58 +0000 (11:01 -0200)]
qemu-ga: qmp_guest_shutdown(): improve error reporting

Most errors are QERR_UNDEFINED_ERROR. Also, adds ga_wait_child() as
a future commit will use it too.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_file_*: improve error reporting
Luiz Capitulino [Tue, 27 Nov 2012 13:01:57 +0000 (11:01 -0200)]
qemu-ga: qmp_guest_file_*: improve error reporting

Use error_setg_errno() when possible with an improved error description.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: qmp_guest_file_close(): fix fclose() error check
Luiz Capitulino [Tue, 27 Nov 2012 13:01:56 +0000 (11:01 -0200)]
qemu-ga: qmp_guest_file_close(): fix fclose() error check

fclose() returns EOF on error.

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoqemu-ga: guest_file_handle_find(): take an Error argument
Luiz Capitulino [Tue, 27 Nov 2012 13:01:55 +0000 (11:01 -0200)]
qemu-ga: guest_file_handle_find(): take an Error argument

Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
*Fixed missing space character in error message

Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
11 years agoMerge remote-tracking branch 'afaerber/qom-cpu' into staging
Anthony Liguori [Tue, 8 Jan 2013 21:15:57 +0000 (15:15 -0600)]
Merge remote-tracking branch 'afaerber/qom-cpu' into staging

* afaerber/qom-cpu:
  target-i386: Explicitly set vendor for each built-in cpudef
  target-i386: Sanitize AMD's ext2_features at realize time
  target-i386: Filter out unsupported features at realize time
  qemu-common.h: Make qemu_init_vcpu() stub static inline
  target-i386: check/enforce: Eliminate check_feat field
  target-i386: check/enforce: Check SVM flag support as well
  target-i386: check/enforce: Check all CPUID.80000001H.EDX bits
  target-i386: check/enforce: Do not ignore "hypervisor" flag
  target-i386: check/enforce: Fix CPUID leaf numbers on error messages
  target-i386: kvm: Enable all supported KVM features for -cpu host
  target-i386: kvm: -cpu host: Use GET_SUPPORTED_CPUID for SVM features
  cpu: Change parent type to Device
  qdev: Don't assume existence of parent bus on unparenting
  qdev: Include qdev code into *-user, too
  libqemustub: sysbus_get_default() stub
  libqemustub: vmstate register/unregister stubs
  libqemustub: Add qemu_[un]register_reset() stubs

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agovfio-pci: Loosen sanity checks to allow future features
Alex Williamson [Tue, 8 Jan 2013 21:10:03 +0000 (14:10 -0700)]
vfio-pci: Loosen sanity checks to allow future features

VFIO_PCI_NUM_REGIONS and VFIO_PCI_NUM_IRQS should never have been
used in this manner as it locks a specific kernel implementation.
Future features may introduce new regions or interrupt entries
(VGA may add legacy ranges, AER might add an IRQ for error
signalling).  Fix this before it gets us into trouble.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Cc: qemu-stable@nongnu.org
11 years agovfio-pci: Make host MSI-X enable track guest
Alex Williamson [Tue, 8 Jan 2013 21:09:03 +0000 (14:09 -0700)]
vfio-pci: Make host MSI-X enable track guest

Guests typically enable MSI-X with all of the vectors in the MSI-X
vector table masked.  Only when the vector is enabled does the vector
get unmasked, resulting in a vector_use callback.  These two points,
enable and unmask, correspond to pci_enable_msix() and request_irq()
for Linux guests.  Some drivers rely on VF/PF or PF/fw communication
channels that expect the physical state of the device to match the
guest visible state of the device.  They don't appreciate lazily
enabling MSI-X on the physical device.

