Skip to contentSkip to conversation

ide_dma_speed() fixes

PATCH In review Sergei Shtylyov 3 participants 6 messages May 12, 2006 – May 15, 2006 libata

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ide_dma_speed() fixes

Hello.

ide_dma_speed() fails to actually honor the IDE drivers' mode support
masks) because of the bogus checks -- thus, selecting the DMA transfer mode that the driver explicitly refuses to support is possible. Additionally, there is no check for validity of the UltraDMA mode data in the drive ID, and the function is misdocumented.

MBR, Sergei

drivers/ide/ide-lib.c +34 -35
Index: linus/drivers/ide/ide-lib.c
===================================================================
@@ -72,66 +72,65 @@ EXPORT_SYMBOL(ide_xfer_verbose);
 /**
  *	ide_dma_speed	-	compute DMA speed
  *	@drive: drive
- *	@mode; intended mode
+ *	@mode:	modes available
  *
  *	Checks the drive capabilities and returns the speed to use
- *	for the transfer. Returns -1 if the requested mode is unknown
- *	(eg PIO)
+ *	for the DMA transfer.  Returns 0 if the drive is incapable
+ *	of DMA transfers.
  */
  
 u8 ide_dma_speed(ide_drive_t *drive, u8 mode)
 {
 	struct hd_driveid *id   = drive->id;
 	ide_hwif_t *hwif	= HWIF(drive);
+	u8 ultra_mask, mwdma_mask, swdma_mask;
 	u8 speed = 0;
 
 	if (drive->media != ide_disk && hwif->atapi_dma == 0)
 		return 0;
 
-	switch(mode) {
-		case 0x04:
-			if ((id->dma_ultra & 0x0040) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+	/* Capable of UltraDMA modes? */
+	if (id->field_valid & 4)
+		ultra_mask = id->dma_ultra & hwif->ultra_mask;
+	else
+		mode = 0;	/* fallback to MW/SW DMA if no UltraDMA */
+
+	switch (mode) {
+		case 4:
+			if (ultra_mask & 0x40)
 				{ speed = XFER_UDMA_6; break; }
-		case 0x03:
-			if ((id->dma_ultra & 0x0020) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+		case 3:
+			if (ultra_mask & 0x20)
 				{ speed = XFER_UDMA_5; break; }
-		case 0x02:
-			if ((id->dma_ultra & 0x0010) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+		case 2:
+			if (ultra_mask & 0x10)
 				{ speed = XFER_UDMA_4; break; }
-			if ((id->dma_ultra & 0x0008) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+			if (ultra_mask & 0x08)
 				{ speed = XFER_UDMA_3; break; }
-		case 0x01:
-			if ((id->dma_ultra & 0x0004) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+		case 1:
+			if (ultra_mask & 0x04)
 				{ speed = XFER_UDMA_2; break; }
-			if ((id->dma_ultra & 0x0002) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+			if (ultra_mask & 0x02)
 				{ speed = XFER_UDMA_1; break; }
-			if ((id->dma_ultra & 0x0001) &&
-			    (id->dma_ultra & hwif->ultra_mask))
+			if (ultra_mask & 0x01)
 				{ speed = XFER_UDMA_0; break; }
-		case 0x00:
-			if ((id->dma_mword & 0x0004) &&
-			    (id->dma_mword & hwif->mwdma_mask))
+		case 0:
+			mwdma_mask = id->dma_mword & hwif->mwdma_mask;
+
+			if (mwdma_mask & 0x04)
 				{ speed = XFER_MW_DMA_2; break; }
-			if ((id->dma_mword & 0x0002) &&
-			    (id->dma_mword & hwif->mwdma_mask))
+			if (mwdma_mask & 0x02)
 				{ speed = XFER_MW_DMA_1; break; }
-			if ((id->dma_mword & 0x0001) &&
-			    (id->dma_mword & hwif->mwdma_mask))
+			if (mwdma_mask & 0x01)
 				{ speed = XFER_MW_DMA_0; break; }
-			if ((id->dma_1word & 0x0004) &&
-			    (id->dma_1word & hwif->swdma_mask))
+
+			swdma_mask = id->dma_1word & hwif->swdma_mask;
+
+			if (swdma_mask & 0x04)
 				{ speed = XFER_SW_DMA_2; break; }
-			if ((id->dma_1word & 0x0002) &&
-			    (id->dma_1word & hwif->swdma_mask))
+			if (swdma_mask & 0x02)
 				{ speed = XFER_SW_DMA_1; break; }
-			if ((id->dma_1word & 0x0001) &&
-			    (id->dma_1word & hwif->swdma_mask))
+			if (swdma_mask & 0x01)
 				{ speed = XFER_SW_DMA_0; break; }
 	}
 
#1 lore raw
Quoted 6 lines · Sergei Shtylyov

ide_dma_speed() fails to actually honor the IDE drivers' mode support
masks) because of the bogus checks -- thus, selecting the DMA transfer mode that the driver explicitly refuses to support is possible. Additionally, there is no check for validity of the UltraDMA mode data in the drive ID, and the function is misdocumented.

drivers/ide/ide-lib.c: In function `ide_dma_speed':
drivers/ide/ide-lib.c:86: warning: `ultra_mask' might be used uninitialized in this function

Looks like a real bug to me - it depends up on the values of `mode' and id->field_value.

