Re: [PATCH v3 1/1] libata: Add ATA_HORKAGE_NONCQ_ON_AMD for Samsung 860 and 870 SSD.

[Damien Le Moal <Damien.LeMoal@wdc.com>]

On 2021/08/30 23:43, Kate Hsuan wrote:
> Many users are reporting that the Samsung 860 and 870 SSD are having
> various issues when combined with AMD SATA controllers and only
> completely disabling NCQ helps to avoid these issues.
> 
> Entire disabling NCQ for Samsugn 860/870 SSD will cause I/O performance
> drop. In this case, a flag ATA_HORKAGE_NONCQ_ON_AMD is introduced to

With "Entire disabling NCQ...", did you mean "Always disabling NCQ" ?
If I understand this issue correctly, the explanation should be something like:

Always disabling NCQ for Samsung 860/870 SSDs regardless of the host SATA
adapter vendor will cause I/O performance degradation with well behaved
adapters. To limit the performance impact to AMD adapters, introduce the
ATA_HORKAGE_NO_NCQ_ON_AMD flag to force disable NCQ only for these adapters.

> used to perform an additional check for these SSDs. If it finds it's
> parent ATA controller is AMD, the NCQ will be disabled. Otherwise, the
> NCQ is kept to enable.
> 
> BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=201693
> Signed-off-by: Kate Hsuan <hpa@redhat.com>
> ---
>  drivers/ata/libata-core.c | 24 ++++++++++++++++++++++--
>  include/linux/libata.h    |  1 +
>  2 files changed, 23 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
> index c861c93d1e84..36c62f758b73 100644
> --- a/drivers/ata/libata-core.c
> +++ b/drivers/ata/libata-core.c
> @@ -2190,6 +2190,8 @@ static int ata_dev_config_ncq(struct ata_device *dev,
>  			       char *desc, size_t desc_sz)
>  {
>  	struct ata_port *ap = dev->link->ap;
> +	struct pci_dev *pcidev = NULL;
> +	struct device *parent_dev = NULL;
>  	int hdepth = 0, ddepth = ata_id_queue_depth(dev->id);
>  	unsigned int err_mask;
>  	char *aa_desc = "";
> @@ -2204,6 +2206,22 @@ static int ata_dev_config_ncq(struct ata_device *dev,
>  		snprintf(desc, desc_sz, "NCQ (not used)");
>  		return 0;
>  	}
> +
> +	if (dev->horkage & ATA_HORKAGE_NONCQ_ON_AMD) {
> +		for (parent_dev = dev->tdev.parent; parent_dev != NULL;
> +		    parent_dev = parent_dev->parent) {
> +			if (dev_is_pci(parent_dev)) {
> +				pcidev = to_pci_dev(parent_dev);
> +				if (pcidev->vendor == PCI_VENDOR_ID_AMD) {
> +					snprintf(desc, desc_sz,
> +						 "NCQ (not used)");
> +					return 0;
> +				}
> +			break;
> +			}
> +		}

It would be really nice to move this hunk into a small helper, something like:

static bool ata_dev_check_adapter(struct ata_device *dev,
				  unsigned short vendor_id)

The "if" code block then becomes:

	if ((dev->horkage & ATA_HORKAGE_NONCQ_ON_AMD) &&
	    ata_dev_check_adapter(dev, PCI_VENDOR_ID_AMD)) {
		snprintf(desc, desc_sz, "NCQ (not used)");
		return 0;
	}


> +	}
> +
>  	if (ap->flags & ATA_FLAG_NCQ) {
>  		hdepth = min(ap->scsi_host->can_queue, ATA_MAX_QUEUE);
>  		dev->flags |= ATA_DFLAG_NCQ;
> @@ -3971,9 +3989,11 @@ static const struct ata_blacklist_entry ata_device_blacklist [] = {
>  	{ "Samsung SSD 850*",		NULL,	ATA_HORKAGE_NO_NCQ_TRIM |
>  						ATA_HORKAGE_ZERO_AFTER_TRIM, },
>  	{ "Samsung SSD 860*",           NULL,   ATA_HORKAGE_NO_NCQ_TRIM |
> -						ATA_HORKAGE_ZERO_AFTER_TRIM, },
> +						ATA_HORKAGE_ZERO_AFTER_TRIM |
> +						ATA_HORKAGE_NONCQ_ON_AMD, },
>  	{ "Samsung SSD 870*",           NULL,   ATA_HORKAGE_NO_NCQ_TRIM |
> -						ATA_HORKAGE_ZERO_AFTER_TRIM, },
> +						ATA_HORKAGE_ZERO_AFTER_TRIM |
> +						ATA_HORKAGE_NONCQ_ON_AMD, },
>  	{ "FCCT*M500*",			NULL,	ATA_HORKAGE_NO_NCQ_TRIM |
>  						ATA_HORKAGE_ZERO_AFTER_TRIM, },
>  
> diff --git a/include/linux/libata.h b/include/linux/libata.h
> index 860e63f5667b..42e16114e91f 100644
> --- a/include/linux/libata.h
> +++ b/include/linux/libata.h
> @@ -426,6 +426,7 @@ enum {
>  	ATA_HORKAGE_NOTRIM	= (1 << 24),	/* don't use TRIM */
>  	ATA_HORKAGE_MAX_SEC_1024 = (1 << 25),	/* Limit max sects to 1024 */
>  	ATA_HORKAGE_MAX_TRIM_128M = (1 << 26),	/* Limit max trim size to 128M */
> +	ATA_HORKAGE_NONCQ_ON_AMD = (1 << 27),	/* Disable NCQ on AMD chipset */

Please add a "_" after "NO", similarly to ATA_HORKAGE_NO_NCQ_TRIM:

	ATA_HORKAGE_NO_NCQ_ON_AMD = (1 << 27),	/* Disable NCQ on AMD chipset */

>  
>  	 /* DMA mask for user DMA control: User visible values; DO NOT
>  	    renumber */
> 


-- 
Damien Le Moal
Western Digital Research