To solve this, enable MSI-X with a single vector when the MSI-X
capability is enabled and immediate disable the vector.  This leaves
the physical device in exactly the same state between host and guest.
Furthermore, the brief gap where we enable vector 0, it fires into
userspace, not KVM, so the guest doesn't get spurious interrupts.
Ideally we could call VFIO_DEVICE_SET_IRQS with the right parameters
to enable MSI-X with zero vectors, but this will currently return an
error as the Linux MSI-X interfaces do not allow it.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
Cc: qemu-stable@nongnu.org
11 years agotarget-i386: Explicitly set vendor for each built-in cpudef
Igor Mammedov [Fri, 28 Dec 2012 20:01:18 +0000 (21:01 +0100)]
target-i386: Explicitly set vendor for each built-in cpudef

Since cpudef config is not supported anymore and all remaining sources
now always set x86_def_t.vendor[123] fields, remove setting default
vendor to simplify future re-factoring.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: Sanitize AMD's ext2_features at realize time
Igor Mammedov [Fri, 28 Dec 2012 20:01:17 +0000 (21:01 +0100)]
target-i386: Sanitize AMD's ext2_features at realize time

When CPU properties are implemented, ext2_features may change
between object_new(CPU) and cpu_realize_fn(). Sanitizing
ext2_features for AMD based CPU at realize() time will keep
current behavior after CPU features are converted to properties.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: Filter out unsupported features at realize time
Igor Mammedov [Fri, 28 Dec 2012 20:01:16 +0000 (21:01 +0100)]
target-i386: Filter out unsupported features at realize time

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agoqemu-common.h: Make qemu_init_vcpu() stub static inline
Andreas Färber [Sat, 5 Jan 2013 13:44:08 +0000 (14:44 +0100)]
qemu-common.h: Make qemu_init_vcpu() stub static inline

Turn the *-user macro into a no-op inline function to avoid
unused-variable warnings and band-aiding #ifdef'ery.

This allows to drop an #ifdef for alpha and avoids more for unicore32
and other upcoming trivial realizefn implementations.

Suggested-by: Lluís Vilanova <vilanova@ac.upc.edu>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: Eduardo Habkost <ehabkost@redhat.com>
11 years agotarget-i386: check/enforce: Eliminate check_feat field
Eduardo Habkost [Fri, 4 Jan 2013 22:01:10 +0000 (20:01 -0200)]
target-i386: check/enforce: Eliminate check_feat field

Now that all entries have check_feat=~0 in
kvm_check_features_against_host(), we can eliminate check_feat entirely
and make the code check all bits.

This patch shouldn't introduce any behavior change, as check_feat is set
to ~0 on all entries.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: check/enforce: Check SVM flag support as well
Eduardo Habkost [Fri, 4 Jan 2013 22:01:09 +0000 (20:01 -0200)]
target-i386: check/enforce: Check SVM flag support as well

When nested SVM is supported, the kernel returns the SVM flag on
GET_SUPPORTED_CPUID[1], so we can check the SVM flag safely in
kvm_check_features_against_host().

I don't know why the original code ignored the SVM flag. Maybe it was
because kvm_cpu_fill_host() used the CPUID instruction directly instead
of GET_SUPPORTED_CPUID

[1] Older kernels (before v2.6.37) returned the SVM flag even if nested
    SVM was _not_ supported. So the only cases where this patch should
    change behavior is when SVM is being requested by the user or the
    CPU model, but not supported by the host. And on these cases we
    really want QEMU to abort if the "enforce" option is set.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: check/enforce: Check all CPUID.80000001H.EDX bits
Eduardo Habkost [Fri, 4 Jan 2013 22:01:08 +0000 (20:01 -0200)]
target-i386: check/enforce: Check all CPUID.80000001H.EDX bits

I have no idea why PPRO_FEATURES was being ignored on the check of the
CPUID.80000001H.EDX bits. I believe it was a mistake, and it was
supposed to be ~(PPRO_FEATURES & CPUID_EXT2_AMD_ALIASES) or just
~CPUID_EXT2_AMD_ALIASES, because some time ago kvm_cpu_fill_host() used
the CPUID instruction directly (instead of
kvm_arch_get_supported_cpuid()).

