Changeset c520034 in mainline for kernel/generic/src/proc/thread.c

Timestamp:

2011-12-31T18:19:35Z (13 years ago)

Author:

Jakub Jermar <jakub@…>

Branches:

lfn, master, serial, ticket/834-toolchain-update, topic/msim-upgrade, topic/simplify-dev-export

Children:

295f658, 77c2b02, 96cd5b4

Parents:

852052d (diff), 22f0561 (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Support for kernel non-identity mappings, phase I.

• identity/non-identity kernel memory split on all architectures
• low/high physical memory split on all architectures
• frame allocator understands low/high memory
• high physical memory currently unused (Phase II)
• more compact frame_t
• zone conf frames, pte_t, kernel stacks allocated from low memory
• lockless TLB-miss handlers everywhere (newly sparc64, ia64)
• preallocate PTL1 page tables for non-identity and prevent their deallocation
• hw_map() unification
• new resource allocator used for allocating kernel virtual addresses

Breakage:

• sparc64/sun4v creates too large kernel identity; not fixed because of lack of testing hw
• ppc32's tlb_refill() seems wrong as it creates too large kernel identity, but appears unused and the architecture works normally

Not implemented yet (phase II):

• allow high memory to be used for common kernel allocations

File:

• kernel/generic/src/proc/thread.c (modified) (1 diff)

kernel/generic/src/proc/thread.c

@@ -173 +173 @@
 #endif /* CONFIG_FPU */
         
+        /*
+         * Allocate the kernel stack from the low-memory to prevent an infinite
+         * nesting of TLB-misses when accessing the stack from the part of the
+         * TLB-miss handler written in C.
+         *
+         * Note that low-memory is safe to be used for the stack as it will be
+         * covered by the kernel identity mapping, which guarantees not to
+         * nest TLB-misses infinitely (either via some hardware mechanism or
+         * by the construciton of the assembly-language part of the TLB-miss
+         * handler).
+         *
+         * This restriction can be lifted once each architecture provides
+         * a similar guarantee, for example by locking the kernel stack
+         * in the TLB whenever it is allocated from the high-memory and the
+         * thread is being scheduled to run.
+         */
+        kmflags |= FRAME_LOWMEM;
+        kmflags &= ~FRAME_HIGHMEM;
+
         thread->kstack = (uint8_t *) frame_alloc(STACK_FRAMES, FRAME_KA | kmflags);
         if (!thread->kstack) {