It might be related to the local reset behavior of CorePac. Please take a look at the following thread, which seems to discuss the similar issue:
http://e2e.ti.com/support/dsp/c6000_multi-core_dsps/f/639/t/294703.aspx?pi267198=1
David mentioned some workaround to flush the cache in the example DSP local reset code provided by Advantech EVM driver.
"I found Advantech DSPC-8681E has DSP local reset functionality provided both in Windows driver and Linux driver. The download page is: DSPC8681 driver link. The latest version for windows is V0.7.8, the local reset has been implemented in do_dsp_local_reset() function in dsp_loader(Lightning_PCIE_0_7_8\Project\dsp_loader), the API call into Windows driver. The demo script is in Lightning_PCIE_0_7_8\examples\script\DSPC8681E\dspreset.bat.
For the DSP local rest implementation, you can refer to the Linux source code V0.7 in Lightning_PCIE_0_7\dsp_loader\app. It is still done by do_dsp_local_reset() in dsp_loader.c 7002.dsp_loader.c. Two major function calls are dio_put_dsp_in_reset() and dio_bring_dsp_out_reset(). In the second function, you can find this comments: “The following code is a work around to flush the cache, without this Any dirty cache lines in L1D may cause corruption of the downloaded image”. It wrote 8 idle instructions into MSMC, the DSP cores are bring out of reset once to execute this code. Then, the DSP cores are put into reset again, load your real application code and bring out of reset. So I think this is quite relavant to issue you are seeing."
You may also want to check if the PLL is set correctly as discussed in the later part of the thread above as well.
Hope it could help with your problem.