add driver for Atmel QSPI controller
Subsystems (5)
v5 cover add driver for Atmel QSPI controller
ChangeLog
v5:
- remove unused inline functions qspi_read[bw]() and qspi_write[bw](), keep only qspi_readl() and qspi_writel().
- use reinit_completion() instead of init_completion() during run time, call init_completion() once for all in the probe().
- add a dev_warn() when trying to tune the number of dummy cycles for spi-nor of a not supported manufacturer then fall back to the default framework value.
- reword some comments.
- add "Acked-by: Marek Vasut <[email protected]>" for patches 1 and 4.
- add "Acked-by: Bean Huo <[email protected]>" for patch 1.
v4:
- add "OF && HAS_DMA" dependency in Kconfig for Atmel Quad SPI driver.
- return -ENOMEM instead of the return code of dma_mapping_error() as this function returns a boolean on ARM achitecture.
- add "Acked-by: Nicolas Ferre <[email protected]>" for Atmel Quad SPI driver and its DT binding documentation.
v3:
- reword the comment which explains that spi_nor_set_protocol() is used by the spi-nor framework to notify lower layers, especially the (Q)SPI controller about a protocol change.
- change the definitions of register/bitfield macros in the Atmel QSPI controller driver: get rid of concatenation operator and use BIT and GENMASK macros when possible.
- use #define[SPACE] instead of #define[TAB]
v2:
- remove the patches to set the "latency code" of Spansion QSPI memories (support of Spansion memories may be submitted in later series).
- rename "qspi" node into "spi" in the DT example to fit ePAPR standard.
- remove the useless "qspi0" label from the DT node example.
- remove the leading 0 from the size of the second memory region to make it consistent with the size of the first memory region.
- indent the DT bindings documentation to make it more readable.
- remove the useless ".bus = &platform_bus_type," line from the
platform driver definition.
v1:
This series of patches add support for the new Atmel QSPI controller embedded inside sama5d2x SoCs.
These patches were first developped for linux-3.18-at91 and tested on a sama5d27 Xplained ultra board, which embeds a Micron n25q128a13 QSPI NOR flash memory. Then the series was adapted for mainline.
Cyrille Pitchen (5):
mtd: spi-nor: notify (Q)SPI controller about protocol change
Documentation: mtd: add a DT property to set the number of dummy
cycles
mtd: spi-nor: allow to tune the number of dummy cycles
Documentation: atmel-quadspi: add binding file for Atmel QSPI driver
mtd: atmel-quadspi: add driver for Atmel QSPI controller
.../devicetree/bindings/mtd/atmel-quadspi.txt | 29 + .../devicetree/bindings/mtd/jedec,spi-nor.txt | 6 + drivers/mtd/spi-nor/Kconfig | 7 + drivers/mtd/spi-nor/Makefile | 1 + drivers/mtd/spi-nor/atmel-quadspi.c | 859 +++++++++++++++++++++ drivers/mtd/spi-nor/spi-nor.c | 123 ++- include/linux/mtd/spi-nor.h | 15 + 7 files changed, 1021 insertions(+), 19 deletions(-) create mode 100644 Documentation/devicetree/bindings/mtd/atmel-quadspi.txt create mode 100644 drivers/mtd/spi-nor/atmel-quadspi.c
Signature · 1 line
1.8.2.2
PATCH v5 1/5 mtd: spi-nor: notify (Q)SPI controller about protocol change
Once the Quad SPI mode has been enabled on a Micron flash memory, this device expects ALL the following commands to use the SPI 4-4-4 protocol. The (Q)SPI controller needs to be notified about the protocol change so it can adapt and keep on dialoging with the Micron memory.
--- drivers/mtd/spi-nor/spi-nor.c | 21 +++++++++++++++++++++ include/linux/mtd/spi-nor.h | 13 +++++++++++++ 2 files changed, 34 insertions(+)
drivers/mtd/spi-nor/spi-nor.c +21 -0
@@ -165,6 +165,22 @@ static inline int write_disable(struct spi_nor *nor)return nor->write_reg (nor,SPINOR_OP_WRDI ,NULL ,0 ,0 ); } +/* +* Let the spi-nor framework notify lower layers, especially the driver of the +* (Q)SPI controller, about the new protocol to be used. Indeed, once the +* spi-nor framework has sent manufacturer specific commands to a memory to +* enable its Quad SPI mode, it should immediately after tell the QSPI +* controller to use the very same Quad SPI protocol as expected by the memory. +*/ +static inline int spi_nor_set_protocol (struct spi_nor *nor, +enum spi_protocol proto) +{ +if (nor->set_protocol) +return nor->set_protocol (nor, proto); + +return 0 ; +} +static inline struct spi_nor *mtd_to_spi_nor (struct mtd_info *mtd) {return mtd->priv;
@@ -940,6 +956,11 @@ static int micron_quad_enable(struct spi_nor *nor)return ret; } +/* switch protocol to Quad CMD 4-4-4 */ + ret =spi_nor_set_protocol (nor,SPI_PROTO_4_4_4 ); +if (ret) +return ret; + ret =spi_nor_wait_till_ready (nor);if (ret)return ret;
include/linux/mtd/spi-nor.h +13 -0
@@ -87,6 +87,16 @@ enum read_mode { SPI_NOR_QUAD, }; +enum spi_protocol { +SPI_PROTO_1_1_1 ,/* SPI */ +SPI_PROTO_1_1_2 ,/* Dual Output */ +SPI_PROTO_1_1_4 ,/* Quad Output */ +SPI_PROTO_1_2_2 ,/* Dual IO */ +SPI_PROTO_1_4_4 ,/* Quad IO */ +SPI_PROTO_2_2_2 ,/* Dual Command */ +SPI_PROTO_4_4_4 ,/* Quad Command */ +}; +/** * struct spi_nor_xfer_cfg - Structure for defining a Serial Flash transfer * @wren: command for "Write Enable", or 0x00 for not required
@@ -149,6 +159,7 @@ enum spi_nor_option_flags {* read/write/erase/lock/unlock operations * @read_xfer: [OPTIONAL] the read fundamental primitive * @write_xfer: [OPTIONAL] the writefundamental primitive +* @set_protocol: [OPTIONAL] notify about protocol change * @read_reg: [DRIVER-SPECIFIC] read out the register * @write_reg: [DRIVER-SPECIFIC] write data to the register * @read: [DRIVER-SPECIFIC] read data from the SPI NOR
@@ -185,6 +196,8 @@ struct spi_nor {int (*write_reg)(struct spi_nor *nor,u8 opcode,u8 *buf,int len,int write_enable); +int (*set_protocol)(struct spi_nor *nor,enum spi_protocol proto); +int (*read)(struct spi_nor *nor,loff_t from,size_t len,size_t *retlen,u_char *read_buf); void (*write)(struct spi_nor *nor, loff_t to,
Signature · 1 line
1.8.2.2
PATCH v5 2/5 Documentation: mtd: add a DT property to set the number of dummy cycles
In reply to Cyrille PitchenDepending on the SPI clock frequency, the Fast Read op code and the Single/Dual Data Rate mode, the number of dummy cycles can be tuned to improve transfer speed.
