Hi Saravana,
On Thu, May 19, 2022 at 10:00 PM Saravana Kannan <saravanak@xxxxxxxxxx> wrote:
> There's still work to be done that might make this easier/cleaner in the future:
> 1. Adding a DT property that explicitly marks device A as not
> dependent on B (Rob was already open to this -- with additional
> context I don't want to type up here).
> 2. Adding kernel command line options that might allow people to say
> stuff like "Device A doesn't depend on Device B independent of what DT
> might say".
There are clearly cases where the hardware defines if the property
is optional or required, and a DT property would work. However, in
general relaxing such dependencies involves a complex mix of hardware
capabilities, driver support, and system policies.
Examples:
- A hardware block may support both DMA and PIO, or can require DMA,
- DMA-capable devices can typically work without an IOMMU,
unless all RAM lies outside the address space addressable by
the DMA controller,
- A driver may fall back to PIO if DMA is not available (yet),
but doing so may not meet the required performance target,
- Not using the IOMMU may violate anti-tampering policies.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
