In commit 71b9a45330fe220d1 we changed the condition we use to
determine whether we need to refill the TLB in get_page_addr_code() to
if (unlikely(env->tlb_table[mmu_idx][index].addr_code !=
(addr & (TARGET_PAGE_MASK | TLB_INVALID_MASK)))) {
This isn't the right check (it will falsely fail if the
input addr happens to have the low bit corresponding to
TLB_INVALID_MASK set, for instance).
This patchset first factors out the "check for a hit" logic
into some new functions tlb_hit() and tlb_hit_page() (the
latter is for when the address is known to be page-aligned),
uses those functions in the various places that do TLB hit tests,
and then uses tlb_hit() to replace the erroneous code in
get_page_addr_code().
I noticed this while trying to debug Laurent's m68k test case:
it meant that we would come into get_page_addr_code() for a
TLB hit, falsely decide it was a miss, and then fish an
older entry out of the TLB victim cache...
Peter Maydell (2):
tcg: Define and use new tlb_hit() and tlb_hit_page() functions
accel/tcg: Correct "is this a TLB miss" check in get_page_addr_code()
accel/tcg/softmmu_template.h | 16 ++++++----------
include/exec/cpu-all.h | 23 +++++++++++++++++++++++
include/exec/cpu_ldst.h | 3 +--
accel/tcg/cputlb.c | 18 ++++++------------
4 files changed, 36 insertions(+), 24 deletions(-)
--
2.17.1