The actual number of dummy cycles is specific for each memory model and is provided by the manufacturer thanks to the memory datasheet.
--- Documentation/devicetree/bindings/mtd/jedec,spi-nor.txt | 6 ++++++ 1 file changed, 6 insertions(+)
Documentation/devicetree/bindings/mtd/jedec,spi-nor.txt +6 -0
@@ -19,6 +19,11 @@ Optional properties: all chips and support for it can not be detected at runtime. Refer to your chips' datasheet to check if this is supported by your chip. +- m25p,num-dummy-cycles : Set the number of dummy cycles for Fast Read commands. + Depending on the manufacturer additional dedicated + commands are sent to the flash memory so the + controller and the memory can agree on the number of + dummy cycles to use. Example:
@@ -29,4 +34,5 @@ Example: reg = <0>; spi-max-frequency = <40000000>; m25p,fast-read; + m25p,num-dummy-cycles = <8>; };
Signature · 1 line
1.8.2.2
PATCH v5 3/5 mtd: spi-nor: allow to tune the number of dummy cycles
In reply to Cyrille PitchenThe number of dummy cycles used during Fast Read commands can be reduced to improve transfer performances. Each manufacturer has a dedicated set of registers to provide the memory with the exact number of dummy cycles it should expect. Both the memory and the (Q)SPI controller must agree on this number of dummy cycles.
The number of dummy cycles can be found into the memory datasheet and mostly depends on the SPI clock frequency, the Fast Read op code and the Single/Dual Data Rate mode.
Probing JEDEC Serial Flash Discoverable Parameters (SFDP) tables would only provide the driver with a high enough number of dummy cycles for each Fast Read command to be used for all clock frequencies: this solution would not be optimized.
--- drivers/mtd/spi-nor/spi-nor.c | 102 ++++++++++++++++++++++++++++++++++-------- include/linux/mtd/spi-nor.h | 2 + 2 files changed, 85 insertions(+), 19 deletions(-)
drivers/mtd/spi-nor/spi-nor.c +83 -19
@@ -121,24 +121,6 @@ static int read_cr(struct spi_nor *nor) }/* -* Dummy Cycle calculation for different type of read. -* It can be used to support more commands with -* different dummy cycle requirements. -*/ -static inline int spi_nor_read_dummy_cycles (struct spi_nor *nor) -{ -switch (nor->flash_read) { -case SPI_NOR_FAST : -case SPI_NOR_DUAL : -case SPI_NOR_QUAD : -return 8 ; -case SPI_NOR_NORMAL : -return 0 ; - } -return 0 ; -} - -/* * Write status register 1 byte * Returns negative if error occurred. */
@@ -1008,6 +990,86 @@ static int set_quad_mode(struct spi_nor *nor, const struct flash_info *info) } } +static int micron_set_dummy_cycles (struct spi_nor *nor) +{ +int ret; +u8 val, mask; + +/* Read the Volatile Configuration Register (VCR). */ + ret = nor->read_reg (nor,SPINOR_OP_RD_VCR , &val,1 ); +if (ret <0 ) { +dev_err (nor->dev,"error %d reading VCR\n" , ret); +return ret; + } + +write_enable (nor); + +/* Update the number of dummy into the VCR. */ + mask =GENMASK (7 ,4 ); + val &= ~mask; + val |= (nor->read_dummy <<4 ) & mask; + ret = nor->write_reg (nor,SPINOR_OP_WR_VCR , &val,1 ,0 ); +if (ret <0 ) { +dev_err (nor->dev,"error while writing VCR register\n" ); +return ret; + } + + ret =spi_nor_wait_till_ready (nor); +if (ret) +return ret; + +return 0 ; +} + +/* +* Dummy Cycle calculation for different type of read. +* It can be used to support more commands with +* different dummy cycle requirements. +*/ +static int spi_nor_read_dummy_cycles (struct spi_nor *nor, +const struct flash_info *info) +{ +struct device_node *np = nor->dev->of_node; +struct device *dev = nor->dev; +u32 num_dummy_cycles; + +if (np && !of_property_read_u32 (np,"m25p,num-dummy-cycles" , + &num_dummy_cycles)) { + nor->read_dummy = num_dummy_cycles; + +/* +* This switch block might be moved at the end of the function, +* once nor->read_dummy has been set, but it was not tested with +* all Micron memories. +* Now the "m25p,num-dummy-cycles" property needs to be +* explicitly set in the device tree so the switch statement is +* executed. This should avoid unwanted side effects and keep +* backward compatibility. +*/ +switch (JEDEC_MFR (info)) { +case CFI_MFR_ST : +return micron_set_dummy_cycles (nor); +default : +dev_warn (dev, +"Tuning of the number of dummy cycles is not implemented for spi-nor of this manufacturer,\n" +"ignoring the DT property value and falling back to the framework default settings.\n" ); +break ; + } + } + +/* Fallback to legacy code. */ +switch (nor->flash_read) { +case SPI_NOR_FAST : +case SPI_NOR_DUAL : +case SPI_NOR_QUAD : + nor->read_dummy =8 ; +case SPI_NOR_NORMAL : + nor->read_dummy =0 ; + } + +return 0 ; +} +static int spi_nor_check (struct spi_nor *nor) { if (!nor->dev || !nor->read || !nor->write ||
