Skip to contentSkip to conversation

qla2xxx: improve error returns in qla2x00_process_vendor_specific

PATCH In review Stephen Rothwell 2 participants 2 messages Feb 18, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 qla2xxx: improve error returns in qla2x00_process_vendor_specific

These changes mean that the request sg will be unmapped if the response sg map fails, both the dma_allocs are freed and the dma_allocs only get freed if necessary. There are still two places that return from this routine without doing either the unmaps or the dma_frees.

This gets rid of these warnings:

drivers/scsi/qla2xxx/qla_attr.c: In function 'qla2x00_process_vendor_specific':
drivers/scsi/qla2xxx/qla_attr.c:2150: warning: 'req_data' may be used uninitialized in this function drivers/scsi/qla2xxx/qla_attr.c:2152: warning: 'req_data_len' may be used uninitialized in this function

---
 drivers/scsi/qla2xxx/qla_attr.c |   20 ++++++++++++--------
 1 files changed, 12 insertions(+), 8 deletions(-)

I have only build tested these changes as I do not have the hardware involved. Also, I am not sure if some of the infrastructure does the dma_free_coherent() for the rsp_data.

drivers/scsi/qla2xxx/qla_attr.c +12 -8
index 5a19aea..d8c5bf3 100644
@@ -2173,7 +2173,7 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);
         if (!elreq.rsp_sg_cnt) {
 		rval = -ENOMEM;
-                goto done;
+                goto done_unmap_sg_req;
 	}
 
 	if ((elreq.req_sg_cnt !=  bsg_job->request_payload.sg_cnt) ||
@@ -2224,7 +2224,7 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 				DEBUG2(qla_printk(KERN_INFO, ha, "scsi(%ld)"
 				"loopback option:0x%x not supported\n", vha->host_no, elreq.options));
 				rval = -EINVAL;
-				goto done_unmap_sg;
+				goto done_dma_free;
 			}
 
 			DEBUG2(qla_printk(KERN_INFO, ha,
@@ -2304,17 +2304,21 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 	}
 	bsg_job->job_done(bsg_job);
 
+done_dma_free:
+	dma_free_coherent(&ha->pdev->dev, rsp_data_len,
+		rsp_data, rsp_data_dma);
+	dma_free_coherent(&ha->pdev->dev, req_data_len,
+		req_data, req_data_dma);
+
 done_unmap_sg:
+	dma_unmap_sg(&ha->pdev->dev,
+	    bsg_job->reply_payload.sg_list,
+	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);
 
-	if(req_data)
-		dma_free_coherent(&ha->pdev->dev, req_data_len,
-			req_data, req_data_dma);
+done_unmap_sg_req:
 	dma_unmap_sg(&ha->pdev->dev,
 	    bsg_job->request_payload.sg_list,
 	    bsg_job->request_payload.sg_cnt, DMA_TO_DEVICE);
-	dma_unmap_sg(&ha->pdev->dev,
-	    bsg_job->reply_payload.sg_list,
-	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);
 
 done:
         return rval;
Signature · 6 lines
1.6.6.2

-- 
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#1 lore raw

Thanks for pointing this.

The patch does not apply cleanly with the latest BSG changes. I will make the necessary changes and re-send the patch.

-- Giri

Quoted 77 lines · Stephen Rothwell

These changes mean that the request sg will be unmapped if the response sg map fails, both the dma_allocs are freed and the dma_allocs only get freed if necessary. There are still two places that return from this routine without doing either the unmaps or the dma_frees.

This gets rid of these warnings:

drivers/scsi/qla2xxx/qla_attr.c: In function 'qla2x00_process_vendor_specific':
drivers/scsi/qla2xxx/qla_attr.c:2150: warning: 'req_data' may be used uninitialized in this function drivers/scsi/qla2xxx/qla_attr.c:2152: warning: 'req_data_len' may be used uninitialized in this function

---
drivers/scsi/qla2xxx/qla_attr.c | 20 ++++++++++++--------

1 files changed, 12 insertions(+), 8 deletions(-)

I have only build tested these changes as I do not have the hardware involved. Also, I am not sure if some of the infrastructure does the dma_free_coherent() for the rsp_data.

drivers/scsi/qla2xxx/qla_attr.c +8 -2
index 5a19aea..d8c5bf3 100644
@@ -2173,7 +2173,7 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);
        if (!elreq.rsp_sg_cnt) {
 		rval = -ENOMEM;
-                goto done;
+                goto done_unmap_sg_req;
 	}
 
 	if ((elreq.req_sg_cnt !=  bsg_job->request_payload.sg_cnt) ||
@@ -2224,7 +2224,7 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 				DEBUG2(qla_printk(KERN_INFO, ha, "scsi(%ld)"
 				"loopback option:0x%x not supported\n", vha->host_no, elreq.options));
 				rval = -EINVAL;
-				goto done_unmap_sg;
+				goto done_dma_free;
 			}
 
 			DEBUG2(qla_printk(KERN_INFO, ha,
@@ -2304,17 +2304,21 @@ qla2x00_process_vendor_specific(struct fc_bsg_job *bsg_job)
 	}
 	bsg_job->job_done(bsg_job);
 
+done_dma_free:
+	dma_free_coherent(&ha->pdev->dev, rsp_data_len,
+		rsp_data, rsp_data_dma);
+	dma_free_coherent(&ha->pdev->dev, req_data_len,
+		req_data, req_data_dma);
+

done_unmap_sg:

drivers/scsi/qla2xxx/qla_attr.c +4 -3
+	dma_unmap_sg(&ha->pdev->dev,
+	    bsg_job->reply_payload.sg_list,
+	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);
 
-	if(req_data)
-		dma_free_coherent(&ha->pdev->dev, req_data_len,
-			req_data, req_data_dma);
+done_unmap_sg_req:
	dma_unmap_sg(&ha->pdev->dev,
	    bsg_job->request_payload.sg_list,
	    bsg_job->request_payload.sg_cnt, DMA_TO_DEVICE);
drivers/scsi/qla2xxx/qla_attr.c +0 -3
-	dma_unmap_sg(&ha->pdev->dev,
-	    bsg_job->reply_payload.sg_list,
-	    bsg_job->reply_payload.sg_cnt, DMA_FROM_DEVICE);

done:

        return rval;
Signature · 6 lines
1.6.6.2

-- 
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/