Re: [PATCH] [MTD] [NAND] Add prefetch and dma support for omap2/3NAND driver

[Tony Lindgren]

* Singh, Vimal <IMCEAEX-_O=TI_OU=BD_CN=RECIPIENTS_CN=X0094262@xxxxxxxxxxxxxxxxx> [090609 08:24]:
> 
> 
> On Tue, Jun 9, 2009 at 3:49 PM, Tony Lindgren<tony@xxxxxxxxxxx> wrote:
> > * Singh, Vimal <IMCEAEX-_O=TI_OU=BD_CN=RECIPIENTS_CN=X0094262@xxxxxxxxxxxxxxxxx> [090608 09:01]:
> >>
> >> On Mon, Jun 8, 2009 at 5:08 PM, Tony Lindgren<tony@xxxxxxxxxxx> wrote:
> >> > * Singh, Vimal <IMCEAEX-_O=TI_OU=BD_CN=RECIPIENTS_CN=X0094262@xxxxxxxxxxxxxxxxx> [090605 05:00]:
> >> >>
> >> >>
> >> >> On Thu, Jun 4, 2009 at 8:58 PM, Tony Lindgren<tony@xxxxxxxxxxx> wrote:
> >> >> > * vimal singh <vimalsingh@xxxxxx> [090604 02:34]:
> >> >> >> On Wed, Jun 3, 2009 at 10:03 PM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
> >> >> >> > * Singh, Vimal
> >> >> >> <IMCEAEX-_O=TI_OU=BD_CN=RECIPIENTS_CN=X0094262@xxxxxxxxxxxxxxxxx> [090602
> >> >> >> 23:46]:
> >> >> >> >>
> >> >> >> >> On Wed, Jun 3, 2009 at 2:06 AM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
> >> >> >> >> > * vimal singh <vimalsingh@xxxxxx> [090602 05:40]:
> >> >> >> >> >> This patch adds prefetch support to access nand flash in both mpu and dma
> >> >> >> mode.
> >> >> >> >> >> This patch also adds 8-bit nand support (omap_read/write_buf8).
> >> >> >> >> >> Prefetch can be used for both 8- and 16-bit devices.
> >> >> >> >> >
> >> >> >> >> > This should be reviewed on the linux-omap@xxxxxxxxxxxxxxx list for sure.
> >> >> >> >> > One other comment below.
> >> >> >> >> >
> >> >> >> >> >> Signed-off-by: Vimal Singh <vimalsingh@xxxxxx>
> >> >> >> >> >> ---
> >> >> >> >> >> I prepared this patch on top of "OMAP2 / OMAP3 NAND driver" patch:
> >> >> >> >> >> http://lists.infradead.org/pipermail/linux-mtd/2009-May/025562.html
> >> >> >> >> >>
> >> >> >> >> >> ---
> >> >> >> >> >>  arch/arm/mach-omap2/gpmc.c             |  102 ++++++++++
> >> >> >> >> >>  arch/arm/plat-omap/include/mach/gpmc.h |    4
> >> >> >> >> >>  drivers/mtd/nand/Kconfig               |   17 +
> >> >> >> >> >>  drivers/mtd/nand/omap2.c               |  308
> >> >> >> ++++++++++++++++++++++++++++++++-
> >> >> >> >> >>  4 files changed, 422 insertions(+), 9 deletions(-)
> >> >> >> >> >>
> >> >> >> >> >> Index: mtd-2.6/arch/arm/mach-omap2/gpmc.c
> >> >> >> >> >> ===================================================================
> >> >> >> >> >> --- mtd-2.6.orig/arch/arm/mach-omap2/gpmc.c
> >> >> >> >> >> +++ mtd-2.6/arch/arm/mach-omap2/gpmc.c
> >> >> >> >> >> @@ -54,6 +54,12 @@
> >> >> >> >> >>  #define GPMC_CHUNK_SHIFT     24              /* 16 MB */
> >> >> >> >> >>  #define GPMC_SECTION_SHIFT   28              /* 128 MB */
> >> >> >> >> >>
> >> >> >> >> >> +#ifdef CONFIG_MTD_NAND_OMAP_PREFETCH
> >> >> >> >> >> +#define CS_NUM_SHIFT         24
> >> >> >> >> >> +#define ENABLE_PREFETCH              7
> >> >> >> >> >> +#define DMA_MPU_MODE         2
> >> >> >> >> >> +#endif
> >> >> >> >> >> +
> >> >> >> >> >>  static struct resource       gpmc_mem_root;
> >> >> >> >> >>  static struct resource       gpmc_cs_mem[GPMC_CS_NUM];
> >> >> >> >> >>  static DEFINE_SPINLOCK(gpmc_mem_lock);
> >> >> >> >> >> @@ -383,6 +389,99 @@ void gpmc_cs_free(int cs)
> >> >> >> >> >>  }
> >> >> >> >> >>  EXPORT_SYMBOL(gpmc_cs_free);
> >> >> >> >> >>
> >> >> >> >> >> +#ifdef CONFIG_MTD_NAND_OMAP_PREFETCH
> >> >> >> >> >> +/**
> >> >> >> >> >> + * gpmc_prefetch_init - configures default configuration for prefetch
> >> >> >> engine
> >> >> >> >> >> + */
> >> >> >> >> >> +static void gpmc_prefetch_init(void)
> >> >> >> >> >> +{
> >> >> >> >> >> +     /* Setting the default threshold to 64 */
> >> >> >> >> >> +     gpmc_write_reg(GPMC_PREFETCH_CONTROL, 0x0);
> >> >> >> >> >> +     gpmc_write_reg(GPMC_PREFETCH_CONFIG1, 0x40  << 8);
> >> >> >> >> >> +     gpmc_write_reg(GPMC_PREFETCH_CONFIG2, 0x0);
> >> >> >> >> >> +}
> >> >> >> >> >
> >> >> >> >> > Why would you want to have NAND specific init code int gpmc.c?
> >> >> >> >> >
