- Various comment fixes and additions.

- Add 2 explicit (paranoid?) memory barriers in the
  interrupt code (After the reading of the `flag' and
  prior to looking at the data, of course. :-) ).
- Remove obsolete informations from the README.sym file.

This commit actually results in no object difference
for IA32, but 2x`mb' added for Alpha.

2 files changed, 62 insertions, 267 deletions

diff --git a/sys/dev/sym/README.sym b/sys/dev/sym/README.sym
index fb277c7..a4e19ed 100644
--- a/sys/dev/sym/README.sym
+++ b/sys/dev/sym/README.sym
@@ -2,13 +2,12 @@
  *  Device driver optimized for the Symbios/LSI 53C896/53C895A/53C1010 
  *  PCI-SCSI controllers.
  *
- *  Copyright (C) 1999  Gerard Roudier <groudier@club-internet.fr>
+ *  Copyright (C) 1999-2000  Gerard Roudier <groudier@club-internet.fr>
  *
  *  This driver also supports the following Symbios/LSI PCI-SCSI chips:
- *	53C810A, 53C825A, 53C860, 53C875, 53C876, 53C885, 53C895.
+ *	53C810A, 53C825A, 53C860, 53C875, 53C876, 53C885, 53C895,
+ *	53C810,  53C815,  53C825 and the 53C1510D is 53C8XX mode.
  *
- *  but does not support earlier chips as the following ones:
- *	53C810, 53C815, 53C825.
  *  
  *  This driver for FreeBSD-CAM is derived from the Linux sym53c8xx driver.
  *  Copyright (C) 1998-1999  Gerard Roudier
@@ -58,242 +57,22 @@
  * $FreeBSD$
  */
 
