@@ -88,6 +88,10 @@ static void isp_save_ctx(struct isp_device *isp);
static void isp_restore_ctx(struct isp_device *isp);
+static int isp_attach_iommu(struct isp_device *isp);
+
+static void isp_detach_iommu(struct isp_device *isp);
+
static const struct isp_res_mapping isp_res_maps[] = {
{
.isp_rev = ISP_REVISION_2_0,
@@ -1407,6 +1411,14 @@ static struct isp_device *__omap3isp_get(struct isp_device *isp, bool irq)
__isp = NULL;
goto out;
}
+ /* IOMMU */
+ if (isp_attach_iommu(isp) < 0) {
+ dev_err(isp->dev, "unable to attach to IOMMU\n");
+ isp_disable_clocks(isp);
+ __isp = NULL;
+ goto out;
+ }
+
/* We don't want to restore context before saving it! */
if (isp->has_context)
@@ -1453,6 +1465,7 @@ static void __omap3isp_put(struct isp_device *isp, bool save_ctx)
if (!media_entity_enum_empty(&isp->crashed) ||
isp->stop_failure)
isp_reset(isp);
+ isp_detach_iommu(isp);
isp_disable_clocks(isp);
}
mutex_unlock(&isp->isp_mutex);
@@ -1999,10 +2012,6 @@ static int isp_remove(struct platform_device *pdev)
isp_cleanup_modules(isp);
isp_xclk_cleanup(isp);
- __omap3isp_get(isp, false);
- isp_detach_iommu(isp);
- __omap3isp_put(isp, false);
-
media_entity_enum_cleanup(&isp->crashed);
v4l2_async_notifier_cleanup(&isp->notifier);
@@ -2313,19 +2322,12 @@ static int isp_probe(struct platform_device *pdev)
isp->mmio_hist_base_phys =
mem->start + isp_res_maps[m].offset[OMAP3_ISP_IOMEM_HIST];
- /* IOMMU */
- ret = isp_attach_iommu(isp);
- if (ret < 0) {
- dev_err(&pdev->dev, "unable to attach to IOMMU\n");
- goto error_isp;
- }
-
/* Interrupt */
ret = platform_get_irq(pdev, 0);
if (ret <= 0) {
dev_err(isp->dev, "No IRQ resource\n");
ret = -ENODEV;
- goto error_iommu;
+ goto error_isp;
}
isp->irq_num = ret;
@@ -2339,7 +2341,7 @@ static int isp_probe(struct platform_device *pdev)
/* Entities */
ret = isp_initialize_modules(isp);
if (ret < 0)
- goto error_iommu;
+ goto error_isp;
ret = isp_register_entities(isp);
if (ret < 0)
On the GTA04, current consumption rose by about 30mA when the omap3_isp module was loaded and the v4l device was not accessed and even no camera attached. Module removal fixed it again. Slowing down the removal process reveals that calling isp_detach_iommu() is required to have low current. So isp_attach/detach_iommu() to moved to the get()/put() functions. This all has strange side effects. The hwmod seems to be accessible using /dev/mem if the iommu calls in their original place. With the modified placement it is not. In a very old setup with a 3.7 kernel which has the iommu calls at the same place as our current kernel, the memory is not accessible. Note: isp_get()/put() calls seem to be balanced. But at the current wonky gta04 setup (with a not upstreamed image sensor driver)which also has other problems, CAM reports address holes. So I have no clear idea whether this patch is right or not. Signed-off-by: Andreas Kemnade <andreas@kemnade.info> --- drivers/media/platform/omap3isp/isp.c | 28 +++++++++++++++------------- 1 file changed, 15 insertions(+), 13 deletions(-)