Re: [PATCH 14/15] ide: rework the code for selecting the best DMAtransfer mode

From: Bartlomiej Zolnierkiewicz
Date: Fri Feb 02 2007 - 18:54:08 EST

Next message: Adam Kropelin: "Re: MSI failure on nForce 430 (WAS: intel 82571EB gigabit fails to see link on 2.6.20-rc5 in-tree e1000 driver (regression))"
Previous message: Andrew Morton: "Re: [patch 9/9] mm: fix pagecache write deadlocks"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi,

Sergei Shtylyov wrote:
>
> Hello.
>
> Bartlomiej Zolnierkiewicz wrote:
>
>> [PATCH] ide: rework the code for selecting the best DMA transfer mode
>
> Here's another portion of comments...
>
>> Depends on the "ide: fix UDMA/MWDMA/SWDMA masks" patch.
>
>> * add ide_hwif_t.filter_udma_mask hook for filtering UDMA mask
>
> Erm, maybe a shorter method name like udma_filter would go with the others
> better. But well, that's my taste. :-)

done

>> (use it in alim15x3, hpt366, siimage and serverworks drivers)
>> * add ide_max_dma_mode() for finding best DMA mode for the device
>> (loosely based on some older libata-core.c code)
>> * convert ide_dma_speed() users to use ide_max_dma_mode()
>> * make ide_rate_filter() take "ide_drive_t *drive" as an argument instead
>> of "u8 mode" and teach it to how to use UDMA mask to do filtering
>> * use ide_rate_filter() in hpt366 driver
>> * remove no longer needed ide_dma_speed() and *_ratemask()
>> * unexport eighty_ninty_three()
>
>> Index: b/drivers/ide/ide-dma.c
>> ===================================================================
>> --- a/drivers/ide/ide-dma.c
>> +++ b/drivers/ide/ide-dma.c
>> @@ -705,6 +705,80 @@ int ide_use_dma(ide_drive_t *drive)
>>
>> EXPORT_SYMBOL_GPL(ide_use_dma);
>>
>> +static const u8 xfer_mode_bases[] = {
>> + XFER_UDMA_0,
>> + XFER_MW_DMA_0,
>> + XFER_SW_DMA_0,
>> +};
>> +
>> +static unsigned int ide_get_mode_mask(ide_drive_t *drive, u8 base)
>> +{
>> + struct hd_driveid *id = drive->id;
>> + ide_hwif_t *hwif = drive->hwif;
>> + unsigned int mask = 0;
>> +
>> + switch(base) {
>> + case XFER_UDMA_0:
>> + if ((id->field_valid & 4) == 0)
>> + break;
>> +
>> + mask = id->dma_ultra & hwif->ultra_mask;
>> +
>> + if (hwif->filter_udma_mask)
>> + mask &= hwif->filter_udma_mask(drive);
>> +
>> + if ((mask & 0x78) && (eighty_ninty_three(drive) == 0))
>> + mask &= 0x07;
>> + break;
>> + case XFER_MW_DMA_0:
>> + mask = id->dma_mword & hwif->mwdma_mask;
>> + break;
>> + case XFER_SW_DMA_0:
>> + mask = id->dma_1word & hwif->swdma_mask;
>> + break;
>> + default:
>> + BUG();
>> + break;
>> + }
>> +
>> + return mask;
>> +}
>> +
>> +/**
>> + * ide_max_dma_mode - compute DMA speed
>> + * @drive: IDE device
>> + *
>> + * Checks the drive capabilities and returns the speed to use
>> + * for the DMA transfer. Returns 0 if the drive is incapable
>> + * of DMA transfers.
>> + */
>> +
>> +u8 ide_max_dma_mode(ide_drive_t *drive)
>> +{
>> + ide_hwif_t *hwif = drive->hwif;
>> + unsigned int mask;
>> + int x, i;
>> + u8 mode = 0;
>> +
>> + if (drive->media != ide_disk && hwif->atapi_dma == 0)
>> + return 0;
>> +
>> + for (i = 0; i < ARRAY_SIZE(xfer_mode_bases); i++) {
>> + mask = ide_get_mode_mask(drive, xfer_mode_bases[i]);
>> + x = fls(mask) - 1;
>> + if (x >= 0) {
>> + mode = xfer_mode_bases[i] + x;
>> + break;
>> + }
>> + }
>> +
>> + printk(KERN_DEBUG "%s: selected mode 0x%x\n", drive->name, mode);
>> +
>> + return mode;
>> +}
>> +
>> +EXPORT_SYMBOL_GPL(ide_max_dma_mode);
>> +
>
> I didn't quite like the array/loop approach but well, that's my taste (I'd
> rather put the mode-from-mask evaluation to the function and call it thrice)...

