With this change, on SYSCALL64 code path we are now populating
pt_regs->cs, pt_regs->ss and pt_regs->rcx unconditionally and
therefore don't need to do that in FIXUP_TOP_OF_STACK.
We lose a number of large instructions there:
text data bss dec hex filename
13298 0 0 13298 33f2 entry_64_before.o
12978 0 0 12978 32b2 entry_64.o
What's more important, we convert two "MOVQ $imm,off(%rsp)" to
"PUSH $imm" (the ones which fill pt_regs->cs,ss).
Before this patch, placing them on fast path was slowing it down
by two cycles: this form of MOV is very large, 12 bytes, and
this probably reduces decode bandwidth to one instruction per cycle
when CPU sees them.
Therefore they were living in FIXUP_TOP_OF_STACK instead (away
from fast path).
"PUSH $imm" is a small 2-byte instruction. Moving it to fast path does
not slow it down in my measurements.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Borislav Petkov <bp@suse.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/1426785469-15125-3-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
PER_CPU_VAR(kernel_stack) was set up in a way where it points
five stack slots below the top of stack.
Presumably, it was done to avoid one "sub $5*8,%rsp"
in syscall/sysenter code paths, where iret frame needs to be
created by hand.
Ironically, none of them benefits from this optimization,
since all of them need to allocate additional data on stack
(struct pt_regs), so they still have to perform subtraction.
This patch eliminates KERNEL_STACK_OFFSET.
PER_CPU_VAR(kernel_stack) now points directly to top of stack.
pt_regs allocations are adjusted to allocate iret frame as well.
Hopefully we can merge it later with 32-bit specific
PER_CPU_VAR(cpu_current_top_of_stack) variable...
Net result in generated code is that constants in several insns
are changed.
This change is necessary for changing struct pt_regs creation
in SYSCALL64 code path from MOV to PUSH instructions.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Borislav Petkov <bp@suse.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/1426785469-15125-2-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This changes the THREAD_INFO() definition and all its callsites
so that they do not count stack position from
(top of stack - KERNEL_STACK_OFFSET), but from top of stack.
Semi-mysterious expressions THREAD_INFO(%rsp,RIP) - "why RIP??"
are now replaced by more logical THREAD_INFO(%rsp,SIZEOF_PTREGS)
- "calculate thread_info's address using information that
rsp is SIZEOF_PTREGS bytes below top of stack".
While at it, replace "(off)-THREAD_SIZE(reg)" with equivalent
"((off)-THREAD_SIZE)(reg)". The form without parentheses
falsely looks like we invoke THREAD_SIZE() macro.
Improve comment atop THREAD_INFO macro definition.
This patch does not change generated code (verified by objdump).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Borislav Petkov <bp@suse.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/1426785469-15125-1-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Having syscall32/sysenter32 initialization in a separate tiny
function, called from within a function that is already syscall
init specific, serves no real purpose.
Its existense also caused an unintended effect of having
wrmsrl(MSR_CSTAR) performed twice: once we set it to a dummy
function returning -ENOSYS, and immediately after
(if CONFIG_IA32_EMULATION), we set it to point to the proper
syscall32 entry point, ia32_cstar_target.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Both the execve() and sigreturn() family of syscalls have the
ability to change registers in ways that may not be compatabile
with the syscall path they were called from.
In particular, SYSRET and SYSEXIT can't handle non-default %cs and %ss,
and some bits in eflags.
These syscalls have stubs that are hardcoded to jump to the IRET path,
and not return to the original syscall path.
The following commit:
76f5df43ca ("Always allocate a complete "struct pt_regs" on the kernel stack")
recently changed this for some 32-bit compat syscalls, but introduced a bug where
execve from a 32-bit program to a 64-bit program would fail because it still returned
via SYSRETL. This caused Wine to fail when built for both 32-bit and 64-bit.
This patch sets TIF_NOTIFY_RESUME for execve() and sigreturn() so
that the IRET path is always taken on exit to userspace.
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/1426978461-32089-1-git-send-email-brgerst@gmail.com
[ Improved the changelog and comments. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Before the patch, the 'tss_struct::stack' field was not referenced anywhere.
It was used only to set SYSENTER's stack to point after the last byte
of tss_struct, thus the trailing field, stack[64], was used.
But grep would not know it. You can comment it out, compile,
and kernel will even run until an unlucky NMI corrupts
io_bitmap[] (which is also not easily detectable).
This patch changes code so that the purpose and usage of this
field is not mysterious anymore, and can be easily grepped for.
This does change generated code, for a subtle reason:
since tss_struct is ____cacheline_aligned, there happens to be
5 longs of padding at the end. Old code was using the padding
too; new code will strictly use it only for SYSENTER_stack[].
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/1425912738-559-2-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The comment in the signal code says that apps can save/restore
other segments on their own. It's true that apps can *save* SS
on their own, but there's no way for apps to restore it: SYSCALL
effectively resets SS to __USER_DS, so any value that user code
tries to load into SS gets lost on entry to sigreturn.
This recycles two padding bytes in the segment selector area for SS.
While we're at it, we need a second change to make this useful.
If the signal we're delivering is caused by a bad SS value,
saving that value isn't enough. We need to remove that bad
value from the regs before we try to deliver the signal. Oddly,
the i386 code already got this right.
