This closes some arcane holes in single-step handling that can arise
only when user programs set TF directly (via popf or sigreturn) and
then use vDSO (syscall/sysenter) system call entry. In those entry
paths, the clear_TF_reenable case hits and we must check TIF_SINGLESTEP
to be sure our bookkeeping stays correct wrt the user's view of TF.
Signed-off-by: Roland McGrath <roland@redhat.com>
This unifies and cleans up the syscall tracing code on i386 and x86_64.
Using a single function for entry and exit tracing on 32-bit made the
do_syscall_trace() into some terrible spaghetti. The logic is clear and
simple using separate syscall_trace_enter() and syscall_trace_leave()
functions as on 64-bit.
The unification adds PTRACE_SYSEMU and PTRACE_SYSEMU_SINGLESTEP support
on x86_64, for 32-bit ptrace() callers and for 64-bit ptrace() callers
tracing either 32-bit or 64-bit tasks. It behaves just like 32-bit.
Changing syscall_trace_enter() to return the syscall number shortens
all the assembly paths, while adding the SYSEMU feature in a simple way.
Signed-off-by: Roland McGrath <roland@redhat.com>
This unifies the treatment of TIF_SINGLESTEP on i386 and x86_64.
The bit is now excluded from _TIF_WORK_MASK on i386 as it has been
on x86_64. This means the do_notify_resume() path using it is never
used, so TIF_SINGLESTEP is not cleared on returning to user mode.
Both now leave TIF_SINGLESTEP set when returning to user, so that
it's already set on an int $0x80 system call entry. This removes
the need for testing TF on the system_call path. Doing it this way
fixes the regression for PTRACE_SINGLESTEP into a sigreturn syscall,
introduced by commit 1e2e99f0e4.
The clear_TF_reenable case that sets TIF_SINGLESTEP can only happen
on a non-exception kernel entry, i.e. sysenter/syscall instruction.
That will always get to the syscall exit tracing path.
Signed-off-by: Roland McGrath <roland@redhat.com>
The standard ticket spinlocks are very expensive in a virtual
environment, because their performance depends on Xen's scheduler
giving vcpus time in the order that they're supposed to take the
spinlock.
This implements a Xen-specific spinlock, which should be much more
efficient.
The fast-path is essentially the old Linux-x86 locks, using a single
lock byte. The locker decrements the byte; if the result is 0, then
they have the lock. If the lock is negative, then locker must spin
until the lock is positive again.
When there's contention, the locker spin for 2^16[*] iterations waiting
to get the lock. If it fails to get the lock in that time, it adds
itself to the contention count in the lock and blocks on a per-cpu
event channel.
When unlocking the spinlock, the locker looks to see if there's anyone
blocked waiting for the lock by checking for a non-zero waiter count.
If there's a waiter, it traverses the per-cpu "lock_spinners"
variable, which contains which lock each CPU is waiting on. It picks
one CPU waiting on the lock and sends it an event to wake it up.
This allows efficient fast-path spinlock operation, while allowing
spinning vcpus to give up their processor time while waiting for a
contended lock.
[*] 2^16 iterations is threshold at which 98% locks have been taken
according to Thomas Friebel's Xen Summit talk "Preventing Guests from
Spinning Around". Therefore, we'd expect the lock and unlock slow
paths will only be entered 2% of the time.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Christoph Lameter <clameter@linux-foundation.org>
Cc: Petr Tesarik <ptesarik@suse.cz>
Cc: Virtualization <virtualization@lists.linux-foundation.org>
Cc: Xen devel <xen-devel@lists.xensource.com>
Cc: Thomas Friebel <thomas.friebel@amd.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Implement a version of the old spinlock algorithm, in which everyone
spins waiting for a lock byte. In order to be compatible with the
ticket-lock's use of a zero initializer, this uses the convention of
'0' for unlocked and '1' for locked.
This algorithm is much better than ticket locks in a virtual
envionment, because it doesn't interact badly with the vcpu scheduler.
If there are multiple vcpus spinning on a lock and the lock is
released, the next vcpu to be scheduled will take the lock, rather
than cycling around until the next ticketed vcpu gets it.
To use this, you must call paravirt_use_bytelocks() very early, before
any spinlocks have been taken.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Christoph Lameter <clameter@linux-foundation.org>
Cc: Petr Tesarik <ptesarik@suse.cz>
Cc: Virtualization <virtualization@lists.linux-foundation.org>
Cc: Xen devel <xen-devel@lists.xensource.com>
Cc: Thomas Friebel <thomas.friebel@amd.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Ticket spinlocks have absolutely ghastly worst-case performance
characteristics in a virtual environment. If there is any contention
for physical CPUs (ie, there are more runnable vcpus than cpus), then
ticket locks can cause the system to end up spending 90+% of its time
spinning.