But now kvm_cpu_fill_host() uses kvm_arch_get_supported_cpuid(), and
kvm_arch_get_supported_cpuid() returns all supported bits for
CPUID.80000001H.EDX, even the AMD aliases (that are explicitly copied
from CPUID.01H.EDX), so we can make the code check/enforce all the
CPUID.80000001H.EDX bits.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: check/enforce: Do not ignore "hypervisor" flag
Eduardo Habkost [Fri, 4 Jan 2013 22:01:07 +0000 (20:01 -0200)]
target-i386: check/enforce: Do not ignore "hypervisor" flag

We don't need any hack to ignore CPUID_EXT_HYPERVISOR anymore, because
kvm_arch_get_supported_cpuid() now sets CPUID_EXT_HYPERVISOR properly.
So, this shouldn't introduce any behavior change, but it makes the code
simpler.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: check/enforce: Fix CPUID leaf numbers on error messages
Eduardo Habkost [Fri, 4 Jan 2013 22:01:06 +0000 (20:01 -0200)]
target-i386: check/enforce: Fix CPUID leaf numbers on error messages

The -cpu check/enforce warnings are printing incorrect information about the
missing flags. There are no feature flags on CPUID leaves 0 and 0x80000000, but
there were references to 0 and 0x80000000 in the table at
kvm_check_features_against_host().

This changes the model_features_t struct to contain the register number as
well, so the error messages print the correct CPUID leaf+register information,
instead of wrong CPUID leaf numbers.

This also changes the format of the error messages, so they follow the
"CPUID.<leaf>.<register>.<name> [bit <offset>]" convention used in Intel
documentation. Example output:

    $ qemu-system-x86_64 -machine pc-1.0,accel=kvm -cpu Opteron_G4,+ia64,enforce
    warning: host doesn't support requested feature: CPUID.01H:EDX.ia64 [bit 30]
    warning: host doesn't support requested feature: CPUID.01H:ECX.xsave [bit 26]
    warning: host doesn't support requested feature: CPUID.01H:ECX.avx [bit 28]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.abm [bit 5]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.sse4a [bit 6]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.misalignsse [bit 7]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.3dnowprefetch [bit 8]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.xop [bit 11]
    warning: host doesn't support requested feature: CPUID.80000001H:ECX.fma4 [bit 16]
    Unable to find x86 CPU definition
    $

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: kvm: Enable all supported KVM features for -cpu host
Eduardo Habkost [Fri, 4 Jan 2013 22:01:05 +0000 (20:01 -0200)]
target-i386: kvm: Enable all supported KVM features for -cpu host

When using -cpu host, we don't need to use the kvm_default_features
variable, as the user is explicitly asking QEMU to enable all feature
supported by the host.

This changes the kvm_cpu_fill_host() code to use GET_SUPPORTED_CPUID to
initialize the kvm_features field, so we get all host KVM features
enabled.

This will also allow us to properly check/enforce KVM features inside
kvm_check_features_against_host() later. For example, we will be able to
make this:

  $ qemu-system-x86_64 -cpu ...,+kvm_pv_eoi,enforce

refuse to start if kvm_pv_eoi is not supported by the host (after we fix
kvm_check_features_against_host() to check KVM flags as well).

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agotarget-i386: kvm: -cpu host: Use GET_SUPPORTED_CPUID for SVM features
Eduardo Habkost [Fri, 4 Jan 2013 22:01:04 +0000 (20:01 -0200)]
target-i386: kvm: -cpu host: Use GET_SUPPORTED_CPUID for SVM features

The existing -cpu host code simply sets every bit inside svm_features
(initializing it to -1), and that makes it impossible to make the
enforce/check options work properly when the user asks for SVM features
explicitly in the command-line.

So, instead of initializing svm_features to -1, use GET_SUPPORTED_CPUID
to fill only the bits that are supported by the host (just like we do
for all other CPUID feature words inside kvm_cpu_fill_host()).

This will keep the existing behavior (as filter_features_for_kvm()
already uses GET_SUPPORTED_CPUID to filter svm_features), but will allow
us to properly check for KVM features inside
kvm_check_features_against_host() later.

