Update ACPI to illumos/20201113
To address a combination of illumos #12169 and other changes, it's time to dust off illumos/acpica and update to a more recent acpica release. Gordon chose to start with the 20201113 release as there had been some additional refactoring since then that he wanted to let settle. With that in mind, we've gone through and rebased our patches in illumos/acpica.
The following summarizes systems tested against this to date (this will be updated over time):
|ASRockRack EPYCD8||AMD EPYC 7282||-|
Updated by Gordon Ross 14 days ago
Running this under VMware ESXi the high-count leaks described in #12169 appear to be fixed.
After running a while, I see only one block leaked from the acpi code:
kmem_alloc_80 leak: 1 buffer, 80 bytes ADDR BUFADDR TIMESTAMP THREAD CACHE LASTLOG CONTENTS fffffe0b25e00658 fffffe0b25e25018 0 fffffffffbc58060 fffffe0b0f62e000 fffffe0b1216db40 0 kmem_cache_alloc_debug+0x2fc kmem_cache_alloc+0x2bb kmem_zalloc+0x47 AcpiOsAllocate+0x1c AcpiOsAllocateZeroed+0x24 AcpiOsAcquireObject+0x12b AcpiUtAllocateObjectDescDbg+0x4f AcpiUtCreateInternalObjectDbg+0x6d AcpiDsCreateOperand+0x286 AcpiDsCreateOperands+0xfe AcpiDsLoad2EndOp+0x66e AcpiDsExecEndOp+0x438 AcpiPsParseLoop+0x25a AcpiPsParseAml+0x236 AcpiPsExecuteTable+0x14d
And note that's from ...ObjectDescDbg and I find that the debug support (anything...Dbg) tends to lack caution about leaks.
If I get time I'll try a non-debug build. That might make this particular leak disappear.