QEMU tracks whether a vcpu is halted using CPUState::halted. E.g.,
after initialization or reset, halted is 0 for the BSP (vcpu 0)
and 1 for the APs (vcpu 1, 2, ...). A halted vcpu should not be
handed to the hypervisor to run (e.g. hax_vcpu_run()).
Under HAXM, Android Emulator sometimes boots into a "vcpu shutdown
request" error while executing in SeaBIOS, with the HAXM driver
logging a guest triple fault in vcpu 1, 2, ... at RIP 0x3. That is
ultimately because the HAX accelerator asks HAXM to run those APs
when they are still in the halted state.
Normally, the vcpu thread for an AP will start by looping in
qemu_wait_io_event(), until the BSP kicks it via a pair of IPIs
(INIT followed by SIPI). But because the HAX accelerator does not
honor cpu->halted, it allows the AP vcpu thread to proceed to
hax_vcpu_run() as soon as it receives any kick, even if the kick
does not come from the BSP. It turns out that emulator has a
worker thread which periodically kicks every vcpu thread (possibly
to collect CPU usage data), and if one of these kicks comes before
those by the BSP, the AP will start execution from the wrong RIP,
resulting in the aforementioned SMP boot failure.
The solution is inspired by the KVM accelerator (credit to
Chuanxiao Dong <chuanxiao.dong@intel.com> for the pointer):
1. Get rid of questionable logic that unconditionally resets
cpu->halted before hax_vcpu_run(). Instead, only reset it at the
right moments (there are only a few "unhalt" events).
2. Add a check for cpu->halted before hax_vcpu_run().
Note that although the non-Unrestricted Guest (!ug_platform) code
path also forcibly resets cpu->halted, it is left untouched,
because only the UG code path supports SMP guests.
The patch is first merged to android emulator with Change-Id:
I9c5752cc737fd305d7eace1768ea12a07309d716
Cc: Yu Ning <yu.ning@intel.com>
Cc: Chuanxiao Dong <chuanxiao.dong@intel.com>
Signed-off-by: Colin Xu <colin.xu@intel.com>
Message-Id: <20190610021939.13669-1-colin.xu@intel.com>
Now that we have both ArchCPU and CPUArchState, we can define
this generically instead of via macro in each target's cpu.h.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Acked-by: Alistair Francis <alistair.francis@wdc.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
CPUClass method dump_statistics() takes an fprintf()-like callback and
a FILE * to pass to it. Most callers pass fprintf() and stderr.
log_cpu_state() passes fprintf() and qemu_log_file.
hmp_info_registers() passes monitor_fprintf() and the current monitor
cast to FILE *. monitor_fprintf() casts it right back, and is
otherwise identical to monitor_printf().
The callback gets passed around a lot, which is tiresome. The
type-punning around monitor_fprintf() is ugly.
Drop the callback, and call qemu_fprintf() instead. Also gets rid of
the type-punning, since qemu_fprintf() takes NULL instead of the
current monitor cast to FILE *.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20190417191805.28198-15-armbru@redhat.com>
And also the g_malloc doesn't need check return value,
remove it.
Cc: qemu-trivial@nongnu.org
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Li Qiang <liq3ea@gmail.com>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <1543226179-5135-1-git-send-email-liq3ea@gmail.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Since HAX_VM_IOCTL_ALLOC_RAM takes a 32-bit size, it cannot handle
RAM blocks of 4GB or larger, which is why HAXM can only run guests
with less than 4GB of RAM. Solve this problem by utilizing the new
HAXM API, HAX_VM_IOCTL_ADD_RAMBLOCK, which takes a 64-bit size, to
register RAM blocks with the HAXM kernel module. The new API is
first added in HAXM 7.0.0, and its availablility and be confirmed
by the presence of the HAX_CAP_64BIT_RAMBLOCK capability flag.
When the guest RAM size reaches 7GB, QEMU will ask HAXM to set up a
memory mapping that covers a 4GB region, which will fail, because
HAX_VM_IOCTL_SET_RAM also takes a 32-bit size. Work around this
limitation by splitting the large mapping into small ones and
calling HAX_VM_IOCTL_SET_RAM multiple times.
Bug: https://bugs.launchpad.net/qemu/+bug/1735576
Signed-off-by: Yu Ning <yu.ning@intel.com>
Message-Id: <1515752555-12784-1-git-send-email-yu.ning@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
System headers should be included with <...>, our own headers with
"...". Offenders tracked down with an ugly, brittle and probably
buggy Perl script. Previous iteration was commit a9c94277f0.
Delete inclusions of "string.h" and "strings.h" instead of fixing them
to <string.h> and <strings.h>, because we always include these via
osdep.h.
