static U32 PDrvCryptoUpdateHashWithDMA(PUBLIC_CRYPTO_HASH_CONTEXT * pHashCtx,
U8 * pData,
U32 dataLength,
U32 dmaUse)
{
/*
* Note: The DMA only sees physical addresses !
*/
U32 bufIn_phys = virt_to_phys(pData);
U32 res = PUBLIC_CRYPTO_OPERATION_SUCCESS;
U32 nHWAlgo = 0;
U32 vInitValue = 0;
U32 vDirectProcess = 0;
if (pHashCtx->vNbBytesProcessed == 0) {
vInitValue = 1;
}
nHWAlgo = DIGEST_CTRL_GET_ALGO(INREG32(&g_pSha1Md5Reg_t->CTRL));
dprintk(KERN_INFO "PDrvCryptoUpdateHASHWithDMA: Use=%u, Len=%u, In=0x%08x\n",
(unsigned int)dmaUse, (unsigned int)dataLength, (unsigned int)bufIn_phys);
if (dataLength == 0) {
/* No need of setting the dma and crypto-processor */
dprintk(KERN_INFO "PDrvCryptoUpdateHASHWithDMA: Nothing to process\n");
return PUBLIC_CRYPTO_OPERATION_SUCCESS;
}
if (pData == 0) {
dprintk(KERN_ERR "PDrvCryptoUpdateHASHWithDMA: bufIn_phys/bufOut_phys NULL\n");
return PUBLIC_CRYPTO_ERR_BAD_PARAMETERS;
}
/* Makes sure buffers are 4-bytes aligned */
if (!IS_4_BYTES_ALIGNED((int)bufIn_phys)) {
dprintk(KERN_ERR "PDrvCryptoUpdateHASHWithDMA: bufIn_phys/Out not 4 bytes aligned\n");
return PUBLIC_CRYPTO_ERR_ALIGNMENT;
}
if (pHashCtx->vChunckIndex == 0) {
vDirectProcess = 1;
}
if ((IS_4_BYTES_ALIGNED(HASH_CHUNK_BYTES_LENGTH-pHashCtx->vChunckIndex)) && (dataLength > HASH_CHUNK_BYTES_LENGTH-pHashCtx->vChunckIndex)) {
/* Fill the chunk */
U32 vLengthToComplete = HASH_CHUNK_BYTES_LENGTH - pHashCtx->vChunckIndex;
/* So we fill the chunk buffer with the new data to complete to HASH_CHUNK_BYTES_LENGTH */
memcpy(pHashCtx->pChunckBuffer + pHashCtx->vChunckIndex, pData, vLengthToComplete);
pHashCtx->vChunckIndex = HASH_CHUNK_BYTES_LENGTH;
pData += vLengthToComplete;
dataLength -= vLengthToComplete;
// Process the chunk first with no dma
if (HASH_CHUNK_BYTES_LENGTH == HASH_BLOCK_BYTES_LENGTH) {
res = static_Hash_HwPerform64bDigest((U32*)pHashCtx->pChunckBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1));
} else {
res = static_Hash_HwPerformDmaDigest((U32*)pHashCtx->pChunckBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1), dmaUse);
}
if (res != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
return res;
}
vInitValue = 0;
pHashCtx->vNbBytesProcessed+=HASH_CHUNK_BYTES_LENGTH;
pHashCtx->vChunckIndex = 0;
vDirectProcess = 1;
}
if ((vDirectProcess == 1) && (dataLength > HASH_BLOCK_BYTES_LENGTH)) {
U32 vDmaProcessSize = dataLength & 0xFFFFFFC0;
if (vDmaProcessSize == dataLength) {
vDmaProcessSize = vDmaProcessSize - HASH_BLOCK_BYTES_LENGTH; // We keep one block for the final
}
res = static_Hash_HwPerformDmaDigest((U32*)pData, pHashCtx->vNbBytesProcessed,
(vDmaProcessSize << 5) | (vInitValue << 3) | (nHWAlgo << 1), dmaUse);
if (res != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
return res;
}
vInitValue = 0;
pHashCtx->vNbBytesProcessed += vDmaProcessSize;
pData += vDmaProcessSize;
dataLength -= vDmaProcessSize;
}
/******************************** TO DO ******************************/
// Look if we can fill the chunk. If it is filled then process the dma
// on the HASH_CHUNK_BYTES_LENGTH length. And put the remaining in the
// chunk. Else put the remaining in the chunk and return the function.
