Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On Fri, 1 Feb 2019 14:49:27 + wrote: > > +#define QSPI_IFR_APBTFRTYP_READ BIT(24) > > > > And this one would be > > > > define QSPI_IFR_SAM9X60_READ_TRSFR BIT(24) > > I prefer letting this bit named as in the datasheet, QSPI_IFR_APBTFRTYP_READ, > and change it if future versions of the IP will modify its sense. It is a READ > transfer done on APB, it is more generic this way. If you have a strong > opinion > on this, please let me know. As you wish.
Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On 02/01/2019 09:07 AM, tudor.amba...@microchip.com wrote: cut >>> diff --git a/drivers/spi/atmel-quadspi.c b/drivers/spi/atmel-quadspi.c cut >>> +static int atmel_sam9x60_qspi_set_cfg(void __iomem *base, >>> + const struct spi_mem_op *op, >>> + struct atmel_qspi_cfg *cfg) >>> +{ >>> + int ret = atmel_qspi_set_mode(cfg, op); >>> + >>> + if (ret) >>> + return ret; >>> + >>> + ret = atmel_qspi_set_address_mode(cfg, op); >>> + if (ret) >>> + return ret; >>> + >>> + cfg->ifr |= QSPI_IFR_INSTEN; >>> + cfg->icr |= QSPI_ICR_INST(op->cmd.opcode); >>> + >>> + /* Set data enable */ >>> + if (op->data.nbytes) >>> + cfg->ifr |= QSPI_IFR_DATAEN; >>> + >>> + if (!op->addr.nbytes) { >>> + cfg->ifr |= QSPI_IFR_TFRTYP_TRSFR_REG; >>> + if (op->data.dir == SPI_MEM_DATA_OUT) >>> + cfg->ifr |= QSPI_IFR_APBTFRTYP_WRITE; >>> + else >>> + cfg->ifr |= QSPI_IFR_APBTFRTYP_READ; >>> + } else { >>> + cfg->ifr |= QSPI_IFR_TFRTYP_TRSFR_MEM; >> >> Can you try doing only regular transfers and let me know if it still >> works. Support for mem transfers can then be added along with dirmap >> support. > > should work. Will try and let you know. you were right, it works. I will let mem transfer logic for dirmap support. Cheers, ta
Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On 02/01/2019 09:57 AM, Boris Brezillon wrote: > On Fri, 1 Feb 2019 07:07:40 + > wrote: > >>> #define QSPI_IFR_TFRTYP_MASKGENMASK(13, 12) #define QSPI_IFR_TFRTYP_TRSFR_READ (0 << 12) #define QSPI_IFR_TFRTYP_TRSFR_READ_MEM (1 << 12) >>> >>> Looks like the read/write flag is on bit 13. Can we just add >> >> for sama5d2 only > > Feel free to prefix macros with the SoC name to make it clear: > > #define QSPI_IFR_SAMA5D2_WRITE_TRSFR BIT(13) agreed > cut +#define QSPI_IFR_APBTFRTYP_READ BIT(24) > > And this one would be > > define QSPI_IFR_SAM9X60_READ_TRSFRBIT(24) I prefer letting this bit named as in the datasheet, QSPI_IFR_APBTFRTYP_READ, and change it if future versions of the IP will modify its sense. It is a READ transfer done on APB, it is more generic this way. If you have a strong opinion on this, please let me know. ta
Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On Fri, 1 Feb 2019 07:07:40 + wrote: > > > >> #define QSPI_IFR_TFRTYP_MASKGENMASK(13, 12) > >> #define QSPI_IFR_TFRTYP_TRSFR_READ (0 << 12) > >> #define QSPI_IFR_TFRTYP_TRSFR_READ_MEM (1 << 12) > > > > Looks like the read/write flag is on bit 13. Can we just add > > for sama5d2 only Feel free to prefix macros with the SoC name to make it clear: #define QSPI_IFR_SAMA5D2_WRITE_TRSFRBIT(13) > > > > > #define QSPI_IFR_TFRTYP_TRSFR_WRITE BIT(13) > > > > and drop all others def? This way the implementation is consistent > > between sam9x60 and sama5d2. > > BIT(13) has no meaning for sam9x60. I can drop the macros with zero value for > sama5d2 in a separate patch. > > > >> +#define QSPI_IFR_APBTFRTYP_READ BIT(24) And this one would be define QSPI_IFR_SAM9X60_READ_TRSFR BIT(24) > >> > >> /* Bitfields in QSPI_SMR (Scrambling Mode Register) */ > >> #define QSPI_SMR_SCREN BIT(0) > >> @@ -137,16 +144,37 @@ > >> #define QSPI_WPSR_WPVSRC(src) (((src) << 8) & QSPI_WPSR_WPVSRC) > >> > >> > >> +/* Describes register values. */ > >> +struct atmel_qspi_cfg { > >> + u32 icr; > >> + u32 iar; > >> + u32 ifr; > >> +}; > >> + > >> +struct atmel_qspi_caps; > >> + > >> struct atmel_qspi { > >>void __iomem*regs; > >>void __iomem*mem; > >>struct clk *clk; > > > > Can we rename that on pclk? > > will rename it, together with the support for unnamed clock of sama5d2 in a > separate > patch. The dt-bindings patch that imposes "pclk" for sama5d2 should be > separated too. Sounds good.
Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On 01/31/2019 06:32 PM, Boris Brezillon wrote: > On Thu, 31 Jan 2019 16:15:51 + > wrote: > >> From: Tudor Ambarus >> >> The sam9x60 qspi controller uses 2 clocks, one for the peripheral register >> access, the other for the qspi core and phy. Both are mandatory. It uses >> dedicated register for Read Instruction Code Register (RICR) and >> Write Instruction Code Register (WICR). ICR/RICR/WICR have identical >> fields. >> >> Tested with sst26vf064b jedec,spi-nor flash. Backward compatibility test >> done on sama5d2 qspi controller and mx25l25635e jedec,spi-nor flash. >> >> Signed-off-by: Tudor Ambarus >> --- >> v2: >> - rework clock handling >> - reorder setting of register values in set_cfg() calls -> move functions >> that can fail in the upper part of the function body. >> >> drivers/spi/atmel-quadspi.c | 296 >> +++- >> 1 file changed, 239 insertions(+), 57 deletions(-) >> >> diff --git a/drivers/spi/atmel-quadspi.c b/drivers/spi/atmel-quadspi.c >> index d3e76acf8517..80c934f3e479 100644 >> --- a/drivers/spi/atmel-quadspi.c >> +++ b/drivers/spi/atmel-quadspi.c >> @@ -19,6 +19,7 @@ >> #include >> #include >> #include >> +#include >> #include >> #include >> >> @@ -35,7 +36,9 @@ >> >> #define QSPI_IAR 0x0030 /* Instruction Address Register */ >> #define QSPI_ICR 0x0034 /* Instruction Code Register */ >> +#define QSPI_WICR0x0034 /* Write Instruction Code Register */ >> #define QSPI_IFR 0x0038 /* Instruction Frame Register */ >> +#define QSPI_RICR0x003C /* Read Instruction Code Register */ >> >> #define QSPI_SMR 0x0040 /* Scrambling Mode Register */ >> #define QSPI_SKR 0x0044 /* Scrambling Key Register */ >> @@ -88,7 +91,7 @@ >> #define QSPI_SCR_DLYBS_MASK GENMASK(23, 16) >> #define QSPI_SCR_DLYBS(n) (((n) << 16) & QSPI_SCR_DLYBS_MASK) >> >> -/* Bitfields in QSPI_ICR (Instruction Code Register) */ >> +/* Bitfields in QSPI_ICR (Read/Write Instruction Code Register) */ >> #define QSPI_ICR_INST_MASK GENMASK(7, 0) >> #define QSPI_ICR_INST(inst) (((inst) << 0) & QSPI_ICR_INST_MASK) >> #define QSPI_ICR_OPT_MASK GENMASK(23, 16) >> @@ -113,6 +116,8 @@ >> #define QSPI_IFR_OPTL_4BIT (2 << 8) >> #define QSPI_IFR_OPTL_8BIT (3 << 8) >> #define QSPI_IFR_ADDRL BIT(10) >> +#define QSPI_IFR_TFRTYP_TRSFR_MEM BIT(12) >> +#define QSPI_IFR_TFRTYP_TRSFR_REG (0 << 12) > > You don't need to define TRSFR_REG, just set QSPI_IFR_TFRTYP_TRSFR_MEM > when you do a mem transfer and do nothing when this is a regular > transfer. I chose to introduce macros with zero value for better code readability. I would expect that the NOP operations to be optimized at compile time. I will remove them if you prefer, it will result in fewer lines of code. > >> #define QSPI_IFR_TFRTYP_MASKGENMASK(13, 12) >> #define QSPI_IFR_TFRTYP_TRSFR_READ (0 << 12) >> #define QSPI_IFR_TFRTYP_TRSFR_READ_MEM (1 << 12) > > Looks like the read/write flag is on bit 13. Can we just add for