Anyway, I'll drop it, please review and fix. I assume that warning was occurring for you as well - please spend more time over these things. Especially when working on IDE, where bugs are slow to show themselves and have particularly bad consequences.

#2 lore raw

Hello.

Quoted 6 lines · Andrew Morton
Quoted 2 lines

ide_dma_speed() fails to actually honor the IDE drivers' mode support
masks) because of the bogus checks -- thus, selecting the DMA transfer mode that the driver explicitly refuses to support is possible. Additionally, there is no check for validity of the UltraDMA mode data in the drive ID, and the function is misdocumented.

drivers/ide/ide-lib.c: In function `ide_dma_speed':
drivers/ide/ide-lib.c:86: warning: `ultra_mask' might be used uninitialized in this function

Quoted 6 lines

Looks like a real bug to me - it depends up on the values of `mode' and id->field_value.

Anyway, I'll drop it, please review and fix. I assume that warning was
occurring for you as well - please spend more time over these things. Especially when working on IDE, where bugs are slow to show themselves and have particularly bad consequences.

That's what gcc thinks. The code is 100% correct -- it will never reach
the switch statement with mode > 0 (in which case ultra_mask isn't used) and ultra_mask unitialized. I may add an explicit initializer in the declaration if you like...

MBR, Sergei

#3 lore raw
Quoted 4 lines · Sergei Shtylyov

That's what gcc thinks. The code is 100% correct -- it will never reach
the switch statement with mode > 0 (in which case ultra_mask isn't used) and ultra_mask unitialized. I may add an explicit initializer in the declaration if you like...

Please leave the warning, this was discussed recently in other threads and hiding warnings is very dangerous.

#4 lore raw
In reply to Sergei Shtylyov
Quoted 17 lines · Sergei Shtylyov

Hello.

Quoted 11 lines · Andrew Morton
Quoted 5 lines

ide_dma_speed() fails to actually honor the IDE drivers' mode support
masks) because of the bogus checks -- thus, selecting the DMA transfer mode that the driver explicitly refuses to support is possible. Additionally, there is no check for validity of the UltraDMA mode data in the drive ID, and the function is misdocumented.

drivers/ide/ide-lib.c: In function `ide_dma_speed':
drivers/ide/ide-lib.c:86: warning: `ultra_mask' might be used uninitialized in this function

Looks like a real bug to me - it depends up on the values of `mode' and id->field_value.

Quoted 8 lines
Quoted 4 lines

Anyway, I'll drop it, please review and fix. I assume that warning was occurring for you as well - please spend more time over these things. Especially when working on IDE, where bugs are slow to show themselves and have particularly bad consequences.

That's what gcc thinks. The code is 100% correct -- it will never reach
the switch statement with mode > 0 (in which case ultra_mask isn't used) and ultra_mask unitialized.

hm, OK.

Quoted 1 line

I may add an explicit initializer in the declaration if you like...

Opinions vary. I think that's less bad than spitting a warning at all developers for all time.

But we can redo things like below and save a little code in the process. Look OK?

drivers/ide/ide-lib.c renamed from drivers/ide/ide-lib.c~ide_dma_speed-fixes-warning-fix +3 -3
@@ -90,9 +90,9 @@ u8 ide_dma_speed(ide_drive_t *drive, u8 
 		return 0;
 
 	/* Capable of UltraDMA modes? */
-	if (id->field_valid & 4)
-		ultra_mask = id->dma_ultra & hwif->ultra_mask;
-	else
+	ultra_mask = id->dma_ultra & hwif->ultra_mask;
+
+	if (!(id->field_valid & 4))
 		mode = 0;	/* fallback to MW/SW DMA if no UltraDMA */
 
 	switch (mode) {

_

#5 lore raw

drivers/ide/ide-lib.c u8 ide_dma_speed(ide_drive_t *drive, u8 · line 96

Quoted 10 lines · Andrew Morton
drivers/ide/ide-lib.c renamed from drivers/ide/ide-lib.c~ide_dma_speed-fixes-warning-fix +3 -3
 	/* Capable of UltraDMA modes? */
-	if (id->field_valid & 4)
-		ultra_mask = id->dma_ultra & hwif->ultra_mask;
-	else
+	ultra_mask = id->dma_ultra & hwif->ultra_mask;
+
+	if (!(id->field_valid & 4))
 		mode = 0;	/* fallback to MW/SW DMA if no UltraDMA */

Looks fine to me, id-> is always the full 512 bytes of data so its safe to do that.