@@ -1208,7 +1270,9 @@ int spi_nor_scan(struct spi_nor *nor, const char *name, enum read_mode mode) nor->addr_width =3 ; } - nor->read_dummy =spi_nor_read_dummy_cycles (nor); + ret =spi_nor_read_dummy_cycles (nor, info); +if (ret) +return ret;dev_info (dev,"%s (%lld Kbytes)\n" , info->name, (long long )mtd->size >>10 );
include/linux/mtd/spi-nor.h +2 -0
@@ -59,6 +59,8 @@/* Used for Micron flashes only. */ #define SPINOR_OP_RD_EVCR 0x65/* Read EVCR register */ #define SPINOR_OP_WD_EVCR 0x61/* Write EVCR register */ +#define SPINOR_OP_RD_VCR 0x85/* Read VCR register */ +#define SPINOR_OP_WR_VCR 0x81/* Write VCR register */ /* Status Register bits. */ #define SR_WIP 1/* Write in progress */
Signature · 1 line
1.8.2.2
PATCH v5 4/5 Documentation: atmel-quadspi: add binding file for Atmel QSPI driver
In reply to Cyrille PitchenThis patch documents the DT bindings for the driver of the Atmel QSPI controller embedded inside sama5d2x SoCs.
--- .../devicetree/bindings/mtd/atmel-quadspi.txt | 29 ++++++++++++++++++++++ 1 file changed, 29 insertions(+) create mode 100644 Documentation/devicetree/bindings/mtd/atmel-quadspi.txt
Documentation/devicetree/bindings/mtd/atmel-quadspi.txt new file +29 -0
@@ -0,0 +1,29 @@ +* Atmel Quad Serial Peripheral Interface (QSPI) + +Required properties: +- compatible: should be "atmel,sama5d2-qspi" +- reg: the first contains the register location and length, + the second contains the memory mapping address and length +- interrupts: should contain the interrupt for the device +- clocks: the phandle of the clock needed by the QSPI controller +- #address-cells: should be <1> +- #size-cells: should be <0> + +Example: + +spi at f0020000 { + compatible = "atmel,sama5d2-qspi"; + reg = <0xf0020000 0x100>, + <0xd0000000 0x8000000>; + interrupts = <52 IRQ_TYPE_LEVEL_HIGH 7>; + clocks = <&spi0_clk>; + #address-cells = <1>; + #size-cells = <0>; + pinctrl-names = "default"; + pinctrl-0 = <&pinctrl_spi0_default>; + status = "okay"; + + m25p80 at 0 { + ... + }; +};
Signature · 1 line
1.8.2.2
PATCH v5 5/5 mtd: atmel-quadspi: add driver for Atmel QSPI controller
In reply to Cyrille PitchenThis driver add support to the new Atmel QSPI controller embedded into sama5d2x SoCs. It expects a NOR memory to be connected to the QSPI controller.
--- drivers/mtd/spi-nor/Kconfig | 7 + drivers/mtd/spi-nor/Makefile | 1 + drivers/mtd/spi-nor/atmel-quadspi.c | 859 ++++++++++++++++++++++++++++++++++++ 3 files changed, 867 insertions(+) create mode 100644 drivers/mtd/spi-nor/atmel-quadspi.c
drivers/mtd/spi-nor/Kconfig +7 -0
@@ -29,6 +29,13 @@ config SPI_FSL_QUADSPI This controller does not support generic SPI. It only supports SPI NOR. +config SPI_ATMEL_QUADSPI + tristate "Atmel Quad SPI Controller" + depends on OF && HAS_DMA && (ARCH_AT91 || COMPILE_TEST) + help + This enables support for the Quad SPI controller in master mode. + We only connect the NOR to this controller now. + config SPI_NXP_SPIFI tristate "NXP SPI Flash Interface (SPIFI)" depends on OF && (ARCH_LPC18XX || COMPILE_TEST)
drivers/mtd/spi-nor/Makefile +1 -0
@@ -1,3 +1,4 @@ obj-$(CONFIG_MTD_SPI_NOR) += spi-nor.o obj-$(CONFIG_SPI_FSL_QUADSPI) += fsl-quadspi.o +obj-$(CONFIG_SPI_ATMEL_QUADSPI) += atmel-quadspi.o obj-$(CONFIG_SPI_NXP_SPIFI) += nxp-spifi.o
drivers/mtd/spi-nor/atmel-quadspi.c new file +859 -0
859 lines not on this page · raw message
Signature · 1 line
1.8.2.2
Re: [PATCH linux-next v5 1/5] mtd: spi-nor: notify (Q)SPI controller about protocol change
In reply to Cyrille PitchenQuoted 4 lines · Cyrille Pitchen
Once the Quad SPI mode has been enabled on a Micron flash memory, this device expects ALL the following commands to use the SPI 4-4-4 protocol. The (Q)SPI controller needs to be notified about the protocol change so it can adapt and keep on dialoging with the Micron memory.
Doesn't that mean you need to disable quad mode on removal? Else the following will break/fail:
insmod atmel-quadspi.ko ~> spi-nor attaches -> sees micron -> enables quad mode rmmod atmel-quadspi.ko ~> spi-nor detaches
insmod atmel-quadspi.ko ~> spi-nor attaches -> fails to read the id because flash is still in quad mode.