sama5d2 only > > #define QSPI_IFR_TFRTYP_TRSFR_WRITE BIT(13) > > and drop all others def? This way the implementation is consistent > between sam9x60 and sama5d2. BIT(13) has no meaning for sam9x60. I can drop the macros with zero value for sama5d2 in a separate patch. > >> @@ -121,6 +126,8 @@ >> #define QSPI_IFR_CRMBIT(14) >> #define QSPI_IFR_NBDUM_MASK GENMASK(20, 16) >> #define QSPI_IFR_NBDUM(n) (((n) << 16) & QSPI_IFR_NBDUM_MASK) >> +#define QSPI_IFR_APBTFRTYP_WRITE(0 << 24) > > As for the other defs, I don't think you need to define _WRITE. understood > >> +#define QSPI_IFR_APBTFRTYP_READ BIT(24) >> >> /* Bitfields in QSPI_SMR (Scrambling Mode Register) */ >> #define QSPI_SMR_SCREN BIT(0) >> @@ -137,16 +144,37 @@ >> #define QSPI_WPSR_WPVSRC(src) (((src) << 8) & QSPI_WPSR_WPVSRC) >> >> >> +/* Describes register values. */ >> +struct atmel_qspi_cfg { >> +u32 icr; >> +u32 iar; >> +u32 ifr; >> +}; >> + >> +struct atmel_qspi_caps; >> + >> struct atmel_qspi { >> void __iomem*regs; >> void __iomem*mem; >> struct clk *clk; > > Can we rename that on pclk? will rename it, together with the support for unnamed clock of sama5d2 in a separate patch. The dt-bindings patch that imposes "pclk" for sama5d2 should be separated too. > >> +struct clk *qspick; >> struct platform_device *pdev; >> +const struct atmel_qspi_caps *caps; >> u32 pending; >> u32 mr; >> struct completion cmd_completion; >> }; >> > > ... > >> + >> +static int atmel_sam9x60_qspi_set_cfg(void __iomem *base, >> + const struct spi_mem_op *op, >> +
Re: [PATCH v2 10/10] spi: atmel-quadspi: add support for sam9x60 qspi controller
On Thu, 31 Jan 2019 16:15:51 + wrote: > From: Tudor Ambarus > > The sam9x60 qspi controller uses 2 clocks, one for the peripheral register > access, the other for the qspi core and phy. Both are mandatory. It uses > dedicated register for Read Instruction Code Register (RICR) and > Write Instruction Code Register (WICR). ICR/RICR/WICR have identical > fields. > > Tested with sst26vf064b jedec,spi-nor flash. Backward compatibility test > done on sama5d2 qspi controller and mx25l25635e jedec,spi-nor flash. > > Signed-off-by: Tudor Ambarus > --- > v2: > - rework clock handling > - reorder setting of register values in set_cfg() calls -> move functions > that can fail in the upper part of the function body. > > drivers/spi/atmel-quadspi.c | 296 > +++- > 1 file changed, 239 insertions(+), 57 deletions(-) > > diff --git a/drivers/spi/atmel-quadspi.c b/drivers/spi/atmel-quadspi.c > index d3e76acf8517..80c934f3e479 100644 > --- a/drivers/spi/atmel-quadspi.c > +++ b/drivers/spi/atmel-quadspi.c > @@ -19,6 +19,7 @@ > #include > #include > #include > +#include > #include > #include > > @@ -35,7 +36,9 @@ > > #define QSPI_IAR 0x0030 /* Instruction Address Register */ > #define QSPI_ICR 0x0034 /* Instruction Code Register */ > +#define QSPI_WICR0x0034 /* Write Instruction Code Register */ > #define QSPI_IFR 