Skip to contentSkip to conversation

: Fix IDE host_flags regression.

PATCH In review David Miller 3 participants 3 messages Oct 22, 2007 – Oct 23, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 : Fix IDE host_flags regression.

Bart, please test your changes.

Thank you.

[IDE]: Expand hwif->host_flags so that it fits new flags.

This fixes regressions added by:
238e4f142c33bb34440cc64029dde7b9fbc4e65f

include/linux/ide.h +1 -1
index 2e4b8dd..4ed4777 100644
@@ -667,7 +667,7 @@ typedef struct hwif_s {
 	u8 straight8;	/* Alan's straight 8 check */
 	u8 bus_state;	/* power state of the IDE bus */
 
-	u16 host_flags;
+	u32 host_flags;
 
 	u8 pio_mask;
 
#1 lore raw

Hi,

Thanks for debugging & fixing this.

Quoted 2 lines · David Miller

Bart, please test your changes.

Well, I do. However due to hardware diversity it is not practically possible to test them all myself...

Quoted 6 lines

Thank you.

[IDE]: Expand hwif->host_flags so that it fits new flags.

This fixes regressions added by:
238e4f142c33bb34440cc64029dde7b9fbc4e65f

Yeah, sorry for that. ->host flags were expanded in struct ide_pci_device_s (now struct ide_port_info) but not in struct hwif_s. This went unnoticed because majority of new flags are used only through struct ide_pci_device_s.

The exceptions are:
- IDE_HFLAG_NO_LBA48 (only user is trm290, exotic hardware)
- IDE_HFLAG_NO_LBA48_DMA (only user is alim15x3, revisions <= 0xC4)
- IDE_HFLAG_ERROR_STOPS_FIFO (the error path of pdc202xx_{old,new})

What worries me a bit is that commit 238e4f142c33bb34440cc64029dde7b9fbc4e65f was in -mm since 2.6.23-rc4-mm1 (released on Sep 1 2007) and nobody reported the issue earlier...

Quoted 14 lines
include/linux/ide.h +1 -1
index 2e4b8dd..4ed4777 100644
@@ -667,7 +667,7 @@ typedef struct hwif_s {
 	u8 straight8;	/* Alan's straight 8 check */
 	u8 bus_state;	/* power state of the IDE bus */
 
-	u16 host_flags;
+	u32 host_flags;
 
 	u8 pio_mask;
#2 lore raw
Quoted 1 line

- IDE_HFLAG_ERROR_STOPS_FIFO (the error path of pdc202xx_{old,new})

I've been playing with this a bit in libata - it seems the pdc202xx is a bit more complicated. An error doesn't stop the FIFO, a reset does.

The problem thus seems to be triggered ultimately by the fact we call the timeout method before doing a FIFO drain in some cases, and also that old IDE tries to use PIO to flush a DRQ when the failing taskfile is a DMA operation not PIO..

Alan