The mode-from-mask approach is indeed nicer. If you send me a patch to
use the mode-from-mask evaluation I'll happily apply it. :)

However the array/loop approach is also definitively an improvement
over the current code.

>> Index: b/drivers/ide/ide-lib.c
>> ===================================================================
>> --- a/drivers/ide/ide-lib.c
>> +++ b/drivers/ide/ide-lib.c
>> @@ -69,123 +69,34 @@ char *ide_xfer_verbose (u8 xfer_rate)
>> EXPORT_SYMBOL(ide_xfer_verbose);
>>
>> /**
>> - * ide_dma_speed - compute DMA speed
>> - * @drive: drive
>> - * @mode: modes available
>> - *
>> - * Checks the drive capabilities and returns the speed to use
>> - * for the DMA transfer. Returns 0 if the drive is incapable
>> - * of DMA transfers.
>> - */
>> -
>> -u8 ide_dma_speed(ide_drive_t *drive, u8 mode)
>
> [...]
>
>> -EXPORT_SYMBOL(ide_dma_speed);
>
> Alas, my ide_dma_speed() fix is going to be oudated rather quickly... :-)

C'est la vie :)

>> Index: b/drivers/ide/pci/hpt366.c
>> ===================================================================
>> --- a/drivers/ide/pci/hpt366.c
>> +++ b/drivers/ide/pci/hpt366.c
>> @@ -513,43 +513,31 @@ static int check_in_drive_list(ide_drive
>> return 0;
>> }
>>
>> -static u8 hpt3xx_ratemask(ide_drive_t *drive)
>> -{
>> - struct hpt_info *info = pci_get_drvdata(HWIF(drive)->pci_dev);
>> - u8 mode = info->max_mode;
>> -
>> - if (!eighty_ninty_three(drive) && mode)
>> - mode = min(mode, (u8)1);
>> - return mode;
>> -}
>> -
>> /*
>> * Note for the future; the SATA hpt37x we must set
>> * either PIO or UDMA modes 0,4,5
>> */
>> -
>> -static u8 hpt3xx_ratefilter(ide_drive_t *drive, u8 speed)
>> +
>> +static u8 hpt3xx_filter_udma_mask(ide_drive_t *drive)
>> {
>> struct hpt_info *info = pci_get_drvdata(HWIF(drive)->pci_dev);
>> u8 chip_type = info->chip_type;
>> - u8 mode = hpt3xx_ratemask(drive);
>> -
>> - if (drive->media != ide_disk)
>> - return min(speed, (u8)XFER_PIO_4);
>> + u8 mode = info->max_mode;
>> + u8 mask;
>>
>> switch (mode) {
>> case 0x04:
>> - speed = min_t(u8, speed, XFER_UDMA_6);
>> + mask = 0x7f;
>> break;
>> case 0x03:
>> - speed = min_t(u8, speed, XFER_UDMA_5);
>> + mask = 0x3f;
>> if (chip_type >= HPT374)
>> break;
>> if (!check_in_drive_list(drive, bad_ata100_5))
>> goto check_bad_ata33;
>> /* fall thru */
>> case 0x02:
>> - speed = min_t(u8, speed, XFER_UDMA_4);
>> + mask = 0x1f;
>>
>> /*
>> * CHECK ME, Does this need to be changed to HPT374 ??
>> @@ -560,13 +548,13 @@ static u8 hpt3xx_ratefilter(ide_drive_t
>> !check_in_drive_list(drive, bad_ata66_4))
>> goto check_bad_ata33;
>>
>> - speed = min_t(u8, speed, XFER_UDMA_3);
>> + mask = 0x0f;
>> if (HPT366_ALLOW_ATA66_3 &&
>> !check_in_drive_list(drive, bad_ata66_3))
>> goto check_bad_ata33;
>> /* fall thru */
>> case 0x01:
>> - speed = min_t(u8, speed, XFER_UDMA_2);
>> + mask = 0x07;
>>
>> check_bad_ata33:
>> if (chip_type >= HPT370A)
>> @@ -576,10 +564,10 @@ static u8 hpt3xx_ratefilter(ide_drive_t
>> /* fall thru */
>> case 0x00:
>> default:
>> - speed = min_t(u8, speed, XFER_MW_DMA_2);
>> + mask = 0x00;
>> break;
>> }
>> - return speed;
>> + return mask;
>> }
>
> Erm, I see. This driver will need some redesign because 'struct hpt_info'
> was fitted for the old rate filtering model. Looks like the 'max_mode' field
> should be replaced by 'ultra_mask' there...

yes...

