### Version 2.10.1 Wed Dec 3 15:34:42 EST 2003 - Atul Mukker 1. pci_dma_sync_sg(), 2nd argument is pointer to scatter-gather list. All previous drivers have a pointer which is incremented beyond the end of the scatter-gather list 2. Remove 'ipdev', pci device pointer, to allocate memory for internal commands. pci_alloc_consistent() guaranteed to allocate memory below 4GB, which is a requirement for these commands - Tom Coughlan 3. Added support for LSI SATA PCI-X controllers 4. Advanced Server 3.0 version of the driver is the reference driver now. For all other kernels, we create patches. 5. Removed superfluous white spaces. 6. Corrected some comments ### Version 2.10.0 Tue Nov 4 14:33:43 EST 2003 - Atul Mukker 1. Added vendor ids and device ids for PCI-Express controllers, PERC4E/Si, PERC4E/Di, PERC4E/DC, PERC4E/SC 2. Backport some minor changes from 2.6 kernel version of the driver. ### Version 2.00.9a Fri Oct 3 18:04:29 EDT 2003 - Atul Mukker 1. Minor changes which are brough in when sync'ing with kernel 2.9-test6. 2. Use sizeof(mbox_t) in synchronous routines instead of hard-coded value of 16 bytes. 3. De-couple adapter->host->pci_dev. Replace with adapter->pdev ### Version 2.00.9 Thu Sep 4 17:49:42 EDT 2003 - Atul Mukker i. For extended passthru commands, 64-bit scatter-gather list and 64-bit mailbox address must only be used if the controller supports extended CDBs and 64-bit addressing and the kernel is configured to support memory beyond 4GB. With 2.00.8 and previous drivers, if the controllers supports extended CDBs but the kernel does not support high memory - the driver prepares a 32-bit sg list but the mailbox address is chosen to be 64-bit. This causes FW to incorrectly read 32-bit (8 bytes) sg list as 64-bit (12 bytes) sglist. This would cause IO from non-disk devices to fail ### Version 2.00.8 Wed Aug 27 16:40:05 EDT 2003 - Atul Mukker i. Make sure the value of number of statuses, completed command id array, and the mailbox status fields are updated in host memory before we read and interpret them. For this to happen - mailbox numstatus and command id array's first element are invalidated. The ISR busy waits till valid values are obtained in these two fields. Now since the status field is between these two fields it is assumed that status values is sane when write to numstatus and completed id array is complete ii. while returning from the ioctl handler for the SCSI passthru commands a direct access was made to the user address. Now the user structure is copied in before the required field is accessed. iii. Remove redundant volatile casts from pending command counter pend_cmds. ### Version 2.00.7 Fri Aug 1 11:01:11 EDT 2003 - Atul Mukker i. Adapter lock re-definition so that patch for kernels w/o per host lock is less intrusive - Jens Axboe ii. While in abort and reset handling, check for non-empty pending list is invalid. The intent is to wait for pending commands in FW to complete, not the pending commands with the driver - Atul Mukker ### Version 2.00.6 Wed Jul 30 11:35:31 EDT 2003 - Sreenivas Bagalkote i. Declare the function prototypes used for "/proc" within the compiler directives #ifdef CONFIG_PROC_FS. Similary, move global/local variable declarations and code fragments related to /proc within the directives. - Mark Haverkamp ii. Initialize host->lock with adapter->lock. iii. Wait for mailbox status to become valid - Haruo Tomita iv. Right structure is passed to FW for 4-span read configuration command. - Sreenivas Bagalkote ### Version 2.00.5 Thu Apr 24 14:06:55 EDT 2003 - Atul Mukker i. Do not use repeated allocations for "pci_dev" for internal allocations. Allocate the handle at load time and set the DMA mask for allocations below 4GB. ii. Remove superfluous definitions, mid-layer /proc entry and synchronous commands iii. Logical drive numbers were incremented in passthru structure after random-deletion operation! Also, the logical drive deletion command is issued in polled mode because of the racy nature of interrupt based operation. iv. Add support for Intel's subsystem vendor id ### Version 2.00.4 Thu Apr 17 15:58:58 EDT 2003 i. Do not put the completed SCSI commands in a list. Complete them to mid-layer as soon as we have completed them and reclaimed our associated resources ii. Break ISR functionality in two portions. The lower half serves as interrupt acknowledgment sequence for the firmware. This half can also be called from other places in the driver, e.g., the abort and reset handlers. iii. New abort and reset handling. In these situations, driver allows more time for the firmware to complete outstanding requests instead of failing handlers right-away. ### Version 2.00.3 Wed Jan 29 09:13:44 EST 200 - Atul Mukker i. Change the handshake in ISR while acknowledging interrupts. Write the valid interrupt pattern 0x10001234 as soon as it is read from the outdoor register. In existing driver and on certain platform, invalid command ids were being returned. Also, do not wait on status be become 0xFF, since FW can return this status in certain circumstances. Initialize the numstatus field of mailbox to 0xFF so that we can wait on this wait in next interrupt. Firmware does not change its value unless there are some status to be posted ii. Specify the logical drive number while issuing the RESERVATION_STATUS iii. Reduce the default mailbox busy wait time from 300us to 10us. This is done to avoid a possible deadlock in FW because of longer bust waits. iv. The max outstanding commands are reduced to 126 because that's the safest value on all FW. v. Number of sectors per IO are reduced to 128 (64kb), because FW needs resources in special circumstances like check consistency, rebuilds etc. vi. max_commands is no longer a module parameter because of iv. ### Version: 2.00.2 i. Intermediate release with kernel specific code ### Version: 2.00.1i Wed Dec 4 14:34:51 EST 2002 - Atul Mukker i. Making the older IO based controllers to work with this driver ### Version 2.00.1 Fri Nov 15 10:59:44 EST 2002 - Atul Mukker i. Release host lock before issuing internal command to reset reservations in megaraid_reset() and reacquire after internal command is completed.