Re: [PATCH 14/15] scsi: qla2xxx: add back SRR support

[Tony Battersby <tonyb@cybernetics.com>]

On 9/25/25 13:00, Tony Battersby wrote:
> On 9/25/25 12:04, Tony Battersby wrote:
>> On 9/25/25 11:30, Xose Vazquez Perez wrote:
>>> On 9/25/25 2:49 PM, Xose Vazquez Perez wrote:
>>>
>>>> If you want to review the firmware changelog, mainly: FCD-1183 (FCD-371, ER147301), FCD-259, ER146998
>>>> (from 9.00.00 to 9.15.05 [06/10/25]):
>>>> https://www.marvell.com/content/dam/marvell/en/drivers/2025-06-10-release/fw_release_notes/Fibre_Channel_Firmware_Release_Notes.pdf
>>>>
>>>> It's look like all 2{678}xx devices/chips are affected by this bug.
>>>> Perhaps the Marvel crew could provide more information on this.
>>> 267x, or older, is still on 8.08, so apparently it's free of this bug:
>>> https://www.marvell.com/content/dam/marvell/en/drivers/release-matrix/release-matrix-qlogic-fc-sw-posting-by-release-matrix.pdf
>>>
>>> 2870 / 2770 :        9.15.06 FW
>>> 2740 / 2760 / 269x : 9.15.01 FW
>>> 267x :               8.08.231 FW
>> I am still trying to figure out what macros to use to test for the
>> affected HBAs.  So far I have:
>>
>> if (IS_QLA27XX(ha) || IS_QLA28XX(ha))
>>
>> But all the ISP numbers are pretty confusing.  I have a number of 8, 16,
>> 32, and 64 Gbps HBAs lying around to test, but I am sure I don't have
>> every possible model.
>>
>> There are a number of items under "Changes and Fixes from v9.08.00 to
>> v9.09.00" that might be related to the problem that I was experiencing. 
>> For example FCD-1076 sounds similar except that the SRR problem was with
>> the CTIO queue rather than the ATIO queue.  I could expand the "bad
>> firmware" versions to include v9.04.00 - v9.08.00, since v9.04.00
>> introduced ER147301 and v9.09.00 fixed FCD-1183.
>>
>> Thanks,
>> Tony
>>
> This is what I found by checking the PCI IDs on some of my HBAs:
>
> not affected
> QLE2672 16 Gbps ISP2031 8.08.231 FW
>
> affected
> QLE2694 16 Gpbs ISP2071 (tested)
> QLE2742 32 Gbps ISP2261 (not tested)
> QLE2872 64 Gpbs ISP2281 (not tested)
>
> So the following check should cover them:
>
> if (IS_QLA27XX(ha) || IS_QLA28XX(ha))
>
> Tony
>
Here is the updated version of the function:

/*
 * Return true if the HBA firmware version is known to have bugs that
 * prevent Sequence Level Error Recovery (SLER) / Sequence Retransmission
 * Request (SRR) from working.
 *
 * Some bad versions are based on testing and some are based on "Marvell Fibre
 * Channel Firmware Release Notes".
 */
static bool qlt_has_sler_fw_bug(struct qla_hw_data *ha)
{
	bool has_sler_fw_bug = false;

	if (IS_QLA27XX(ha) || IS_QLA28XX(ha)) {
		/*
		 * In the fw release notes:
		 *   ER147301 was added to v9.05.00 causing SLER regressions
		 *   FCD-259  was fixed in v9.08.00
		 *   FCD-371  was fixed in v9.08.00
		 *   FCD-1183 was fixed in v9.09.00
		 *
		 * QLE2694L (ISP2071) known bad firmware (tested):
		 *   9.06.02
		 *   9.07.00
		 *   9.08.02
		 *   SRRs trigger hundreds of bogus entries in the response
		 *   queue and various other problems.
		 *
		 * QLE2694L known good firmware (tested):
		 *   8.08.05
		 *   9.09.00
		 *
		 * Suspected bad firmware (not confirmed by testing):
		 *   v9.05.xx
		 *
		 * unknown firmware:
		 *   9.00.00 - 9.04.xx
		 */
		if (ha->fw_major_version == 9 &&
		    ha->fw_minor_version >= 5 &&
		    ha->fw_minor_version <= 8)
			has_sler_fw_bug = true;
	}

	return has_sler_fw_bug;
}