- 29 Aug, 2017 38 commits
-
-
Thomas Gleixner authored
Initialize the IST based traps via a table. Signed-off-by:
Thomas Gleixner <tglx@linutronix.de> Cc: Andy Lutomirski <luto@kernel.org> 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: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/20170828064959.091328949@linutronix.deSigned-off-by:
Ingo Molnar <mingo@kernel.org>
-
Thomas Gleixner authored
Add the debug_idt init table and make use of it. Signed-off-by:
-
Thomas Gleixner authored
Add the initialization table for the early trap setup and replace the early trap init code. Signed-off-by:
-
Thomas Gleixner authored
The IDT setup code is handled in several places. All of them use variants of set_intr_gate() inlines. This can be done with a table based initialization, which allows to reduce the inline zoo and puts all IDT related code and information into a single place. Add the infrastructure. Signed-off-by:
-
Thomas Gleixner authored
The early IDT setup can be done in C code like it's done on 64-bit kernels. Reuse the 64-bit version. Signed-off-by:
-
Thomas Gleixner authored
The early IDT handler setup is done in C entry code on 64-bit kernels and in ASM entry code on 32-bit kernels. Move the 64-bit variant to the IDT code so it can be shared with 32-bit in the next step. Signed-off-by:
-
Thomas Gleixner authored
kexec and reboot have both code to invalidate IDT. Create a common function and use it. Signed-off-by:
-
Thomas Gleixner authored
This inline is not used at all. Signed-off-by:
-
Thomas Gleixner authored
32-bit kernels have the idt_descr defined in the low level assembly entry code, but there is no good reason for that. Move it into the C file and use the 64-bit version of it. Signed-off-by:
-
Thomas Gleixner authored
IDT related code lives scattered around in various places. Create a new source file in arch/x86/kernel/idt.c to hold it. Move the idt_tables and descriptors to it for a start. Follow up patches will gradually move more code over. Signed-off-by:
-
Thomas Gleixner authored
Like the IDT descriptors, the LDT/TSS descriptors are pointlessly different on 32 and 64 bit kernels. Unify them and get rid of the duplicated code. Signed-off-by:
-
Thomas Gleixner authored
The GDT entry related code uses two ways to access entries via union fields: - bitfields - macros which initialize the two 16-bit parts of the entry by magic shift and mask operations. Clean it up and only use the bitfields to initialize and access entries. ( The old access patterns were partly done due to GCC optimizing bitfield accesses in a horrible way - that's mostly fixed these days and clarity of code in such low level accessors is very important. ) Signed-off-by:
-
Thomas Gleixner authored
The union inside of desc_struct which allows access to the raw u32 parts of the descriptors. This raw access part is about to go away. Replace the few code parts which access those fields. Signed-off-by:
Boris Ostrovsky <boris.ostrovsky@oracle.com> Cc: Andy Lutomirski <luto@kernel.org> 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: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Juergen Gross <jgross@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/20170828064958.120214366@linutronix.deSigned-off-by:
-
Thomas Gleixner authored
desc_struct is a union of u32 fields and bitfields. The access to the u32 fields is done with magic macros. Convert it to use the bitfields and replace the macro magic with parseable inline functions. Signed-off-by:
-
Thomas Gleixner authored
The IDT cleanup is about to remove pack_descriptor(). The GDT setup for the per-cpu storage can be achieved with the static initializer as well. Replace it. Signed-off-by:
-
Thomas Gleixner authored
The first 32 bits of gate struct are the same for 32 and 64 bit kernels. The 32-bit version uses desc_struct and no designated data structure, so we need different accessors for 32 and 64 bit kernels. Aside of that the macros which are necessary to build the 32-bit gate descriptor are horrible to read. Unify the gate structs and switch all code fiddling with it over. Signed-off-by:
-
Thomas Gleixner authored
The tracepoint macro magic emits code for all tracepoints in a event header file. That code stays around even if the tracepoint is not used at all. The linker does not discard it. Build the various irq_vector tracepoints dependent on the appropriate CONFIG switches. Signed-off-by:
-
Thomas Gleixner authored
The irq work interrupt vector is only installed when CONFIG_X86_LOCAL_APIC is enabled, but the interrupt handler is compiled in unconditionally. Compile the cruft out when the APIC is disabled. Signed-off-by:
-
Thomas Gleixner authored
The platform IPI vector is only installed when the local APIC is enabled. All users of it depend on the local APIC anyway. Make the related code conditional on CONFIG_X86_LOCAL_APIC=y. Signed-off-by:
-
Thomas Gleixner authored
The pagefault and the resched IPI handler are the only ones where it is worth to optimize the code further in case tracepoints are disabled. But it makes no sense to have a single static key for both. Seperate the static keys so the facilities are handled seperately. Signed-off-by:
-
Thomas Gleixner authored
Some of the entry function defines for i386 were explictely using the BUILD_INTERRUPT3() macro to prevent that the extra trace entry got added via BUILD_INTERRUPT(). No that the trace cruft is gone, the file can be cleaned up and converted to use BUILD_INTERRUPT() which avoids the ugly line breaks. Signed-off-by:
-
Thomas Gleixner authored
No more users of the tracing IDT. All exception tracepoints have been moved into the regular handlers. Get rid of the mess which shouldn't have been created in the first place. Signed-off-by:
-
Thomas Gleixner authored
It's worth to avoid the extra irq_enter()/irq_exit() pair in the case that the reschedule interrupt tracepoints are disabled. Use the static key which indicates that exception tracing is enabled. For now this key is global. It will be optimized in a later step. Signed-off-by:
Steven Rostedt (VMware) <rostedt@goodmis.org> Cc: Andy Lutomirski <luto@kernel.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/20170828064957.299808677@linutronix.deSigned-off-by:
-
Thomas Gleixner authored
Two NOP5s are really a good tradeoff vs. the unholy IDT switching mess, which duplicates code all over the place. The rescheduling interrupt gets optimized in a later step. Make the ordering of function call and statistics increment the same as in other places. Calculate stats first, then do the function call. Signed-off-by:
-
Thomas Gleixner authored
Machine checks are not really high frequency events. The extra two NOP5s for the disabled tracepoints are noise vs. the heavy lifting which needs to be done in the MCE handler. Signed-off-by:
-
Thomas Gleixner authored
Two NOP5s are a reasonable tradeoff to avoid duplicated code and the requirement to switch the IDT. Signed-off-by:
-
Thomas Gleixner authored
The error and the spurious interrupt are really rare events and not at all performance sensitive: two NOP5s can be tolerated when tracing is disabled. Remove the complication. Signed-off-by:
-
Thomas Gleixner authored
Two NOP5s are really a good tradeoff vs. the unholy IDT switching mess, which duplicates code all over the place. Signed-off-by:
-
Thomas Gleixner authored
Accessing the per cpu data via per_cpu(, smp_processor_id()) is pointless. Use this_cpu_ptr() instead. Signed-off-by:
-
Thomas Gleixner authored
The two NOP5s are noise in the rest of the work which is done by the timer interrupt and modern CPUs are pretty good in optimizing NOPs anyway. Get rid of the interrupt handler duplication and move the tracepoints into the regular handler. Signed-off-by:
-
Thomas Gleixner authored
Make use of the new irqvector tracing static key and remove the duplicated trace_do_pagefault() implementation. If irq vector tracing is disabled, then the overhead of this is a single NOP5, which is a reasonable tradeoff to avoid duplicated code and the unholy macro mess. Signed-off-by:
-
Thomas Gleixner authored
Switching the IDT just for avoiding tracepoints creates a completely impenetrable macro/inline/ifdef mess. There is no point in avoiding tracepoints for most of the traps/exceptions. For the more expensive tracepoints, like pagefaults, this can be handled with an explicit static key. Preparatory patch to remove the tracing IDT. Signed-off-by:
-
Thomas Gleixner authored
EISA has absolutely nothing to do with traps, so move it out of traps.c into its own eisa.c file. Furthermore, the EISA bus detection does not need to run during very early boot, it's good enough to run it before the EISA bus and drivers are initialized. I.e. instead of calling it from the very early trap_init() code, make it a subsys_initcall(). Signed-off-by:
-
Thomas Gleixner authored
Also remove the unparseable comment in the other place while at it. Signed-off-by:
-
Thomas Gleixner authored
This variable is beyond pointless. Nothing allocates a vector via alloc_gate() below FIRST_SYSTEM_VECTOR. So nothing can change first_system_vector. If there is a need for a gate below FIRST_SYSTEM_VECTOR then it can be added to the vector defines and FIRST_SYSTEM_VECTOR can be adjusted accordingly. Signed-off-by:
-
Thomas Gleixner authored
No modular users. Signed-off-by:
-
Thomas Gleixner authored
Last user (lguest) is gone. Remove it. Signed-off-by:
-
Ingo Molnar authored
Signed-off-by:
-
- 28 Aug, 2017 2 commits
-
-
Linus Torvalds authored
Commit 3510ca20 ("Minor page waitqueue cleanups") made the page queue code always add new waiters to the back of the queue, which helps upcoming patches to batch the wakeups for some horrid loads where the wait queues grow to thousands of entries. However, I forgot about the nasrt add_page_wait_queue() special case code that is only used by the cachefiles code. That one still continued to add the new wait queue entries at the beginning of the list. Fix it, because any sane batched wakeup will require that we don't suddenly start getting new entries at the beginning of the list that we already handled in a previous batch. [ The current code always does the whole list while holding the lock, so wait queue ordering doesn't matter for correctness, but even then it's better to add later entries at the end from a fairness standpoint ] Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Tejun Heo authored
When !NUMA, cpumask_of_node(@node) equals cpu_online_mask regardless of @node. The assumption seems that if !NUMA, there shouldn't be more than one node and thus reporting cpu_online_mask regardless of @node is correct. However, that assumption was broken years ago to support DISCONTIGMEM and whether a system has multiple nodes or not is separately controlled by NEED_MULTIPLE_NODES. This means that, on a system with !NUMA && NEED_MULTIPLE_NODES, cpumask_of_node() will report cpu_online_mask for all possible nodes, indicating that the CPUs are associated with multiple nodes which is an impossible configuration. This bug has been around forever but doesn't look like it has caused any noticeable symptoms. However, it triggers a WARN recently added to workqueue to verify NUMA affinity configuration. Fix it by reporting empty cpumask on non-zero nodes if !NUMA. Signed-off-by:
Tejun Heo <tj@kernel.org> Reported-and-tested-by:
Geert Uytterhoeven <geert@linux-m68k.org> Cc: stable@vger.kernel.org Signed-off-by:
-