-Status:
-   BETA-RELEASE
-   This driver is currently developped under the following O/S:
-         FreeBSD 4.0 SNAPSHOT July 5th 1999
-         FreeBSD 4.0 current  October 1999
-   Starting with version 0.9.0, the driver contains conditionnal code 
-   for using pre-4.0 BUS stuff for PCI. I have successfully tested driver 
-   version 0.9.0 under:
-         FreeBSD 3.2 RELEASE
-         FreeBSD 3.3 RELEASE
-
-Latest revision:
-   sym-0.12.0-19991127
-   Files to download:
-      SYM-0.9.0-19991024.tar.gz
-      PATCH-SYM-0.10.0-19991111.gz
-      PATCH-SYM-0.11.0-19991120.gz
-      PATCH-SYM-0.12.0-19991127.gz
-
-Supported SCSI features:
-  - Initiator mode
-  - Wide 16 SCSI BUS
-  - FAST10 up to FAST80-DT synchronous data transfers
-    (depends on controller capabilities)
-  - 64 luns per target
-  - 256 tags per lun
-  - MDP (1)
-  - BUS DEVICE RESET message
-  - ABORT, ABORT TAG message
-
-Supported generic chip features:
-  - On chip RAM
-  - PCI optimized transactions.
-
-Supported features of 896/895A/1010:
-  - Full SCRIPTS in on-board RAM (8K)
-  - Phase mismatch handling from SCRIPTS.
-
-Others:
-  - SYMBIOS NVRAM layout using 24c16 EEPROM
-  - TEKRAM  NVRAM layout using 24c16 EEPROM (2)
-  - TEKRAM  NVRAM layout using 93c46 EEPROM (2)
-  - Transfer residuals
-  - 0(1) for both SCRIPTS and C code, allowing 
-    100% scalability.
-
-Notes:
-(1) Not yet reported. May not work as expected.
-   
-Files:
-  README.sym		this file
-  sym_conf.h		driver configuration
-  sym_defs.h		driver definitions
-  sym_hipd.c		driver code
-  sym_sys.patch		kernel patch for FreeBSD 4
-  sym_sys_3.patch	kernel patch for FreeBSD 3
-
-These patches apply a change to the ncr.c driver that forces it to 
-ignore SYM53C8XX chips that support LOAD/STORE. These devices will 
-be attached by the sym_hipd.c driver.
-
-Installation:
-  1) Untar SYM-0.9.0-19991024.tar.gz
-  2) Create the /usr/src/sys/dev/sym directory
-  3) Copy README.sym, sym_conf.h, sym_defs.h and sym_hipd.c to 
-     /usr/src/sys/dev/sym/
-  4) Change to /usr/src/sys/ directory
-  5) If FreeBSD-4, apply the unified patch sym_sys.patch
-         (patch -p0 <sym_sys.patch)
-     Otherwise,
-     If FreeBSD-3, apply the unified patch sym_sys_3.patch
-         (patch -p0 <sym_sys_3.patch)
-     Otherwise, leave here. ;)
-  6) Change to /usr/src/sys/dev/sym directory
-  7) Update to SYM-0.11.0-19991120
-     gzip -d <PATCH-SYM-0.10.0-19991111.gz | patch -p0
-     gzip -d <PATCH-SYM-0.11.0-19991120.gz | patch -p0
-     gzip -d <PATCH-SYM-0.12.0-19991127.gz | patch -p0
-  8) Add sym0 to your kernel configuration as indicated in patched LINT file.
-     Configure and make your kernel.
-
-As seen from the shell history file (driver material assumed from /tmp), 
-installation should proceed as follow (modulo mistakes from me:)):
-
-cd /tmp
-tar zxvf SYM-0.9.0-19991024.tar.gz
-cp SYM-0.9.0-19991024/* /usr/src/sys/dev/sym
-cd /usr/src/sys
-patch -p0 <dev/sym/sym_sys.patch    # use sym_sys_3.patch if FreeBSD-3.X
-cd dev/sym
-gzip -d </tmp/PATCH-SYM-0.10.0-19991111.gz | patch -p0
-gzip -d </tmp/PATCH-SYM-0.11.0-19991120.gz | patch -p0
-gzip -d </tmp/PATCH-SYM-0.12.0-19991127.gz | patch -p0
-
-Configuring Ultra-3 DT data transfer support.
----------------------------------------------
-If you have a board using the SYM53C1010 (only evaluation boards seem to 
-be available for now), you may want to give Ultra-3 DT transfer a try.
-For the reasons that FreeBSD-CAM is not yet ready for Ultra-3 and that 
-early SYM53C1010 chips need some work-around for DT transfers to work 
-reliably, you must apply manually the following additionnal patch to your 
-kernel tree:
-