/*********************************************************************/
/* Is there any data in the chunk? If yes is it possible to make a HASH_CHUNK_BYTES_LENGTH buffer
with the new data passed ? */
if ((pHashCtx->vChunckIndex != 0) && (pHashCtx->vChunckIndex + dataLength >= HASH_CHUNK_BYTES_LENGTH)) {
U32 vLengthToComplete = HASH_CHUNK_BYTES_LENGTH - pHashCtx->vChunckIndex;
/* So we fill the chunk buffer with the new data to complete to HASH_CHUNK_BYTES_LENGTH */
memcpy(pHashCtx->pChunckBuffer + pHashCtx->vChunckIndex, pData, vLengthToComplete);
if (pHashCtx->vChunckIndex + dataLength == HASH_CHUNK_BYTES_LENGTH) {
/* We'll keep some data for the final */
pHashCtx->vChunckIndex = HASH_CHUNK_BYTES_LENGTH;
return PUBLIC_CRYPTO_OPERATION_SUCCESS;
}
/* Then we send this buffer to the hash hardware */
if (HASH_CHUNK_BYTES_LENGTH == HASH_BLOCK_BYTES_LENGTH) {
res = static_Hash_HwPerform64bDigest((U32*)pHashCtx->pChunckBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1));
} else {
res = static_Hash_HwPerformDmaDigest((U32*)pHashCtx->pChunckBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1), dmaUse);
}
vInitValue = 0;
if (res != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
return res;
}
pHashCtx->vNbBytesProcessed += HASH_CHUNK_BYTES_LENGTH;
/* We have flushed the chunk so it is empty now */
pHashCtx->vChunckIndex = 0;
/* Then we have less data to proceed */
pData += vLengthToComplete;
dataLength -= vLengthToComplete;
}
/* (2) We process all the HASH_CHUNK_BYTES_LENGTH buffer that we can */
/* Dirty patch to confirm : keep the last full block of HASH_CHUNK_BYTES_LENGTH for the final.
We must keep some data for the final. To remove the patch replace > by >= */
while (dataLength > HASH_CHUNK_BYTES_LENGTH) {
U8 pTempAlignedBuffer[HASH_CHUNK_BYTES_LENGTH];
/*
* We process a HASH_CHUNK_BYTES_LENGTH buffer
*/
/* We copy the data to process to an aligned buffer !LINUX only ??! */
memcpy(pTempAlignedBuffer, pData, HASH_CHUNK_BYTES_LENGTH);
/* Then we send this buffer to the hash hardware */
if (HASH_CHUNK_BYTES_LENGTH == HASH_BLOCK_BYTES_LENGTH) {
res = static_Hash_HwPerform64bDigest((U32*)pTempAlignedBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1));
} else {
res = static_Hash_HwPerformDmaDigest((U32*)pTempAlignedBuffer, pHashCtx->vNbBytesProcessed,
(HASH_CHUNK_BYTES_LENGTH << 5) | (vInitValue << 3) | (nHWAlgo << 1), dmaUse);
}
vInitValue = 0;
if (res != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
return res;
}
pHashCtx->vNbBytesProcessed += HASH_CHUNK_BYTES_LENGTH;
/* Then we decrease the remaining data of HASH_CHUNK_BYTES_LENGTH */
pData += HASH_CHUNK_BYTES_LENGTH;
dataLength -= HASH_CHUNK_BYTES_LENGTH;
}
/* (3) We look if we have some data that could not be processed yet because it is
not large enough to fill a buffer of HASH_CHUNK_BYTES_LENGTH */
if (dataLength > 0) {
if (pHashCtx->vChunckIndex + dataLength > HASH_CHUNK_BYTES_LENGTH) {
/* Should never be in this case => Serious problem in the code !!! */
return PUBLIC_CRYPTO_ERR_BAD_PARAMETERS;
} else {
/* So we fill the chunk buffer with the new data */
memcpy(pHashCtx->pChunckBuffer + pHashCtx->vChunckIndex, pData, dataLength);
pHashCtx->vChunckIndex += dataLength;
}
}
dprintk(KERN_INFO "PDrvCryptoUpdateHASHWithDMA: Success\n");
return PUBLIC_CRYPTO_OPERATION_SUCCESS;
}
static U32 PDrvCryptoUpdateDESWithDMA(PUBLIC_CRYPTO_DES_DES3_CONTEXT * pDesCtx,
U8 * pSrc, U8 * pDest, U32 nbBlocks, U32 dmaUse)
{
/*
* Note: The DMA only sees physical addresses !
