Most recent kernel where this bug did *NOT* occur: None Distribution: Ubuntu 6.10 Edgy Eft Hardware Environment: Toshiba Satellite P100 MA5 laptop (model PSPA3C MA502C, system BIOS V2.40) Problem Description: Decompiling the DSDT using iasl reveals the following method where OS detection occurs. Method (_INI, 0, NotSerialized) { If (DTSE) { TRAP (0x47) } Store (0x07D0, OSYS) If (CondRefOf (_OSI, Local0)) { If (_OSI ("Linux")) { Store (0x03E8, OSYS) } Else { Store (0x07D1, OSYS) If (_OSI ("Windows 2001 SP2")) { Store (0x07D2, OSYS) } If (_OSI ("Windows 2001.1")) { Store (0x07D3, OSYS) } If (_OSI ("Windows 2001.1 SP1")) { Store (0x07D4, OSYS) } If (_OSI ("Windows 2006")) { Store (0x07D6, OSYS) } If (LAnd (MPEN, LEqual (OSYS, 0x07D1))) { TRAP (0x3D) } } } } There is no further OS ACPI interface checking once the OS confesses to supporting the Linux ACPI interface which causes certain laptop devices to not be enabled (e.g. Intel HDA sound). Disabling the OSI functionality also gives the same result. My complete winding road leading to this bug report can be found at https://launchpad.net/ubuntu/+source/linux-source-2.6.17/+bug/75434
Perhaps Linux should never "confess" to being Linux? (via _OSI) Also, the bios does not follow the bios writers guide recommendations. The correct bios implementation should not have an else clause - i.e. the code should be inline successive.
Yes, removing Linux as a supported ACPI interface should solve the problem. Would this be supported through a new kernel parameter? (e.g. acpi_osi_unsupported="Linux") Is this better than only supporting a single ACPI interface as did the allow acpi_osi="Winows XY" overrides proposal? (see http://marc.theaimsgroup.com/?l=linux-acpi&m=115563247228643&w=2). I assume so. True, the bios does not follow the bios writers guide recommendations but unless Toshiba is persuaded to release a fixed bios the work around will have to occur in the kernel.
The definition of the _OSI method differs from _OS in that it is supposed to report ALL of the interfaces and features that the host supports, not just a single one.
Too late to remove "Linux" from _OSI for all systems. There are BIOS' that depend on it today (against my recommendation) for doing things like enabling video re-post on S3 resume. The proper workaround for this BIOS bug is to create: 1. a boot param allowing one to delete any _OSI entry 2. a boot param allowing one to add any _OSI entry 3. a workaround for the broken Toshiba BIOS at hand to invoke #1 with "Linux" automatically. Today we have "acpi_osi=", which disables _OSI if no params, and it ignores any params if they are given. This is somewhat confusing. Lets do this: Remove acpi_osi= Add acpi_osi_disable Add acpi_osi_add=a,b,c,... Add acpi_osi_delete=a,b,c,... Patrick, Please verify that you're running the latest BIOS and attach the complete output from acpidump here for reference. For the benefit of #3, please attach the output from dmidecode.
This sounds like a good general solution.
I've confirmed that my BIOS (V2.40) is the same as the latest BIOS available from the Toshiba support website.