----------------------- Cut Here ------------------------
---- cam/scsi/scsi_all.c.00	1999/08/29 16:21:44
-+++ cam/scsi/scsi_all.c	1999/11/23 22:11:22
-@@ -2394,6 +2394,7 @@
-         u_int period_factor;
-         u_int period;	/* in 10ths of ns */
- } scsi_syncrates[] = {
-+        { 0x09, 125 },
-         { 0x0a, 250 },
-         { 0x0b, 303 },
-         { 0x0c, 500 }
---- pci/sym_conf.h.00	Fri Nov 26 22:58:59 1999
-+++ pci/sym_conf.h	Fri Nov 26 22:58:38 1999
-@@ -69,6 +69,7 @@
-  *  corresponding code will get useless.
-  */
- /* #define	SYMCONF_BROKEN_U3EN_SUPPORT */
-+#define	SYMCONF_BROKEN_U3EN_SUPPORT
- 
- /*
-  *  Use Normal IO instead of MMIO.
----------------------- Cut Here ------------------------
-
-This change consist in allowing sync factor 9 support to be handled 
-by scsi_all.c and a compilation option to be defined in sym_conf.h.
-The driver may only start a PPR negotiation if sync factor is 9.
-This change has been only tested on FreeBSD-3.3 for the moment.
-
-For the PPR negotiation to occur at system startup time, you need to 
-configure Ultra3 hard disks in the NVRAM for 80 Mega-transfers per second 
-Wide.
-If you prefer to use camcontrol for such a negotiation to be performed after 
-system startup, then you have to rebuild either the libcam shareable library 
-with the patched version of scsi_all.c, on to rebuild camcontrol using 
-the new static version of the libcam library that includes the patched 
-version of scsi_all.o.
-
-Warnings:
-- This driver uses LOAD/STORE instructions from SCRIPTS and therefore does 
-  not support earliest NCR chips (NCR53C810, NCR53C815, NCR53C825).
-
-Versionning:
-- The version number consists in 3 numbers X.Y.Z, starting with 0.0.0.
-- Z will not be used for incremental patches, but may be used for 
-  quick fixes.
-- Incremental diff files are of the following pattern:
-            PATCH-SYM-X.Y.Z-YYYYMMDD
-  They must be applied from directory: /usr/src/sys/dev/sym/
-- Version 1.0.0 will happen when the driver will be tested enough but this 
-  driver version will probably not officially support the C1010 (Ultra-3) 
-  since the testing of Ultra3 has just started and FreeBSD-CAM is not yet  
-  ready for the support of Ultra-3. Some 1.X.0 version will support 
-  DT data transfer for the C1010.
-- Version 2.0.0 is not planned for now, but will add support for host target 
-  mode if it will ever exist.
-
-Change log:
-* SYM-0.0.0-19990915
-  Initial release.
-
-* SYM-0.1.0-19990919  (diff file PATCH-SYM-0.1.0-19990919
-  Add NVRAM support for latest Tekram boards using 24c16 EEPROM.
-
-* SYM-0.2.0-19990922  (diff file PATCH-SYM-0.2.0-19990922
-  Add PPR negotiation and Ultra3 DT transfers for the LSI53C1010.
-  This corresponding code is untested since I haven't yet the 
-  hardware. But the driver is not broken for current chips.
-
-* SYM-0.3.0-19990925  (diff file PATCH-SYM-0.3.0-19990925
-  Source review.
-  Testing of the QUEUE FULL handling (some fixes applied).
-
-* SYM-0.4.0-19990925  (diff file PATCH-SYM-0.4.0-19990928)
-  Fixes, notably the QUEUE FULL handling that requeued everything 
-  without telling the XPT about the error. Note that this did not 
-  break anything, but the queue depth was never reduced.
-
-* SYM-0.5.0-19991001  (diff file PATCH-SYM-0.5.0-19991001)
-  Problem of not discovering LUNs != 0 hopefully fixed.
-  Some other minor fixes.
-
-* SYM-0.6.0-19991003  (diff file PATCH-SYM-0.6.0-19991003)
-  Test error recovery pathes and residual calculation.
-  Improve Symbios NVRAM support by applying the SCAN_LUNS flag. The driver 
-  answers DEV_NOT_THERE to INQUIRYs for LUNs that are not to be scanned.
-  Switch the driver status for SPI2 support to BETA-RELEASE.