Quoted 20 lines
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Marek Vasut <[email protected]>Acked-by Bean Huo <[email protected]>--- drivers/mtd/spi-nor/spi-nor.c | 21 +++++++++++++++++++++ include/linux/mtd/spi-nor.h | 13 +++++++++++++ 2 files changed, 34 insertions(+)drivers/mtd/spi-nor/spi-nor.c +4
-0@@ -165,6 +165,22 @@ static inline int write_disable(struct spi_nor *nor)return nor->write_reg (nor,SPINOR_OP_WRDI ,NULL ,0 ,0 ); } +/* +* Let the spi-nor framework notify lower layers, especially the driver of the +* (Q)SPI controller, about the new protocol to be used. Indeed, once the +* spi-nor framework has sent manufacturer specific commands to a memory to
drivers/mtd/spi-nor/spi-nor.c static inline int write_disable(struct spi_nor *nor) · line 181
Quoted 10 lines
drivers/mtd/spi-nor/spi-nor.c +10
-0+* enable its Quad SPI mode, it should immediately after tell the QSPI +* controller to use the very same Quad SPI protocol as expected by the memory. +*/ +static inline int spi_nor_set_protocol (struct spi_nor *nor, +enum spi_protocol proto) +{ +if (nor->set_protocol) +return nor->set_protocol (nor, proto); + +return 0 ;
Shouldn't the default assumption be that it won't support it? Also it might make sense to first check if it's supported before enabling it in the chip, so that we don't enable something just to then find out we can't back out of it.
I also wonder if we need an extra flag for that as at least SPI has RX_{DUAL,QUAD} and TX_{DUAL,QUAD} separated, so in theory there could be a controller that supports quad read, but not quad write, so we shouldn't be using the quad mode in that case. m25p80 currently sets SPI_NOR_{DUAL,QUAD} only based on SPI_RX_{DUAL,QUAD}, so that would then fail.
At least the n25q32 seems to support the "boring" 1_1_2 and 1_1_4 commands, so these should work in case the spi controller does not support quad tx, but quad rx.
So maybe an additional flag for the "full" dual/quad modes would be in order.
Last but not least, the protocol probably should be stored somewhere in the nor struct, so that users don't have to store it themselves (I assume they will need to check it for each command again to know if the cmd/address should be send in serial or quad mode).
Jonas
Re: [PATCH linux-next v5 3/5] mtd: spi-nor: allow to tune the number of dummy cycles
In reply to Cyrille PitchenQuoted 10 lines · Cyrille Pitchen
The number of dummy cycles used during Fast Read commands can be reduced to improve transfer performances. Each manufacturer has a dedicated set of registers to provide the memory with the exact number of dummy cycles it should expect. Both the memory and the (Q)SPI controller must agree on this number of dummy cycles.
The number of dummy cycles can be found into the memory datasheet and mostly depends on the SPI clock frequency, the Fast Read op code and the Single/Dual Data Rate mode.
Quoted 6 lines
Probing JEDEC Serial Flash Discoverable Parameters (SFDP) tables would only provide the driver with a high enough number of dummy cycles for each Fast Read command to be used for all clock frequencies: this solution would not be optimized.
Signed-off-by Cyrille Pitchen <[email protected]>
Best regards,
Marek Vasut
Re: [PATCH linux-next v5 5/5] mtd: atmel-quadspi: add driver for Atmel QSPI controller
In reply to Cyrille PitchenQuoted 6 lines · Cyrille Pitchen
This driver add support to the new Atmel QSPI controller embedded into sama5d2x SoCs. It expects a NOR memory to be connected to the QSPI controller.
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Nicolas Ferre <[email protected]>
Best regards,
Marek Vasut
Re: [PATCH linux-next v5 1/5] mtd: spi-nor: notify (Q)SPI controller about protocol change
In reply to Jonas GorskiHi Jonas,
Le 26/08/2015 16:02, Jonas Gorski a ?crit :
Quoted 7 lines
Quoted 4 lines · Cyrille Pitchen
Once the Quad SPI mode has been enabled on a Micron flash memory, this device expects ALL the following commands to use the SPI 4-4-4 protocol. The (Q)SPI controller needs to be notified about the protocol change so it can adapt and keep on dialoging with the Micron memory.
Quoted 8 lines
Doesn't that mean you need to disable quad mode on removal? Else the following will break/fail:
insmod atmel-quadspi.ko ~> spi-nor attaches -> sees micron -> enables quad mode rmmod atmel-quadspi.ko ~> spi-nor detaches
insmod atmel-quadspi.ko ~> spi-nor attaches -> fails to read the id because flash is still in quad mode.
Indeed you're right such an issue does exist. So as you said one solution could be create a new function to "clean" what have been done by spi_nor_scan() then call it from remove() function of each driver calling spi_nor_scan() from its probe().
However we could also enhance the probing of the memory. It is true that Micron spi nor memories only accept SPI 4-4-4 commands once switched in quad mode but actually they also provide a new command for this purpose:
"Multiple I/O Read ID" (0xAF).
Hence we could first try to probe using the regular Read ID (0x9F) command then change the protocol, for instance to SPI 4-4-4, and try the 0xAF command. I don't think all combinations for command/protocol need to be tested: for Micron memories, their datasheets claim the 0x9F command is only supported in "extended spi mode" so for the SPI 1-1-1 protocol whereas the 0xAF command only works in dual or quad modes.
On the other hand Spansion memories still reply to the regular 0x9F command in SPI 1-1-1 protocol even after their quad mode had been enabled. For other manufacturers, well... I don't know!
Some of the advantages of the probe solution as compared to the remove one are:
- we don't need to patch all drivers using spi_nor_scan() to call a new function from their remove().
- it doesn't rely on the assumption that the spi-nor memory starts in SPI 1-1-1 protocol. As a matter of fact the remove() won't be called for built-in modules or in many (all ?) cases of reset. Moreover some bootloaders may have already enabled the quad mode before starting the Linux kernel. This is what the sama5d2 romcode does when it is configured to boot from a QSPI memory.
