• Len Brown's avatar
    ACPICA: simplify SCI_EN workaround · b430acbd
    Len Brown authored
    acpi_hw_set_mode() double checks its effectiveness
    by calling acpi_hw_get_mode() -- polling up to 3 seconds.
    
    It would be more logical for its caller, acpi_enable()
    acpi_enable() to do the double-checking.  (lets assume
    that acpi_disable() isn't interesting)
    
    The ACPI specification is unclear on this point.
    Some parts say that the BIOS sets SCI_EN and then returns to the OS,
    but one part says "OSPM polls the SCI_EN bit until it is sampled SET".
    
    The systems I have on hand do the former,
    SCI_EN is observed to be set upon return from the BIOS.
    
    So we move the check up out of acpi_hw_set_mode()
    up into acpi_enable() where it makes logical sense.
    
    Then we replace the 3-second polling loop
    with a single check.  If this check fails, we'll see:
    
    	"Hardware did not enter ACPI mode"
    
    and the system will bail out of ACPI initialization
    and likely fail to boot.  If we see that in practice,
    we can restore the polling, but put it into acpi_enable.
    
    This patch is important if acpi_enable() is used in
    the resume from S3 path.  Many systems today are seen
    coming back from S3 with SCI_EN off, and then failing
    to set SCI_EN in response to acpi_enable().  Those systems
    will take 3 seconds longer to resume due to this loop.
    
    However, it is possible that we will not use acpi_enable()
    in the S3 resume path, and bang SCI_EN directly, which
    would make the loop harmless, as it would be invisible
    to all systems except those that need it.
    Signed-off-by: default avatarLen Brown <len.brown@intel.com>
    b430acbd
evxfevnt.c 24.4 KB