CAAM operation fails with DMA Error when using global buffers

[dhanushkadangam]

I'm using CAAM through Linux Crypto API as shown in kernel source tree crypto/tcrypt.c in a kernel module.

See the code segment below. I've changed the test_acipher_speed taken from tcrypt.c in my kernel module implementation to use BUFFSIZE instead of PAGE_SIZE.

When I build this to use memory allocated by kmalloc, the encryption succeeds, but when I build it to use the global array, buf, the encryption fails with the following error:

caam_jr 2142000.jr1: 40001016: DECO: desc idx 16: DMA Error

I'm running this on an iMX6UL with a 3.14.38 kernel. Why does global array test fail? Thank you.

//#define USE_KMALLOC

#define TVMEMSIZE 4

#define BUFFSIZE  128

static char *tvmem[TVMEMSIZE];

static u32 block_sizes[] = { 16, 0 };

static u8 speed_template_16[] = {16};

static char buf[BUFFSIZE*TVMEMSIZE];

static int __init testmodule_init(void)

{

   int i;

#ifndef  USE_KMALLOC

   char* buffer = buf;

#else // USE_KMALLOC

   char* buffer = (char*)kmalloc(BUFFSIZE*TVMEMSIZE, GFP_KERNEL);

#endif   // USE_KMALLOC

   for (i = 0; i < TVMEMSIZE; i++)

   {

      tvmem[i] = buffer + i*BUFFSIZE;

   }

   test_acipher_speed("ecb(aes)", ENCRYPT, 0, NULL, 0, speed_template_16);

#ifdef  USE_KMALLOC

   if (buffer)

   {

      kfree(buffer);

   }

#endif   // USE_KMALLOC

   return 0;

}

[dhanushkadangam]

I think I figured out the reason for DMA Error.

In linux kernel documentation https://www.kernel.org/doc/Documentation/DMA-API-HOWTO.txt, What memory is DMA'able? section it says

"If you acquired your memory via the page allocator (i.e. __get_free_page*()) or the generic memory allocators (i.e. kmalloc() or kmem_cache_alloc()) then you may DMA to/from that memory using the addresses returned from those routines.

:

This rule also means that you may use neither kernel image addresses (items in data/text/bss segments), nor module image addresses, nor stack addresses for DMA"

[Yuri · Accepted solution]

Hello,

  Yes, it is needed to use the GFP_DMA flag of the kmalloc, which is used to specify that the allocator
must satisfy the request from ZONE_DMA. This flag is used by device drivers, which need DMA-able
memory for their devices.

Have a great day,
Yuri

-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

[dhanushkadangam]

Hi Yuri,

Now I'm trying to use the secure-memory as the source and destination with CAAM async-block ciphers through Linux crypto-API. I obtain the source and destination slots from secure-memory using sm_keystore_slot_alloc as shown below in the code segment. Error checking has been omitted for brevity here, but I do check for errors in the real code.

The crypto operation completes successfully, but when I check the dst location, it's contents remain the same, as if no crypto operation occurred. Why is this happening?

Thank you!

   :

   u8 __iomem *src, *dst;

   struct caam_drv_private_sm *kspriv = dev_get_drvdata(ksdev);

   sm_establish_keystore(ksdev, 1);

   sm_keystore_slot_alloc(ksdev, 1, slotsize, &keyslot1);

   src = kspriv->slot_get_address(ksdev, 1, keyslot1);

   sm_keystore_slot_alloc(ksdev, 1, slotsize, &keyslot2);

   dst = kspriv->slot_get_address(ksdev, 1, keyslot2);

   // use src and dst with CAAM

   :

[Yuri]

Hello,

perhaps we have the same cache coherence issue ?

Are there CAAM errors ?

Regards,

Yuri.

[dhanushkadangam]

Hi Yuri,

CAAM doesn't report any error. It says the operation is successful.

Can you please explain how cache coherency could come in here?

Thanks

[Yuri]

Hello,

  As for coherency : all buffers should not be cached.

Regards,

Yuri.