On 08/21/2012 12:49 PM, Guenter Roeck wrote:

On Fri, May 11, 2012 at 08:34:46PM -0000, David Daney wrote:

quoted

From: David Daney <redacted>

Add the driver, link it into the kbuild system and provide device tree
binding documentation.

Signed-off-by: David Daney <redacted>
Acked-by: Grant Likely <redacted>

[ ... ]

quoted

+
+static int __devexit octeon_spi_remove(struct platform_device *pdev)
+{
+	struct octeon_spi *p = platform_get_drvdata(pdev);
+	struct spi_master *master = p->my_master;
+
+	spi_unregister_master(master);
+

I know it is a bit late, but ...

In this case, just in time.  I am now finally getting back to fixing the 
issues with this driver, and looking to merging it in the near future.

David Daney

The call to spi_unregister_master() frees the memory associated with master,
ie 'p', and the spi_master_put() below without matching spi_master_get() is
unnecessary/wrong. One possible fix would be to use

	struct spi_master *master = spi_master_get(p->my_master);

above. That protects master and p while it is still being used, and makes use
of the call to spi_master_put() below. Another option might be to move
cvmx_write_csr() ahead of the call to spi_unregister_master() and drop the
call to spi_master_put().

Guenter

quoted

+	/* Clear the CSENA* and put everything in a known state. */
+	cvmx_write_csr(p->register_base + OCTEON_SPI_CFG, 0);
+	spi_master_put(master);
+	return 0;
+}
+