-
-* SYM-0.7.0-19991006  (diff file PATCH-SYM-0.7.0-19991006)
-  Add support of the LSI1510D that emulates a 895 (infos sent by Compaq 
-  that seems to be the only supplier of this chip)
-  Return CAM_REQ_ABORTED on HS_ABORT host status.
-  Fix a tiny bug that let the driver miss the TAG setting.
-
-* SYM-0.8.0-19991016  (diff file PATCH-SYM-0.8.0-19991016)
-  Heavy rewrite/rewamping of the error handling code in order to conform 
-  to FreeBSD-CAM expectation of commands having to be requeued and device 
-  queue to be frozen on any kind of error. Driver status switched back 
-  to EXPERIMENTAL due to these changes.
-  Warn user about targets that are not flagged for SCAN AT BOOT in the 
-  NVRAM, since the driver will not allow the corresponding devices to be 
-  discovered by XPT.
-  And numerous minor changes.
-
-* SYM-0.9.0-19991024  (Tar file SYM-0.9.0-19991024.tar.gz)
-  Add support for previous PCI BUS stuff, for the driver to be useable 
-  under FreeBSD 3.X RELEASES. The source is #ifdefed for using the old 
-  PCI BUS code under __FreeBSD_version < 400000 and the new one otherwise.
-  Tested under RELENG_3_2_0_RELEASE, but should also be just fine for 3.3.
-  Other changes against version 0.8.0 are only minor buglet fixes.
-
-* SYM-0.10.0-19991111  (diff file PATCH-SYM-0.10.0-19991111)
-  Add support for Alpha - UNTESTED. Consists in some minor changes picked 
-  mostly from the ncr driver and some others as ahc_pci.c.
-  Fix the LED support through GPIO0 for pre-896 chips. In fact some code 
-  was wrongly conditionned by something that was never met.
-  Switch driver status to BETA-RELEASE given that this driver has been rock 
-  solid even if a small number of users seems to actually use it.
-
-* SYM-0.11.0-19991120  (diff file PATCH-SYM-0.11.0-19991120)
-  Rearrange a bit the initialisation code that deals with IO registers.
-  Fix a bug in MODIFY_DP/IGN_RESIDUE handling (May-be this code will 
-  never be used).
-  Fix a tiny problem in Wide nego. The driver didn't start the Sync nego 
-  after a Wide nego if synchronous parameters weren't also changed.
-  Apply fixes that should allow to successfully link the kernel on Alpha.
-  LSI53C1010 tested with Asynchronous, FAST5, FAST10, FAST20 and FAST40 
-  data transfers.
-
-* SYM-0.12.0-19991127  (diff file PATCH-SYM-0.12.0-19991127)
-  Some testing in Ultra3 FAST-80 DT mode using a SYM53C1010 connected 
-  to an ATLAS 4 Ultra3 disk. Required some tiny fix in the ppr nego code.
-  Some cosmetic changes in messages displayed under DEBUG.
-  Fix the chip table (and code) that made the driver wrongly attach 
-  810 and 825 devices.
-
-November 27 1999.          "Gerard Roudier"<groudier@club-internet.fr>
+README.sym
+----------
+This file was bundled with the `sym' driver source files, when 
+the driver was in development stage and not included yet into 
+the FreeBSD repository. It described how to install the driver, 
+gave informations on driver features and summerized important 
+changes between versions. These informations are now available 
+using `man sym' for driver usage and features description, and 
+`cvs log' for driver changes.
+
+I haven't removed this README file from the repository since 
+I may have to provide additional documentation to people 
+interested in understanding the driver sources and/or in helping 
+driver development and maintainance.
+
+As a result, this file is now under reconstruction. :-)
+
+--
+"Gerard Roudier"<groudier@club-internet.fr>
diff --git a/sys/dev/sym/sym_hipd.c b/sys/dev/sym/sym_hipd.c
index ca60b0c..d5d7cd8 100644
--- a/sys/dev/sym/sym_hipd.c
+++ b/sys/dev/sym/sym_hipd.c
@@ -57,7 +57,7 @@
 