For example, we will be able to make this:

  $ qemu-system-x86_64 -cpu ...,+pfthreshold,enforce

refuse to start if the SVM "pfthreshold" feature is not supported by the
host (after we fix kvm_check_features_against_host() to check SVM flags
as well).

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agocpu: Change parent type to Device
Eduardo Habkost [Wed, 5 Dec 2012 16:49:13 +0000 (14:49 -0200)]
cpu: Change parent type to Device

This finally makes the CPU class a subclass of the Device class,
allowing us to start using DeviceState properties on CPU subclasses.

It has no_user=1, as creating CPUs using -device doesn't work yet.

Signed-off-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agoqdev: Don't assume existence of parent bus on unparenting
Andreas Färber [Fri, 4 Jan 2013 17:13:00 +0000 (18:13 +0100)]
qdev: Don't assume existence of parent bus on unparenting

Commit 667d22d1ae59da46b4c1fbd094ca61145f19b8c3 (qdev: move bus removal
to object_unparent) made the assumption that at unparenting time
parent_bus is not NULL. This assumption is unjustified since
object_unparent() may well be called directly after object_initialize(),
without any qdev_set_parent_bus().

This did not cause any issues yet because qdev_[try_]create() does call
qdev_set_parent_bus(), falling back to SysBus if unsupplied.

While at it, ensure that this new function uses the device_ prefix and
make the name more neutral in light of this semantic change.

Reported-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
Tested-by: Igor Mammedov <imammedo@redhat.com>
11 years agoqdev: Include qdev code into *-user, too
Eduardo Habkost [Wed, 5 Dec 2012 16:49:12 +0000 (14:49 -0200)]
qdev: Include qdev code into *-user, too

The code depends on some functions from qemu-option.o, so add
qemu-option.o to universal-obj-y to make sure it's included.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agolibqemustub: sysbus_get_default() stub
Eduardo Habkost [Wed, 5 Dec 2012 16:49:09 +0000 (14:49 -0200)]
libqemustub: sysbus_get_default() stub

The stub will be used on cases where sysbus.c is not compiled in (e.g.
*-user).

Note that code that uses NULL as the bus with qdev{_try,}_create()
implicitly uses sysbus_get_default() as the bus, and will still require
sysbus.c to be compiled in.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agolibqemustub: vmstate register/unregister stubs
Eduardo Habkost [Wed, 5 Dec 2012 16:49:08 +0000 (14:49 -0200)]
libqemustub: vmstate register/unregister stubs

Add vmstate stub functions, so that qdev.o can be used without savevm.o
when vmstate support is not necessary (i.e. by *-user).

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agolibqemustub: Add qemu_[un]register_reset() stubs
Eduardo Habkost [Wed, 5 Dec 2012 16:49:07 +0000 (14:49 -0200)]
libqemustub: Add qemu_[un]register_reset() stubs

This will be useful for code that don't call qemu_devices_reset() (e.g.
*-user). If qemu_devices_reset() is never called, it means we don't need
to keep track of the reset handler list.

Signed-off-by: Eduardo Habkost <ehabkost@redhat.com>
Signed-off-by: Andreas Färber <afaerber@suse.de>
11 years agoMerge remote-tracking branch 'kraxel/usb.75' into staging
Anthony Liguori [Tue, 8 Jan 2013 16:36:20 +0000 (10:36 -0600)]
Merge remote-tracking branch 'kraxel/usb.75' into staging