>> static u32 get_speed_setting(u8 speed, struct hpt_info *info)
>> @@ -607,12 +595,19 @@ static int hpt36x_tune_chipset(ide_drive
>> ide_hwif_t *hwif = HWIF(drive);
>> struct pci_dev *dev = hwif->pci_dev;
>> struct hpt_info *info = pci_get_drvdata(dev);
>> - u8 speed = hpt3xx_ratefilter(drive, xferspeed);
>> + u8 speed = ide_rate_filter(drive, xferspeed);
>> u8 itr_addr = drive->dn ? 0x44 : 0x40;
>> - u32 itr_mask = speed < XFER_MW_DMA_0 ? 0x30070000 :
>> - (speed < XFER_UDMA_0 ? 0xc0070000 : 0xc03800ff);
>> - u32 new_itr = get_speed_setting(speed, info);
>> u32 old_itr = 0;
>> + u32 itr_mask, new_itr;
>> +
>> + /* TODO: move this to ide_rate_filter() [ check ->atapi_dma ] */
>> + if (drive->media != ide_disk)
>> + speed = min_t(u8, speed, XFER_PIO_4);
>> +
>
> When I think about it, it seems quite stupid to set a PIO mode instead of
> a requested DMA one. So, this is actually a questionable piece of code in
> this driver...

This can safely vanish when core code gets fixed to correctly check device+host
supported PIO/DMA modes before calling ->tuneproc/->speedproc for user space
originated code change requests.

As for now IMHO it is the best to just leave it alone...

> Well, I must note the sorrow fact that both the IDE susbsytem as a whole
> doesn't keep track of PIO/DMA modes separately (having only current_mode) and
> the many drivers also fail to handle the timing registers shared by PIO/DMA
> modes correctly (well, it's actually also a hardware design issue :-).

Actually, I have an unfinished patch to fix it but it depends on
many other unfinished patches... *sigh*...

>> + itr_mask = speed < XFER_MW_DMA_0 ? 0x30070000 :
>> + (speed < XFER_UDMA_0 ? 0xc0070000 : 0xc03800ff);
>> +
>> + new_itr = get_speed_setting(speed, info);
>
> Well, I liked this code where it was. But anyway, it's going to be
> replaced RSN...
>
>> @@ -632,12 +627,19 @@ static int hpt37x_tune_chipset(ide_drive
>> ide_hwif_t *hwif = HWIF(drive);
>> struct pci_dev *dev = hwif->pci_dev;
>> struct hpt_info *info = pci_get_drvdata(dev);
>> - u8 speed = hpt3xx_ratefilter(drive, xferspeed);
>> + u8 speed = ide_rate_filter(drive, xferspeed);
>> u8 itr_addr = 0x40 + (drive->dn * 4);
>> - u32 itr_mask = speed < XFER_MW_DMA_0 ? 0x303c0000 :
>> - (speed < XFER_UDMA_0 ? 0xc03c0000 : 0xc1c001ff);
>> - u32 new_itr = get_speed_setting(speed, info);
>> u32 old_itr = 0;
>> + u32 itr_mask, new_itr;
>> +
>> + /* TODO: move this to ide_rate_filter() [ check ->atapi_dma ] */
>> + if (drive->media != ide_disk)
>> + speed = min_t(u8, speed, XFER_PIO_4);
>> +
>> + itr_mask = speed < XFER_MW_DMA_0 ? 0x303c0000 :
>> + (speed < XFER_UDMA_0 ? 0xc03c0000 : 0xc1c001ff);
>> +
>> + new_itr = get_speed_setting(speed, info);
>
> Same comments here...
>
>> Index: b/drivers/ide/pci/serverworks.c
>> ===================================================================
>> --- a/drivers/ide/pci/serverworks.c
>> +++ b/drivers/ide/pci/serverworks.c
>> @@ -65,16 +65,16 @@ static int check_in_drive_lists (ide_dri
>> return 0;
>> }
>>
>> -static u8 svwks_ratemask (ide_drive_t *drive)
>> +static u8 svwks_filter_udma_mask(ide_drive_t *drive)
>> {
>> struct pci_dev *dev = HWIF(drive)->pci_dev;
>> - u8 mode = 0;
>> + u8 mask = 0;
>
> Hm, this looks like it needs rework...

