Hello,
On 4/29/2013 11:21 PM, Marc C wrote:
/**
- dma_contiguous_reserve() - reserve area(s) for contiguous memory handling
- @limit: End address of the reserved memory (optional, 0 for any).
@@ -149,6 +200,10 @@ void __init dma_contiguous_reserve(phys_addr_t limit)
pr_debug("%s(limit %08lx)\n", __func__, (unsigned long)limit);+#ifdef CONFIG_OF
of_scan_flat_dt(cma_fdt_scan, NULL);+#endif
What is your expectation with the contention between default region setup via the kernel config (CONFIG_CMA_SIZE_SEL_*) and via the DT? Could the call to 'of_scan_flat_dt()' be done before the setup of the kernel config early regions, followed by a return code check 'fail-over' scheme, or were you intending for the default region setups to be mutually-exclusive?
In the proposed patch the default/global cma region setup from device tree had higher priority than kernel command line parameter and .config saved values, but now as I think of this, it looks that it would make more sense to have the following priority for setting up the default cma region:
1. kernel cmd line - if not available, then use: 2. device tree - if not available, then use: 3. kernel compiled .config
I will update this in the next version of the CMA DT patches.
Best regards