I suspect that 64-bit programs that try to run 16-bit code and
use signals will have a lot of trouble without this.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Borislav Petkov <bp@suse.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/405594361340a2ec32f8e2b115c142df0e180d8e.1426193719.git.luto@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
math_state_restore() assumes it is called with irqs disabled,
but this is not true if the caller is __restore_xstate_sig().
This means that if ia32_fxstate == T and __copy_from_user()
fails, __restore_xstate_sig() returns with irqs disabled too.
This triggers:
BUG: sleeping function called from invalid context at kernel/locking/rwsem.c:41
dump_stack
___might_sleep
? _raw_spin_unlock_irqrestore
__might_sleep
down_read
? _raw_spin_unlock_irqrestore
print_vma_addr
signal_fault
sys32_rt_sigreturn
Change __restore_xstate_sig() to call set_used_math()
unconditionally. This avoids enabling and disabling interrupts
in math_state_restore(). If copy_from_user() fails, we can
simply do fpu_finit() by hand.
[ Note: this is only the first step. math_state_restore() should
not check used_math(), it should set this flag. While
init_fpu() should simply die. ]
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: <stable@vger.kernel.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Pekka Riikonen <priikone@iki.fi>
Cc: Quentin Casasnovas <quentin.casasnovas@oracle.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Suresh Siddha <sbsiddha@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150307153844.GB25954@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
By the nature of the TEST operation, it is often possible to test
a narrower part of the operand:
"testl $3, mem" -> "testb $3, mem",
"testq $3, %rcx" -> "testb $3, %cl"
This results in shorter instructions, because the TEST instruction
has no sign-entending byte-immediate forms unlike other ALU ops.
Note that this change does not create any LCP (Length-Changing Prefix)
stalls, which happen when adding a 0x66 prefix, which happens when
16-bit immediates are used, which changes such TEST instructions:
[test_opcode] [modrm] [imm32]
to:
[0x66] [test_opcode] [modrm] [imm16]
where [imm16] has a *different length* now: 2 bytes instead of 4.
This confuses the decoder and slows down execution.
REX prefixes were carefully designed to almost never hit this case:
adding REX prefix does not change instruction length except MOVABS
and MOV [addr],RAX instruction.
This patch does not add instructions which would use a 0x66 prefix,
code changes in assembly are:
-48 f7 07 01 00 00 00 testq $0x1,(%rdi)
+f6 07 01 testb $0x1,(%rdi)
-48 f7 c1 01 00 00 00 test $0x1,%rcx
+f6 c1 01 test $0x1,%cl
-48 f7 c1 02 00 00 00 test $0x2,%rcx
+f6 c1 02 test $0x2,%cl
-41 f7 c2 01 00 00 00 test $0x1,%r10d
+41 f6 c2 01 test $0x1,%r10b
-48 f7 c1 04 00 00 00 test $0x4,%rcx
+f6 c1 04 test $0x4,%cl
-48 f7 c1 08 00 00 00 test $0x8,%rcx
+f6 c1 08 test $0x8,%cl
Linus further notes:
"There are no stalls from using 8-bit instruction forms.
Now, changing from 64-bit or 32-bit 'test' instructions to 8-bit ones
*could* cause problems if it ends up having forwarding issues, so that
instead of just forwarding the result, you end up having to wait for
it to be stable in the L1 cache (or possibly the register file). The
forwarding from the store buffer is simplest and most reliable if the
read is done at the exact same address and the exact same size as the
write that gets forwarded.
But that's true only if:
(a) the write was very recent and is still in the write queue. I'm
not sure that's the case here anyway.
(b) on at least most Intel microarchitectures, you have to test a
different byte than the lowest one (so forwarding a 64-bit write
to a 8-bit read ends up working fine, as long as the 8-bit read
is of the low 8 bits of the written data).
A very similar issue *might* show up for registers too, not just
memory writes, if you use 'testb' with a high-byte register (where
instead of forwarding the value from the original producer it needs to
go through the register file and then shifted). But it's mainly a
problem for store buffers.
But afaik, the way Denys changed the test instructions, neither of the
above issues should be true.
The real problem for store buffer forwarding tends to be "write 8
bits, read 32 bits". That can be really surprisingly expensive,
because the read ends up having to wait until the write has hit the
cacheline, and we might talk tens of cycles of latency here. But
"write 32 bits, read the low 8 bits" *should* be fast on pretty much
all x86 chips, afaik."
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Andy Lutomirski <luto@amacapital.net>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Will Drewry <wad@chromium.org>
Link: http://lkml.kernel.org/r/1425675332-31576-1-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
I broke 32-bit kernels. The implementation of sp0 was correct
as far as I can tell, but sp0 was much weirder on x86_32 than I
realized. It has the following issues:
- Init's sp0 is inconsistent with everything else's: non-init tasks
are offset by 8 bytes. (I have no idea why, and the comment is unhelpful.)
- vm86 does crazy things to sp0.
Fix it up by replacing this_cpu_sp0() with
current_top_of_stack() and using a new percpu variable to track
the top of the stack on x86_32.
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 75182b1632 ("x86/asm/entry: Switch all C consumers of kernel_stack to this_cpu_sp0()")
Link: http://lkml.kernel.org/r/d09dbe270883433776e0cbee3c7079433349e96d.1425692936.git.luto@amacapital.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