*/
U32 bufIn_phys = virt_to_phys(pSrc);
U32 bufOut_phys = virt_to_phys(pDest);
U32 dma_ch0 = OMAP3430_SMC_DMA_CH_0;
U32 dma_ch1 = OMAP3430_SMC_DMA_CH_1;
SCX_DMA_CHANNEL_PARAM ch0_parameters;
SCX_DMA_CHANNEL_PARAM ch1_parameters;
U32 nLength = nbBlocks * DES_BLOCK_SIZE;
U32 returnCode;
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Use=%u, Len=%u, In=0x%08x, Out=0x%08x\n",
(unsigned int)dmaUse, (unsigned int)nLength, (unsigned int)bufIn_phys, (unsigned int)bufOut_phys);
if (nLength == 0) {
/* No need of setting the dma and crypto-processor */
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Nothing to process\n");
return PUBLIC_CRYPTO_OPERATION_SUCCESS;
}
if ((bufIn_phys == 0) || (bufOut_phys == 0)) {
dprintk(KERN_ERR "PDrvCryptoUpdateDESWithDMA: bufIn_phys/bufOut_phys NULL\n");
return PUBLIC_CRYPTO_ERR_BAD_PARAMETERS;
}
/* Makes sure buffers are 4-bytes aligned */
if (!IS_4_BYTES_ALIGNED((int)bufIn_phys) || !IS_4_BYTES_ALIGNED((int)bufOut_phys)) {
dprintk(KERN_ERR "PDrvCryptoUpdateDESWithDMA: bufIn_phys/Out not 4 bytes aligned\n");
return PUBLIC_CRYPTO_ERR_ALIGNMENT;
}
/*
* Only one segment of the sg list to proceed --> no need of scatter gather algo
*/
/* Makes sure that if the dma channels that will need to be used are currently active,
one can reprogram it (them) */
scxPublicDMADisableChannel(dma_ch0);
scxPublicDMADisableChannel(dma_ch1);
if (dmaUse == PUBLIC_CRYPTO_DMA_USE_IRQ) {
/* Reset DMA int (DMA CTRL) - The DMA int (INT CTRL) is reset by the OS */
scxPublicDMADisableL3IRQ();
scxPublicDMAClearL3IRQ();
}
/* DMA used for Input and Output */
OUTREG32(&g_pDESReg_t->DES_MASK,
INREG32(&g_pDESReg_t->DES_MASK)
| DES_MASK_DMA_REQ_OUT_EN_BIT
| DES_MASK_DMA_REQ_IN_EN_BIT);
/* DMA1: Mem -> DES */
ch0_parameters.data_type = DMA_CSDP_Srce_Endian_little
| DMA_CSDP_Srce_Endian_Lock_off
| DMA_CSDP_Dest_Endian_little
| DMA_CSDP_Dest_Endian_Lock_off
| DMA_CSDP_Write_Mode_none_posted
| DMA_CSDP_Dest_Burst_off
| DMA_CSDP_Dest_packed_off
| DMA_CSDP_WR_Add_Trslt
| DMA_CSDP_Src_Burst_off
| DMA_CSDP_Src_packed_off
| DMA_CSDP_RD_Add_Trslt
| DMA_CSDP_Data_32b;
ch0_parameters.elem_count = DMA_CEN_Elts_per_Frame_DES;
ch0_parameters.frame_count = nbBlocks;
ch0_parameters.src_amode = 1; /* post increment */
ch0_parameters.src_start = bufIn_phys;
ch0_parameters.src_ei = DMA_CSEI_Default;
ch0_parameters.src_fi = DMA_CSFI_Default;
ch0_parameters.dst_amode = 0; /* const */
ch0_parameters.dst_start = DES1_REGS_HW_ADDR + 0x24;
ch0_parameters.dst_ei = DMA_CDEI_Default;
ch0_parameters.dst_fi = DMA_CDFI_Default; /* source frame index */
ch0_parameters.trigger = DMA_CCR_Mask_Channel(DMA_CCR_Channel_Mem2DES);
ch0_parameters.sync_mode = 0x2; /* FS =1, BS=0 => An entire frame is
transferred once a DMA request is made */
ch0_parameters.src_or_dst_synch = 0; /* Transfert is triggered by the Dest */
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: scxPublicDMASetParams(ch0)\n");
scxPublicDMASetParams(dma_ch0, &ch0_parameters);
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Start DMA channel %d\n", (unsigned int)dma_ch0);
scxPublicDMAStart(dma_ch0, OMAP_DMA_DROP_IRQ);
/* DMA2: DES -> Mem */
