2.5.35+ IDE problems on old pre-PCI HW
One of my test boxes has serious problems with the 2.5 IDE code since 2.5.35. It's a 486 with a QD6580 VLB dual-channel controller card. No PCI in sight :-(
2.5.33/.34 worked with CONFIG_BLK_DEV_QD65XX && !CONFIG_PCI, and "ide0=qd65xx idebus=33 ide0=autotune ide1=autotune" kernel parameters. 2.4 and 2.2 + Andre's IDE patch also work.
2.5.35/.36/.38 all reboot instantly after loading the kernel, if I pass "ide0=qd65xx" as a kernel parameter. Martin's 2.5 IDE code also had this problem. Skipping "ide0=qd65xx" allows .35 and .36 to boot.
2.5.38 boots w/o "ide0=qd65xx" but hangs near the end of INIT. (With the ide_toggle_bounce() patch applied to prevent an oops.)
OTOH, current 2.5 works fine on my PIIX/ICH2 boxes.
Below is a patch for a long-standing bug which affects PCI-less kernels. The early setup passes a partially uninitialised local variable to ide_register_hw(), which copies the uninitialised fields. Among other things, this puts garbage in the "chipset" field, which can prevent chipset-selection options like "ide0=qd65xx" from succeeding.
/Mikael
include/asm-i386/ide.h renamed from include/asm-i386/ide.h.~1~ +1 -0
@@ -70,6 +70,7 @@int index;for (index =0 ; index <MAX_HWIFS ; index++) { +memset (&hw,0 ,sizeof hw);ide_init_hwif_ports (&hw,ide_default_io_base (index),0 ,NULL ); hw.irq =ide_default_irq (ide_default_io_base (index));ide_register_hw (&hw,NULL );