> >> >> >> >> > The purpose if gpmc.c is to provide access to configuring the
> >> >> >> >> > General Purpose Memory Controller (GPMC). You should just provide
> >> >> >> >> > functions in gpmc.c for the platform init code to use, and then
> >> >> >> >> > the drivers can stay platform independent.
> >> >> >> >>
> >> >> >> >> In my understanding, this 'prefetch' engine is part of GPMC itself, it is a
> >> >> >> >> kind of feature provided by GPMC which can be utilized by NAND driver.
> >> >> >> >> So, to me, it makes sens to get initialized prefetch by GPMC itself so that
> >> >> >> >> NAND driver can use it.
> >> >> >> >
> >> >> >> > But it should not have a dependency to NAND.
> >> >> >>
> >> >> >> This engine, in GPMC, is dedicated for NAND devices only.
> >> >> >>
> >> >> >> >
> >> >> >> >> Another reason was that all read / write to GPMC register are done by
> >> >> >> >> functions 'gpmc_read_reg' / 'gpmc_write_reg', which have been made
> >> >> >> >> 'static' in nature.
> >> >> >> >
> >> >> >> > That's why you need to provide a generic function in gpmc.c to enable
> >> >> >> > prefetch that the platform code for any driver can use.
> >> >> >>
> >> >> >> Exactly, and whenever a platform code uses gpmc init call, gpmc initializes
> >> >> >> this engine too. Since prefetch engine is the part of GPMC, IMHOP, it should
> >> >> >> get initialized as part of GPMC initialization.
> >> >> >
> >> >> > No, the driver needing it should initialize it. There should not be any
> >> >> > ifdefs needed in the gpmc.c for that.
> >> >>
> >> >> Are you suggesting to move gpmc prefetch specific code to 'drivers/mtd/nand/omap2.c'?
> >> >> This can be done but then, we will be accessing gpmc registers outside gpmc.c. Is that ok?
> >> >
> >> > Well I'm thinking that you just basically need something like this in
> >> > gpmc.c:
> >> >
> >> > int gpmc_prefetch_request(int cs);
> >> > int gpmc_prefetch_enable(int cs);
> >> > int gpmc_prefetch_reset(int cs);
> >> > ...
> >> >
> >> > And gpmc_prefetch_request would then return an error if already taken.
> >> > No need then to tinker with the GPMC prefetch registers in the NAND
> >> > driver.
> >>
> >> This seems exactly what in this patch has been done, other than renaming the functions and suggestion that gpmc prefetch init call should happen from NAND driver and should return success or failure based on whether it has already been initialized or not. Am I correct?
> >
> > Yes, let's take another look once those have been updated.
> 
> Plz refer to my comments below.
> 
> >
> >> But since, the prefetch and write-posting engine is a single-context engine that can be allocated to only one chip-select at a time for a read prefetch or a write-posting process, I feel it is still OK if we initialize gpmc prefetch engine only once in gpmc.c only and then whenever there is a request for gmpc prefetch start/enable we should check if engine is already being used by some other chip-select by verifying GPMC_PREFETCH_CONFIG1.ENABLEENGINE (bit '7').
> >>
> >> This was not done because there is only NAND device, as of now, in all the boards we are using.
> >>
> >> Let me know what you think about this.
> >
> > There's absolutely no reason to add any NAND specific code to gpmc.c.
> 
> Yes, and this patch has no NAND specific code in gpmc.c. But it has functions required for gpmc prefetch:
> 
> static void gpmc_prefetch_init(void)
> void gpmc_prefetch_start(int cs, int dma_mode,
>                                      unsigned int u32_count, int is_write)
> void gpmc_prefetch_stop(void)
> int gpmc_prefetch_status(void)
> 
> 
> Which are similar to functions you are suggesting:
> int gpmc_prefetch_request(int cs);       -  gpmc_prefetch_init
> int gpmc_prefetch_enable(int cs);      -  gpmc_prefetch_start
> int gpmc_prefetch_reset(int cs);      -  gpmc_prefetch_stop
> ...
> 
> As per my understanding this patch already implements your comments, Plz let me know if I am missing something. 

