The actual tracking of dirty bitmap updates is happening in architecture code. So far, the aarch64 hvf code has not updated QEMU's dirty bitmap at all. The net result of that is that the VGA device's framebuffer would not update.
This patch adds simplistic dirty bitmap updates. Unfortunately hvf can only set permissions per full region, so we have to mark the complete region as dirty when only a single byte was modified inside. We also handle the write protect update logic before we handle any writes. This allows us to even handle non-ISV instructions on dirty logging enabled memory regions: Once we flip the region to writable again, we just rerun the trapping instruction. Signed-off-by: Alexander Graf <[email protected]> --- target/arm/hvf/hvf.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) diff --git a/target/arm/hvf/hvf.c b/target/arm/hvf/hvf.c index 0dc96560d3..92ad0d29c4 100644 --- a/target/arm/hvf/hvf.c +++ b/target/arm/hvf/hvf.c @@ -1163,6 +1163,28 @@ int hvf_vcpu_exec(CPUState *cpu) break; } + /* + * Dirty log updates work without isv as well. We just run the write + * again with write permissions set. So handle them before the assert. + */ + if (iswrite) { + uint64_t gpa = hvf_exit->exception.physical_address; + hvf_slot *slot = hvf_find_overlap_slot(gpa, 1); + + if (slot && slot->flags & HVF_SLOT_LOG) { + /* + * HVF can only set a full region's permissions, so let's just + * mark the full region as dirty. + */ + memory_region_set_dirty(slot->region, 0, slot->size); + hv_vm_protect(slot->start, slot->size, HV_MEMORY_READ | + HV_MEMORY_WRITE | HV_MEMORY_EXEC); + + /* Run the same instruction again, without write faulting */ + break; + } + } + assert(isv); if (iswrite) { -- 2.32.0 (Apple Git-132)