Anyway you're right and the issue need to be addressed but maybe in another dedicated patch ?
Quoted 43 lines
Quoted 20 lines
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Marek Vasut <[email protected]>Acked-by Bean Huo <[email protected]>--- drivers/mtd/spi-nor/spi-nor.c | 21 +++++++++++++++++++++ include/linux/mtd/spi-nor.h | 13 +++++++++++++ 2 files changed, 34 insertions(+)drivers/mtd/spi-nor/spi-nor.c +4
-0@@ -165,6 +165,22 @@ static inline int write_disable(struct spi_nor *nor)return nor->write_reg (nor,SPINOR_OP_WRDI ,NULL ,0 ,0 ); } +/* +* Let the spi-nor framework notify lower layers, especially the driver of the +* (Q)SPI controller, about the new protocol to be used. Indeed, once the +* spi-nor framework has sent manufacturer specific commands to a memory to drivers/mtd/spi-nor/spi-nor.c static inline int write_disable(struct spi_nor *nor) · line 181
Quoted 10 lines
drivers/mtd/spi-nor/spi-nor.c +10
-0+* enable its Quad SPI mode, it should immediately after tell the QSPI +* controller to use the very same Quad SPI protocol as expected by the memory. +*/ +static inline int spi_nor_set_protocol (struct spi_nor *nor, +enum spi_protocol proto) +{ +if (nor->set_protocol) +return nor->set_protocol (nor, proto); + +return 0 ;Shouldn't the default assumption be that it won't support it? Also it might make sense to first check if it's supported before enabling it in the chip, so that we don't enable something just to then find out we can't back out of it.
I also wonder if we need an extra flag for that as at least SPI has RX_{DUAL,QUAD} and TX_{DUAL,QUAD} separated, so in theory there could be a controller that supports quad read, but not quad write, so we shouldn't be using the quad mode in that case. m25p80 currently sets SPI_NOR_{DUAL,QUAD} only based on SPI_RX_{DUAL,QUAD}, so that would then fail.
Quoted 6 lines
At least the n25q32 seems to support the "boring" 1_1_2 and 1_1_4 commands, so these should work in case the spi controller does not support quad tx, but quad rx.
So maybe an additional flag for the "full" dual/quad modes would be in order.
My first thought was to return -ENOSYS when the set_protocol() callback is not implemented but logically none of the already existing drivers implements it. So I made this new callback optional. This way, micron_quad_enable() works the exact same way as before for all the existing drivers so no regression or side- effect should be introduced by this patch.
Besides, the purpose of this callback is to notify the SPI controller that the protocol change has been done at the memory side but not to decide whether such a change is possible. Indeed, the capabilities of both the controller and the memory are checked before calling set_quad_mode() so the decision has already been taken when micron_quad_enable() is eventually called. Once the Quad Mode bit set in the Enhanced Volatile Configuration Register of Micron memory, it's too late: the memory now expects commands in SPI 4-4-4 protocol whether or not the controller supports this protocol.
So more accurate checks should be done before calling set_quad_mode(). Maybe the range of values for the mode parameter of spi_nor_scan() is too small. SPI_NOR_QUAD doesn't allow to make the difference between SPI protocols 1-1-4, 1-4-4 or 4-4-4. Knowing the exact set of protocols supported by both the controller and the memory could help making the right decision and choosing the best suited Fast Read command.
The RX_{DUAL_QUAD} and TX_{DUAL,QUAD} flags from the spi->mode in the m25p80 driver provide more information than the limited "mode" argument of spi_nor_scan().
Quoted 6 lines
Last but not least, the protocol probably should be stored somewhere in the nor struct, so that users don't have to store it themselves (I assume they will need to check it for each command again to know if the cmd/address should be send in serial or quad mode).
I agree with you, this could be an improvement for some spi controller drivers :)
Quoted 2 lines
Jonas
thanks for your review!
Best regards,
Cyrille
Re: [PATCH linux-next v5 1/5] mtd: spi-nor: notify (Q)SPI controller about protocol change
Hi,
Quoted 28 lines · Cyrille Pitchen
Hi Jonas,
Le 26/08/2015 16:02, Jonas Gorski a ?crit :
Quoted 7 lines
Quoted 4 lines · Cyrille Pitchen
Once the Quad SPI mode has been enabled on a Micron flash memory, this device expects ALL the following commands to use the SPI 4-4-4 protocol. The (Q)SPI controller needs to be notified about the protocol change so it can adapt and keep on dialoging with the Micron memory.
Quoted 8 lines
Doesn't that mean you need to disable quad mode on removal? Else the following will break/fail:
insmod atmel-quadspi.ko ~> spi-nor attaches -> sees micron -> enables quad mode rmmod atmel-quadspi.ko ~> spi-nor detaches
insmod atmel-quadspi.ko ~> spi-nor attaches -> fails to read the id because flash is still in quad mode.Indeed you're right such an issue does exist. So as you said one solution could be create a new function to "clean" what have been done by
spi_nor_scan()then call it fromremove()function of each driver callingspi_nor_scan()from itsprobe().
However we could also enhance the probing of the memory. It is true that Micron spi nor memories only accept SPI 4-4-4 commands once switched in quad mode but actually they also provide a new command for this purpose:
"Multiple I/O Read ID" (0xAF).
Hence we could first try to probe using the regular Read ID (0x9F) command then
Quoted 8 lines
change the protocol, for instance to SPI 4-4-4, and try the
0xAFcommand. I don't think all combinations for command/protocol need to be tested: for Micron memories, their datasheets claim the0x9Fcommand is only supported in "extended spi mode" so for the SPI 1-1-1 protocol whereas the0xAFcommand only works in dual or quad modes.
On the other hand Spansion memories still reply to the regular0x9Fcommand in SPI 1-1-1 protocol even after their quad mode had been enabled. For other manufacturers, well... I don't know!
