Re: [PATCH] of: reserved_mem: Restructure call site for dma_contiguous_early_fixup()

[Oreoluwa Babatunde <oreoluwa.babatunde@oss.qualcomm.com>]

On 7/8/2025 10:17 AM, Robin Murphy wrote:
> On 08/07/2025 5:56 pm, Oreoluwa Babatunde wrote:
>> Restructure the call site for dma_contiguous_early_fixup() to
>> where the reserved_mem nodes are being parsed from the DT so that
>> dma_mmu_remap[] is populated before dma_contiguous_remap() is called.
>>
>> Fixes: 8a6e02d0c00e ("of: reserved_mem: Restructure how the reserved memory regions are processed")
>> Signed-off-by: Oreoluwa Babatunde <oreoluwa.babatunde@oss.qualcomm.com>
>> ---
>>   drivers/of/of_reserved_mem.c | 14 +++++++++++---
>>   kernel/dma/contiguous.c      |  2 --
>>   2 files changed, 11 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/of/of_reserved_mem.c b/drivers/of/of_reserved_mem.c
>> index 77016c0cc296..132d2c66cafc 100644
>> --- a/drivers/of/of_reserved_mem.c
>> +++ b/drivers/of/of_reserved_mem.c
>> @@ -25,6 +25,7 @@
>>   #include <linux/memblock.h>
>>   #include <linux/kmemleak.h>
>>   #include <linux/cma.h>
>> +#include <linux/dma-map-ops.h>
>>     #include "of_private.h"
>>   @@ -175,13 +176,17 @@ static int __init __reserved_mem_reserve_reg(unsigned long node,
>>           base = dt_mem_next_cell(dt_root_addr_cells, &prop);
>>           size = dt_mem_next_cell(dt_root_size_cells, &prop);
>>   -        if (size &&
>> -            early_init_dt_reserve_memory(base, size, nomap) == 0)
>> +        if (size && early_init_dt_reserve_memory(base, size, nomap) == 0) {
>> +            /* Architecture specific contiguous memory fixup. */
>> +            if (of_flat_dt_is_compatible(node, "shared-dma-pool"))
> 
> Surely this is now going to call the fixup for *all* "shared-dma-pool" reservations, rather than just the ones which are specifically CMA regions? That seems potentially bad in a number of ways.
> 
> Thanks,
> Robin.

Hi Robin,

Thanks for pointing that out. I believe if we add a check for
of_get_flat_dt_prop(node, "reusable", NULL) that should help
limit this to only the regions that are meant to be cma.

Please let me know if there's something else that I'm missing.

Thanks,
Oreoluwa

> 
>> +                dma_contiguous_early_fixup(base, size);
>> +
>>               pr_debug("Reserved memory: reserved region for node '%s': base %pa, size %lu MiB\n",
>>                   uname, &base, (unsigned long)(size / SZ_1M));
>> -        else
>> +        } else {
>>               pr_err("Reserved memory: failed to reserve memory for node '%s': base %pa, size %lu MiB\n",
>>                      uname, &base, (unsigned long)(size / SZ_1M));
>> +        }
>>             len -= t_len;
>>       }
>> @@ -472,6 +477,9 @@ static int __init __reserved_mem_alloc_size(unsigned long node, const char *unam
>>                  uname, (unsigned long)(size / SZ_1M));
>>           return -ENOMEM;
>>       }
>> +    /* Architecture specific contiguous memory fixup. */
>> +    if (of_flat_dt_is_compatible(node, "shared-dma-pool"))
>> +        dma_contiguous_early_fixup(base, size);
>>         /* Save region in the reserved_mem array */
>>       fdt_reserved_mem_save_node(node, uname, base, size);
>> diff --git a/kernel/dma/contiguous.c b/kernel/dma/contiguous.c
>> index 8df0dfaaca18..9e5d63efe7c5 100644
>> --- a/kernel/dma/contiguous.c
>> +++ b/kernel/dma/contiguous.c
>> @@ -480,8 +480,6 @@ static int __init rmem_cma_setup(struct reserved_mem *rmem)
>>           pr_err("Reserved memory: unable to setup CMA region\n");
>>           return err;
>>       }
>> -    /* Architecture specific contiguous memory fixup. */
>> -    dma_contiguous_early_fixup(rmem->base, rmem->size);
>>         if (default_cma)
>>           dma_contiguous_default_area = cma;