 /* $FreeBSD$ */
 
-#define SYM_DRIVER_NAME	"sym-1.6.3-20000626"
+#define SYM_DRIVER_NAME	"sym-1.6.4-20000701"
 
 /* #define SYM_DEBUG_GENERIC_SUPPORT */
 
@@ -305,7 +305,7 @@ static __inline struct sym_quehead *sym_remque_tail(struct sym_quehead *head)
 }
 
 /*
- *  This one may be usefull.
+ *  This one may be useful.
  */
 #define FOR_EACH_QUEUED_ELEMENT(head, qp) \
 	for (qp = (head)->flink; qp != (head); qp = qp->flink)
@@ -446,9 +446,9 @@ static void MDELAY(int ms) { while (ms--) UDELAY(1000); }
  *  This allocator has been developped for the Linux sym53c8xx  
  *  driver, since this O/S does not provide naturally aligned 
  *  allocations.
- *  It has the vertue to allow the driver to use private pages 
- *  of memory that will be useful if we ever need to deal with 
- *  IO MMU for PCI.
+ *  It has the advantage of allowing the driver to use private 
+ *  pages of memory that will be useful if we ever need to deal 
+ *  with IO MMUs for PCI.
  */
 
 #define MEMO_SHIFT	4	/* 16 bytes minimum memory chunk */
@@ -1209,7 +1209,7 @@ struct sym_nvram {
 /*
  *  Device quirks.
  *  Some devices, for example the CHEETAH 2 LVD, disconnects without 
- *  saving the DATA POINTER then reconnect and terminates the IO.
+ *  saving the DATA POINTER then reselects and terminates the IO.
  *  On reselection, the automatic RESTORE DATA POINTER makes the 
  *  CURRENT DATA POINTER not point at the end of the IO.
  *  This behaviour just breaks our calculation of the residual.
@@ -1543,7 +1543,7 @@ struct sym_ccbh {
 struct sym_dsb {
 	/*
 	 *  CCB header.
-	 *  Also Assumed at offset 0 of the sym_ccb structure.
+	 *  Also assumed at offset 0 of the sym_ccb structure.
 	 */
 /*0*/	struct sym_ccbh head;
 
@@ -1716,7 +1716,7 @@ struct sym_hcb {
 		rv_ctest5, rv_stest2, rv_ccntl0, rv_ccntl1, rv_scntl4;
 
 	/*
-	 *  Target data used by the CPU.
+	 *  Target data.
 	 */
 	struct sym_tcb	target[SYM_CONF_MAX_TARGET];
 
@@ -2026,7 +2026,7 @@ sym_fw1_patch(hcb_p np)
 #endif	/* SYM_CONF_GENERIC_SUPPORT */
 
 /*
- *  Patch routine for firmware 2.
+ *  Patch routine for firmware #2.
  */
 static void
 sym_fw2_patch(hcb_p np)
@@ -2172,7 +2172,7 @@ sym_fw1_setup(hcb_p np, struct sym_fw *fw)
 #endif	/* SYM_CONF_GENERIC_SUPPORT */
 
 /*
- *  Setup routine for firmware 2.
+ *  Setup routine for firmware #2.
  */
 static void 
 sym_fw2_setup(hcb_p np, struct sym_fw *fw)
@@ -2527,8 +2527,6 @@ static void PRINT_ADDR (ccb_p cp)
 
 /*
  *  Take into account this ccb in the freeze count.
- *  The flag that tells user about avoids doing that 
- *  more than once for a ccb.
  */	
 static void sym_freeze_cam_ccb(union ccb *ccb)
 {
@@ -2614,7 +2612,6 @@ static u32 div_10M[] = {2*_5M, 3*_5M, 4*_5M, 6*_5M, 8*_5M, 12*_5M, 16*_5M};
  *
  *  For PCI 32 bit data transfers each transfer is a DWORD.
  *  It is a QUADWORD (8 bytes) for PCI 64 bit data transfers.
- *  Only the 896 is able to perform 64 bit data transfers.
  *
  *  We use log base 2 (burst length) as internal code, with 
  *  value 0 meaning "burst disabled".
@@ -2924,8 +2921,8 @@ static int sym_prepare_setting(hcb_p np, struct sym_nvram *nvram)
 	/*
 	 *  Set LED support from SCRIPTS.
 	 *  Ignore this feature for boards known to use a 
-	 *  specific GPIO wiring and for the 895A or 896 
-	 *  that drive the LED directly.
+	 *  specific GPIO wiring and for the 895A, 896 
+	 *  and 1010 that drive the LED directly.
 	 */
 	if ((SYM_SETUP_SCSI_LED || nvram->type == SYM_SYMBIOS_NVRAM) &&
 	    !(np->features & FE_LEDC) && !(np->sv_gpcntl & 0x01))
@@ -3266,6 +3263,10 @@ out:
 
 /*
  *  The chip may have completed jobs. Look at the DONE QUEUE.
+ *
+ *  On architectures that may reorder LOAD/STORE operations, 
+ *  a memory barrier may be needed after the reading of the 
+ *  so-called `flag' and prior to dealing with the data.
  */
 static int sym_wakeup_done (hcb_p np)
 {
@@ -3285,6 +3286,7 @@ static int sym_wakeup_done (hcb_p np)
 
 		cp = sym_ccb_from_dsa(np, dsa);
 		if (cp) {
+			MEMORY_BARRIER();
 			sym_complete_ok (np, cp);
 			++n;
 		}
@@ -3567,8 +3569,8 @@ sym_getsync(hcb_p np, u_char dt, u_char sfac, u_char *divp, u_char *fakp)
 		kpc <<= 1;
 
