Subject: [PATCH RESEND 1/2] scsi: vmw_pvscsi: Fix swiotlb operation

From: Thomas Hellstrom <[email protected]>

With swiotlb, the first byte of the sense buffer may in some cases be
uninitialized since we use DMA_FROM_DEVICE, and the device incorrectly
doesn't clear it. In those cases, clear it after DMA unmapping.

Cc: "James E.J. Bottomley" <[email protected]>
Cc: "Martin K. Petersen" <[email protected]>
Suggested-by: Vishal Bhakta <[email protected]>
Signed-off-by: Thomas Hellstrom <[email protected]>
Acked-by: Jim Gill <[email protected]>
---
drivers/scsi/vmw_pvscsi.c | 14 ++++++++++++++
1 file changed, 14 insertions(+)

diff --git a/drivers/scsi/vmw_pvscsi.c b/drivers/scsi/vmw_pvscsi.c
index 70008816c91f..8a09d184a320 100644
--- a/drivers/scsi/vmw_pvscsi.c
+++ b/drivers/scsi/vmw_pvscsi.c
@@ -402,6 +402,17 @@ static int pvscsi_map_buffers(struct pvscsi_adapter *adapter,
return 0;
}

+/*
+ * The device incorrectly doesn't clear the first byte of the sense
+ * buffer in some cases. We have to do it ourselves.
+ * Otherwise we run into trouble when SWIOTLB is forced.
+ */
+static void pvscsi_patch_sense(struct scsi_cmnd *cmd)
+{
+ if (cmd->sense_buffer)
+ cmd->sense_buffer[0] = 0;
+}
+
static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter,
struct pvscsi_ctx *ctx)
{
@@ -544,6 +555,8 @@ static void pvscsi_complete_request(struct pvscsi_adapter *adapter,
cmd = ctx->cmd;
abort_cmp = ctx->abort_cmp;
pvscsi_unmap_buffers(adapter, ctx);
+ if (sdstat != SAM_STAT_CHECK_CONDITION)
+ pvscsi_patch_sense(cmd);
pvscsi_release_context(adapter, ctx);
if (abort_cmp) {
/*
@@ -873,6 +886,7 @@ static void pvscsi_reset_all(struct pvscsi_adapter *adapter)
scmd_printk(KERN_ERR, cmd,
"Forced reset on cmd %p\n", cmd);
pvscsi_unmap_buffers(adapter, ctx);
+ pvscsi_patch_sense(cmd);
pvscsi_release_context(adapter, ctx);
cmd->result = (DID_RESET << 16);
cmd->scsi_done(cmd);
--
2.21.0


2019-12-20 03:45:19

by Martin K. Petersen

[permalink] [raw]
Subject: Re: [PATCH RESEND 1/2] scsi: vmw_pvscsi: Fix swiotlb operation


Thomas,

> With swiotlb, the first byte of the sense buffer may in some cases be
> uninitialized since we use DMA_FROM_DEVICE, and the device incorrectly
> doesn't clear it. In those cases, clear it after DMA unmapping.

Applied 1+2 to 5.6/scsi-queue, thanks!

--
Martin K. Petersen Oracle Linux Engineering

2019-12-20 08:57:23

by Thomas Hellstrom

[permalink] [raw]
Subject: Re: [PATCH RESEND 1/2] scsi: vmw_pvscsi: Fix swiotlb operation

On 12/20/19 4:44 AM, Martin K. Petersen wrote:
> Thomas,
>
>> With swiotlb, the first byte of the sense buffer may in some cases be
>> uninitialized since we use DMA_FROM_DEVICE, and the device incorrectly
>> doesn't clear it. In those cases, clear it after DMA unmapping.
> Applied 1+2 to 5.6/scsi-queue, thanks!
>
Thanks!

Thomas