- 19 May, 2015 40 commits
-
-
Ingo Molnar authored
Introduce a simple fpu->fpstate_active flag in the fpu context data structure and use that instead of PF_USED_MATH in task->flags. Testing for this flag byte should be slightly more efficient than testing a bit in a bitmask, but the main advantage is that most FPU functions can now be performed on a 'struct fpu' alone, they don't need access to 'struct task_struct' anymore. There's a slight linecount increase, mostly due to the 'fpu' local variables and due to extra comments. The local variables will go away once we move most of the FPU methods to pure 'struct fpu' parameters. Reviewed-by:
Borislav Petkov <bp@alien8.de> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Signed-off-by:
Ingo Molnar <mingo@kernel.org>
-
Ingo Molnar authored
Explain its usage and also document a TODO item. Reviewed-by:
-
Ingo Molnar authored
PF_USED_MATH is used directly, but also in a handful of helper inlines. To ease the elimination of PF_USED_MATH, convert all inline helpers to open-coded PF_USED_MATH usage. Reviewed-by:
-
Ingo Molnar authored
Migrate this function to pure 'struct fpu' usage. Reviewed-by:
-
Ingo Molnar authored
Migrate this function to pure 'struct fpu' usage. Reviewed-by:
-
Ingo Molnar authored
Migrate this function to pure 'struct fpu' usage. Reviewed-by:
-
Ingo Molnar authored
Track the FPU owner context instead of the owner task: this change, together with other changes, will allow in subsequent patches the elimination of 'struct task_struct' usage in various FPU code: we'll be able to use 'struct fpu' only. There's no change in code size: text data bss dec hex filename 13066467 2545248 1626112 17237827 1070743 vmlinux.before 13066467 2545248 1626112 17237827 1070743 vmlinux.after Reviewed-by: -
Ingo Molnar authored
Move it closer to other per-cpu FPU data structures. This also unifies the 32-bit and 64-bit code. Reviewed-by:
-
Ingo Molnar authored
We do this to make the code more readable, and also to be able to eliminate task_struct usage from most of the FPU code. Reviewed-by:
-
Ingo Molnar authored
Start migrating FPU methods towards using 'struct fpu *fpu' directly. __thread_has_fpu() is just a trivial wrapper around fpu->has_fpu, eliminate it. Reviewed-by:
-
Ingo Molnar authored
Ever since the kernel started defaulting to eager FPU switches on modern Intel CPUs it's not been obvious whether a given system is using the lazy or the eager FPU context switching logic. So generate a boot message about which mode the FPU code is in: x86/fpu: Using 'lazy' FPU context switches. or: x86/fpu: Using 'eager' FPU context switches. Reviewed-by:
-
Ingo Molnar authored
Also add a bit of documentation. Reviewed-by:
-
Ingo Molnar authored
Move fpu_copy() where its only user is. Beyond readability this also speeds up compilation, as fpu-internal.h is included in over a dozen .c files. Reviewed-by:
-
Ingo Molnar authored
__save_init_fpu() is just a trivial wrapper around fpu_save_init(). Remove the extra layer of obfuscation. Reviewed-by:
-
Ingo Molnar authored
Instead of open-coded in_kernel_fpu access, Use kernel_fpu_disabled() in interrupted_kernel_fpu_idle(), matching the other kernel_fpu_*() methods. Also add some documentation for in_kernel_fpu. Reviewed-by:
-
Ingo Molnar authored
We are not supposed to call kernel_fpu_disable() if we have not previously enabled it. Also use kernel_fpu_disable()/enable() in the __kernel_fpu_begin/end() primitives, instead of writing to in_kernel_fpu directly, so that we get the debugging checks. Reviewed-by:
-
Ingo Molnar authored
This allows the compiler to inline them and to eliminate them: arch/x86/kernel/fpu/core.o: text data bss dec hex filename 6741 4 8 6753 1a61 core.o.before 6716 4 8 6728 1a48 core.o.after Reviewed-by:
-
Ingo Molnar authored
It's now local to fpu/core.c, make it static. Reviewed-by:
-
Ingo Molnar authored
Reviewed-by:
-
Ingo Molnar authored
Introduce fpu__copy() and use it in arch_dup_task_struct(), thus moving another chunk of FPU logic to fpu/core.c. Reviewed-by:
-
Ingo Molnar authored
This code was historically in process.c, now we have FPU core internals in fpu/core.c instead - move it there. Reviewed-by:
-
Ingo Molnar authored
It's a pointless wrapper now - use xsave_state(). Reviewed-by:
-
Ingo Molnar authored
These functions (xsave_state() and xsave_state_booting()) have a 'mask' argument that is always -1. Propagate this into the functions instead and eliminate the extra argument. Does not change the generated code, because these were inlined functions. Reviewed-by:
-
Ingo Molnar authored
Move the boot-time FPU bug detection code to the other FPU boot time init code in fpu/init.c. No change in code size: text data bss dec hex filename 13044568 1884440 1130496 16059504 f50c70 vmlinux.before 13044568 1884440 1130496 16059504 f50c70 vmlinux.after Reviewed-by:
-
Ingo Molnar authored
Move to the new fpu__*() namespace. Reviewed-by:
-
Ingo Molnar authored
It's another piece of FPU internals that is better off close to the other FPU internals. Reviewed-by:
-
Ingo Molnar authored
flush_thread() open codes a lot of FPU internals - create a separate function for it in fpu/core.c. Turns out that this does not hurt performance: text data bss dec hex filename 11843039 1884440 1130496 14857975 e2b6f7 vmlinux.before 11843039 1884440 1130496 14857975 e2b6f7 vmlinux.after and since this is a slowpath clarity comes first anyway. We can reconsider inlining decisions after the FPU code has been cleaned up. Reviewed-by:
-
Ingo Molnar authored
Use fpstate_free() directly to manage FPU state. Only process.c was using this method, so this is a speedup as well, as it removes the extra function call and related clobbers. Reviewed-by:
-
Ingo Molnar authored
Both no_387() and fpu__detect() run at boot time, so they belong into init.c. Reviewed-by:
-
Ingo Molnar authored
fpu/core.c includes a lot of files for mostly historic reasons. It only needs fpu-internal.h, which already includes all the required headers. Reviewed-by:
-
Ingo Molnar authored
Move boot time FPU initialization code into init.c, to better isolate it into its own domain. Reviewed-by:
-
Ingo Molnar authored
Fix a minor header file dependency bug in asm/fpu-internal.h: it relies on i387.h but does not include it. All users of fpu-internal.h included it explicitly. Also remove unnecessary includes, to reduce compilation time. This also makes it easier to use it as a standalone header file for FPU internals, such as an upcoming C module in arch/x86/kernel/fpu/. Reviewed-by:
-
Ingo Molnar authored
Create a new subdirectory for the FPU support code in arch/x86/kernel/fpu/. Rename 'i387.c' to 'core.c' - as this really collects the core FPU support code, nothing i387 specific. We'll better organize this directory in later patches. Reviewed-by:
-
Ingo Molnar authored
- add header guards - standardize vertical alignment - add comments about MPX No code changed. Reviewed-by:
-
Ingo Molnar authored
Move the FPU details to asm/fpu_types.h, to further factor out the FPU code. ( As an added bonus, the 'struct orig_ist' definition now moves next to its other data types - the FPU definitions were slapped in the middle of them for some mysterious reason. ) No code changed. Reviewed-by:
-
Ingo Molnar authored
This was pretty hard to read, improve it. Reviewed-by:
-
Ingo Molnar authored
This field is kept separate from the main FPU state structure for no good reason. Reviewed-by:
-
Ingo Molnar authored
So init_thread_xstate() is a misnomer in that it's not really related to a specific thread - it determines, once during initial bootup, the size of the xstate context. Also improve the comments. Reviewed-by:
-
Ingo Molnar authored
fpu_init() is a bit of a misnomer in that it (falsely) creates the impression that it's related to the (old) fpu_finit() function, which initializes FPU ctx state. Rename it to fpu__cpu_init() to make its boot time initialization clear, and to move it to the fpu__*() namespace. Also fix and extend its comment block to point out that it's called not only on the boot CPU, but on secondary CPUs as well. Reviewed-by:
-
Ingo Molnar authored
Make it clear that we are initializing the in-memory FPU context area, no the FPU registers. Also move it to the fpu__*() namespace. Reviewed-by:
-
