summaryrefslogtreecommitdiffstats
path: root/arch/sparc/kernel/tsb.S
diff options
context:
space:
mode:
authorDavid S. Miller <davem@davemloft.net>2013-02-20 07:34:10 +0100
committerDavid S. Miller <davem@davemloft.net>2013-02-20 18:46:08 +0100
commit0fbebed682ff2788dee58e8d7f7dda46e33aa10b (patch)
tree62aecd278c28c2f5007ece44c5cf34429807e8ae /arch/sparc/kernel/tsb.S
parentsparc64: Handle hugepage TSB being NULL. (diff)
downloadlinux-0fbebed682ff2788dee58e8d7f7dda46e33aa10b.tar.xz
linux-0fbebed682ff2788dee58e8d7f7dda46e33aa10b.zip
sparc64: Fix tsb_grow() in atomic context.
If our first THP installation for an MM is via the set_pmd_at() done during khugepaged's collapsing we'll end up in tsb_grow() trying to do a GFP_KERNEL allocation with several locks held. Simply using GFP_ATOMIC in this situation is not the best option because we really can't have this fail, so we'd really like to keep this an order 0 GFP_KERNEL allocation if possible. Also, doing the TSB allocation from khugepaged is a really bad idea because we'll allocate it potentially from the wrong NUMA node in that context. So what we do is defer the hugepage TSB allocation until the first TLB miss we take on a hugepage. This is slightly tricky because we have to handle two unusual cases: 1) Taking the first hugepage TLB miss in the window trap handler. We'll call the winfix_trampoline when that is detected. 2) An initial TSB allocation via TLB miss races with a hugetlb fault on another cpu running the same MM. We handle this by unconditionally loading the TSB we see into the current cpu even if it's non-NULL at hugetlb_setup time. Reported-by: Meelis Roos <mroos@ut.ee> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc/kernel/tsb.S')
-rw-r--r--arch/sparc/kernel/tsb.S39
1 files changed, 35 insertions, 4 deletions
diff --git a/arch/sparc/kernel/tsb.S b/arch/sparc/kernel/tsb.S
index d4bdc7a62375..a313e4a9399b 100644
--- a/arch/sparc/kernel/tsb.S
+++ b/arch/sparc/kernel/tsb.S
@@ -136,12 +136,43 @@ tsb_miss_page_table_walk_sun4v_fastpath:
nop
/* It is a huge page, use huge page TSB entry address we
- * calculated above.
+ * calculated above. If the huge page TSB has not been
+ * allocated, setup a trap stack and call hugetlb_setup()
+ * to do so, then return from the trap to replay the TLB
+ * miss.
+ *
+ * This is necessary to handle the case of transparent huge
+ * pages where we don't really have a non-atomic context
+ * in which to allocate the hugepage TSB hash table. When
+ * the 'mm' faults in the hugepage for the first time, we
+ * thus handle it here. This also makes sure that we can
+ * allocate the TSB hash table on the correct NUMA node.
*/
TRAP_LOAD_TRAP_BLOCK(%g7, %g2)
- ldx [%g7 + TRAP_PER_CPU_TSB_HUGE_TEMP], %g2
- cmp %g2, -1
- movne %xcc, %g2, %g1
+ ldx [%g7 + TRAP_PER_CPU_TSB_HUGE_TEMP], %g1
+ cmp %g1, -1
+ bne,pt %xcc, 60f
+ nop
+
+661: rdpr %pstate, %g5
+ wrpr %g5, PSTATE_AG | PSTATE_MG, %pstate
+ .section .sun4v_2insn_patch, "ax"
+ .word 661b
+ SET_GL(1)
+ nop
+ .previous
+
+ rdpr %tl, %g3
+ cmp %g3, 1
+ bne,pn %xcc, winfix_trampoline
+ nop
+ ba,pt %xcc, etrap
+ rd %pc, %g7
+ call hugetlb_setup
+ add %sp, PTREGS_OFF, %o0
+ ba,pt %xcc, rtrap
+ nop
+
60:
#endif