At least from the two spansion datasheets I looked at, there is no 2_ or 4_ mode for spansion flashes, and the quad mode only enables the 1_x_4 commands, where as on micron the 1_x_4 commands are always available, and you only need to switch to DIO or QIO if you want to use 2_2_2 or 4_4_4.
The question is how to detect if the READID command failed. Hopefully it will result in an invalid command, and we get 0xff... (or 0x00...) for the the id. And from there on we can first test 4_4_4 MIORDID (so it won't a a full command on a DIO-enabled flash), and if that fails, 2_2_2. We need to tell the spi-nor controller to send these accordingly though. I wonder if it might not be better to pass the c_a_d as an additional parameter to the read_reg/write_reg etc call backs, because they might change depending on the command used. E.g. when using SPI_NOR_QUAD, we might want to make use of the quad page program command (which is 1_1_4), but there is no 1_1_2 version, so we cannot rely on a global "protocol" member for that, unless we want to do a set_protocol before every command.
Quoted 11 lines
Some of the advantages of the probe solution as compared to the remove one are:
- we don't need to patch all drivers usingspi_nor_scan()to call a new function from theirremove().
- it doesn't rely on the assumption that the spi-nor memory starts in SPI 1-1-1 protocol. As a matter of fact theremove()won't be called for built-in modules or in many (all ?) cases of reset. Moreover some bootloaders may have already enabled the quad mode before starting the Linux kernel. This is what the sama5d2 romcode does when it is configured to boot from a QSPI memory.
Quoted 2 lines
Anyway you're right and the issue need to be addressed but maybe in another dedicated patch ?
Yes, probably as a prerequisite to the quad mode on micron flashes. I think we should also add a separate flag for the DIO/QIO modes for micron flashes, as these are something separate from just the DOR/QOR commands usually implied by SPI_NOR_DUAL/QUAD (well they are the same, but work differently)
Quoted 47 lines
Quoted 47 lines
Quoted 20 lines
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Marek Vasut <[email protected]>Acked-by Bean Huo <[email protected]>--- drivers/mtd/spi-nor/spi-nor.c | 21 +++++++++++++++++++++ include/linux/mtd/spi-nor.h | 13 +++++++++++++ 2 files changed, 34 insertions(+)drivers/mtd/spi-nor/spi-nor.c +4
-0@@ -165,6 +165,22 @@ static inline int write_disable(struct spi_nor *nor)return nor->write_reg (nor,SPINOR_OP_WRDI ,NULL ,0 ,0 ); } +/* +* Let the spi-nor framework notify lower layers, especially the driver of the +* (Q)SPI controller, about the new protocol to be used. Indeed, once the +* spi-nor framework has sent manufacturer specific commands to a memory to drivers/mtd/spi-nor/spi-nor.c static inline int write_disable(struct spi_nor *nor) · line 181
Quoted 10 lines
drivers/mtd/spi-nor/spi-nor.c +10
-0+* enable its Quad SPI mode, it should immediately after tell the QSPI +* controller to use the very same Quad SPI protocol as expected by the memory. +*/ +static inline int spi_nor_set_protocol (struct spi_nor *nor, +enum spi_protocol proto) +{ +if (nor->set_protocol) +return nor->set_protocol (nor, proto); + +return 0 ;Shouldn't the default assumption be that it won't support it? Also it might make sense to first check if it's supported before enabling it in the chip, so that we don't enable something just to then find out we can't back out of it.
I also wonder if we need an extra flag for that as at least SPI has RX_{DUAL,QUAD} and TX_{DUAL,QUAD} separated, so in theory there could be a controller that supports quad read, but not quad write, so we shouldn't be using the quad mode in that case. m25p80 currently sets SPI_NOR_{DUAL,QUAD} only based on SPI_RX_{DUAL,QUAD}, so that would then fail.
At least the n25q32 seems to support the "boring" 1_1_2 and 1_1_4 commands, so these should work in case the spi controller does not support quad tx, but quad rx.
Quoted 8 lines
Quoted 2 lines
So maybe an additional flag for the "full" dual/quad modes would be in order.
My first thought was to return
-ENOSYSwhen theset_protocol()callback is not implemented but logically none of the already existing drivers implements it. So I made this new callback optional. This way,micron_quad_enable()works the exact same way as before for all the existing drivers so no regression or side- effect should be introduced by this patch.
If they don't implement the callback, then they won't know that they are supposed to use a different protocol, so change will definitely have broken them. Therefore an error return value and or a dev_warn would be IMHO better.
Quoted 9 lines
So more accurate checks should be done before calling
set_quad_mode(). Maybe the range of values for the mode parameter ofspi_nor_scan()is too small.SPI_NOR_QUADdoesn't allow to make the difference between SPI protocols 1-1-4, 1-4-4 or 4-4-4. Knowing the exact set of protocols supported by both the controller and the memory could help making the right decision and choosing the best suited Fast Read command.
The RX_{DUAL_QUAD} and TX_{DUAL,QUAD} flags from thespi->modein the m25p80 driver provide more information than the limited "mode" argument ofspi_nor_scan().
so from the controller side we need to know whether it can do RX_{DUAL,QUAD} and TX_{DUAL,QUAD}; but for the flash supported commands we have a bit more variety (the x_y_z you mentioned).
Funny enough, once we are on QIO on spansion, there is no difference anymore between FAST (0x0b), QOR (0x6b) and QIOR (0xeb), they all work in the 4_4_4 mode.
Quoted 8 lines
Quoted 6 lines
Last but not least, the protocol probably should be stored somewhere in the nor struct, so that users don't have to store it themselves (I assume they will need to check it for each command again to know if the cmd/address should be send in serial or quad mode).
I agree with you, this could be an improvement for some spi controller drivers :)
Quoted 8 lines
Quoted 2 lines
Jonas
thanks for your review!
