earlyprintk=efi,keep will cause kernel hangs while freeing initmem like
below:
VFS: Mounted root (ext4 filesystem) readonly on device 254:2.
devtmpfs: mounted
Freeing unused kernel memory: 880K (ffffffff817d4000 - ffffffff818b0000)
It is caused by efi earlyprintk use __init function which will be freed
later. Such as early_efi_write is marked as __init, also it will use
early_ioremap which is init function as well.
To fix this issue, I added early initcall early_efi_map_fb which maps
the whole efi fb for later use. OTOH, adding a wrapper function
early_efi_map which calls early_ioremap before ioremap is available.
With this patch applied efi boot ok with earlyprintk=efi,keep console=efi
Signed-off-by: Dave Young <dyoung@redhat.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
We're currently passing the file handle for the root file system to
efi_file_read() and efi_file_close(), instead of the file handle for the
file we wish to read/close.
While this has worked up until now, it seems that it has only been by
pure luck. Olivier explains,
"The issue is the UEFI Fat driver might return the same function for
'fh->read()' and 'h->read()'. While in our case it does not work with
a different implementation of EFI_SIMPLE_FILE_SYSTEM_PROTOCOL. In our
case, we return a different pointer when reading a directory and
reading a file."
Fixing this actually clears up the two functions because we can drop one
of the arguments, and instead only pass a file 'handle' argument.
Reported-by: Olivier Martin <olivier.martin@arm.com>
Reviewed-by: Olivier Martin <olivier.martin@arm.com>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Cc: Leif Lindholm <leif.lindholm@linaro.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
code32_start should point at the start of the protected mode code, and
*not* at the beginning of the bzImage. This is much easier to do in
assembly so document that callers of make_boot_params() need to fill out
code32_start.
The fallout from this bug is that we would end up relocating the image
but copying the image at some offset, resulting in what appeared to be
memory corruption.
Reported-by: Thomas Bächler <thomas@archlinux.org>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
commit 54b52d8726 ("x86/efi: Build our own EFI services pointer
table") introduced a regression because the 64-bit file_size()
implementation passed a pointer to a 32-bit data object, instead of a
pointer to a 64-bit object.
Because the firmware treats the object as 64-bits regardless it was
reading random values from the stack for the upper 32-bits.
This resulted in people being unable to boot their machines, after
seeing the following error messages,
Failed to get file info size
Failed to alloc highmem for files
Reported-by: Dzmitry Sledneu <dzmitry.sledneu@gmail.com>
Reported-by: Koen Kooi <koen@dominion.thruhere.net>
Tested-by: Koen Kooi <koen@dominion.thruhere.net>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The ARM EFI boot stub doesn't need to care about the efi_early
infrastructure that x86 requires in order to do mixed mode thunking. So
wrap everything up in an efi_call_early() macro.
This allows x86 to do the necessary indirection jumps to call whatever
firmware interface is necessary (native or mixed mode), but also allows
the ARM folks to mask the fact that they don't support relocation in the
boot stub and need to pass 'sys_table_arg' to every function.
[ hpa: there are no object code changes from this patch ]
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Link: http://lkml.kernel.org/r/20140326091011.GB2958@console-pimps.org
Cc: Roy Franz <roy.franz@linaro.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
In the thunk patches the 'attr' argument was dropped to
query_variable_info(). Restore it otherwise the firmware will return
EFI_INVALID_PARAMETER.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Dan reported that phys_efi_get_time() is doing kmalloc(..., GFP_KERNEL)
under a spinlock which is very clearly a bug. Since phys_efi_get_time()
has no users let's just delete it instead of trying to fix it.
Note that since there are no users of phys_efi_get_time(), it is not
possible to actually trigger a GFP_KERNEL alloc under the spinlock.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Nathan Zimmer <nzimmer@sgi.com>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Jan Beulich <JBeulich@suse.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
I was triggering a #GP(0) from userland when running with
CONFIG_EFI_MIXED and CONFIG_IA32_EMULATION, from what looked like
register corruption. Turns out that the mixed mode code was trashing the
contents of %ds, %es and %ss in __efi64_thunk().
