Re: am335x: GPMC: reading speed with prefetch mode

Re: am335x: GPMC: reading speed with prefetch mode

[Roger Quadros]

+Tony and l-o

On 20/03/15 15:37, Yegor Yefremov wrote:
> On Fri, Mar 20, 2015 at 1:37 PM, Roger Quadros <rogerq@ti.com> wrote:
>> On 20/03/15 13:24, Yegor Yefremov wrote:
>>> On Thu, Mar 19, 2015 at 4:56 PM, Daniel Mack <zonque@gmail.com> wrote:
>>>> On 03/19/2015 04:13 PM, Yegor Yefremov wrote:
>>>>> Strange. Have tried with "nand read" command, but still the same
>>>>> result with and without CONFIG_NAND_OMAP_GPMC_PREFETCH :
>>>>>
>>>>> [2.150655 0.001006] NAND read: device 0 offset 0x260000, size 0x1200000
>>>>> [15.978943 13.828288]  18874368 bytes read: OK
>>>>
>>>> What about adding some debug prints to the prefetch setup function and
>>>> see if it is executed at all?
>>>
>>> I2C:   ready
>>> DRAM:  256 MiB
>>> NAND: prefetch enabled
>>> NAND: 256 MiB
>>> MMC:   OMAP SD/MMC: 0, OMAP SD/MMC: 1
>>> Using default environment
>>>
>>> I've added "NAND: prefetch enabled" output for:
>>>
>>> #ifdef CONFIG_NAND_OMAP_GPMC_PREFETCH
>>>         else {
>>>                 printf("NAND: prefetch enabled\n");
>>>                 nand->read_buf = omap_nand_read_prefetch8;
>>>         }
>>> #else
>>>
>>> I've also put printf() into omap_nand_read_prefetch8() just to make
>>> sure it is called - it was called.
>>>
>>> Further ideas?
>>>
>>>>> Daniel, do you have the numbers? Images size and load time? What can I expect?
>>>>
>>>> I don't currently have the setup at hand, sorry. But the number I recall
>>>> from an email conversation back then is: The time from power-on, loading
>>>> SPL, loading U-Boot, leeching a 6MB uImage, jumping into it waiting for
>>>> the console to start dumping the kernel boot messages was less than 5
>>>> seconds in total.
>>>
>>> I Linux I had ti,nand-xfer-type = "polled";. After replacing it with
>>> ti,nand-xfer-type = "prefetch-polled"; I now get
>>>
>>> # dd if=/dev/mtdblock5 of=/dev/null bs=2M count=8
>>> 8+0 records in
>>> 8+0 records out
>>> 16777216 bytes (17 MB) copied, 2.58744 s, 6.5 MB/s
>>>
>>> instead of:
>>>
>>> # dd if=/dev/mtdblock5 of=/dev/null bs=2M count=8
>>> 8+0 records in
>>> 8+0 records out
>>> 16777216 bytes (17 MB) copied, 6.05157 s, 2.8 MB/s
>>>
>>> Do I see it right, that DMA support is not implemented in am33xx.dtsi?
>>
>> DMA support will have to be enabled in the board dts. e.g. am335x-bone.dts.
>>
>> if ti,nand-xfer-type is not specified (like in mainline kernel) then it
>> defaults to prefetch-polled
> 
> I get following error:
> 
> omap-gpmc 50000000.gpmc: GPMC revision 6.0
> nand: device found, Manufacturer ID: 0x2c, Chip ID: 0xda
> nand: Micron MT29F2G08ABAEAWP
> nand: 256MiB, SLC, page size: 2048, OOB size: 64
> omap2-nand omap2-nand.0: DMA engine request failed
> 
> Kernel: 3.18.1
> 
> GPMC config:
> 
> &gpmc {
>         pinctrl-names = "default";
>         pinctrl-0 = <&nandflash_pins_s0>;
>         ranges = <0 0 0x08000000 0x10000000>;   /* CS0: NAND */
>         status = "okay";
> 
>         nand@0,0 {
>                 reg = <0 0 4>; /* CS0, offset 0 */
>                 nand-bus-width = <8>;
>                 ti,nand-ecc-opt = "bch8";
>                 ti,nand-xfer-type = "prefetch-dma";
> 
>                 gpmc,device-nand = "true";
>                 gpmc,device-width = <1>;
>                 gpmc,sync-clk-ps = <0>;
>                 gpmc,cs-on-ns = <0>;
>                 gpmc,cs-rd-off-ns = <44>;
>                 gpmc,cs-wr-off-ns = <44>;
>                 gpmc,adv-on-ns = <6>;
>                 gpmc,adv-rd-off-ns = <34>;
>                 gpmc,adv-wr-off-ns = <44>;
>                 gpmc,we-on-ns = <0>;
>                 gpmc,we-off-ns = <40>;
>                 gpmc,oe-on-ns = <0>;
>                 gpmc,oe-off-ns = <54>;
>                 gpmc,access-ns = <64>;
>                 gpmc,rd-cycle-ns = <82>;
>                 gpmc,wr-cycle-ns = <82>;
>                 gpmc,wait-on-read = "true";
>                 gpmc,wait-on-write = "true";
>                 gpmc,bus-turnaround-ns = <0>;
>                 gpmc,cycle2cycle-delay-ns = <0>;
>                 gpmc,clk-activation-ns = <0>;
>                 gpmc,wait-monitoring-ns = <0>;
>                 gpmc,wr-access-ns = <40>;
>                 gpmc,wr-data-mux-bus-ns = <0>;
> 
>                 #address-cells = <1>;
>                 #size-cells = <1>;
>                 elm_id = <&elm>;
>         };
> };
> 
> Any idea?
> 
> Btw. where DMA support should be configured? GPMC config has no "dmas"
> like other components like mmc, uart, USB etc.