The problem is that (v)cpus waiting on a ticket spinlock will be
granted access to the lock in strict order they got their tickets. If
the hypervisor scheduler doesn't give the vcpus time in that order,
they will burn timeslices waiting for the scheduler to give the right
vcpu some time. In the worst case it could take O(n^2) vcpu scheduler
timeslices for everyone waiting on the lock to get it, not counting
new cpus trying to take the lock while the log-jam is sorted out.
These hooks allow a paravirt backend to replace the spinlock
implementation.
At the very least, this could revert the implementation back to the
old lock algorithm, which allows the next scheduled vcpu to take the
lock, and has basically fairly good performance.
It also allows the spinlocks to take advantages of the hypervisor
features to make locks more efficient (spin and block, for example).
The cost to native execution is an extra direct call when using a
spinlock function. There's no overhead if CONFIG_PARAVIRT is turned
off.
The lock structure is fixed at a single "unsigned int", initialized to
zero, but the spinlock implementation can use it as it wishes.
Thanks to Thomas Friebel's Xen Summit talk "Preventing Guests from
Spinning Around" for pointing out this problem.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Christoph Lameter <clameter@linux-foundation.org>
Cc: Petr Tesarik <ptesarik@suse.cz>
Cc: Virtualization <virtualization@lists.linux-foundation.org>
Cc: Xen devel <xen-devel@lists.xensource.com>
Cc: Thomas Friebel <thomas.friebel@amd.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
AMD only supports "syscall" from 32-bit compat usermode.
Intel and Centaur(?) only support "sysenter" from 32-bit compat usermode.
Set the X86 feature bits accordingly, and set up the vdso in
accordance with those bits. On the offchance we run on in a 64-bit
environment which supports neither syscall nor sysenter from 32-bit
mode, then fall back to the int $0x80 vdso.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The Xen hypercall interface is allowed to trash any or all of the
argument registers, so we need to be careful that the kernel state
isn't damaged. On 32-bit kernels, the hypercall parameter registers
same as a regparm function call, so we've got away without explicit
clobbering so far. The 64-bit ABI defines lots of caller-save
registers, so save them all for safety. We can trim this set later by
re-distributing the responsibility for saving all these registers.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Use callback_op hypercall to register callbacks in a 32/64-bit
independent way (64-bit doesn't need a code segment, but that detail
is hidden in XEN_CALLBACK).
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
When building initial pagetables in 64-bit kernel the pud/pmd pointer may
be in ioremap/fixmap space, so we need to walk the pagetable to look up the
physical address.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Set up the initial pagetables to map the kernel mapping into the
physical mapping space. This makes __va() usable, since it requires
physical mappings.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
As a stopgap until Mike Travis's x86-64 gs-based percpu patches are
ready, provide workaround functions for x86_read/write_percpu for
Xen's use.
Specifically, this means that we can't really make use of vcpu
placement, because we can't use a single gs-based memory access to get
to vcpu fields. So disable all that for now.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
The 64-bit calling convention for hypercalls uses different registers
from 32-bit. Annoyingly, gcc's asm syntax doesn't have a way to
specify one of the extra numeric reigisters in a constraint, so we
must use explicitly placed register variables. Given that we have to
do it for some args, may as well do it for all.
Also fix syntax gcc generates for the call instruction itself. We
need a plain direct call, but the asm expansion which works on 32-bit
generates a rip-relative addressing mode in 64-bit, which is treated
as an indirect call. The alternative is to pass the hypercall page
offset into the asm, and have it add it to the hypercall page start
address to generate the call.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Stephen Tweedie <sct@redhat.com>
Cc: Eduardo Habkost <ehabkost@redhat.com>
Cc: Mark McLoughlin <markmc@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
AS arch/x86/lib/csum-copy_64.o
arch/x86/lib/csum-copy_64.S: Assembler messages:
arch/x86/lib/csum-copy_64.S:48: Error: Macro `ignore' was already defined
make[1]: *** [arch/x86/lib/csum-copy_64.o] Error 1
make: *** [arch/x86/lib] Error 2
It appears that csum-copy_64.S and dwarf2.h both define an ignore macro.
I would expect one of them can be renamed quite easily, unless they
are references elsewhere.
Caused-by-commit: 392a0fc96b
x86: merge dwarf2 headers
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Now that IRQ2 is never made available to the I/O APIC, there is no need
to special-case it and mask as a workaround for broken systems. Actually,
because of the former, mask_IO_APIC_irq(2) is a no-op already.
Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Andreas Herrmann <andreas.herrmann3@amd.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
got this on a test-system:
calling numaq_tsc_disable+0x0/0x39
NUMAQ: disabling TSC
initcall numaq_tsc_disable+0x0/0x39 returned 0 after 0 msecs
that's because we should not be using arch_initcall to call numaq_tsc_disable.
need to call it in setup_arch before time_init()/tsc_init()
and call it in init_intel() to make the cpu feature bits right.
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In dwarf2_32.h, test for CONFIG_AS_CFI instead of
CONFIG_UNWIND_INFO. Turns out that searching for UNWIND_INFO
returns no match in any Kconfig or Makefile, so we're really
just throwing everything away regarding dwarf frames for i386.
The test that generates CONFIG_AS_CFI does not have anything
x86_64-specific, and right now, checking V=1 builds shows me
that the flags is there anyway, although unused.
Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
In dwarf_64.h header, use the "ignore" macro the way
i386 does.
Signed-off-by: Glauber Costa <gcosta@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
i spent a fair amount of time chasing a 64-bit bootup crash that manifested
itself as bootup segfaults:
S10network[1825]: segfault at 7f3e2b5d16b8 ip 00000031108748c9 sp 00007fffb9c14c70 error 4 in libc-2.7.so[3110800000+14d000]
eventually causing init to die and panic the system:
Kernel panic - not syncing: Attempted to kill init!
Pid: 1, comm: init Not tainted 2.6.26-rc9-tip #13878
after a maratonic bisection session, the bad commit turned out to be:
| b7675791859075418199c7af86a116ea34eaf5bd is first bad commit
| commit b7675791859075418199c7af86a116ea34eaf5bd
| Author: Jeremy Fitzhardinge <jeremy@goop.org>
| Date: Wed Jun 25 00:19:00 2008 -0400
|
| x86: remove open-coded save/load segment operations
|
| This removes a pile of buggy open-coded implementations of savesegment
| and loadsegment.
after some more bisection of this patch itself, it turns out that what
makes the difference are the savesegment() changes to __switch_to().
Taking a look at this portion of arch/x86/kernel/process_64.o revealed
this crutial difference:
| good: 99c: 8c e0 mov %fs,%eax
| 99e: 89 45 cc mov %eax,-0x34(%rbp)
|
| bad: 99c: 8c 65 cc mov %fs,-0x34(%rbp)
which is due to:
| unsigned fsindex;
| - asm volatile("movl %%fs,%0" : "=r" (fsindex));
| + savesegment(fs, fsindex);
savesegment() is implemented as:
#define savesegment(seg, value) \
asm("mov %%" #seg ",%0":"=rm" (value) : : "memory")
note the "m" modifier - it allows GCC to generate the segment move
into a memory operand as well.
But regarding segment operands there's a subtle detail in the x86
instruction set: the above 16-bit moves are zero-extend, but only
if it goes to a register.
If it goes to a memory operand, -0x34(%rbp) in the above case, there's
no zero-extend to 32-bit and the instruction will only save 16 bits
instead of the intended 32-bit.
The other 16 bits is random data - which can cause problems when that
value is used later on.
The solution is to only allow segment operands to go to registers.
This fix allows my test-system to boot up without crashing.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Add pseudo-feature bits to describe whether the CPU supports sysenter
and/or syscall from ia32-compat userspace. This removes a hardcoded
test in vdso32-setup.
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
As other IOMMUs do, this puts dummy pci_swiotlb_init() in swiotlb.h
and remove ifdef CONFIG_SWIOTLB in pci-dma.c.
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Our way to handle gart_* functions for CONFIG_GART_IOMMU and
!CONFIG_GART_IOMMU cases is inconsistent.
We have some dummy gart_* functions in !CONFIG_GART_IOMMU case and
also use ifdef CONFIG_GART_IOMMU tricks in pci-dma.c to call some
gart_* functions in only CONFIG_GART_IOMMU case.
This patch removes ifdef CONFIG_GART_IOMMU in pci-dma.c and always use
dummy gart_* functions in iommu.h.
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
gart.h has only GART-specific stuff. Only GART code needs it. Other
IOMMU stuff should include iommu.h instead of gart.h.
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Acked-by: Muli Ben-Yehuda <muli@il.ibm.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
NR_IRQS: let VISWS be just a sub-case of the generic code.
This can create a somewhat larger irq_desc[] array if NR_CPUS is high
but that should not worry VisWS which has 4 CPUs at most.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
move the SGIVW definitions from setup_arch.h into its own header file.
preparation for turning VISWS into a generic PC architecture.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
move the include/asm-x86/mach-visws/ VISWS specific hardware
details include files into include/asm-x86/visws, to be used from
generic code.
No code changed.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
