Subversion Repositories HelenOS-historic

Rev

Go to most recent revision | Show changed files | Directory listing | RSS feed

Filtering Options

Rev Age Author Path Log message Diff
796 6694 d 8 h palkovsky /kernel/trunk/ Gxemul support for mips userspace. MIPS now _must_ be started
always with userspace, there is (almost) no way how to pass parameters
to kernel.
 
794 6694 d 9 h palkovsky /kernel/trunk/ Added MIPS userspace. Currently can't make it work in gxemul.  
793 6694 d 10 h jermar /kernel/trunk/ Use hash_table_get_instance instead of list_get_instance.
Rename page_operations to page_mapping_operations.
Rename page_pt_operations to pt_mapping_operations.
Rename page_ht_operations to ht_mapping_operations.
 
792 6694 d 10 h jermar /kernel/trunk/ Page hash table architectures now use generic hash table to manage
mappings.
 
791 6694 d 23 h palkovsky /kernel/trunk/generic/ better utilization of memory in slab alloc
removed malloc-8 slab
 
790 6695 d 3 h jermar /kernel/trunk/ Generic chaining hash table.  
789 6695 d 3 h palkovsky /kernel/trunk/generic/ Early SLAB initialization, the cpu-cache is initialized later.
If you want to use slab_cache_create befor slab_cpu_enable, add
a flag SLAB_CACHE_MAGDEFERRED.
 
788 6695 d 7 h jermar /kernel/trunk/ Move list and fifo data types to adt/.  
787 6695 d 8 h palkovsky /kernel/trunk/generic/ Added constructor/destructor calls to SLAB.
Changed allocation of thread_t structure to use SLAB.
 
786 6695 d 9 h bondari /kernel/trunk/ frame_alloc() functions variations in order to avoid problems with frame_alloc_generic() changes.
IMPORTANT: Parameter order for frame_alloc() was changed
 
785 6695 d 10 h jermar /kernel/trunk/generic/src/ Comments clarification.  
784 6695 d 10 h palkovsky /kernel/trunk/generic/src/proc/ Cleanup of scheduler code  
783 6695 d 11 h palkovsky /kernel/trunk/generic/ Fixed some typos in slab allocator.
Scheduler now has better algorithm on load balancing.
Unfortunately it reveals deadlock in slab allocator :-/
 
782 6695 d 12 h palkovsky /kernel/trunk/generic/src/mm/ Small locking performance tweak.  
781 6695 d 23 h palkovsky /kernel/trunk/generic/ Still better locking - does not hold global spinlock when
destroying cache.
 
780 6696 d 3 h palkovsky /kernel/trunk/ Small fix in slab allocator for large-sized slabs.  
779 6696 d 4 h jermar /kernel/trunk/generic/ Simplify load balancing activation.
This in fact removes the livelock that prevented cpu_sleep() to be called in SMP systems.
 
778 6696 d 6 h palkovsky /kernel/trunk/ Fixed boot process not to call malloc when slab not initialized.
Added simics workaround.
 
777 6696 d 6 h palkovsky /kernel/trunk/generic/ added thread list to kconsole.  
776 6696 d 10 h palkovsky /kernel/trunk/generic/ Added more granular locking to slab allocator and thus fix
hopefully last race condition.
 
775 6696 d 11 h palkovsky /kernel/trunk/generic/ Added scheduler queues output. The scheduler is buggy - on SMP
the cpus never get tu cpu_sleep, in slab2 test on 4 cpus everything
is on the first cpu.
The slab allocator passes tests in this configuration, but in slightly
different(more efficient) locking order it panics. TODO: Find out why
does it panic.
 
773 6696 d 23 h palkovsky /kernel/trunk/ Added slab stress test.
Fixed race condition in slab allocator.
Moved initialization of slab to the point where we know correct config.cpu_count
Correctly passes tests on amd64, ia32 in both bochs & qemu.
 
772 6697 d 0 h palkovsky /kernel/trunk/ Fixed non-initialized cpu-cache.  
771 6697 d 1 h palkovsky /kernel/trunk/generic/ Implemented malloc/free as SLABs.  
770 6697 d 8 h palkovsky /kernel/trunk/generic/include/ Fix context_save, so that it works even with inlines disabled (-O0).  
769 6697 d 22 h palkovsky /kernel/trunk/ SLAB allocator now uses itself for all its internal structures.
Added description of allocator.
Removed messy_stack_trace from amd64, as it would scroll away important
part of exception.
 
768 6698 d 1 h palkovsky /kernel/trunk/ Debugged slab allocator. It currently supports per-CPU cache on 1 cpu.  
767 6698 d 2 h palkovsky /kernel/trunk/ Partially working SLAB CPU cache.  
766 6698 d 3 h palkovsky /kernel/trunk/ Basic SLAB (without CPU-cache) passes test.  
765 6698 d 3 h jermar /kernel/trunk/ Add PAGE_WIDTH to aid divisions by PAGE_SIZE.  
764 6698 d 8 h palkovsky /kernel/trunk/ Tested basic non-cached slab allocation.  
763 6698 d 10 h jermar /kernel/trunk/ Modify frame.h to use shifting instead of multiplication and division.
Define FRAME_WIDTH for all architectures.
 
762 6698 d 11 h palkovsky /kernel/trunk/ Currently not-working SLAB allocator.
- slightly changed interface to frame_alloc, allow zone preference
 
761 6698 d 12 h jermar /kernel/trunk/doc/ Remove document describing more relaxed preemption model than the one used in the current kernel.  
760 6698 d 12 h jermar /kernel/trunk/ mips32 is not supposed to allocate page table.
This is done by the generic code now.
Remove PTL0 pointer as it is not needed.

Remove GET_PTL0_ADDRESS from kernel.

Update sparc64 comments in barrier.h.
 
759 6698 d 23 h palkovsky /kernel/trunk/ Skeleton of SLAB allocator.  
758 6699 d 2 h jermar /kernel/trunk/arch/ sparc64 bugfix.
When disabling IMMU and DMMU the kernel has to perform synchronization operation
(e.g flush %r or membar #Sync instruction). There is no guarantee that the address
contained in %r is in DTLB and therefore the flush instruction can fault. Normally
this would be recognized and fixed by the OpenFirmware Fast Data MMU fault handler.
However, this handler lives in virtually mapped memory and an attempt to execute
there while the MMUs are disabled would result in a nested trap leading to error state.
Replacing flush %r instruction with membar #Sync, wich is sufficient in this case,
fixes this problem.
 
757 6699 d 4 h jermar /kernel/trunk/ Blacklist addresses between 0xa0000 and 0xfffff on ia64 for frame allocator.
This area contains VGA text frame buffer and should be avoided.
falloc2 test now passes on ia64.
 
756 6700 d 1 h jermar /kernel/trunk/ Memory management work.
Remove the last (i.e. 'root') argument from page_mapping_insert() and page_mapping_find().
Page table address is now extracted from the first (i.e. 'as') argument.
Add a lot of infrastructure to make the above possible.
sparc64 is now broken, most likely because of insufficient identity mapping of physical memory.
 
755 6701 d 0 h jermar /kernel/trunk/ Change page_mapping_find/insert interfaces to take as_t * as first argument
and not asid_t as second argument. This change was necessitated by the
removal of mapping array from as_area_t and the fact that an address
space doesn't have an ASID when it is created.