Save and restore the contents of these segment registers across the call
to __efi64_thunk() so that we don't corrupt the CPU context.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The kbuild test robot reported the following errors, introduced with
commit 54b52d8726 ("x86/efi: Build our own EFI services pointer
table"),
arch/x86/boot/compressed/head_32.o: In function `efi32_config':
>> (.data+0x58): undefined reference to `efi_call_phys'
arch/x86/boot/compressed/head_64.o: In function `efi64_config':
>> (.data+0x90): undefined reference to `efi_call6'
Wrap the efi*_config structures in #ifdef CONFIG_EFI_STUB so that we
don't make references to EFI functions if they're not compiled in.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The kbuild test robot reported the following errors that were introduced
with commit 993c30a04e ("x86, tools: Consolidate #ifdef code"),
arch/x86/boot/tools/build.c: In function 'update_pecoff_setup_and_reloc':
>> arch/x86/boot/tools/build.c:252:1: error: parameter name omitted
static inline void update_pecoff_setup_and_reloc(unsigned int) {}
^
arch/x86/boot/tools/build.c: In function 'update_pecoff_text':
>> arch/x86/boot/tools/build.c:253:1: error: parameter name omitted
static inline void update_pecoff_text(unsigned int, unsigned int) {}
^
>> arch/x86/boot/tools/build.c:253:1: error: parameter name omitted
arch/x86/boot/tools/build.c: In function 'main':
>> arch/x86/boot/tools/build.c:372:2: warning: implicit declaration of function 'efi_stub_entry_update' [-Wimplicit-function-declaration]
efi_stub_entry_update();
^
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The trace_do_page_fault function trigger tracepoint
and then handles the actual page fault.
This could lead to error if the tracepoint caused page
fault. The original cr2 value gets lost and the original
page fault handler kills current process with SIGSEGV.
This happens if you record page faults with callchain
data, the user part of it will cause tracepoint handler
to page fault:
# perf record -g -e exceptions:page_fault_user ls
Fixing this by saving the original cr2 value
and using it after tracepoint handler is done.
v2: Moving the cr2 read before exception_enter, because
it could trigger tracepoint as well.
Reported-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Tested-by: Vince Weaver <vincent.weaver@maine.edu>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Seiji Aguchi <seiji.aguchi@hds.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Link: http://lkml.kernel.org/r/alpine.DEB.2.10.1402211701380.6395@vincent-weaver-1.um.maine.edu
Link: http://lkml.kernel.org/r/20140228160526.GD1133@krava.brq.redhat.com
causes a crash during boot - Borislav Petkov
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1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=YIab
-----END PGP SIGNATURE-----
Merge tag 'efi-urgent' into x86/urgent
* Disable the new EFI 1:1 virtual mapping for SGI UV because using it
causes a crash during boot - Borislav Petkov
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Alex reported hitting the following BUG after the EFI 1:1 virtual
mapping work was merged,
kernel BUG at arch/x86/mm/init_64.c:351!
invalid opcode: 0000 [#1] SMP
Call Trace:
[<ffffffff818aa71d>] init_extra_mapping_uc+0x13/0x15
[<ffffffff818a5e20>] uv_system_init+0x22b/0x124b
[<ffffffff8108b886>] ? clockevents_register_device+0x138/0x13d
[<ffffffff81028dbb>] ? setup_APIC_timer+0xc5/0xc7
[<ffffffff8108b620>] ? clockevent_delta2ns+0xb/0xd
[<ffffffff818a3a92>] ? setup_boot_APIC_clock+0x4a8/0x4b7
[<ffffffff8153d955>] ? printk+0x72/0x74
[<ffffffff818a1757>] native_smp_prepare_cpus+0x389/0x3d6
[<ffffffff818957bc>] kernel_init_freeable+0xb7/0x1fb
[<ffffffff81535530>] ? rest_init+0x74/0x74
[<ffffffff81535539>] kernel_init+0x9/0xff
[<ffffffff81541dfc>] ret_from_fork+0x7c/0xb0
[<ffffffff81535530>] ? rest_init+0x74/0x74
Getting this thing to work with the new mapping scheme would need more
work, so automatically switch to the old memmap layout for SGI UV.
Acked-by: Russ Anderson <rja@sgi.com>
Cc: Alex Thorlton <athorlton@sgi.com
Signed-off-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Some firmware appears to enable interrupts during boot service calls,
even if we've explicitly disabled them prior to the call. This is
actually allowed per the UEFI spec because boottime services expect to
be called with interrupts enabled.