Best regards,
Cyrille
Regards
Jonas
Re: [PATCH linux-next v5 1/5] mtd: spi-nor: notify (Q)SPI controller about protocol change
Hi Jonas,
taking your comments into account I'm about to test a new series with additional patches to handle the Read ID command in multiple I/O protocols and relying on new members in the struct spi_nor:
* @erase_proto: the SPI protocol used by erase operations
* @read_proto: the SPI protocol used by read operations
* @write_proto: the SPI protocol used by write operations
* @reg_proto the SPI protocol used by read_reg/write_reg operations
enum spi_protocol erase_proto; enum spi_protocol read_proto; enum spi_protocol write_proto; enum spi_protocol reg_proto;
This way, the read(), write(), erase(), read_reg() and write_reg() hooks can check the relevant protocol member so the spi-nor framework doesn't need to call spi_nor_set_protocol() before any command.
Also the op codes for read, page program and erase commands will be tuned depending on the memory manufacturer and the selected SPI protocol.
I'm likely to publish this new series tomorrow after my tests on a Micron memory.
Best Regards,
Cyrille
Le 31/08/2015 21:22, Jonas Gorski a ?crit :
Quoted 180 lines
Hi,
Quoted 28 lines · Cyrille Pitchen
Hi Jonas,
Le 26/08/2015 16:02, Jonas Gorski a ?crit :
Quoted 7 lines
Quoted 4 lines · Cyrille Pitchen
Once the Quad SPI mode has been enabled on a Micron flash memory, this device expects ALL the following commands to use the SPI 4-4-4 protocol. The (Q)SPI controller needs to be notified about the protocol change so it can adapt and keep on dialoging with the Micron memory.
Quoted 8 lines
Doesn't that mean you need to disable quad mode on removal? Else the following will break/fail:
insmod atmel-quadspi.ko ~> spi-nor attaches -> sees micron -> enables quad mode rmmod atmel-quadspi.ko ~> spi-nor detaches
insmod atmel-quadspi.ko ~> spi-nor attaches -> fails to read the id because flash is still in quad mode.Indeed you're right such an issue does exist. So as you said one solution could be create a new function to "clean" what have been done by
spi_nor_scan()then call it fromremove()function of each driver callingspi_nor_scan()from itsprobe().
However we could also enhance the probing of the memory. It is true that Micron spi nor memories only accept SPI 4-4-4 commands once switched in quad mode but actually they also provide a new command for this purpose:
"Multiple I/O Read ID" (0xAF).
Hence we could first try to probe using the regular Read ID (0x9F) command thenQuoted 8 lines
change the protocol, for instance to SPI 4-4-4, and try the
0xAFcommand. I don't think all combinations for command/protocol need to be tested: for Micron memories, their datasheets claim the0x9Fcommand is only supported in "extended spi mode" so for the SPI 1-1-1 protocol whereas the0xAFcommand only works in dual or quad modes.
On the other hand Spansion memories still reply to the regular0x9Fcommand in SPI 1-1-1 protocol even after their quad mode had been enabled. For other manufacturers, well... I don't know!At least from the two spansion datasheets I looked at, there is no 2_ or 4_ mode for spansion flashes, and the quad mode only enables the 1_x_4 commands, where as on micron the 1_x_4 commands are always available, and you only need to switch to DIO or QIO if you want to use 2_2_2 or 4_4_4.
The question is how to detect if the READID command failed. Hopefully it will result in an invalid command, and we get
0xff... (or0x00...) for the the id. And from there on we can first test 4_4_4 MIORDID (so it won't a a full command on a DIO-enabled flash), and if that fails, 2_2_2. We need to tell the spi-nor controller to send these accordingly though. I wonder if it might not be better to pass thec_a_das an additional parameter to the read_reg/write_reg etc call backs, because they might change depending on the command used. E.g. when usingSPI_NOR_QUAD, we might want to make use of the quad page program command (which is 1_1_4), but there is no 1_1_2 version, so we cannot rely on a global "protocol" member for that, unless we want to do aset_protocolbefore every command.Quoted 11 lines
Some of the advantages of the probe solution as compared to the remove one are:
- we don't need to patch all drivers usingspi_nor_scan()to call a new function from theirremove().
- it doesn't rely on the assumption that the spi-nor memory starts in SPI 1-1-1 protocol. As a matter of fact theremove()won't be called for built-in modules or in many (all ?) cases of reset. Moreover some bootloaders may have already enabled the quad mode before starting the Linux kernel. This is what the sama5d2 romcode does when it is configured to boot from a QSPI memory.Quoted 2 lines
Anyway you're right and the issue need to be addressed but maybe in another dedicated patch ?
Yes, probably as a prerequisite to the quad mode on micron flashes. I think we should also add a separate flag for the DIO/QIO modes for micron flashes, as these are something separate from just the DOR/QOR commands usually implied by SPI_NOR_DUAL/QUAD (well they are the same, but work differently)
Quoted 47 lines
Quoted 47 lines
Quoted 20 lines
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Marek Vasut <[email protected]>Acked-by Bean Huo <[email protected]>--- drivers/mtd/spi-nor/spi-nor.c | 21 +++++++++++++++++++++ include/linux/mtd/spi-nor.h | 13 +++++++++++++ 2 files changed, 34 insertions(+)drivers/mtd/spi-nor/spi-nor.c +4
-0@@ -165,6 +165,22 @@ static inline int write_disable(struct spi_nor *nor)return nor->write_reg (nor,SPINOR_OP_WRDI ,NULL ,0 ,0 ); } +/* +* Let the spi-nor framework notify lower layers, especially the driver of the +* (Q)SPI controller, about the new protocol to be used. Indeed, once the +* spi-nor framework has sent manufacturer specific commands to a memory to drivers/mtd/spi-nor/spi-nor.c static inline int write_disable(struct spi_nor *nor) · line 181
Quoted 10 lines
drivers/mtd/spi-nor/spi-nor.c +10
-0+* enable its Quad SPI mode, it should immediately after tell the QSPI +* controller to use the very same Quad SPI protocol as expected by the memory. +*/ +static inline int spi_nor_set_protocol (struct spi_nor *nor, +enum spi_protocol proto) +{ +if (nor->set_protocol) +return nor->set_protocol (nor, proto); + +return 0 ;Shouldn't the default assumption be that it won't support it? Also it might make sense to first check if it's supported before enabling it in the chip, so that we don't enable something just to then find out we can't back out of it.
