[ros-dev] Merging our x86 HALs

[Thomas Faber]

Now the only problem is that neither our APIC nor MP HAL actually
work...


On 2017-12-12 18:31, David Quintana (gigaherz) wrote:
> I have to agree that reducing it to 2 HALs (one ACPI with multiprocessor
> and such, that maybe is also used for single-cpu systems with ACPI), and a
> legacy one for systems unable to handle ACPI+MP, sounds like a great idea.
> 
> On 12 December 2017 at 18:13, Javier Agustìn Fernàndez Arroyo <
> elhoir at gmail.com> wrote:
> 
>> Win8 does not support old hardware as ReactOS do!
>>
>> El 12 dic. 2017 17:52, "Alex Ionescu" <ionucu at videotron.ca> escribió:
>>
>>> I would move to the Win8+ HAL Model -- a single HAL for APIC, ACPI with
>>> runtime support for UEFI (if present) and MP (if present).
>>>
>>> If people still want to run on a PIC VM (why???) or old computer, then we
>>> can also maintain the HAL PIC x86 for UP.
>>>
>>> Hence there would only be 2 HALs.
>>>
>>> Best regards,
>>> Alex Ionescu
>>>
>>> On Mon, Dec 11, 2017 at 1:07 AM, Colin Finck <colin at reactos.org> wrote:
>>>
>>>> Am 11.12.2017 um 01:18 schrieb Hermès BÉLUSCA-MAÏTO:> If you basically
>>>> put all the HALs into one, then you obtain bloated stuff (which remains
>>>> in memory for the whole life of the OS). Example: standard HAL is 1MB
>>>> vs. ACPI HAL which is few kBHave you actually checked what makes up this
>>>> difference?
>>>> Hint: hal/halx86/legacy/bus/pci_vendors.ids
>>>>
>>>>
>>>>> Note that if Windows nowadays has only one hal, it's because they now
>>>> support basically only one "architecture"/platform, namely, ACPI
>>>> multiprocessor (to put it simple). It has its pros, but also a lot of cons.
>>>>
>>>> That doesn't mean we need to do the same. We can have one HAL for all
>>>> (Pentium and newer) x86 platforms. The overhead of additional checks at
>>>> boot-up is negligible. That should be a solution for 99% of the people
>>>> out there. The rest may still go and trim down our HAL to their needs.
>>>>
>>>> But let's not pretend we can maintain multiple x86 HALs for all x86
>>>> computers out there. Do you really want to test X HALs with Y different
>>>> systems? Ensure that a legacy HAL runs on a modern ACPI system? What
>>>> would be the point?
>>>>
>>>>
>>>>> Besides this, I've a question about your observation that in the APIC
>>>> hal (not ACPI) there's different implementation of
>>>> HalpCalibrateStallExecution and HalpInitializePICs /
>>>> HalpInitializeLegacyPIC . Isn't it precisely because these stuff are
>>>> completely different from the standard PICs used in platforms for which the
>>>> standard HAL (and possibly the ACPI HAL) are used?
>>>>
>>>> Absolutely not! You need to reprogram the standard PICs also on an APIC
>>>> system, and this is precisely what both functions do. Put them into a
>>>> diff tool to see for yourself.
>>>>
>>>> The same goes for timers. Even with the introduction of ACPI Timers,
>>>> Local APIC Timers, and Time-Stamp Counters, you still need a traditional
>>>> one (like RTC or PIT) for calibration at system startup. Simply because
>>>> the newer ones don't run at a known fixed frequency.
>>>> The Legacy HAL successfully employs an algorithm based on the RTC while
>>>> the APIC HAL unsuccessfully tries to use the PIT.
>>>>
>>>>
>>>>> Actually we should, because the detection might not work (of course in
>>>> our simple case "ACPI UP/MP" vs. "Standard", it's simple, but think about
>>>> other platforms where there can be subtle differences)
>>>>
>>>> Tell me about a single one we cannot detect and which is worth to
>>>> support. I don't recall that we ever recommended our testers to choose a
>>>> different HAL at setup.
>>>>
>>>>
>>>>> And normally it's not the setup that decides about the HAL, but the
>>>> bootloader.
>>>>
>>>> That defies your previous point about the setup initializing the
>>>> registry depending on the HAL.
>>>> If we can let the user select a Legacy HAL in the boot loader after
>>>> installing with an ACPI HAL, it is also technically possible to have one
>>>> HAL that encompasses both.
>>>>
>>>>
>>>> - Colin