So that's fine, we just need to ensure that we disable them again in
efi_enter32() before switching to a 64-bit GDT, otherwise an interrupt
may fire causing a 32-bit IRQ handler to run after we've left
compatibility mode.
Despite efi_enter32() being called both for boottime and runtime
services, this really only affects boottime because the runtime services
callchain is executed with interrupts disabled. See efi_thunk().
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Some EFI firmware makes use of the FPU during boottime services and
clearing X86_CR4_OSFXSR by overwriting %cr4 causes the firmware to
crash.
Add the PAE bit explicitly instead of trashing the existing contents,
leaving the rest of the bits as the firmware set them.
Cc: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Add the Kconfig option and bump the kernel header version so that boot
loaders can check whether the handover code is available if they want.
The xloadflags field in the bzImage header is also updated to reflect
that the kernel supports both entry points by setting both of
XLF_EFI_HANDOVER_32 and XLF_EFI_HANDOVER_64 when CONFIG_EFI_MIXED=y.
XLF_CAN_BE_LOADED_ABOVE_4G is disabled so that the kernel text is
guaranteed to be addressable with 32-bits.
Note that no boot loaders should be using the bits set in xloadflags to
decide which entry point to jump to. The entire scheme is based on the
concept that 32-bit bootloaders always jump to ->handover_offset and
64-bit loaders always jump to ->handover_offset + 512. We set both bits
merely to inform the boot loader that it's safe to use the native
handover offset even if the machine type in the PE/COFF header claims
otherwise.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Setup the runtime services based on whether we're booting in EFI native
mode or not. For non-native mode we need to thunk from 64-bit into
32-bit mode before invoking the EFI runtime services.
Using the runtime services after SetVirtualAddressMap() is slightly more
complicated because we need to ensure that all the addresses we pass to
the firmware are below the 4GB boundary so that they can be addressed
with 32-bit pointers, see efi_setup_page_tables().
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The EFI handover code only works if the "bitness" of the firmware and
the kernel match, i.e. 64-bit firmware and 64-bit kernel - it is not
possible to mix the two. This goes against the tradition that a 32-bit
kernel can be loaded on a 64-bit BIOS platform without having to do
anything special in the boot loader. Linux distributions, for one thing,
regularly run only 32-bit kernels on their live media.
Despite having only one 'handover_offset' field in the kernel header,
EFI boot loaders use two separate entry points to enter the kernel based
on the architecture the boot loader was compiled for,
(1) 32-bit loader: handover_offset
(2) 64-bit loader: handover_offset + 512
Since we already have two entry points, we can leverage them to infer
the bitness of the firmware we're running on, without requiring any boot
loader modifications, by making (1) and (2) valid entry points for both
CONFIG_X86_32 and CONFIG_X86_64 kernels.
To be clear, a 32-bit boot loader will always use (1) and a 64-bit boot
loader will always use (2). It's just that, if a single kernel image
supports (1) and (2) that image can be used with both 32-bit and 64-bit
boot loaders, and hence both 32-bit and 64-bit EFI.
(1) and (2) must be 512 bytes apart at all times, but that is already
part of the boot ABI and we could never change that delta without
breaking existing boot loaders anyhow.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Implement the transition code to go from IA32e mode to protected mode in
the EFI boot stub. This is required to use 32-bit EFI services from a
64-bit kernel.
Since EFI boot stub is executed in an identity-mapped region, there's
not much we need to do before invoking the 32-bit EFI boot services.
However, we do reload the firmware's global descriptor table
(efi32_boot_gdt) in case things like timer events are still running in
the firmware.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
It's not possible to dereference the EFI System table directly when
booting a 64-bit kernel on a 32-bit EFI firmware because the size of
pointers don't match.
In preparation for supporting the above use case, build a list of
function pointers on boot so that callers don't have to worry about
converting pointer sizes through multiple levels of indirection.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
The traditional approach of using machine-specific types such as
'unsigned long' does not allow the kernel to interact with firmware
running in a different CPU mode, e.g. 64-bit kernel with 32-bit EFI.
Add distinct EFI structure definitions for both 32-bit and 64-bit so
that we can use them in the 32-bit and 64-bit code paths.
