# Re: \[RFT PATCH v3 12/27\] of/address: Add infrastructure to declare MMIO as non-posted

- From: Hector Martin \<hidden\>
- Date: 2021-03-05 17:20:15 
- List: linux-devicetree
- Page: https://ratatoskr.run/linux-devicetree/2021/03/6290202
- Thread: https://ratatoskr.run/linux-arm-kernel/2021/03/6288800/t.md

```text
On 06/03/2021 01.43, Arnd Bergmann wrote:
> - setting ioremap() on PCI buses non-posted only makes them
>    only slower but not more reliable, because the non-posted flag
>    on the bus is discarded by the PCI host bridge.

Note that this doesn't work here *anyway*. The fabric is picky in both directions: thou shalt use nGnRnE for on-SoC MMIO and nGnRE for PCIe windows, or else, SError.

Since these devices can support *any* PCI device via Thunderbolt, making PCI drivers be the oddball ones needing special APIs would mean hundreds of changes needed - the vast majority of PCI drivers in the kernel use plain ioremap variants that don't have any flags to look at.

-- 
Hector Martin (marcan@marcan.st)
Public Key: https://mrcn.st/pub
```
