- 07 Jul, 2015 9 commits
-
-
Andy Lutomirski authored
The current x86 entry and exit code, written in a mixture of assembly and C code, is incomprehensible due to being open-coded in a lot of places without coherent documentation. It appears to work primary by luck and duct tape: i.e. obvious runtime failures were fixed on-demand, without re-thinking the design. Due to those reasons our confidence level in that code is low, and it is very difficult to incrementally improve. Add new code written in C, in preparation for simply deleting the old entry code. prepare_exit_to_usermode() is a new function that will handle all slow path exits to user mode. It is called with IRQs disabled and it leaves us in a state in which it is safe to immediately return to user mode. IRQs must not be re-enabled at any point after prepare_exit_to_usermode() returns and user mode is actually entered. (We can, of course, fail to enter user mode and treat that failure as a fresh entry to kernel mode.) All callers of do_notify_resume() will be migrated to call prepare_exit_to_usermode() instead; prepare_exit_to_usermode() needs to do everything that do_notify_resume() does today, but it also takes care of scheduling and context tracking. Unlike do_notify_resume(), it does not need to be called in a loop. syscall_return_slowpath() is exactly what it sounds like: it will be called on any syscall exit slow path. It will replace syscall_trace_leave() and it calls prepare_exit_to_usermode() on the way out. Signed-off-by:
Andy Lutomirski <luto@kernel.org> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: Denys Vlasenko <vda.linux@googlemail.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: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: paulmck@linux.vnet.ibm.com Link: http://lkml.kernel.org/r/c57c8b87661a4152801d7d3786eac2d1a2f209dd.1435952415.git.luto@kernel.org [ Improved the changelog a bit. ] Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
Andy Lutomirski authored
Changing the x86 context tracking hooks is dangerous because there are no good checks that we track our context correctly. Add a helper to check that we're actually in CONTEXT_USER when we enter from user mode and wire it up for syscall entries. Subsequent patches will wire this up for all non-NMI entries as well. NMIs are their own special beast and cannot currently switch overall context tracking state. Instead, they have their own special RCU hooks. This is a tiny speedup if !CONFIG_CONTEXT_TRACKING (removes a branch) and a tiny slowdown if CONFIG_CONTEXT_TRACING (adds a layer of indirection). Eventually, we should fix up the core context tracking code to supply a function that does what we want (and can be much simpler than user_exit), which will enable us to get rid of the extra call. Signed-off-by:
-
Andy Lutomirski authored
Other than the super-atomic exception entries, all exception entries are supposed to switch our context tracking state to CONTEXT_KERNEL. Assert that they do. These assertions appear trivial at this point, as exception_enter() is the function responsible for switching context, but I'm planning on reworking x86's exception context tracking, and these assertions will help make sure that all of this code keeps working. Signed-off-by:
-
Andy Lutomirski authored
The entry and exit C helpers were confusingly scattered between ptrace.c and signal.c, even though they aren't specific to ptrace or signal handling. Move them together in a new file. This change just moves code around. It doesn't change anything. Signed-off-by:
-
Andy Lutomirski authored
Low-level arch entries often call notify_die(), and it's easy for arch code to fail to exit an RCU quiescent state first. Assert that we're not quiescent in notify_die(). Signed-off-by:
-
Andy Lutomirski authored
This will let us sprinkle sanity checks around the kernel without making too much of a mess. Signed-off-by:
-
Ingo Molnar authored
Once x86 exports its do_signal(), the prototypes will clash. Fix the clash and also improve the code a bit: remove the unnecessary kern_do_signal() indirection. This allows interrupt_end() to share the 'regs' parameter calculation. Also remove the unused return code to match x86. Minimally build and boot tested. Signed-off-by:
-
Andy Lutomirski authored
If user code does SYSCALL32 or SYSENTER without a valid stack, then our attempt to determine the syscall args will result in a failed uaccess fault. Previously, we would try to recover by jumping to the syscall exit code, but we'd run the syscall exit work even though we never made it to the syscall entry work. Clean it up by treating the failure path as a non-syscall entry and exit pair. This fixes strace's output when running the syscall_arg_fault test. Without this fix, strace would get out of sync and would fail to associate syscall entries with syscall exits. Signed-off-by:
-
Andy Lutomirski authored
This test passes on 4.0 and fails on some newer kernels. Fortunately, the failure is likely not a big deal. This test will make sure that we don't break it further (e.g. OOPSing) as we clean up the entry code and that we eventually fix the regression. There's arguably no need to preserve the old ABI here -- anything that makes it into a fast (vDSO) syscall with a bad stack is about to crash no matter what we do. Signed-off-by:
-
- 06 Jul, 2015 31 commits
-
-
Brian Gerst authored
The x32 ABI is now independent of the ia32 compat ABI. Common code is now conditional on CONFIG_COMPAT, but unshared code like syscall entry, signal handling, and the VDSO are under separate config options. 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> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1434974121-32575-13-git-send-email-brgerst@gmail.comSigned-off-by:
-
Brian Gerst authored
Merge the 32-bit compat config setting for HAVE_UID16 with the 32-bit native one. Signed-off-by:
-
Brian Gerst authored
x32 does not need CONFIG_ARCH_WANT_OLD_COMPAT_IPC=y. Signed-off-by:
-
Brian Gerst authored
Including sys_ia32.h is not needed in signal.c. Signed-off-by:
-
Brian Gerst authored
perf_callchain_user32() is not needed for x32. Signed-off-by:
-
Brian Gerst authored
Change this to CONFIG_COMPAT so both 32-bit compat and x32 will do the check. Signed-off-by:
-
Brian Gerst authored
Build the 32-bit vdso only for native 32-bit or 32-bit compat is enabled. x32 should not force it to build. Signed-off-by:
-
Brian Gerst authored
Move the ia32-specific code in compat_arch_ptrace() into its own function. Signed-off-by:
-
Brian Gerst authored
This function is shared between the 32-bit compat and x32 ABIs. Signed-off-by:
-
Brian Gerst authored
ia32.h should only contain the code for 32-bit compatability. Signed-off-by:
-
Brian Gerst authored
TIF_ADDR32 is set for both ia32 and x32 tasks, so change from CONFIG_IA32_EMULATION to CONFIG_COMPAT. Use config_enabled() to make the function more readable. Signed-off-by:
-
Brian Gerst authored
copy_siginfo_to_user32() and copy_siginfo_from_user32() are used by both the 32-bit compat and x32 ABIs. Move them to signal_compat.c. Signed-off-by:
-
George Spelvin authored
Before, the code to do RDTSC looked like: rdtsc shl $0x20, %rdx mov %eax, %eax or %rdx, %rax The "mov %eax, %eax" is required to clear the high 32 bits of RAX. By declaring low and high as 64-bit variables, the code is simplified to: rdtsc shl $0x20,%rdx or %rdx,%rax Yes, it's a 2-byte instruction that's not on a critical path, but there are principles to be upheld. Every user of EAX_EDX_RET has been checked. I tried to check users of EAX_EDX_ARGS, but there weren't any, so I deleted it to be safe. ( There's no benefit to making "high" 64 bits, but it was the simplest way to proceed. ) Signed-off-by:
George Spelvin <linux@horizon.com> Signed-off-by:
Borislav Petkov <bp@suse.de> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: jacob.jun.pan@linux.intel.com Link: http://lkml.kernel.org/r/20150618075906.4615.qmail@ns.horizon.comSigned-off-by:
-
Andy Lutomirski authored
All callers have been converted to rdtsc_ordered(). Signed-off-by:
-
Andy Lutomirski authored
__pvclock_read_cycles() used to have two barriers, one of which was unnecessary, which got removed after an initial version of this patch was sent. But the barrier is still open-coded unnecessarily - get rid of that barrier and clean up the code by just using rdtsc_ordered(). Signed-off-by:
-
Andy Lutomirski authored
There are two logical changes here. First, this removes a check for cpu_has_tsc. That check is unnecessary, as we don't register the TSC as a clocksource on systems that have no TSC. Second, it adds a barrier, thus preventing observable non-monotonicity. I suspect that the missing barrier was never a problem in practice because system calls themselves were heavy enough barriers to prevent user code from observing time warps due to speculation. (Without the corresponding barrier in the vDSO, however, non-monotonicity is easy to detect.) Signed-off-by:
-
Andy Lutomirski authored
Using get_cycles was unnecessary: check_tsc_warp() is not called on TSC-less systems. Replace rdtsc_barrier(); get_cycles() with rdtsc_ordered(). While we're at it, make the somewhat more dangerous change of removing barrier_before_rdtsc after RDTSC in the TSC warp check code. This should be okay, though -- the vDSO TSC code doesn't have that barrier, so, if removing the barrier from the warp check would cause us to detect a warp that we otherwise wouldn't detect, then we have a genuine bug. Signed-off-by:
-
Andy Lutomirski authored
rdtsc_barrier(); rdtsc() is an unnecessary mouthful and requires more thought than should be necessary. Add an rdtsc_ordered() helper and replace the trivial call sites with it. This should not change generated code. The duplication of the fence asm is temporary. Signed-off-by:
-
Andy Lutomirski authored
Now that there is no paravirt TSC, the "native" is inappropriate. The function does RDTSC, so give it the obvious name: rdtsc(). Suggested-by:
-
Andy Lutomirski authored
It has no more callers, and it was never a very sensible interface to begin with. Users of the TSC should either read all 64 bits or explicitly throw out the high bits. Signed-off-by:
-
Andy Lutomirski authored
It's unclear to me why this code exists in the first place. Signed-off-by:
Dmitry Torokhov <dmitry.torokhov@gmail.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Huang Rui <ray.huang@amd.com> Cc: John Stultz <john.stultz@linaro.org> Cc: Len Brown <lenb@kernel.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: kvm ML <kvm@vger.kernel.org> Cc: linux-input@vger.kernel.org Link: http://lkml.kernel.org/r/9e058e72f4cf1f13c6483c1360b39c3d188a2c2a.1434501121.git.luto@kernel.orgSigned-off-by:
-
Andy Lutomirski authored
This timing code is hideous, and this doesn't help. It gets rid of one of the last users of rdtscl(), though. Signed-off-by:
-
Andy Lutomirski authored
It wasn't compiled in by default. I suspect that the driver was and still is broken, though -- it's calling udelay with a parameter that's derived from loops_per_jiffy. Signed-off-by:
-
Andy Lutomirski authored
This is only used if BAYCOM_DEBUG is defined. Signed-off-by:
Walter Harms <wharms@bfs.de> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Huang Rui <ray.huang@amd.com> Cc: John Stultz <john.stultz@linaro.org> Cc: Len Brown <lenb@kernel.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: kvm ML <kvm@vger.kernel.org> Cc: linux-hams@vger.kernel.org Link: http://lkml.kernel.org/r/1195ce0c7f34169ff3006341b77806184a46b9bf.1434501121.git.luto@kernel.orgSigned-off-by:
-
Andy Lutomirski authored
This code is timing 100k indirect calls, so the added overhead of counting the number of cycles elapsed as a 64-bit number should be insignificant. Drop the optimization of using a 32-bit count. Signed-off-by:
-
Andy Lutomirski authored
As a very minor optimization, delay_tsc() was only using the low 32 bits of the TSC. It's a delay function, so just use the whole thing. Signed-off-by:
-
Andy Lutomirski authored
They have no users. Leave native_read_tscp() which seems potentially useful despite also having no callers. Signed-off-by:
-
Andy Lutomirski authored
Now that the ->read_tsc() paravirt hook is gone, rdtscll() is just a wrapper around native_read_tsc(). Unwrap it. Signed-off-by:
-
Andy Lutomirski authored
We've had ->read_tsc() and ->read_tscp() paravirt hooks since the very beginning of paravirt, i.e., d3561b7f ("[PATCH] paravirt: header and stubs for paravirtualisation"). AFAICT, the only paravirt guest implementation that ever replaced these calls was vmware, and it's gone. Arguably even vmware shouldn't have hooked RDTSC -- we fully support systems that don't have a TSC at all, so there's no point for a paravirt implementation to pretend that we have a TSC but to replace it. I also doubt that these hooks actually worked. Calls to rdtscl() and rdtscll(), which respected the hooks, were used seemingly interchangeably with native_read_tsc(), which did not. Just remove them. If anyone ever needs them again, they can try to make a case for why they need them. Before, on a paravirt config: text data bss dec hex filename 12618257 1816384 1093632 15528273 ecf151 vmlinux After: text data bss dec hex filename 12617207 1816384 1093632 15527223 eced37 vmlinux Signed-off-by:
-
Andy Lutomirski authored
The only caller was KVM's read_tsc(). The only difference between vget_cycles() and native_read_tsc() was that vget_cycles() returned zero instead of crashing on TSC-less systems. KVM already checks vclock_mode() before calling that function, so the extra check is unnecessary. Also, KVM (host-side) requires the TSC to exist. Signed-off-by:
Paolo Bonzini <pbonzini@redhat.com> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Borislav Petkov <bp@alien8.de> Cc: Brian Gerst <brgerst@gmail.com> Cc: Denys Vlasenko <dvlasenk@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Huang Rui <ray.huang@amd.com> Cc: John Stultz <john.stultz@linaro.org> Cc: Len Brown <lenb@kernel.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: kvm ML <kvm@vger.kernel.org> Link: http://lkml.kernel.org/r/20615df14ae2eb713ea7a5f5123c1dc4c7ca993d.1434501121.git.luto@kernel.orgSigned-off-by:
-
Andy Lutomirski authored
In the following commit: cdc7957d ("x86: move native_read_tsc() offline") ... native_read_tsc() was moved out of line, presumably for some now-obsolete vDSO-related reason. Undo it. The entire rdtsc, shl, or sequence is only 11 bytes, and calls via rdtscl() and similar helpers were already inlined. Signed-off-by:
-
