scsi: mesh: Fix panic after host or bus reset

[ Upstream commit edd7dd22 ]

Booting Linux with a Conner CP3200 drive attached to the MESH SCSI bus
results in EH measures and a panic:

[   25.499838] mesh: configured for synchronous 5 MB/s
[   25.787154] mesh: performing initial bus reset...
[   29.867115] scsi host0: MESH
[   29.929527] mesh: target 0 synchronous at 3.6 MB/s
[   29.998763] scsi 0:0:0:0: Direct-Access     CONNER   CP3200-200mb-3.5 4040 PQ: 0 ANSI: 1 CCS
[   31.989975] sd 0:0:0:0: [sda] 415872 512-byte logical blocks: (213 MB/203 MiB)
[   32.070975] sd 0:0:0:0: [sda] Write Protect is off
[   32.137197] sd 0:0:0:0: [sda] Mode Sense: 5b 00 00 08
[   32.209661] sd 0:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
[   32.332708]  sda: [mac] sda1 sda2 sda3
[   32.417733] sd 0:0:0:0: [sda] Attached SCSI disk
... snip ...
[   76.687067] mesh_abort((ptrval))
[   76.743606] mesh: state at (ptrval), regs at (ptrval), dma at (ptrval)
[   76.810798]     ct=6000 seq=86 bs=4017 fc= 0 exc= 0 err= 0 im= 7 int= 0 sp=85
[   76.880720]     dma stat=84e0 cmdptr=1f73d000
[   76.941387]     phase=4 msgphase=0 conn_tgt=0 data_ptr=24576
[   77.005567]     dma_st=1 dma_ct=0 n_msgout=0
[   77.065456]     target 0: req=(ptrval) goes_out=0 saved_ptr=0
[   77.130512] mesh_abort((ptrval))
[   77.187670] mesh: state at (ptrval), regs at (ptrval), dma at (ptrval)
[   77.255594]     ct=6000 seq=86 bs=4017 fc= 0 exc= 0 err= 0 im= 7 int= 0 sp=85
[   77.325778]     dma stat=84e0 cmdptr=1f73d000
[   77.387239]     phase=4 msgphase=0 conn_tgt=0 data_ptr=24576
[   77.453665]     dma_st=1 dma_ct=0 n_msgout=0
[   77.515900]     target 0: req=(ptrval) goes_out=0 saved_ptr=0
[   77.582902] mesh_host_reset
[   88.187083] Kernel panic - not syncing: mesh: double DMA start !
[   88.254510] CPU: 0 PID: 358 Comm: scsi_eh_0 Not tainted 5.6.13-pmac #1
[   88.323302] Call Trace:
[   88.378854] [e16ddc58] [c0027080] panic+0x13c/0x308 (unreliable)
[   88.446221] [e16ddcb8] [c02b2478] mesh_start.part.12+0x130/0x414
[   88.513298] [e16ddcf8] [c02b2fc8] mesh_queue+0x54/0x70
[   88.577097] [e16ddd18] [c02a1848] scsi_send_eh_cmnd+0x374/0x384
[   88.643476] [e16dddc8] [c02a1938] scsi_eh_tur+0x5c/0xb8
[   88.707878] [e16dddf8] [c02a1ab8] scsi_eh_test_devices+0x124/0x178
[   88.775663] [e16dde28] [c02a2094] scsi_eh_ready_devs+0x588/0x8a8
[   88.843124] [e16dde98] [c02a31d8] scsi_error_handler+0x344/0x520
[   88.910697] [e16ddf08] [c00409c8] kthread+0xe4/0xe8
[   88.975166] [e16ddf38] [c000f234] ret_from_kernel_thread+0x14/0x1c
[   89.044112] Rebooting in 180 seconds..

In theory, a panic can happen after a bus or host reset with dma_started
flag set. Fix this by halting the DMA before reinitializing the host.
Don't assume that ms->current_req is set when halt_dma() is invoked as it
may not hold for bus or host reset.

BTW, this particular Conner drive can be made to work by inhibiting
disconnect/reselect with 'mesh.resel_targets=0'.

Link: https://lore.kernel.org/r/3952bc691e150a7128b29120999b6092071b039a.1595460351.git.fthain@telegraphics.com.au
Fixes: 1da177e4 ("Linux-2.6.12-rc2")
Cc: Paul Mackerras <paulus@ozlabs.org>
Reported-and-tested-by: Stan Johnson <userm57@yahoo.com>
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

drivers/scsi/mesh.c

@@ -1044,6 +1044,8 @@ static void handle_error(struct mesh_state *ms)
 		while ((in_8(&mr->bus_status1) & BS1_RST) != 0)
 			udelay(1);
 		printk("done\n");
+		if (ms->dma_started)
+			halt_dma(ms);
 		handle_reset(ms);
 		/* request_q is empty, no point in mesh_start() */
 		return;
@@ -1356,6 +1358,7 @@ static void halt_dma(struct mesh_state *ms)
 		       ms->conn_tgt, ms->data_ptr, scsi_bufflen(cmd),
 		       ms->tgts[ms->conn_tgt].data_goes_out);
 	}
+	if (cmd)
 		scsi_dma_unmap(cmd);
 	ms->dma_started = 0;
 }
@@ -1711,6 +1714,9 @@ static int mesh_host_reset(struct scsi_cmnd *cmd)
 
 	spin_lock_irqsave(ms->host->host_lock, flags);
 
+	if (ms->dma_started)
+		halt_dma(ms);
+
 	/* Reset the controller & dbdma channel */
 	out_le32(&md->control, (RUN|PAUSE|FLUSH|WAKE) << 16);	/* stop dma */
 	out_8(&mr->exception, 0xff);	/* clear all exception bits */