| Message ID | 20180802174042.29234-1-peter.maydell@linaro.org |
|---|---|
| Headers | show |
| Series | hw/net/pcnet-pci: Convert away from old_mmio accessors | expand |
On 08/02/2018 10:40 AM, Peter Maydell wrote: > Peter Maydell (2): > hw/net/pcnet-pci: Convert away from old_mmio accessors > hw/net/pcnet-pci: Unify pcnet_ioport_read/write and > pcnet_mmio_read/write Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
On 4 August 2018 at 02:20, Richard Henderson <richard.henderson@linaro.org> wrote: > On 08/02/2018 10:40 AM, Peter Maydell wrote: >> Peter Maydell (2): >> hw/net/pcnet-pci: Convert away from old_mmio accessors >> hw/net/pcnet-pci: Unify pcnet_ioport_read/write and >> pcnet_mmio_read/write > > Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Thanks for the review. I can put these in via target-arm, though they don't really have anything to do with arm, since nobody else seems very interested in this patchset... thanks -- PMM
This patchset converts the pcnet-pci's MMIO BAR MemoryRegionOps away from the old_mmio accessors. It does it in two stages: * patch 1 is a no-behaviour-change patch which replaces the old split byte/word/long accessor functions with single read and write functions which take the size, and look suspiciously similar but not identical to the existing pcnet_ioport_read/write functions * patch 2 then drops the new read and write functions and just uses the ioport read/write functions My reason for structuring it like this is that I'm pretty sure that the discrepancy between the MMIO BAR accessors and the IO BAR accessors is just a bug (one which we fixed for the IO BAR accessors in commit 7ba79741970 in 2011). But if I'm wrong I want us to be able to revert the behaviour change easily without that bringing back a use of the old_mmio accessors (which might not even compile if we need to revert after we've finally managed to drop those entirely). The bug as fixed in patch 2 is that for the MMIO BAR we were not honouring the DWIO bit (which indicates whether the device is in 16-bit or 32-bit IO mode) when doing accesses to the aprom range 0x0..0xf. We were already honouring DWIO for accesses to the 0x10..0x1f range, and my trawling through datasheets indicates that DWIO applies identically to MMIO and IO BARs (see patch 2's commit message for details and references). This is awkwardly hard to test, though, because Linux's pcnet driver only uses port IO as far as I can tell. (Likely this is why the bug has remained unnoticed for so long.) thanks -- PMM Peter Maydell (2): hw/net/pcnet-pci: Convert away from old_mmio accessors hw/net/pcnet-pci: Unify pcnet_ioport_read/write and pcnet_mmio_read/write hw/net/pcnet-pci.c | 98 +++------------------------------------------ hw/net/trace-events | 6 --- 2 files changed, 6 insertions(+), 98 deletions(-) -- 2.17.1