Please read my comments again, here's the link to the thread:

http://lkml.org/lkml/2009/6/4/145

So far I've asked you to change the following things:

- Get rid of the #ifdef CONFIG_MTD_NAND_OMAP_PREFETCH

- Call the gpmc_prefetch_request() from the nand driver

- Return an error when the gpmc_prefetch_request() fails to avoid multiple drivers
  trying to initialize the prefetch support

- Change the function namings to follow the gpmc.c naming

Regards,

Tony


 
> -vmial
> 
> >
> > Tony
> >
> >>
> >> >
> >> >> >
> >> >> >> And then there are prefetch start, stop and status functions calls have been
> >> >> >> provided to be used by NAND driver.
> >> >> >
> >> >> > These should really be done in the platform init code for the NAND driver.
> >> >> > The NAND driver really should be generic, not omap specific.
> >> >>
> >> >> Are you talking about different omap versions or omap itself, I am bit confused. This NAND driver is
> >> >> omap specific only.
> >> >
> >> > I guess I'm talking in generic terms. What's under drivers/* should in the long
> >> > run be device independent code, and the platform specific code should be under
> >> > arch/arm/*omap*.
> >>
> >> Yes, this is exactly what I was trying to do. GPMC prefetch is platform specific feature and hence code related to this was kept in gmpc.c (i.e. arch/arm/*omap*/.).
> >>
> >> ---
> >> Regards,
> >> Vimal
> >> >
> >> > Regards,
> >> >
> >> > Tony
> >> > --
> >> > To unsubscribe from this list: send the line "unsubscribe linux-omap" in
> >> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> >> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >> >
> >>
> >>
> >>
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-omap" in
> > the body of a message to majordomo@xxxxxxxxxxxxxxx
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/