I've never tested dma on omap-nand in mainline. We've been solely relying on
prefetch-polled mode there.

someone needs to work on it and test dma support.

looks like the driver is hardcoding the DMA channel
                sig = OMAP24XX_DMA_GPMC;
                info->dma = dma_request_channel(mask, omap_dma_filter_fn, &sig);

this will have to be fixed.
I'm not sure what else is missing on DMA side. But you could try to use a
free channel and see if it works?

Tony any insights?

cheers,
-roger

> 
>                 gpmc: gpmc@50000000 {
>                         compatible = "ti,am3352-gpmc";
>                         ti,hwmods = "gpmc";
>                         ti,no-idle-on-init;
>                         reg = <0x50000000 0x2000>;
>                         interrupts = <100>;
>                         gpmc,num-cs = <7>;
>                         gpmc,num-waitpins = <2>;
>                         #address-cells = <2>;
>                         #size-cells = <1>;
>                         status = "disabled";
>                 };
> 
> 
> Yegor
> 

Re: am335x: GPMC: reading speed with prefetch mode

[Tony Lindgren]

* Roger Quadros <rogerq@ti.com> [150320 07:59]:
> On 20/03/15 15:37, Yegor Yefremov wrote:
> > 
> > Btw. where DMA support should be configured? GPMC config has no "dmas"
> > like other components like mmc, uart, USB etc.
> 
> I've never tested dma on omap-nand in mainline. We've been solely relying on
> prefetch-polled mode there.
> 
> someone needs to work on it and test dma support.
> 
> looks like the driver is hardcoding the DMA channel
>                 sig = OMAP24XX_DMA_GPMC;
>                 info->dma = dma_request_channel(mask, omap_dma_filter_fn, &sig);
> 
> this will have to be fixed.
> I'm not sure what else is missing on DMA side. But you could try to use a
> free channel and see if it works?
> 
> Tony any insights?

There are six external DMA request pins on omaps that can be used for
devices on GPMC for DMA. These are the sys_ndmareq0 to 5. So far the only
the old tusb6010 driver is the only one using that, and it's still not
using the DMA engine API. I've been meaning to update it, but have not
had a chance yet.. Anyways, all the new DMA code should use the dmaengine
API instead of the omap legacy DMA calls.

Note that the onenand driver is wrongly using the external gpio request
pin as a GPIO. That's not what the pin is supposed to do, it needs to be
muxed to the external gpio request mode. So I suggest we either rip out
the DMA code from onenand, or update it to use the sys_ndmareq and DMA
engine API.

Regards,

Tony