| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Hand optimised asm code for ARC700 pipeline.
Originally written/optimized by Joern Rennecke
Signed-off-by: Vineet Gupta <[email protected]>
Cc: Joern Rennecke <[email protected]>
|
|
Signed-off-by: Vineet Gupta <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
|
|
* L1_CACHE_SHIFT
* PAGE_SIZE, PAGE_OFFSET
* struct pt_regs, struct user_regs_struct
* struct thread_struct, cpu_relax(), task_pt_regs(), start_thread(), ...
* struct thread_info, THREAD_SIZE, INIT_THREAD_INFO(), TIF_*, ...
* BUG()
* ELF_*
* Elf_*
To disallow user-space visibility into some of the core kernel data-types
such as struct pt_regs, #ifdef __KERNEL__ which also makes the UAPI header
spit (further patch in the series) to NOT export it to asm/uapi/ptrace.h
Signed-off-by: Vineet Gupta <[email protected]>
Cc: Jonas Bonn <[email protected]>
Cc: Al Viro <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
|
|
TBD: do_csum still needs to be written in asm
Signed-off-by: Vineet Gupta <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
|
|
Override asm-generic implementations. We basically gain on 2 fronts
* checks for alignment no longer needed as we are only doing "unit"
sized copies.
(Careful observer could argue that While the kernel buffers are aligned,
the user buffer in theory might not be - however in that case the
user space is already broken when it tries to deref a hword/word
straddling word boundary - so we are not making it any worse).
* __copy_{to,from}_user( ) returns bytes that couldn't be copied,
whereas get_user() returns 0 for success or -EFAULT (not size). Thus
the code to do leftover bytes calculation can be avoided as well.
The savings were significant: ~17k of code.
bloat-o-meter vmlinux_uaccess_pre vmlinux_uaccess_post
add/remove: 0/4 grow/shrink: 8/118 up/down: 1262/-18758 (-17496)
^^^^^^^^^
Signed-off-by: Vineet Gupta <[email protected]>
Acked-by: Arnd Bergmann <[email protected]>
|
|
Signed-off-by: Vineet Gupta <[email protected]>
|
|
This covers the UP / SMP (with no hardware assist for atomic r-m-w) as
well as ARC700 LLOCK/SCOND insns based.
Signed-off-by: Vineet Gupta <[email protected]>
|
|
ARC700 has an in-core intc which provides 2 priorities (a.k.a.) "levels"
of interrupts (per IRQ) hencforth referred to as L1/L2 interrupts.
CPU flags register STATUS32 has Interrupt Enable bits per level (E1/E2)
to globally enable (or disable) all IRQs at a level. Hence the
implementation of arch_local_irq_{save,restore,enable,disable}( )
The STATUS32 reg can be r/w only using the AUX Interface of ARC, hence
the use of LR/SR instructions. Further, E1/E2 bits in there can only be
updated using the FLAG insn.
The intc supports 32 interrupts - and per IRQ enabling is controlled by
a bit in the AUX_IENABLE register, hence the implmentation of
arch_{,un}mask_irq( ) routines.
Signed-off-by: Vineet Gupta <[email protected]>
Cc: Thomas Gleixner <[email protected]>
|
|
Arnd in his review pointed out that arch Kconfig organisation has several
deficiencies:
* Build time entries for things which can be runtime extracted from DT
(e.g. SDRAM size, core clk frequency..)
* Not multi-platform-image-build friendly (choice .. endchoice constructs)
* cpu variants support (750/770) is exclusive.
The first 2 have been fixed in subsequent patches.
Due to the nature of the 750 and 770, it is not possible to build for
both together, w/o special runtime glue code which would hurt
performance.
Signed-off-by: Vineet Gupta <[email protected]>
Cc: Arnd Bergmann <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Acked-by: Sam Ravnborg <[email protected]>
|
|
Signed-off-by: Vineet Gupta <[email protected]>
|