ch1_parameters.data_type = DMA_CSDP_Srce_Endian_little
| DMA_CSDP_Srce_Endian_Lock_off
| DMA_CSDP_Dest_Endian_little
| DMA_CSDP_Dest_Endian_Lock_off
| DMA_CSDP_Write_Mode_none_posted
| DMA_CSDP_Dest_Burst_off
| DMA_CSDP_Dest_packed_off
| DMA_CSDP_WR_Add_Trslt
| DMA_CSDP_Src_Burst_off
| DMA_CSDP_Src_packed_off
| DMA_CSDP_RD_Add_Trslt
| DMA_CSDP_Data_32b;
ch1_parameters.elem_count = DMA_CEN_Elts_per_Frame_DES;
ch1_parameters.frame_count = nbBlocks;
ch1_parameters.src_amode = 0; /* const */
ch1_parameters.src_start = DES1_REGS_HW_ADDR + 0x24;
ch1_parameters.src_ei = DMA_CSEI_Default;
ch1_parameters.src_fi = DMA_CSFI_Default;
ch1_parameters.dst_amode = 1; /* post increment */
ch1_parameters.dst_start = bufOut_phys;
ch1_parameters.dst_ei = DMA_CDEI_Default;
ch1_parameters.dst_fi = DMA_CDFI_Default; /* source frame index */
ch1_parameters.trigger = DMA_CCR_Mask_Channel(DMA_CCR_Channel_DES2Mem);
ch1_parameters.sync_mode = 0x2; /* FS =1, BS=0 => An entire frame is
transferred once a DMA request is made */
ch1_parameters.src_or_dst_synch = 1; /* Transfert is triggered by the Src */
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: scxPublicDMASetParams(ch1)\n");
scxPublicDMASetParams(dma_ch1, &ch1_parameters);
if (dmaUse == PUBLIC_CRYPTO_DMA_USE_IRQ) {
scxPublicDMAEnableL3IRQ();
}
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Start DMA channel %d\n", (unsigned int)dma_ch1);
scxPublicDMAStart(dma_ch1, OMAP_DMA_DROP_IRQ|OMAP_DMA_BLOCK_IRQ);
/*
* The input data may be in the cache only,
* and the DMA is only working with physical addresses.
* So flush the cache to have data coherency.
*/
v7_dma_flush_range((u32)pSrc, (u32)(pSrc + nLength));
/* Start operation */
OUTREG32(&g_pDESReg_t->DES_MASK,
INREG32(&g_pDESReg_t->DES_MASK)|DES_MASK_START_BIT);
if (dmaUse == PUBLIC_CRYPTO_DMA_USE_IRQ) {
/* Suspends the process until the DMA IRQ occurs */
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Waiting for IRQ\n");
returnCode = scxPublicDMAWait();
} else {
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Polling DMA\n");
returnCode = scxPublicDMAPoll(dma_ch1);
}
if (returnCode != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
dprintk(KERN_ERR "PDrvCryptoUpdateDESWithDMA: Timeout\n");
/* Do not exit function but clear properly the operation */
}
if (dmaUse == PUBLIC_CRYPTO_DMA_USE_IRQ) {
/* Acknoledge DMA interrupt */
scxPublicDMADisableL3IRQ();
}
scxPublicDMAClearChannel(dma_ch1);
/*
* The dma transfert is complete
*/
/* Stop clocks */
OUTREG32(&g_pDESReg_t->DES_MASK,
INREG32(&g_pDESReg_t->DES_MASK)
&(~DES_MASK_START_BIT) );
/* Unset DMA synchronisation requests */
OUTREG32(&g_pDESReg_t->DES_MASK,
INREG32(&g_pDESReg_t->DES_MASK)
& (~DES_MASK_DMA_REQ_OUT_EN_BIT)
& (~DES_MASK_DMA_REQ_IN_EN_BIT));
if (returnCode != PUBLIC_CRYPTO_OPERATION_SUCCESS) {
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Error [0x%08x]\n", (unsigned int)returnCode);
return returnCode;
}
/*
* The output data are in the physical memory.
* So invalidate the cache to have data coherency.
*/
v7_dma_inv_range((u32)pDest, (u32)(pDest + nLength));
dprintk(KERN_INFO "PDrvCryptoUpdateDESWithDMA: Success\n");
return PUBLIC_CRYPTO_OPERATION_SUCCESS;
}