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Both efi_free_boot_services() and efi_enter_virtual_mode() are invoked
from init/main.c, but only if the EFI runtime services are available.
This is not the case for non-native boots, e.g. where a 64-bit kernel is
booted with 32-bit EFI firmware.
Delete the dead code.
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Now that we have EFI-specific page tables we need to lookup the pgd when
dumping those page tables, rather than assuming that swapper_pgdir is
the current pgdir.
Remove the double underscore prefix, which is usually reserved for
static functions.
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Instead of littering main() with #ifdef CONFIG_EFI_STUB, move the logic
into separate functions that do nothing if the config option isn't set.
This makes main() much easier to read.
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
handover_offset is now filled out by build.c. Don't set a default value
as it will be overwritten anyway.
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
... into a kexec flavor for better code readability and simplicity. The
original one was getting ugly with ifdeffery.
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Currently, running SetVirtualAddressMap() and passing the physical
address of the virtual map array was working only by a lucky coincidence
because the memory was present in the EFI page table too. Until Toshi
went and booted this on a big HP box - the krealloc() manner of resizing
the memmap we're doing did allocate from such physical addresses which
were not mapped anymore and boom:
http://lkml.kernel.org/r/1386806463.1791.295.camel@misato.fc.hp.com
One way to take care of that issue is to reimplement the krealloc thing
but with pages. We start with contiguous pages of order 1, i.e. 2 pages,
and when we deplete that memory (shouldn't happen all that often but you
know firmware) we realloc the next power-of-two pages.
Having the pages, it is much more handy and easy to map them into the
EFI page table with the already existing mapping code which we're using
for building the virtual mappings.
Thanks to Toshi Kani and Matt for the great debugging help.
Reported-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
We will use it in efi so expose it.
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
This is very useful for debugging issues with the recently added
pagetable switching code for EFI virtual mode.
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
With reusing the ->trampoline_pgd page table for mapping EFI regions in
order to use them after having switched to EFI virtual mode, it is very
useful to be able to dump aforementioned page table in dmesg. This adds
that functionality through the walk_pgd_level() interface which can be
called from somewhere else.
The original functionality of dumping to debugfs remains untouched.
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Tested-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Coalesce formats and remove spaces before tabs.
Move __initdata after the variable declaration.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
For now we only ensure about 5kb free space for avoiding our board
refusing boot. But the comment lies that we retain 50% space.
Signed-off-by: Madper Xie <cxie@redhat.com>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
It makes more sense to set the feature flag in the success path of the
detection function than it does to rely on the caller doing it. Apart
from it being more logical to group the code and data together, it sets
a much better example for new EFI architectures.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
As we grow support for more EFI architectures they're going to want the
ability to query which EFI features are available on the running system.
Instead of storing this information in an architecture-specific place,
stick it in the global 'struct efi', which is already the central
location for EFI state.
While we're at it, let's change the return value of efi_enabled() to be
bool and replace all references to 'facility' with 'feature', which is
the usual word used to describe the attributes of the running system.
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
Pull perf fixes from Ingo Molnar:
"Misc fixes, most of them on the tooling side"
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf tools: Fix strict alias issue for find_first_bit
perf tools: fix BFD detection on opensuse
perf: Fix hotplug splat
perf/x86: Fix event scheduling
perf symbols: Destroy unused symsrcs
perf annotate: Check availability of annotate when processing samples
Pull x86 fixes from Peter Anvin:
"The VMCOREINFO patch I'll pushing for this release to avoid having a
release with kASLR and but without that information.
I was hoping to include the FPU patches from Suresh, but ran into a
problem (see other thread); will try to make them happen next week"
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86, kaslr: add missed "static" declarations
x86, kaslr: export offset in VMCOREINFO ELF notes
virtualization on Intel is broken in 3.13 and fixed by this
pull request.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (GNU/Linux)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=zQTn
-----END PGP SIGNATURE-----
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Pull KVM fixes from Paolo Bonzini:
"Three x86 fixes and one for ARM/ARM64.
