[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170423084133.GE30869@infradead.org>
Date: Sun, 23 Apr 2017 01:41:33 -0700
From: Christoph Hellwig <hch@...radead.org>
To: Alexey Khoroshilov <khoroshilov@...ras.ru>
Cc: "James E.J. Bottomley" <jejb@...ux.vnet.ibm.com>,
"Martin K. Petersen" <martin.petersen@...cle.com>,
linux-scsi@...r.kernel.org, linux-kernel@...r.kernel.org,
ldv-project@...uxtesting.org
Subject: Re: [PATCH] scsi: mvumi: add check for dma mapping errors
On Sat, Apr 22, 2017 at 02:17:50AM +0300, Alexey Khoroshilov wrote:
> } else {
> - scmd->SCp.dma_handle = scsi_bufflen(scmd) ?
> - pci_map_single(mhba->pdev, scsi_sglist(scmd),
> - scsi_bufflen(scmd),
> - (int) scmd->sc_data_direction)
> - : 0;
> + if (!scsi_bufflen(scmd))
> + return -1;
> + scmd->SCp.dma_handle = pci_map_single(mhba->pdev,
> + scsi_sglist(scmd),
> + scsi_bufflen(scmd),
> + (int) scmd->sc_data_direction);
> + if (pci_dma_mapping_error(mhba->pdev, scmd->SCp.dma_handle))
> + return -1;
This looks completely broken. Why would you DMA map the in-memory
struct scatterlist? It has no meaning for the hardware.
In fact this whole branch (and the equivalent in the unmap path)
are dead - SCSI commands that transfer data always have a SG entry.
So the right fix is to remove the !scsi_sg_count(scmd) map/unmap
path.
Powered by blists - more mailing lists