These patches are based on Martin Petersen's 5.20/scsi-queue tree https://git.kernel.org/pub/scm/linux/kernel/git/mkp/scsi.git 5.20/scsi-queue This set of changes consists of: * Remove a device from the OS faster by adding -ENODEV return code check in pqi_lun_reset. This status is set in the io_request->status member. Schedule the rescan worker thread within 5 seconds to initiate the removal. The driver used to retry a reset without checking for a device's removal and initiated 3 more retries. Device resets were taking up to 30 seconds. We also added a check to see if the controller firmware is still responsive during a reset operation. * Add the controller firmware version to the console logs. The firmware version is still in sysfs firmware_version. * Add support for more controllers; Ramaxel, Lenovo, and Adaptec. * Close a few rare read/write ordering issues where a register read could pass a register write. * Add support for multi-actuator devices. Our controllers now support up to 256 LUNs per multi-actuator device. We added a feature bit to check if the controller supports multi-actuator devices and updated support in the driver to support resets, I/O submission, and multi-actuator device removals. * Correct some rare system hangs that can occur when a PCI link-down condition occurs (such as a cable pull). We also fail all outstanding requests when a link-down is detected. * Correct an issue with setting the DMA direction flag for RAID path requests. It should be noted that there are two submission paths for requests in the driver, a RAID path and an Accelerated I/O (AIO) path. Beginning with firmware version 5.0 for Gen1 controllers and 3.01.x for Gen2 controllers, a change was made that removed the SCSI command READ BLOCK LIMITS (0x05) from an internal lookup table for RAID path requests. As a result of this change, the firmware switched to using the DMA direction flag in the request IU, which was incorrect. This caused the command to hang the controller. This patch resolves the hang. The AIO path is unaffected by the controller firmware change. * correct a rare device RAID map access race condition related to configuration changes. We do not access the RAID map until after the new RAID map is valid. * added a module parameter 'disable_managed_interrupts' to allow customers to change IRQ affinity. Multi-queue still works properly. * Updated device removal to using .slave_destroy instead of using our own internal method. * Added another module parameter to reduce the amount of time the driver waits for a controller to become ready. The default wait time is 3 minutes but can be extended to 30 minutes. This change results from customers with large installations requesting a longer wait time. * Updated copyright information. * Bump the driver version to 2.1.18-045 --- Don Brace (2): smartpqi: update copyright to current year. smartpqi: update version to 2.1.18-045 Gilbert Wu (1): smartpqi: add controller fw version to console log Kevin Barnett (4): smartpqi: stop logging spurious PQI reset failures smartpqi: fix RAID map race condition smartpqi: update deleting a LUN via sysfs smartpqi: add ctrl ready timeout module parameter Kumar Meiyappan (1): smartpqi: add driver support for multi-LUN devices Mahesh Rajashekhara (1): smartpqi: fix dma direction for RAID requests Mike McGowen (5): smartpqi: shorten drive visibility after removal smartpqi: close write read holes smartpqi: add PCI-ID for Adaptec SmartHBA 2100-8i smartpqi: add PCI-IDs for Lenovo controllers smartpqi: add module param to disable managed ints Murthy Bhat (1): smartpqi: add PCI-IDs for ramaxel controllers Sagar Biradar (1): smartpqi: fix PCI control linkdown system hang drivers/scsi/smartpqi/Kconfig | 2 +- drivers/scsi/smartpqi/smartpqi.h | 27 +- drivers/scsi/smartpqi/smartpqi_init.c | 405 +++++++++++++----- .../scsi/smartpqi/smartpqi_sas_transport.c | 2 +- drivers/scsi/smartpqi/smartpqi_sis.c | 11 +- drivers/scsi/smartpqi/smartpqi_sis.h | 4 +- 6 files changed, 339 insertions(+), 112 deletions(-) -- Signature