I also wonder if we need an extra flag for that as at least SPI has RX_{DUAL,QUAD} and TX_{DUAL,QUAD} separated, so in theory there could be a controller that supports quad read, but not quad write, so we shouldn't be using the quad mode in that case. m25p80 currently sets SPI_NOR_{DUAL,QUAD} only based on SPI_RX_{DUAL,QUAD}, so that would then fail.
At least the n25q32 seems to support the "boring" 1_1_2 and 1_1_4 commands, so these should work in case the spi controller does not support quad tx, but quad rx.
Quoted 8 lines
Quoted 2 lines
So maybe an additional flag for the "full" dual/quad modes would be in order.
My first thought was to return
-ENOSYSwhen theset_protocol()callback is not implemented but logically none of the already existing drivers implements it. So I made this new callback optional. This way,micron_quad_enable()works the exact same way as before for all the existing drivers so no regression or side- effect should be introduced by this patch.If they don't implement the callback, then they won't know that they are supposed to use a different protocol, so change will definitely have broken them. Therefore an error return value and or a
dev_warnwould be IMHO better.Quoted 9 lines
So more accurate checks should be done before calling
set_quad_mode(). Maybe the range of values for the mode parameter ofspi_nor_scan()is too small.SPI_NOR_QUADdoesn't allow to make the difference between SPI protocols 1-1-4, 1-4-4 or 4-4-4. Knowing the exact set of protocols supported by both the controller and the memory could help making the right decision and choosing the best suited Fast Read command.
The RX_{DUAL_QUAD} and TX_{DUAL,QUAD} flags from thespi->modein the m25p80 driver provide more information than the limited "mode" argument ofspi_nor_scan().so from the controller side we need to know whether it can do RX_{DUAL,QUAD} and TX_{DUAL,QUAD}; but for the flash supported commands we have a bit more variety (the
x_y_zyou mentioned).Funny enough, once we are on QIO on spansion, there is no difference anymore between FAST (
0x0b), QOR (0x6b) and QIOR (0xeb), they all work in the 4_4_4 mode.Quoted 8 lines
Quoted 6 lines
Last but not least, the protocol probably should be stored somewhere in the nor struct, so that users don't have to store it themselves (I assume they will need to check it for each command again to know if the cmd/address should be send in serial or quad mode).
I agree with you, this could be an improvement for some spi controller drivers :)
Quoted 8 lines
Quoted 2 lines
Jonas
thanks for your review!
Best regards,
Cyrille
Regards
Jonas
Re: [PATCH linux-next v5 1/5] mtd: spi-nor: notify (Q)SPI controller about protocol change
Quoted 20 lines · Cyrille Pitchen
Hi Jonas,
taking your comments into account I'm about to test a new series with additional patches to handle the Read ID command in multiple I/O protocols and relying on new members in the struct
spi_nor:* @erase_proto: the SPI protocol used by erase operations
* @read_proto: the SPI protocol used by read operations
* @write_proto: the SPI protocol used by write operations
* @reg_proto the SPI protocol used by read_reg/write_reg operationsenum spi_protocol erase_proto; enum spi_protocol read_proto; enum spi_protocol write_proto; enum spi_protocol reg_proto;This way, the
read(),write(),erase(),read_reg()andwrite_reg()hooks can check the relevant protocol member so the spi-nor framework doesn't need to callspi_nor_set_protocol()before any command.
Quoted 5 lines
Also the op codes for read, page program and erase commands will be tuned depending on the memory manufacturer and the selected SPI protocol.
I'm likely to publish this new series tomorrow after my tests on a Micron memory.
Excellent, please keep me on Cc as I'm already warming up my Spansion part ;-)
Best regards,
Marek Vasut
Re: [PATCH linux-next v5 4/5] Documentation: atmel-quadspi: add binding file for Atmel QSPI driver
In reply to Cyrille PitchenQuoted 6 lines · Cyrille Pitchen
This patch documents the DT bindings for the driver of the Atmel QSPI controller embedded inside sama5d2x SoCs.
Signed-off-by Cyrille Pitchen <[email protected]>Acked-by Nicolas Ferre <[email protected]>Acked-by Marek Vasut <[email protected]>
Quoted 41 lines
--- .../devicetree/bindings/mtd/atmel-quadspi.txt | 29 ++++++++++++++++++++++ 1 file changed, 29 insertions(+) create mode 100644 Documentation/devicetree/bindings/mtd/atmel-quadspi.txtDocumentation/devicetree/bindings/mtd/atmel-quadspi.txt new file +29
-0@@ -0,0 +1,29 @@ +* Atmel Quad Serial Peripheral Interface (QSPI) + +Required properties: +- compatible: should be "atmel,sama5d2-qspi" +- reg: the first contains the register location and length, + the second contains the memory mapping address and length +- interrupts: should contain the interrupt for the device +- clocks: the phandle of the clock needed by the QSPI controller +- #address-cells: should be <1> +- #size-cells: should be <0> + +Example: + +spi at f0020000 { + compatible = "atmel,sama5d2-qspi"; + reg = <0xf0020000 0x100>, + <0xd0000000 0x8000000>; + interrupts = <52 IRQ_TYPE_LEVEL_HIGH 7>; + clocks = <&spi0_clk>; + #address-cells = <1>; + #size-cells = <0>; + pinctrl-names = "default"; + pinctrl-0 = <&pinctrl_spi0_default>; + status = "okay"; + + m25p80 at 0 { + ... + }; +};