[WIP] LoongArch: Add ACPI compatibility for ISA I/O space reservation #19
Conversation
elysia-best
changed the title
|
Eww. This probably calls for a better (even DMI-specific) quirk... |
arch/loongarch/kernel/setup.c
Outdated
| pr_warn("Failed to allocate fwnode for legacy ISA\n"); | ||
| return 0; | ||
| } | ||
| if (add_legacy_isa_io(fwnode, LOONGSON_LIO_BASE, ISA_IOSIZE)) { |
There was a problem hiding this comment.
I don't think this patch can be merged as-is.
Both 7A1000 and 7A2000 manuals state that 0x18000000-0x1800ffff (64K) is for "LPC I/O", but this patch reserves 0x18000000-0x18003fff (16K). The only reference I can find about this 16K range is the 2K1000 ISA controller which is obviously unrelated to any desktop platform. The macro name ("ISA_IOSIZE") also seems something for 2K1000 now.
We need some detailed analysis of the issue and maybe some research on the ACPI specification.
There was a problem hiding this comment.
Right, a fast search for ISA_IOSIZE shows that this marco was introduced in initial commits for LoongArch support at torvalds@7153c3c
It didn't explain why this marco should be 16K rather than 64K. Additionally, ISA_IOSIZE is also used in loongnix's patch for this issue. I agree that this issue needs more analysis :)
elysia-best
force-pushed
the
aosc-6.18.4
branch
from
6dc0f59 to
8a0c322
Compare
|
Add a DMI check for my tested board. This list can be extended in the futuer if more boards need this workaround. |
This patch adds essential compatibility support for ACPI-based systems running on older firmware. Many older computers with old firmware still rely on this mechanism and are unlikely to receive futher firmeware updates. This change ensures proper ISA I/O space reservation across for them. Use DMI to judge if the system needs patching. Signed-off-by: Yinan Qin <a.elysia@proton.me>
elysia-best
force-pushed
the
aosc-6.18.4
branch
from
8a0c322 to
60d943c
Compare
|
If the issue isn't limited to BPI system (i.e. also reproducible with the correct UEFI boot process) we expect to make the quirk upstream. Then don't use I think adding it to arch/loongarch/pci/pci.c as a PCI quirk against some 7A1000/2000 device (which one??1) would be better. Taking twinhead_reserve_killing_zone() in arch/x86/pci/fixup.c as an example. Footnotes
|
Sure, here is the output for that (ssh)elysia@elysia-loong64-pc [ linux@aosc-6.17.7 ] ! lspci -vvv -s "00:17.0"
00:17.0 ISA bridge: Loongson Technology LLC 2K2000 / 7A1000/2000 Chipset LPC Controller (rev 01)
Control: I/O+ Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap- 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Interrupt: pin A routed to IRQ 255
NUMA node: 0 |
|
P.S. Crash log using unpatched kernel is uploaded here panic.log |
3 participants
This patch adds essential compatibility support for ACPI-based systems running on older firmware. Some older computers with Kunlun firmware still rely on this mechanism and are unlikely to receive futher firmeware updates. This change ensures proper ISA I/O space reservation across for them.