On Mon, 27 Apr 2020 12:54:06 +0100 John Garry <john.garry@xxxxxxxxxx> wrote: > On 27/04/2020 12:03, Jiaxun Yang wrote: > > > > > > 于 2020年4月27日 GMT+08:00 下午6:43:09, John Garry > > <john.garry@xxxxxxxxxx> 写到: > >> On 26/04/2020 12:47, Jiaxun Yang wrote: > >>> That would allow platforms reserve some lower address in PIO MMIO > >>> range to deal with legacy drivers with hardcoded I/O ports that > >>> can't be managed by logic_pio. > >> > >> Hi, > >> > >> Is there some reason why the logic_pio code cannot be improved to > >> handle these devices at these "fixed" addresses? Or do you have a > >> plan to improve it? We already support fixed bus address devices > >> in the INDIRECT IO region. > > > > Hi, > > > > The issue about "Fixed Address" is we can't control the ioport > > That driver used to operate devices. > > So any attempt to resolve it in logic_pio seems impossible. > > > > Currently we have i8259, i8042, piix4_smbus, mc146818 rely on this > > assumption. > > Right, and from glancing at a couple of drivers you mentioned, if we > were to register a logic pio region for that legacy region, there > does not seem to be an easy place to fixup to use logic pio addresses > (for those devices). They use hardcoded values. However if all those > drivers were mips specific, you could fixup those drivers to use > logic_pio addresses today through some macro. But not sure on that. > Well, most of these drivers are shared with x86 so.... I guess the conversion needs two or more release cycles. > > So, going back to your change, I have a dilemma wondering whether you > should still register a logic pio region for the legacy region > instead of the carveout reservation, but ensure it is the first > region registered, such that logic pio address base is 0 and no > translation is required. At least then you have a region registered > and it shows in /proc/ioports, but then this whole thing becomes a > bit fragile. Thanks for your solution. So I must register this range as early as possible. As IRQ is the first subsystem using ISA, I'll do it before IRQ init, just at the place I setup iormap for reserved region now. Should be early enough to avoid any collision, as the only logic_pio user on our system is PCI controller. Thanks. > -- Jiaxun Yang
