Add support for GD25Q256, a 32MiB SPI Nor
flash from Gigadevice.
Signed-off-by: Andy Yan <[email protected]>
---
Changes in v4:
- add SPI_NOR_HAS_LOCK and SPI_NOR_HAS_TB
Changes in v3:
- rebase on top of spi-nor tree
- add SPI_NOR_4B_OPCODES flag
Changes in v2:
- drop one line unnecessary modification
drivers/mtd/spi-nor/spi-nor.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
index 196b52f..e4145cd 100644
--- a/drivers/mtd/spi-nor/spi-nor.c
+++ b/drivers/mtd/spi-nor/spi-nor.c
@@ -986,6 +986,11 @@ static const struct flash_info spi_nor_ids[] = {
SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
},
+ {
+ "gd25q256", INFO(0xc84019, 0, 64 * 1024, 512,
+ SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
+ SPI_NOR_4B_OPCODES | SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
+ },
/* Intel/Numonyx -- xxxs33b */
{ "160s33b", INFO(0x898911, 0, 64 * 1024, 32, 0) },
@@ -2365,6 +2370,7 @@ static int spi_nor_init_params(struct spi_nor *nor,
SNOR_HWCAPS_PP_QUAD)) {
switch (JEDEC_MFR(info)) {
case SNOR_MFR_MACRONIX:
+ case SNOR_MFR_GIGADEVICE:
params->quad_enable = macronix_quad_enable;
break;
--
2.7.4
Hi Andy,
Le 25/07/2017 à 12:12, Andy Yan a écrit :
> Add support for GD25Q256, a 32MiB SPI Nor
> flash from Gigadevice.
>
> Signed-off-by: Andy Yan <[email protected]>
>
> ---
>
> Changes in v4:
> - add SPI_NOR_HAS_LOCK and SPI_NOR_HAS_TB
Between v3 and v4, I see that you've also changed the procedure to the
the Quad Enable bit on all Gigadevice memories with QSPI capabilities.
This is not a detail and should have been reported here.
>
> Changes in v3:
> - rebase on top of spi-nor tree
> - add SPI_NOR_4B_OPCODES flag
>
> Changes in v2:
> - drop one line unnecessary modification
>
> drivers/mtd/spi-nor/spi-nor.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
> index 196b52f..e4145cd 100644
> --- a/drivers/mtd/spi-nor/spi-nor.c
> +++ b/drivers/mtd/spi-nor/spi-nor.c
> @@ -986,6 +986,11 @@ static const struct flash_info spi_nor_ids[] = {
> SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
> SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
> },
> + {
> + "gd25q256", INFO(0xc84019, 0, 64 * 1024, 512,
> + SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
> + SPI_NOR_4B_OPCODES | SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
> + },
>
> /* Intel/Numonyx -- xxxs33b */
> { "160s33b", INFO(0x898911, 0, 64 * 1024, 32, 0) },
> @@ -2365,6 +2370,7 @@ static int spi_nor_init_params(struct spi_nor *nor,
> SNOR_HWCAPS_PP_QUAD)) {
> switch (JEDEC_MFR(info)) {
> case SNOR_MFR_MACRONIX:
> + case SNOR_MFR_GIGADEVICE:
> params->quad_enable = macronix_quad_enable;
Here, you've have changed the Quad Enable requirement for *all*
Gigadevice memories with Quad SPI capabilities.
However, I'm reading the GD25Q128 datasheet and it claims that the QE
bit is BIT(1) of the Status Register 2. Hence some
spansion*_quad_enable() should be used, as before your patch.
Then, still according to the datasheet, the GD25Q128 memory is compliant
with the JESD216 specification (minor 0) but neither with rev A (minor
5) nor rev B (minor 6).
So its Basic Flash Parameter Table is limited to 9 DWORDs instead of 16
DWORDs, hence doesn't provide the Quad Enable requirements. It means
that the SFDP tables would not help to select the right _quad_enable()
function by overriding the choice made by the switch() statement above.
tl;dr
This chunk would introduce a regression with some already supported
Gigadevice memories. So I reject this patch, sorry.
Best regards,
Cyrille
> break;
>
>
Hi Cyrille:
On 2017年08月16日 00:04, Cyrille Pitchen wrote:
> Hi Andy,
>
> Le 25/07/2017 à 12:12, Andy Yan a écrit :
>> Add support for GD25Q256, a 32MiB SPI Nor
>> flash from Gigadevice.