In particular, nested virtualization on Intel is broken in 3.13 and
fixed by this pull request"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
kvm, vmx: Really fix lazy FPU on nested guest
kvm: x86: fix emulator buffer overflow (CVE-2014-0049)
arm/arm64: KVM: detect CPU reset on CPU_PM_EXIT
KVM: MMU: drop read-only large sptes when creating lower level sptes
Commit e504c9098e (kvm, vmx: Fix lazy FPU on nested guest, 2013-11-13)
highlighted a real problem, but the fix was subtly wrong.
nested_read_cr0 is the CR0 as read by L2, but here we want to look at
the CR0 value reflecting L1's setup. In other words, L2 might think
that TS=0 (so nested_read_cr0 has the bit clear); but if L1 is actually
running it with TS=1, we should inject the fault into L1.
The effective value of CR0 in L2 is contained in vmcs12->guest_cr0, use
it.
Fixes: e504c9098e
Reported-by: Kashyap Chamarty <kchamart@redhat.com>
Reported-by: Stefan Bader <stefan.bader@canonical.com>
Tested-by: Kashyap Chamarty <kchamart@redhat.com>
Tested-by: Anthoine Bourgeois <bourgeois@bertin.fr>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The problem occurs when the guest performs a pusha with the stack
address pointing to an mmio address (or an invalid guest physical
address) to start with, but then extending into an ordinary guest
physical address. When doing repeated emulated pushes
emulator_read_write sets mmio_needed to 1 on the first one. On a
later push when the stack points to regular memory,
mmio_nr_fragments is set to 0, but mmio_is_needed is not set to 0.
As a result, KVM exits to userspace, and then returns to
complete_emulated_mmio. In complete_emulated_mmio
vcpu->mmio_cur_fragment is incremented. The termination condition of
vcpu->mmio_cur_fragment == vcpu->mmio_nr_fragments is never achieved.
The code bounces back and fourth to userspace incrementing
mmio_cur_fragment past it's buffer. If the guest does nothing else it
eventually leads to a a crash on a memcpy from invalid memory address.
However if a guest code can cause the vm to be destroyed in another
vcpu with excellent timing, then kvm_clear_async_pf_completion_queue
can be used by the guest to control the data that's pointed to by the
call to cancel_work_item, which can be used to gain execution.
Fixes: f78146b0f9
Signed-off-by: Andrew Honig <ahonig@google.com>
Cc: stable@vger.kernel.org (3.5+)
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Vince "Super Tester" Weaver reported a new round of syscall fuzzing (Trinity) failures,
with perf WARN_ON()s triggering. He also provided traces of the failures.
This is I think the relevant bit:
> pec_1076_warn-2804 [000] d... 147.926153: x86_pmu_disable: x86_pmu_disable
> pec_1076_warn-2804 [000] d... 147.926153: x86_pmu_state: Events: {
> pec_1076_warn-2804 [000] d... 147.926156: x86_pmu_state: 0: state: .R config: ffffffffffffffff ( (null))
> pec_1076_warn-2804 [000] d... 147.926158: x86_pmu_state: 33: state: AR config: 0 (ffff88011ac99800)
> pec_1076_warn-2804 [000] d... 147.926159: x86_pmu_state: }
> pec_1076_warn-2804 [000] d... 147.926160: x86_pmu_state: n_events: 1, n_added: 0, n_txn: 1
> pec_1076_warn-2804 [000] d... 147.926161: x86_pmu_state: Assignment: {
> pec_1076_warn-2804 [000] d... 147.926162: x86_pmu_state: 0->33 tag: 1 config: 0 (ffff88011ac99800)
> pec_1076_warn-2804 [000] d... 147.926163: x86_pmu_state: }
> pec_1076_warn-2804 [000] d... 147.926166: collect_events: Adding event: 1 (ffff880119ec8800)
So we add the insn:p event (fd[23]).
At this point we should have:
n_events = 2, n_added = 1, n_txn = 1
> pec_1076_warn-2804 [000] d... 147.926170: collect_events: Adding event: 0 (ffff8800c9e01800)
> pec_1076_warn-2804 [000] d... 147.926172: collect_events: Adding event: 4 (ffff8800cbab2c00)
We try and add the {BP,cycles,br_insn} group (fd[3], fd[4], fd[15]).
