On 2/18/20 1:56 PM, Philippe Mathieu-Daudé wrote:
On 2/18/20 12:24 PM, Peter Maydell wrote:
The address_space_rw() function allows either reads or writes
depending on the is_write argument passed to it; this is useful
when the direction of the access is determined programmatically
(as for instance when handling the KVM_EXIT_MMIO exit reason).
Under the hood it just calls either address_space_write() or
address_space_read_full().
We also use it a lot with a constant is_write argument, though,
which has two issues:
* when reading "address_space_rw(..., 1)" this is less
immediately clear to the reader as being a write than
"address_space_write(...)"
* calling address_space_rw() bypasses the optimization
in address_space_read() that fast-paths reads of a
fixed length
This commit was produced with the included Coccinelle script
scripts/coccinelle/as-rw-const.patch.
Script is "scripts/coccinelle/as_rw_const.cocci".
I plan to respin this patch (fixed) in a larger series.
Two lines in hw/net/dp8393x.c that Coccinelle produced that
were over 80 characters were re-wrapped by hand.
Signed-off-by: Peter Maydell <peter.mayd...@linaro.org>
---
I could break this down into separate patches by submaintainer,
but the patch is not that large and I would argue that it's
better for the project if we can try to avoid introducing too
much friction into the process of doing 'safe' tree-wide
minor refactorings.
v1->v2: put the coccinelle script in scripts/coccinelle rather
than just in the commit message.
---
accel/kvm/kvm-all.c | 6 +--
dma-helpers.c | 4 +-
exec.c | 4 +-
hw/dma/xlnx-zdma.c | 11 ++---
hw/net/dp8393x.c | 68 ++++++++++++++--------------
hw/net/i82596.c | 25 +++++-----
hw/net/lasi_i82596.c | 5 +-
hw/ppc/pnv_lpc.c | 8 ++--
hw/s390x/css.c | 12 ++---
qtest.c | 52 ++++++++++-----------
target/i386/hvf/x86_mmu.c | 12 ++---
scripts/coccinelle/as_rw_const.cocci | 30 ++++++++++++
12 files changed, 133 insertions(+), 104 deletions(-)
create mode 100644 scripts/coccinelle/as_rw_const.cocci
