[U-Boot] [PATCH 3/3] sf: ti_qspi: Enable EDMA for reads in SPL

[Jagan Teki]

On 15 July 2014 at 02:07, Tom Rini <trini at ti.com> wrote:
> On Sat, Jul 12, 2014 at 06:42:31PM +0530, Jagan Teki wrote:
>> On Sat, Jul 12, 2014 at 2:23 AM, Tom Rini <trini at ti.com> wrote:
>> > From: Vinothkumar Rajendran <vinothr at ti.com>
>> >
>> > By default QSPI data through-put in memory mapped mode is ~2.4MB/sec @
>> > 48MHz. Added edma memory copy functionality in spi flash driver to
>> > improve the data through put to 5.1MB/Sec.
> [snip]
>> > +#include <asm/arch/edma.h>
>> > +#include "../dma/ti_edma.h"
>>
>> This looks odd to me - header inclusion, as .h in drivers even.
>
> I could shove this under arch/arm/include/asm/ti-common I suppose.

Any changes in code logic or updates?

>
>> >
>> >  /* ti qpsi register bit masks */
>> >  #define QSPI_TIMEOUT                    2000000
>> > @@ -340,3 +342,72 @@ int spi_xfer(struct spi_slave *slave, unsigned int bitlen, const void *dout,
>> >
>> >         return 0;
>> >  }
>> > +
>> > +#if defined(CONFIG_SPL_DMA_SUPPORT) && defined(CONFIG_TI_EDMA)
>> > +void spi_flash_copy_mmap(void *data, void *offset, size_t len)
>> > +{
>> > +       struct edma_param_entry edma_param;
>> > +       int           b_cnt_value = 1;
>> > +       int           rem_bytes  = 0;
>> > +       int           a_cnt_value = len;
>> > +       unsigned int          addr      = (unsigned int) (data);
>> > +       unsigned int          max_acnt  = 0x7FFFU;
>> > +       unsigned int edma_ch_num = 1;
>> > +
>> > +       if (len > max_acnt)
>> > +       {
>> > +               b_cnt_value = (len / max_acnt);
>> > +               rem_bytes  = (len % max_acnt);
>> > +               a_cnt_value = max_acnt;
>> > +       }
>> > +
>> > +       /* Compute QSPI address and size */
>> > +       edma_param.opt      = 0;
>> > +       edma_param.src_addr  = ((unsigned int) offset);
>> > +       edma_param.dest_addr = addr;
>> > +       edma_param.a_cnt     = a_cnt_value;
>> > +       edma_param.b_cnt     = b_cnt_value;
>> > +       edma_param.c_cnt     = 1;
>> > +       edma_param.src_bidx  = a_cnt_value;
>> > +       edma_param.dest_bidx = a_cnt_value;
>> > +       edma_param.src_cidx  = 0;
>> > +       edma_param.dest_cidx = 0;
>> > +       edma_param.link_addr = 0xFFFF;
>> > +       edma_param.opt     |=
>> > +               (EDMA_TPCC_OPT_TCINTEN_MASK |
>> > +                ((edma_ch_num <<
>> > +                  EDMA_TPCC_OPT_TCC_SHIFT) &
>> > +                 EDMA_TPCC_OPT_TCC_MASK) | EDMA_TPCC_OPT_SYNCDIM_MASK);
>> > +
>> > +       edma_set_param(edma_ch_num, &edma_param);
>> > +       edma_enable_transfer(edma_ch_num);
>> > +
>> > +       while (!(edma_get_intr_status() & (1 << edma_ch_num))) ;
>> > +       edma_clr_intr(edma_ch_num);
>> > +       if (rem_bytes != 0)
>> > +       {
>> > +               /* Compute QSPI address and size */
>> > +               edma_param.opt     = 0;
>> > +               edma_param.src_addr =
>> > +                       (b_cnt_value * max_acnt) + ((unsigned int) offset);
>> > +               edma_param.dest_addr = (addr + (max_acnt * b_cnt_value));
>> > +               edma_param.a_cnt     = rem_bytes;
>> > +               edma_param.b_cnt     = 1;
>> > +               edma_param.c_cnt     = 1;
>> > +               edma_param.src_bidx  = rem_bytes;
>> > +               edma_param.dest_bidx = rem_bytes;
>> > +               edma_param.src_cidx  = 0;
>> > +               edma_param.dest_cidx = 0;
>> > +               edma_param.link_addr = 0xFFFF;
>> > +               edma_param.opt     |=
>> > +                       (EDMA_TPCC_OPT_TCINTEN_MASK |
>> > +                        ((edma_ch_num << EDMA_TPCC_OPT_TCC_SHIFT) & EDMA_TPCC_OPT_TCC_MASK));
>> > +               edma_set_param(edma_ch_num, &edma_param);
>> > +               edma_enable_transfer(edma_ch_num);
>> > +
>> > +               while (!(edma_get_intr_status() & (1 << edma_ch_num))) ;
>> > +               edma_clr_intr(edma_ch_num);
>> > +       }
>> > +       *((unsigned int *) offset) += len;
>> > +}
>> > +#endif
>>
>> I'm some how !OK with this memory or flash change in spi driver.
>> Any better approach to move this - may be in DMA driver itself and
>> picking up the
>> memory attributes from sf layer.
>>
>> I'm not much clear about this now, but will come back again.
>
> Well, are you happy with how drivers/spi/mxs_spi.c works (search around
> on mxs_dma)?  I can re-jigger things along those lines I suppose.

thanks!
-- 
Jagan.