* kraxel/usb.75: (32 commits)
  uhci: stop using portio lists
  usbredir: Add support for buffered bulk input (v2)
  exynos4210: Add EHCI support
  usb/ehci: Add SysBus EHCI device for Exynos4210
  usb/ehci: Move capsbase and opregbase into SysBus EHCI class
  usb/ehci: Clean up SysBus and PCI EHCI split
  xhci: call set-address with dummy usbpacket
  usb-redir: Add debugging to bufpq save / restore
  usbredir: Add usbredir_init_endpoints() helper
  usbredir: Verify we have 32 bits bulk length cap when redirecting to xhci
  usbredir: Add ep_stopped USBDevice method
  usbredir: Add USBEP2I and I2USBEP helper macros
  usbredir: Add an usbredir_stop_ep helper function
  usb: Add an usb_device_ep_stopped USBDevice method
  usb: Fix usb_ep_find_packet_by_id
  hid: Change idle handling to use a timer
  uhci: Maximize how many frames we catch up when behind
  uhci: Limit amount of frames processed in one go
  uhci: Add a QH_VALID define
  uhci: Fix pending interrupts getting lost on migration
  ...

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agoMerge remote-tracking branch 'stefanha/net' into staging
Anthony Liguori [Tue, 8 Jan 2013 16:36:13 +0000 (10:36 -0600)]
Merge remote-tracking branch 'stefanha/net' into staging

* stefanha/net:
  rtl8139: preserve link state across device reset
  e1000: no need auto-negotiation if link was down
  net: clean up network at qemu process termination
  e1000: Discard oversized packets based on SBP|LPE

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
11 years agotarget-mips: Fix helper and tests for dot/cross-dot product instructions
Petar Jovanovic [Wed, 2 Jan 2013 04:08:48 +0000 (05:08 +0100)]
target-mips: Fix helper and tests for dot/cross-dot product instructions

Helper function for dpa_w_ph, dpax_w_ph, dps_w_ph and dpsx_w_ph incorrectly
defines halfword vector elements as unsigned values. This results in wrong
output which is not triggered in the tests as they also follow this logic.

Signed-off-by: Petar Jovanovic <petarj@mips.com>
Reviewed-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Replace macros by inline functions
Stefan Weil [Tue, 1 Jan 2013 18:44:31 +0000 (19:44 +0100)]
target-mips: Replace macros by inline functions

The macros RESTORE_ROUNDING_MODE and RESTORE_FLUSH_MODE silently used
variable env from their callers. Using inline functions with env passed
as a function argument is more transparent.

This modification was proposed by Peter Maydell.

Cc: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agotarget-mips: Allow DSP access to be disabled once enabled.
Eric Johnson [Tue, 8 Jan 2013 06:26:44 +0000 (22:26 -0800)]
target-mips: Allow DSP access to be disabled once enabled.

Clear the DSP hflags at the start of compute_hflags.  Otherwise access
is not properly disabled once enabled.

Signed-off-by: Eric Johnson <ericj@mips.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agolinux-user: fix mips 32-on-64 prealloc case
Alexander Graf [Thu, 3 Jan 2013 13:17:18 +0000 (14:17 +0100)]
linux-user: fix mips 32-on-64 prealloc case

MIPS only supports 31 bits of virtual address space for user space, so let's
make sure we stay within that limit with our preallocated memory block.

This fixes the MIPS user space targets when executed without command line
option.

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
11 years agouhci: stop using portio lists
Gerd Hoffmann [Thu, 3 Jan 2013 11:29:41 +0000 (12:29 +0100)]
uhci: stop using portio lists

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Add support for buffered bulk input (v2)
Hans de Goede [Wed, 19 Dec 2012 14:08:33 +0000 (15:08 +0100)]
usbredir: Add support for buffered bulk input (v2)

Buffered bulk mode is intended for bulk *input* endpoints, where the data is
of a streaming nature (not part of a command-response protocol). These
endpoints' input buffer may overflow if data is not read quickly enough.
So in buffered bulk mode the usb-host takes care of the submitting and
re-submitting of bulk transfers.

Buffered bulk mode is necessary for reliable operation with the bulk in
endpoints of usb to serial convertors. Unfortunatelty buffered bulk input
mode will only work with certain devices, therefor this patch also adds a
usb-id table to enable it for devices which need it, while leaving the
bulk ep handling for other devices unmodified.

Note that the bumping of the required usbredir from 0.5.3 to 0.6 does
not mean that we will now need a newer usbredir release then qemu-1.3,
.pc files reporting 0.5.3 have only ever existed in usbredir builds directly
from git, so qemu-1.3 needs the 0.6 release too.