 	/*
-	 *  For earliest C10, the extra clocks does not apply 
-	 *  to CRC cycles, so it may be safe not to use them.
+	 *  For earliest C10 revision 0, we cannot use extra 
+	 *  clocks for the setting of the SCSI clocking.
 	 *  Note that this limits the lowest sync data transfer 
 	 *  to 5 Mega-transfers per second and may result in
 	 *  using higher clock divisors.
@@ -4003,13 +4005,15 @@ static void sym_intr1 (hcb_p np)
 
 	/*
 	 *  interrupt on the fly ?
+	 *
+	 *  A `dummy read' is needed to ensure that the 
+	 *  clear of the INTF flag reaches the device 
+	 *  before the scanning of the DONE queue.
 	 */
 	istat = INB (nc_istat);
 	if (istat & INTF) {
 		OUTB (nc_istat, (istat & SIGP) | INTF | np->istat_sem);
-#if 1
 		istat = INB (nc_istat);		/* DUMMY READ */
-#endif
 		if (DEBUG_FLAGS & DEBUG_TINY) printf ("F ");
 		(void)sym_wakeup_done (np);
 	};
@@ -4051,6 +4055,12 @@ static void sym_intr1 (hcb_p np)
 			(unsigned)INL(nc_dsp),
 			(unsigned)INL(nc_dbc));
 	/*
+	 *  On paper, a memory barrier may be needed here.
+	 *  And since we are paranoid ... :)
+	 */
+	MEMORY_BARRIER();
+
+	/*
 	 *  First, interrupts we want to service cleanly.
 	 *
 	 *  Phase mismatch (MA) is the most frequent interrupt 
@@ -4556,8 +4566,8 @@ static void sym_int_ma (hcb_p np)
 	 *
 	 *  Look at the PM_SAVE SCRIPT if you want to understand 
 	 *  this stuff. The equivalent code is implemented in 
-	 *  SCRIPTS for the 895A and 896 that are able to handle 
-	 *  PM from the SCRIPTS processor.
+	 *  SCRIPTS for the 895A, 896 and 1010 that are able to 
+	 *  handle PM from the SCRIPTS processor.
 	 */
 	hflags0 = INB (HF_PRT);
 	hflags = hflags0;
@@ -7265,7 +7275,7 @@ static void sym_complete_error (hcb_p np, ccb_p cp)
 	}
 
 	/*
-	 *  Get command, target and lun pointers.
+	 *  Get CAM command pointer.
 	 */
 	csio = &cp->cam_ccb->csio;
 
@@ -7940,6 +7950,12 @@ sym_execute_ccb(void *arg, bus_dma_segment_t *psegs, int nsegs, int error)
 	 */
 	sym_enqueue_cam_ccb(np, ccb);
 
+	/*
+	 *  When `#ifed 1', the code below makes the driver 
+	 *  panic on the first attempt to write to a SCSI device.
+	 *  It is the first test we want to do after a driver 
+	 *  change that does not seem obviously safe. :)
+	 */
 #if 0
 	switch (cp->cdb_buf[0]) {
 	case 0x0A: case 0x2A: case 0xAA:
@@ -9244,8 +9260,8 @@ sym_pci_attach2(pcici_t pci_tag, int unit)
 
 	/*
 	 *  Prepare the bus address array that contains the bus 
-	 *  address of each target control bloc.
-	 *  For now, assume all logical unit are wrong. :)
+	 *  address of each target control block.
+	 *  For now, assume all logical units are wrong. :)
 	 */
 	for (i = 0 ; i < SYM_CONF_MAX_TARGET ; i++) {
 		np->targtbl[i] = cpu_to_scr(vtobus(&np->target[i]));