These events are 0:cycles and 4:br_insn, the BP event isn't x86_pmu so
that's not visible.
group_sched_in()
pmu->start_txn() /* nop - BP pmu */
event_sched_in()
event->pmu->add()
So here we should end up with:
0: n_events = 3, n_added = 2, n_txn = 2
4: n_events = 4, n_added = 3, n_txn = 3
But seeing the below state on x86_pmu_enable(), the must have failed,
because the 0 and 4 events aren't there anymore.
Looking at group_sched_in(), since the BP is the leader, its
event_sched_in() must have succeeded, for otherwise we would not have
seen the sibling adds.
But since neither 0 or 4 are in the below state; their event_sched_in()
must have failed; but I don't see why, the complete state: 0,0,1:p,4
fits perfectly fine on a core2.
However, since we try and schedule 4 it means the 0 event must have
succeeded! Therefore the 4 event must have failed, its failure will
have put group_sched_in() into the fail path, which will call:
event_sched_out()
event->pmu->del()
on 0 and the BP event.
Now x86_pmu_del() will reduce n_events; but it will not reduce n_added;
giving what we see below:
n_event = 2, n_added = 2, n_txn = 2
> pec_1076_warn-2804 [000] d... 147.926177: x86_pmu_enable: x86_pmu_enable
> pec_1076_warn-2804 [000] d... 147.926177: x86_pmu_state: Events: {
> pec_1076_warn-2804 [000] d... 147.926179: x86_pmu_state: 0: state: .R config: ffffffffffffffff ( (null))
> pec_1076_warn-2804 [000] d... 147.926181: x86_pmu_state: 33: state: AR config: 0 (ffff88011ac99800)
> pec_1076_warn-2804 [000] d... 147.926182: x86_pmu_state: }
> pec_1076_warn-2804 [000] d... 147.926184: x86_pmu_state: n_events: 2, n_added: 2, n_txn: 2
> pec_1076_warn-2804 [000] d... 147.926184: x86_pmu_state: Assignment: {
> pec_1076_warn-2804 [000] d... 147.926186: x86_pmu_state: 0->33 tag: 1 config: 0 (ffff88011ac99800)
> pec_1076_warn-2804 [000] d... 147.926188: x86_pmu_state: 1->0 tag: 1 config: 1 (ffff880119ec8800)
> pec_1076_warn-2804 [000] d... 147.926188: x86_pmu_state: }
> pec_1076_warn-2804 [000] d... 147.926190: x86_pmu_enable: S0: hwc->idx: 33, hwc->last_cpu: 0, hwc->last_tag: 1 hwc->state: 0
So the problem is that x86_pmu_del(), when called from a
group_sched_in() that fails (for whatever reason), and without x86_pmu
TXN support (because the leader is !x86_pmu), will corrupt the n_added
state.
Reported-and-Tested-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Dave Jones <davej@redhat.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20140221150312.GF3104@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Read-only large sptes can be created due to read-only faults as
follows:
- QEMU pagetable entry that maps guest memory is read-only
due to COW.
- Guest read faults such memory, COW is not broken, because
it is a read-only fault.
- Enable dirty logging, large spte not nuked because it is read-only.
- Write-fault on such memory causes guest to loop endlessly
(which must go down to level 1 because dirty logging is enabled).
Fix by dropping large spte when necessary.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Include kASLR offset in VMCOREINFO ELF notes to assist in debugging.
[ hpa: pushing this for v3.14 to avoid having a kernel version with
kASLR where we can't debug output. ]
Signed-off-by: Eugene Surovegin <surovegin@google.com>
Link: http://lkml.kernel.org/r/20140123173120.GA25474@www.outflux.net
Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Pull x86 fixes from Thomas Gleixner:
- a bugfix which prevents a divide by 0 panic when the newly introduced
try_msr_calibrate_tsc() fails
- enablement of the Baytrail platform to utilize the newfangled msr
based calibration
* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86: tsc: Add missing Baytrail frequency to the table
x86, tsc: Fallback to normal calibration if fast MSR calibration fails
This patch updates the CBOX PMU filters mapping tables for SNB-EP
and IVT (model 45 and 62 respectively).
The NID umask always comes in addition to another umask.
When set, the NID filter is applied.
The current mapping tables were missing some code/umask
combinations to account for the NID umask. This patch
fixes that.
Cc: mingo@elte.hu
Cc: ak@linux.intel.com
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20140219131018.GA24475@quad
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