Changes in v2:
-Split of quirk handling into quirks.c

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
11 years agoreadline: avoid memcpy() of overlapping regions
Nickolai Zeldovich [Mon, 7 Jan 2013 20:38:39 +0000 (15:38 -0500)]
readline: avoid memcpy() of overlapping regions

memcpy() for overlapping regions is undefined behavior; use memmove()
instead in readline_hist_add().

[Keep tab characters since surrounding code still uses them -- Stefan]

Signed-off-by: Nickolai Zeldovich <nickolai@csail.mit.edu>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
11 years agoexynos4210: Add EHCI support
Liming Wang [Sun, 16 Dec 2012 03:49:46 +0000 (04:49 +0100)]
exynos4210: Add EHCI support

Add EHCI USB host controller to exynos4210.

Signed-off-by: Liming Wang <walimisdev@gmail.com>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Reviewed-by: Igor Mitsyanko <i.mitsyanko@samsung.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb/ehci: Add SysBus EHCI device for Exynos4210
Andreas Färber [Sun, 16 Dec 2012 03:49:45 +0000 (04:49 +0100)]
usb/ehci: Add SysBus EHCI device for Exynos4210

It uses a different capsbase and opregbase than the Xilinx device.

Signed-off-by: Liming Wang <walimisdev@gmail.com>
Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Cc: Igor Mitsyanko <i.mitsyanko@samsung.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb/ehci: Move capsbase and opregbase into SysBus EHCI class
Andreas Färber [Sun, 16 Dec 2012 03:49:44 +0000 (04:49 +0100)]
usb/ehci: Move capsbase and opregbase into SysBus EHCI class

This allows specific derived models to use different values.

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb/ehci: Clean up SysBus and PCI EHCI split
Andreas Färber [Sun, 16 Dec 2012 03:49:43 +0000 (04:49 +0100)]
usb/ehci: Clean up SysBus and PCI EHCI split

SysBus EHCI was introduced in a hurry before 1.3 Soft Freeze.
To use QOM casts in place of DO_UPCAST() / FROM_SYSBUS(), we need an
identifying type. Introduce generic abstract base types for PCI and
SysBus EHCI to allow multiple types to access the shared fields.

While at it, move the state structs being amended with macros to the
header file so that they can be embedded.

The VMSTATE_PCI_DEVICE() macro does not play nice with the QOM
parent_obj naming convention, so defer that cleanup.

Signed-off-by: Andreas Färber <andreas.faerber@web.de>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoxhci: call set-address with dummy usbpacket
Gerd Hoffmann [Fri, 14 Dec 2012 12:10:39 +0000 (13:10 +0100)]
xhci: call set-address with dummy usbpacket

Due to the way devices are addressed with xhci (done by hardware, not
the guest os) there is no packet when invoking the set-address control
request.  Create a dummy packet in that case to avoid null pointer
dereferences.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb-redir: Add debugging to bufpq save / restore
Hans de Goede [Fri, 14 Dec 2012 13:35:46 +0000 (14:35 +0100)]
usb-redir: Add debugging to bufpq save / restore

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Add usbredir_init_endpoints() helper
Hans de Goede [Fri, 14 Dec 2012 13:35:45 +0000 (14:35 +0100)]
usbredir: Add usbredir_init_endpoints() helper

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Verify we have 32 bits bulk length cap when redirecting to xhci
Hans de Goede [Fri, 14 Dec 2012 13:35:44 +0000 (14:35 +0100)]
usbredir: Verify we have 32 bits bulk length cap when redirecting to xhci

The xhci-hcd may submit bulk transfers > 65535 bytes even when not using
bulk-in pipeling, so usbredir can only be used in combination with an xhci
hcd if the client has the 32 bits bulk length capability.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Add ep_stopped USBDevice method
Hans de Goede [Fri, 14 Dec 2012 13:35:43 +0000 (14:35 +0100)]
usbredir: Add ep_stopped USBDevice method