Created attachment 10127 [details] dsdt.dsl Decompiled DSDT by: > cat /proc/acpi/dsdt > dsdt.dat > iasl -d dsdt.dat
Created attachment 10128 [details] dmidecode
*** Bug 8267 has been marked as a duplicate of this bug. ***
Looking over Patrick's DSDT for uses of OPSYS... There is a _WAK SMM hook that _OSI "Linux" will trigger that Windows will not. Does this machine suspend to ram, and does it restore video on resume? Vista uses a mutex when talking to brighness control where others do not -- unclear why Vista has some EC hooks that XP does not have _STA for the HPET will say it is not present for OSI Linux and will say it is present for Windows WMI hooks are not present for Vista, but are otherwise present ----------- Scope (\_SB) Method (_INI, 0, NotSerialized) Store (0x07D0, OSYS) If (_OSI ("Linux")) Store (0x03E8, OSYS) # unclear why this is here # There are no tests for OSYS 038E8 in the DSDT Else Store (0x07D1, OSYS) # 0x07D1 = any _OSI enabled OS besides Linux If (_OSI ("Windows 2001 SP2")) Store (0x07D2, OSYS) # 07D3 = WinXP SP2 If (_OSI ("Windows 2001.1")) Store (0x07D3, OSYS) # 07D4 = WinServer2003 If (_OSI ("Windows 2001.1 SP1")) Store (0x07D4, OSYS) # 07D6 = WinServer2003 SP1 If (_OSI ("Windows 2006")) Store (0x07D6, OSYS) # WinVista ... Method (_WAK, 1, NotSerialized) { If (LOr (LEqual (Arg0, 0x03), LEqual (Arg0, 0x04))) { If (And (CFGD, 0x01000000)) { If (LAnd (And (CFGD, 0xF0), LEqual (OSYS, 0x07D1))) # non-Linux _OSI enabled OS traps to SMM -- why? # could this be a hook for video re-post on resume? { TRAP (0x3D) } } } ... If (LEqual (OSYS, 0x07D2)) # Windows XP SP2 has a uniq PPC notify on _WAK { If (And (CFGD, 0x01)) { If (LGreater (\_PR.CPU0._PPC, 0x00)) { Subtract (\_PR.CPU0._PPC, 0x01, \_PR.CPU0._PPC) PNOT () Add (\_PR.CPU0._PPC, 0x01, \_PR.CPU0._PPC) PNOT () } Else { Add (\_PR.CPU0._PPC, 0x01, \_PR.CPU0._PPC) PNOT () Subtract (\_PR.CPU0._PPC, 0x01, \_PR.CPU0._PPC) PNOT () } } } If (LEqual (OSYS, 0x07D6)) # Windows Vista gets a different EC notify { If (\_SB.PCI0.LPCB.EC0.MBTB) { Notify (\_SB.PCI0.LPCB.EC0.MBTN, 0x02) } } ... Device (PCI0) Device (PEGP) Device (VGA) Device (LCD) Method (_BCM, 1, NotSerialized) If (LEqual (OSYS, 0x07D6)) # Windows Vista uses a mutex in brightness control that others do not { Acquire (\_SB.PCI0.LPCB.EC0.MUT1, 0xFFFF) Store (Local1, \_SB.PCI0.LPCB.EC0.BLVL) Release (\_SB.PCI0.LPCB.EC0.MUT1) } Method (_BQC, 0, NotSerialized) { If (LEqual (OSYS, 0x07D6)) # Windows Vista { Acquire (\_SB.PCI0.LPCB.EC0.MUT1, 0xFFFF) Store (\_SB.PCI0.LPCB.EC0.BLVL, Local0) Release (\_SB.PCI0.LPCB.EC0.MUT1) } Device (GFX0) ... Device (DD04) Method (_BCM, 1, NotSerialized) ... If (LEqual (OSYS, 0x07D6)) # Windows Vista { Acquire (\_SB.PCI0.LPCB.EC0.MUT1, 0xFFFF) Store (Local1, \_SB.PCI0.LPCB.EC0.BLVL) Release (\_SB.PCI0.LPCB.EC0.MUT1) } } Method (_BQC, 0, NotSerialized) { If (LEqual (OSYS, 0x07D6)) # Windows Vista { Acquire (\_SB.PCI0.LPCB.EC0.MUT1, 0xFFFF) Store (\_SB.PCI0.LPCB.EC0.BLVL, Local0) Release (\_SB.PCI0.LPCB.EC0.MUT1) } ... Device (EC0) { Name (_HID, EisaId ("PNP0C09")) # Embedded Controller ... Name (_GPE, 0x17) Name (SEL0, 0xF0) Name (BFLG, 0x00) Method (_REG, 2, NotSerialized) { ... If (\_SB.ECOK) { Acquire (\_SB.PCI0.LPCB.EC0.MUT1, 0xFFFF) If (LEqual (OSYS, 0x07D6)) # Windows Vista has a specific EC flag here { Store (One, \_SB.PCI0.LPCB.EC0.OSTP) \_SB.PHSR (0x0D, 0x00) } Else { Store (Zero, \_SB.PCI0.LPCB.EC0.OSTP) } Store (0x03, \_SB.PCI0.LPCB.EC0.RG59) Store (\_SB.CIRE, \_SB.PCI0.LPCB.EC0.CIRE) Store (\_SB.PHSR (0x05, 0x00), DOFF) Store (\_SB.PCI0.LPCB.EC0.ACDF, \PWRS) If (LEqual (OSYS, 0x07D6)) { # Windows Vista Store (0x3C, \_SB.PCI0.LPCB.EC0.VTMP) } Release (\_SB.PCI0.LPCB.EC0.MUT1) } } ... Device (HPET) { Name (_HID, EisaId ("PNP0103")) Method (_STA, 0, NotSerialized) { If (LGreaterEqual (OSYS, 0x07D1)) # All non-Linux _OSI enabled OS: HPET present { If (HPAE) { Return (0x0F) } # present+enabled+ui+functioning } Else { # else linux and non _OSI If (HPAE) { Return (0x0B) # not present } } Return (0x00) } ... Device (AMW0) { Name (_HID, "*PNP0C14") Name (_UID, 0x00) Method (_STA, 0, NotSerialized) { If (LEqual (OSYS, 0x07D6)) # Windows Vista { Return (Zero) # WMI not present } Else { # non-Windows Vista - WMI present Return (0x0F) } }
Created attachment 10983 [details] 2.6.21-rc5 patch to remove "Linux" from _OSI Patrick, please test this patch, which removes "Linux" from _OSI and report any differences you see in dmesg and sound. Also, if your machine is able to suspend to RAM and wake-up, it would be interesting to see if this changes that behavior since it will change what AML _WAK does. It isn't clear from the DSDT audit above which bit is enabling Sound in the case of Windows and not in the case of Linux.
Created attachment 11552 [details] 2.6.22-rc2 patch extending acpi_osi= parameter Please apply this patch and boot with "acpi_osi=!Linux" This should disable "Linux" from the pre-defined _OSI strings, avoiding BIOS mis-use of the string, and sound should work. You should also get a message about it in dmesg.
Created attachment 11553 [details] 2.6.22-rc2 patch disabling _OSI(Linux) by default please apply this patch on top of the previous one and verify that you no longer need "acpi_osi=!Linux" to get sound to work. Further, there should be a message warning that the system is asking us for _OSI(Linux) and that you should try "acpi_osi=Linux" to enable it. (though on this machine that parameter should make the original sound failure come back rather than making the machine work better)
DMI entry shipped in linux-2.6.22-rc4, Please test that release and verify that it automatically avoids this issue on the Toshiba. Also, it should print a message about it in dmesg. Please re-open this bug if it doesn't work. In 2.6.23 we'll change the default from OSI Linux ENABLED plus DMI entries for machines to turn it OFF... to OSI linux DISABLED and DMI entries for machines to turn it ON
Created attachment 12031 [details] dmidecode As requested I'm attaching dmidecode (acpidump in next attachment) from HP Compaq C518LA F.24 BIOS running on Linux 2.6.22.1
Created attachment 12033 [details] acpidump acpidump from HP Compaq C518LA F.24 BIOS running on Linux 2.6.22.1 as stated during boot sent first to linux-acpi mailing list and then as requested attached here.
Looks like the use of _OSI(Linux) in the HP Compaq C518LA F.24 is a NOP in AML, though conceivably the memory location being set could be used by SMM. OperationRegion (GNVS, SystemMemory, 0x1F693E4C, 0x0100) Field (GNVS, AnyAcc, Lock, Preserve) { OSYS, 16, ... DBGS, 8, LINX, 8, ... If (_OSI ("Linux")) { Store (One, LINX) } Happily, this test doesn't interfere in any way with the tests for various windows interfaces: If (_OSI ("Windows 2001")) { Store (0x07D1, OSYS) } If (_OSI ("Windows 2001 SP1")) { Store (0x07D1, OSYS) } If (_OSI ("Windows 2001 SP2")) { Store (0x07D2, OSYS) } If (_OSI ("Windows 2006")) { Store (0x07D6, OSYS) }