0x0038 /* Instruction Frame Register */ > +#define QSPI_RICR0x003C /* Read Instruction Code Register */ > > #define QSPI_SMR 0x0040 /* Scrambling Mode Register */ > #define QSPI_SKR 0x0044 /* Scrambling Key Register */ > @@ -88,7 +91,7 @@ > #define QSPI_SCR_DLYBS_MASK GENMASK(23, 16) > #define QSPI_SCR_DLYBS(n) (((n) << 16) & QSPI_SCR_DLYBS_MASK) > > -/* Bitfields in QSPI_ICR (Instruction Code Register) */ > +/* Bitfields in QSPI_ICR (Read/Write Instruction Code Register) */ > #define QSPI_ICR_INST_MASK GENMASK(7, 0) > #define QSPI_ICR_INST(inst) (((inst) << 0) & QSPI_ICR_INST_MASK) > #define QSPI_ICR_OPT_MASK GENMASK(23, 16) > @@ -113,6 +116,8 @@ > #define QSPI_IFR_OPTL_4BIT (2 << 8) > #define QSPI_IFR_OPTL_8BIT (3 << 8) > #define QSPI_IFR_ADDRL BIT(10) > +#define QSPI_IFR_TFRTYP_TRSFR_MEMBIT(12) > +#define QSPI_IFR_TFRTYP_TRSFR_REG(0 << 12) You don't need to define TRSFR_REG, just set QSPI_IFR_TFRTYP_TRSFR_MEM when you do a mem transfer and do nothing when this is a regular transfer. > #define QSPI_IFR_TFRTYP_MASKGENMASK(13, 12) > #define QSPI_IFR_TFRTYP_TRSFR_READ (0 << 12) > #define QSPI_IFR_TFRTYP_TRSFR_READ_MEM (1 << 12) Looks like the read/write flag is on bit 13. Can we just add #define QSPI_IFR_TFRTYP_TRSFR_WRITE BIT(13) and drop all others def? This way the implementation is consistent between sam9x60 and sama5d2. > @@ -121,6 +126,8 @@ > #define QSPI_IFR_CRMBIT(14) > #define QSPI_IFR_NBDUM_MASK GENMASK(20, 16) > #define QSPI_IFR_NBDUM(n) (((n) << 16) & QSPI_IFR_NBDUM_MASK) > +#define QSPI_IFR_APBTFRTYP_WRITE (0 << 24) As for the other defs, I don't think you need to define _WRITE. > +#define QSPI_IFR_APBTFRTYP_READ BIT(24) > > /* Bitfields in QSPI_SMR (Scrambling Mode Register) */ > #define QSPI_SMR_SCREN BIT(0) > @@ -137,16 +144,37 @@ > #define QSPI_WPSR_WPVSRC(src) (((src) << 8) & QSPI_WPSR_WPVSRC) > > > +/* Describes register values. */ > +struct atmel_qspi_cfg { > + u32 icr; > + u32 iar; > + u32 ifr; > +}; > + > +struct atmel_qspi_caps; > + > struct atmel_qspi { > void __iomem*regs; > void __iomem*mem; > struct clk *clk; Can we rename that on pclk? > + struct clk *qspick; > struct platform_device *pdev; > + const struct atmel_qspi_caps *caps; > u32 pending; > u32 mr; > struct completion cmd_completion; > }; > ... > + > +static int atmel_sam9x60_qspi_set_cfg(void __iomem *base, > + const struct spi_mem_op *op, > + struct atmel_qspi_cfg *cfg) > +{ > + int ret = atmel_qspi_set_mode(cfg, op); > + > + if (ret) > + return ret; > + > + ret = atmel_qspi_set_address_mode(cfg, op); > + if (ret) > + return ret; > + > + cfg->ifr |= QSPI_IFR_INSTEN; > + cfg->icr |= QSPI_ICR_INST(op->cmd.opcode); > + > + /* Set data enable */ > + if (op->data.nbytes) > + cfg->ifr |= QSPI_IFR_DATAEN; > + > + if (!op->addr.nbytes) { > + cfg->ifr |= QSPI_IFR_TFRTYP_TRSFR_REG; > + if (op->data.dir == SPI_MEM_DATA_OUT) > + cfg->ifr |= QSPI_IFR_APBTFRTYP_WRITE; > + else > + cfg->ifr |= QSPI_IFR_A