To ensure that interrupt receiving is properly stopped when the guest is
no longer interested in an interrupt endpoint.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Add USBEP2I and I2USBEP helper macros
Hans de Goede [Fri, 14 Dec 2012 13:35:42 +0000 (14:35 +0100)]
usbredir: Add USBEP2I and I2USBEP helper macros

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousbredir: Add an usbredir_stop_ep helper function
Hans de Goede [Fri, 14 Dec 2012 13:35:41 +0000 (14:35 +0100)]
usbredir: Add an usbredir_stop_ep helper function

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb: Add an usb_device_ep_stopped USBDevice method
Hans de Goede [Fri, 14 Dec 2012 13:35:40 +0000 (14:35 +0100)]
usb: Add an usb_device_ep_stopped USBDevice method

Some usb devices (host or network redirection) can benefit from knowing when
the guest stops using an endpoint. Redirection may involve submitting packets
independently from the guest (in combination with a fifo buffer between the
redirection code and the guest), to ensure that buffers of the real usb device
are timely emptied. This is done for example for isoc traffic and for interrupt
input endpoints. But when the (re)submission of packets is done by the device
code, then how does it know when to stop this?

For isoc endpoints this is handled by detecting a set interface (change alt
setting) command, which works well for isoc endpoints. But for interrupt
endpoints currently the redirection code never stops receiving data from
the device, which is less then ideal.

However the controller emulation is aware when a guest looses interest, as
then the qh for the endpoint gets unlinked (ehci, ohci, uhci) or the endpoint
is explicitly stopped (xhci). This patch adds a new ep_stopped USBDevice
method and modifies the hcd code to call this on queue unlink / ep stop.

This makes it possible for the redirection code to properly stop receiving
interrupt input (*) data when the guest no longer has interest in it.

*) And in the future also buffered bulk input.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agousb: Fix usb_ep_find_packet_by_id
Hans de Goede [Fri, 14 Dec 2012 13:35:39 +0000 (14:35 +0100)]
usb: Fix usb_ep_find_packet_by_id

usb_ep_find_packet_by_id mistakenly only checks the first packet and if that
is not a match, keeps trying the first packet! This patch fixes this.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agohid: Change idle handling to use a timer
Hans de Goede [Fri, 14 Dec 2012 13:35:38 +0000 (14:35 +0100)]
hid: Change idle handling to use a timer

This leads to cleaner code in usb-hid, and removes up to a 1000 calls / sec to
qemu_get_clock_ns(vm_clock) if idle-time is set to its default value of 0.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agouhci: Maximize how many frames we catch up when behind
Hans de Goede [Fri, 14 Dec 2012 13:35:37 +0000 (14:35 +0100)]
uhci: Maximize how many frames we catch up when behind

If somehow we've gotten behind a lot, simply skip ahead, like the ehci code
does.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agouhci: Limit amount of frames processed in one go
Hans de Goede [Fri, 14 Dec 2012 13:35:36 +0000 (14:35 +0100)]
uhci: Limit amount of frames processed in one go

Before this patch uhci would process an unlimited amount of frames when
behind on schedule, by setting the timer to a time already past, causing the
timer subsys to immediately recall the frame_timer function gain.

This would cause invalid cancellations of bulk queues when the catching up
processed more then 32 frames at a moment when the bulk qh was temporarily
unlinked (which the Linux uhci driver does).

