Bartlomiej Zolnierkiewicz wrote:
hdparm explicitely marks HDIO_[UNREGISTER,SCAN]_HWIF ioctls as DANGEROUS
and given the number of bugs we can assume that there are no real users:
* DMA has no chance of working because DMA resources are released by
ide_unregister() and they are never allocated again.
* Since ide_init_hwif_ports() is used for ->io_ports[] setup the ioctls
don't work for almost all hosts with "non-standard" (== non ISA-like)
layout of IDE taskfile registers (there is a lot of such host drivers).
* ide_port_init_devices() is not called when probing IDE devices so:
- drive->autotune is never set and IDE host/devices are not programmed
for the correct PIO/DMA transfer modes (=> possible data corruption)
- host specific I/O 32-bit and IRQ unmasking settings are not applied
(=> possible data corruption)
- host specific ->port_init_devs method is not called (=> no luck with
ht6560b, qd65xx and opti621 host drivers)
* ->rw_disk method is not preserved (=> no HPT3xxN chipsets support).
* ->serialized flag is not preserved (=> possible data corruption when
using icside, aec62xx (ATP850UF chipset), cmd640, cs5530, hpt366
(HPT3xxN chipsets), rz1000, sc1200, dtc2278 and ht6560b host drivers).
* ->ack_intr method is not preserved (=> needed by ide-cris, buddha,
gayle and macide host drivers).
* ->sata_scr[] and sata_misc[] is cleared by ide_unregister() and it
isn't initialized again (SiI3112 support needs them).
* To issue an ioctl() there need to be at least one IDE device present
in the system.
* ->cable_detect method is not preserved + it is not called when probing
IDE devices so cable detection is broken (however since DMA support is
also broken it doesn't really matter ;-).
* Some objects which may have already been freed in ide_unregister()
are restored by ide_hwif_restore() (i.e. ->hwgroup).
* ide_register_hw() may unregister unrelated IDE ports if free ide_hwifs[]
slot cannot be found.
* When IDE host drivers are modular unregistered port may be re-used by
different host driver that owned it first causing subtle bugs.
Since we now have a proper warm-plug support remove these ioctls,
then remove no longer needed:
- ide_register_hw() and ide_hwif_restore() functions
- 'init_default' and 'restore' arguments of ide_unregister()
- zeroeing of hwif->{dma,extra}_* fields in ide_unregister()
As an added bonus IDE core code size shrinks by ~3kB (x86-32).
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@xxxxxxxxx>
Acked-by: Sergei Shtylyov <sshtylyov@xxxxxxxxxxxxx>
Index: b/drivers/ide/ide-pnp.c
===================================================================
--- a/drivers/ide/ide-pnp.c
+++ b/drivers/ide/ide-pnp.c
[...]
@@ -655,52 +530,6 @@ void ide_init_port_hw(ide_hwif_t *hwif,
}
EXPORT_SYMBOL_GPL(ide_init_port_hw);
-/**
- * ide_register_hw - register IDE interface
- * @hw: hardware registers
- * @quirkproc: quirkproc function
- * @hwifp: pointer to returned hwif
- *
- * Register an IDE interface, specifying exactly the registers etc.
- *
- * Returns -1 on error.
- */
-
-static int ide_register_hw(hw_regs_t *hw, void (*quirkproc)(ide_drive_t *),
- ide_hwif_t **hwifp)
-{
- int index, retry = 1;
- ide_hwif_t *hwif;
- u8 idx[4] = { 0xff, 0xff, 0xff, 0xff };
-
- do {
- hwif = ide_find_port(hw->io_ports[IDE_DATA_OFFSET]);
- index = hwif->index;
- if (hwif)
- goto found;
Hm, I remember there was a patch that fixed the above bug where hwif is
dereferenced before being checked for NULL, I wonder how come it was lost?
WBR, Sergei
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html