Put the cleaned up system header includes first.
While there, separate #include from file comment with exactly one
blank line.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20180201111846.21846-2-armbru@redhat.com>
x86_update_hflags reference env->efer which is updated in hax_get_msrs,
so it has to be called after hax_get_msrs. This fix the bug that sometimes
dump_state show 32 bits regs even in 64 bits mode.
Signed-off-by: Tao Wu <lepton@google.com>
Message-Id: <20180110195056.85403-3-lepton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Change to use x86_update_hflags instead of keeping another copy
at hax side. This also fix bug like HF_CPL_MASK should be SS.DPL,
not CS.DPL.
Signed-off-by: Tao Wu <lepton@google.com>
Message-Id: <20180110195056.85403-2-lepton@google.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
This patch simply replaces the separate boolean field in CPUState that
kvm, hax (and upcoming hvf) have for keeping track of vcpu dirtiness
with a single shared field.
Signed-off-by: Sergio Andres Gomez Del Real <Sergio.G.DelReal@gmail.com>
Message-Id: <20170618191101.3457-1-Sergio.G.DelReal@gmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
While at it, drop the current_cpu assignment since this is a
per-thread variable on modern QEMU.
Cc: Vincent Palatin <vpalatin@chromium.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
As a rule, CPU internal state should never be updated when
!cpu->kvm_vcpu_dirty (or the HAX equivalent). If that is done, then
subsequent calls to cpu_synchronize_state() - usually safe and idempotent -
will clobber state.
However, we routinely do this during a loadvm or incoming migration.
Usually this is called shortly after a reset, which will clear all the cpu
dirty flags with cpu_synchronize_all_post_reset(). Nothing is expected
to set the dirty flags again before the cpu state is loaded from the
incoming stream.
This means that it isn't safe to call cpu_synchronize_state() from a
post_load handler, which is non-obvious and potentially inconvenient.
We could cpu_synchronize_all_state() before the loadvm, but that would be
overkill since a) we expect the state to already be synchronized from the
reset and b) we expect to completely rewrite the state with a call to
cpu_synchronize_all_post_init() at the end of qemu_loadvm_state().
To clear this up, this patch introduces cpu_synchronize_pre_loadvm() and
associated helpers, which simply marks the cpu state as dirty without
actually changing anything. i.e. it says we want to discard any existing
KVM (or HAX) state and replace it with what we're going to load.
Cc: Juan Quintela <quintela@redhat.com>
Cc: Dave Gilbert <dgilbert@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Time to wire up all the call sites that request a shutdown or
reset to use the enum added in the previous patch.
It would have been less churn to keep the common case with no
arguments as meaning guest-triggered, and only modified the
host-triggered code paths, via a wrapper function, but then we'd
still have to audit that I didn't miss any host-triggered spots;
changing the signature forces us to double-check that I correctly
categorized all callers.
Since command line options can change whether a guest reset request
causes an actual reset vs. a shutdown, it's easy to also add the
information to reset requests.
Signed-off-by: Eric Blake <eblake@redhat.com>
Acked-by: David Gibson <david@gibson.dropbear.id.au> [ppc parts]
Reviewed-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk> [SPARC part]
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com> [s390x parts]
Message-Id: <20170515214114.15442-5-eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
That's a forward port of the core HAX interface code from the
emu-2.2-release branch in the external/qemu-android repository as used by
the Android emulator.
The original commit was "target/i386: Add Intel HAX to android emulator"
saying:
"""
Backport of 2b3098ff27bab079caab9b46b58546b5036f5c0c
from studio-1.4-dev into emu-master-dev
Intel HAX (harware acceleration) will enhance android emulator performance
in Windows and Mac OS X in the systems powered by Intel processors with
"Intel Hardware Accelerated Execution Manager" package installed when
user runs android emulator with Intel target.
Signed-off-by: David Chou <david.j.chou@intel.com>
"""
It has been modified to build and run along with the current code base.
The formatting has been fixed to go through scripts/checkpatch.pl,
and the DPRINTF macros have been updated to get the instanciations checked by
the compiler.
The FPU registers saving/restoring has been updated to match the current
QEMU registers layout.
The implementation has been simplified by doing the following modifications:
- removing the code for supporting the hardware without Unrestricted Guest (UG)
mode (including all the code to fallback on TCG emulation).
- not including the Darwin support (which is not yet debugged/tested).
- simplifying the initialization by removing the leftovers from the Android
specific code, then trimming down the remaining logic.
- removing the unused MemoryListener callbacks.
Signed-off-by: Vincent Palatin <vpalatin@chromium.org>
Message-Id: <e1023837f8d0e4c470f6c4a3bf643971b2bca5be.1484045952.git.vpalatin@chromium.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