This patch fixes this by processing maximum 16 frames in one go, and always
setting the timer one ms later, making the code behave more like the ehci
code.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agouhci: Add a QH_VALID define
Hans de Goede [Fri, 14 Dec 2012 13:35:35 +0000 (14:35 +0100)]
uhci: Add a QH_VALID define

Rather then using the magic 32 value in various places.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agouhci: Fix pending interrupts getting lost on migration
Hans de Goede [Fri, 14 Dec 2012 13:35:34 +0000 (14:35 +0100)]
uhci: Fix pending interrupts getting lost on migration

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agouhci: Fix 1 ms delay in interrupt reporting to the guest
Hans de Goede [Fri, 14 Dec 2012 13:35:33 +0000 (14:35 +0100)]
uhci: Fix 1 ms delay in interrupt reporting to the guest

Re-arrange how we process frames / increase frnum / report pending interrupts,
to avoid a 1 ms delay in interrupt reporting to the guest. This increases
the packet throughput for cases where the guest submits a single packet,
then waits for its completion then re-submits from 500 pkts / sec to
1000 pkts / sec. This impacts for example the use of redirected / virtual
usb to serial convertors.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Don't call commit_irq after raising PCD
Hans de Goede [Fri, 14 Dec 2012 13:35:32 +0000 (14:35 +0100)]
ehci: Don't call commit_irq after raising PCD

ehci_raise_irq(s, USBSTS_PCD), gets applied immediately so there is no need
to call commit_irq after it.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Further speedup rescanning if async schedule after raising an interrupt
Hans de Goede [Fri, 14 Dec 2012 13:35:31 +0000 (14:35 +0100)]
ehci: Further speedup rescanning if async schedule after raising an interrupt

I tried lowering the time between raising an interrupt and rescanning the
async schedule to see if the guest has queued a new transfer before, but
that did not have any positive effect. I now believe the cause for this is
that lowering this time made it more likely to hit the 1 ms interrupt
threshold penalty for the next packet, as described in my
"ehci: Use uframe precision for interrupt threshold checking" commit.

Now that we do interrupt threshold handling with uframe precision, futher
lowering this time from .5 to .25 ms gives an extra 15% improvement in speed
(MB/s) reading from a simple USB-2.0 thumb-drive.

While at it also properly set the int_req_by_async flag for short packet
completions.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Use uframe precision for interrupt threshold checking (v2)
Hans de Goede [Tue, 18 Dec 2012 13:17:02 +0000 (14:17 +0100)]
ehci: Use uframe precision for interrupt threshold checking (v2)

Before this patch, the following could happen:
1) Transfer completes, raises interrupt
2) .5 ms later we check if the guest has queued up any new transfers
3) We find and execute a new transfer
4) .2 ms later the new transfer completes
5) We re-run our frame_timer to write back the completion, but less then
   1 ms has passed since our last run, so frindex is not changed, so the
   interrupt threshold code delays the interrupt
6) 1 ms from the re-run our frame-timer runs again and finally delivers
   the interrupt

This leads to unnecessary large delays of interrupts, this code fixes this
by changing frindex to uframe precision and using that for interrupt threshold
control, making the interrupt fire at step 5 for guest which have low interrupt
threshold settings (like Linux).

Note that the guest still sees the frindex move in steps of 8 for migration
compatibility.

This boosts Linux read speed of a simple cheap USB thumb drive by 6 %.

Changes in v2:
-Make the guest see frindex move in steps of 8 by modifying ehci_opreg_read,
 rather then using a shadow variable

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Verify a queue's ep direction does not change
Hans de Goede [Fri, 14 Dec 2012 13:35:29 +0000 (14:35 +0100)]
ehci: Verify a queue's ep direction does not change

ehci_fill_queue assumes that there is a one on one relationship between an ep
and a qh, this patch adds a check to ensure this.

Note I don't expect this to ever trigger, this is just something I noticed
the guest might do while working on other stuff. The only way this check can
trigger is if a guest mixes in and out qtd-s in a single qh for a non
control ep.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Add an ehci_get_pid helper function
Hans de Goede [Fri, 14 Dec 2012 13:35:28 +0000 (14:35 +0100)]
ehci: Add an ehci_get_pid helper function

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
11 years agoehci: Verify qtd for async completed packets
Hans de Goede [Fri, 14 Dec 2012 13:35:27 +0000 (14:35 +0100)]
ehci: Verify qtd for async completed packets

Remove the short-circuiting of fetchqtd in fetchqh, so that the
qtd gets properly verified before completing the transaction.

Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>