Attempt to avoid notdirty_write as much as possible. Please refer to #1838 for the context.
If there is no TB in the page, we can skip the slow path, so mark the TLB as dirty in such case. And PageDesc is allocated only when TB is generated, so we can apply the same logic for the case where PageDesc is not found.
If Unicorn generates a TB, we have to take the slow path for the page. In tb_page_add, if it's the first tb generated, tlb_protect_code will be called, and it will call cpu_physical_memory_test_and_clear_dirty. In this function, we reset the dirty flag so that Unicorn calls notdirty_write when memory is overwritten in the page.
Attempt to avoid
notdirty_write
as much as possible. Please refer to #1838 for the context.If there is no TB in the page, we can skip the slow path, so mark the TLB as
dirty
in such case. AndPageDesc
is allocated only when TB is generated, so we can apply the same logic for the case wherePageDesc
is not found.If Unicorn generates a TB, we have to take the slow path for the page. In
tb_page_add
, if it's the first tb generated,tlb_protect_code
will be called, and it will callcpu_physical_memory_test_and_clear_dirty
. In this function, we reset the dirty flag so that Unicorn callsnotdirty_write
when memory is overwritten in the page.Tested with
tests/regress/x86_self_modifying.py
.