...some time later. 8)

>> if (!svwks_revision)
>> pci_read_config_byte(dev, PCI_REVISION_ID, &svwks_revision);
>>
>> if (dev->device == PCI_DEVICE_ID_SERVERWORKS_HT1000IDE)
>> - return 2;
>> + return 0x1f;
>> if (dev->device == PCI_DEVICE_ID_SERVERWORKS_OSB4IDE) {
>> u32 reg = 0;
>> if (isa_dev)
>> @@ -86,25 +86,31 @@ static u8 svwks_ratemask (ide_drive_t *d
>> if(drive->media == ide_disk)
>> return 0;
>> /* Check the OSB4 DMA33 enable bit */
>> - return ((reg & 0x00004000) == 0x00004000) ? 1 : 0;
>> + return ((reg & 0x00004000) == 0x00004000) ? 0x07 : 0;
>> } else if (svwks_revision < SVWKS_CSB5_REVISION_NEW) {
>> - return 1;
>> + return 0x07;
>> } else if (svwks_revision >= SVWKS_CSB5_REVISION_NEW) {
>> - u8 btr = 0;
>> + u8 btr = 0, mode;
>> pci_read_config_byte(dev, 0x5A, &btr);
>> mode = btr & 0x3;
>> - if (!eighty_ninty_three(drive))
>> - mode = min(mode, (u8)1);
>> +
>> /* If someone decides to do UDMA133 on CSB5 the same
>> issue will bite so be inclusive */
>> if (mode > 2 && check_in_drive_lists(drive, svwks_bad_ata100))
>> mode = 2;
>> +
>> + switch(mode) {
>> + case 2: mask = 0x1f; break;
>> + case 1: mask = 0x07; break;
>> + default: mask = 0x00; break;
>> + }
>> }
>> if (((dev->device == PCI_DEVICE_ID_SERVERWORKS_CSB6IDE) ||
>> (dev->device == PCI_DEVICE_ID_SERVERWORKS_CSB6IDE2)) &&
>> (!(PCI_FUNC(dev->devfn) & 1)))
>> - mode = 2;
>> - return mode;
>> + mask = 0x1f;
>> +
>> + return mask;
>> }
>
> Hm, what was the problem with setting the proper masks based on PCI
> device ID in the init. code?
> That'd have greatly simplified the filter...

I don't see a problem with doing it during initialization but simplifying
this code wasn't the goal of this patch and can be done in the future.

>> Index: b/drivers/ide/pci/siimage.c
>> ===================================================================
>> --- a/drivers/ide/pci/siimage.c
>> +++ b/drivers/ide/pci/siimage.c
>> @@ -116,45 +116,41 @@ static inline unsigned long siimage_seld
>> }
>>
>> /**
>> - * siimage_ratemask - Compute available modes
>> - * @drive: IDE drive
>> + * sil_filter_udma_mask - compute UDMA mask
>> + * @drive: IDE device
>> + *
>> + * Compute the available UDMA speeds for the device on the interface.
>> *
>> - * Compute the available speeds for the devices on the interface.
>> * For the CMD680 this depends on the clocking mode (scsc), for the
>> - * SI3312 SATA controller life is a bit simpler. Enforce UDMA33
>> - * as a limit if there is no 80pin cable present.
>> + * SI3112 SATA controller life is a bit simpler.
>> */
>> -
>> -static byte siimage_ratemask (ide_drive_t *drive)
>> +
>> +static u8 sil_filter_udma_mask(ide_drive_t *drive)
>> {
>> - ide_hwif_t *hwif = HWIF(drive);
>> - u8 mode = 0, scsc = 0;
>> + ide_hwif_t *hwif = drive->hwif;
>> unsigned long base = (unsigned long) hwif->hwif_data;
>> + u8 mask = 0, scsc = 0;
>>
>> if (hwif->mmio)
>> scsc = hwif->INB(base + 0x4A);
>> else
>> pci_read_config_byte(hwif->pci_dev, 0x8A, &scsc);
>>
>> - if(is_sata(hwif))
>> - {
>> - if(strstr(drive->id->model, "Maxtor"))
>> - return 3;
>> - return 4;
>> + if (is_sata(hwif)) {
>> + mask = strstr(drive->id->model, "Maxtor") ? 0x3f : 0x7f;
>> + goto out;
>> }
>> -
>> +
>> if ((scsc & 0x30) == 0x10) /* 133 */
>> - mode = 4;
>> + mask = 0x7f;
>> else if ((scsc & 0x30) == 0x20) /* 2xPCI */
>> - mode = 4;
>> + mask = 0x7f;
>> else if ((scsc & 0x30) == 0x00) /* 100 */
>> - mode = 3;
>> + mask = 0x3f;
>> else /* Disabled ? */
>> BUG();
>
> Most probably this is doable at init. time also...

ditto

patches welcomed (I added this to the IDE tree TODO)

> MBR, Sergei
>
> PS: I understand that the intent wasn not to make this rewrite optimal but
> do it quick and with least affect. And I certainly may handle hpt366.c
> redesign... :-)

Please do.

Thanks,
Bart
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: Adam Kropelin: "Re: MSI failure on nForce 430 (WAS: intel 82571EB gigabit fails to see link on 2.6.20-rc5 in-tree e1000 driver (regression))"
Previous message: Andrew Morton: "Re: [patch 9/9] mm: fix pagecache write deadlocks"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]