>>
>> Signed-off-by: Andy Yan <[email protected]>
>>
>> ---
>>
>> Changes in v4:
>> - add SPI_NOR_HAS_LOCK and SPI_NOR_HAS_TB
> Between v3 and v4, I see that you've also changed the procedure to the
> the Quad Enable bit on all Gigadevice memories with QSPI capabilities.
> This is not a detail and should have been reported here.
Sorry, I will keep this in mind.
>
>> Changes in v3:
>> - rebase on top of spi-nor tree
>> - add SPI_NOR_4B_OPCODES flag
>>
>> Changes in v2:
>> - drop one line unnecessary modification
>>
>> drivers/mtd/spi-nor/spi-nor.c | 6 ++++++
>> 1 file changed, 6 insertions(+)
>>
>> diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
>> index 196b52f..e4145cd 100644
>> --- a/drivers/mtd/spi-nor/spi-nor.c
>> +++ b/drivers/mtd/spi-nor/spi-nor.c
>> @@ -986,6 +986,11 @@ static const struct flash_info spi_nor_ids[] = {
>> SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>> SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>> },
>> + {
>> + "gd25q256", INFO(0xc84019, 0, 64 * 1024, 512,
>> + SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>> + SPI_NOR_4B_OPCODES | SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>> + },
>>
>> /* Intel/Numonyx -- xxxs33b */
>> { "160s33b", INFO(0x898911, 0, 64 * 1024, 32, 0) },
>> @@ -2365,6 +2370,7 @@ static int spi_nor_init_params(struct spi_nor *nor,
>> SNOR_HWCAPS_PP_QUAD)) {
>> switch (JEDEC_MFR(info)) {
>> case SNOR_MFR_MACRONIX:
>> + case SNOR_MFR_GIGADEVICE:
>> params->quad_enable = macronix_quad_enable;
> Here, you've have changed the Quad Enable requirement for *all*
> Gigadevice memories with Quad SPI capabilities.
>
> However, I'm reading the GD25Q128 datasheet and it claims that the QE
> bit is BIT(1) of the Status Register 2. Hence some
> spansion*_quad_enable() should be used, as before your patch.
>
> Then, still according to the datasheet, the GD25Q128 memory is compliant
> with the JESD216 specification (minor 0) but neither with rev A (minor
> 5) nor rev B (minor 6).
> So its Basic Flash Parameter Table is limited to 9 DWORDs instead of 16
> DWORDs, hence doesn't provide the Quad Enable requirements. It means
> that the SFDP tables would not help to select the right _quad_enable()
> function by overriding the choice made by the switch() statement above.
>
> tl;dr
> This chunk would introduce a regression with some already supported
> Gigadevice memories. So I reject this patch, sorry.
After check some other Gigadevice memories, I found it's true as
you mentioned.
Some memories use S9 as the QE bit, but some use S6.
Do you have some ideas for this case? Add a check for the full
jedec_id or encode
the QE bit in the flash_info?
I am a new bee in the flash failed, very appreciate for your advice.
>
> Best regards,
>
> Cyrille
>
>> break;
>>
>>
>
>
>
Hi Andy,
Le 16/08/2017 à 05:40, Andy Yan a écrit :
> Hi Cyrille:
>
>
> On 2017年08月16日 00:04, Cyrille Pitchen wrote:
>> Hi Andy,
>>
>> Le 25/07/2017 à 12:12, Andy Yan a écrit :
>>> Add support for GD25Q256, a 32MiB SPI Nor
>>> flash from Gigadevice.
>>>
>>> Signed-off-by: Andy Yan <[email protected]>
>>>
>>> ---
>>>
>>> Changes in v4:
>>> - add SPI_NOR_HAS_LOCK and SPI_NOR_HAS_TB
>> Between v3 and v4, I see that you've also changed the procedure to the
>> the Quad Enable bit on all Gigadevice memories with QSPI capabilities.
>> This is not a detail and should have been reported here.
>
> Sorry, I will keep this in mind.
>>
>>> Changes in v3:
>>> - rebase on top of spi-nor tree
>>> - add SPI_NOR_4B_OPCODES flag
>>>
>>> Changes in v2:
>>> - drop one line unnecessary modification
>>>
>>> drivers/mtd/spi-nor/spi-nor.c | 6 ++++++
>>> 1 file changed, 6 insertions(+)
>>>
>>> diff --git a/drivers/mtd/spi-nor/spi-nor.c
>>> b/drivers/mtd/spi-nor/spi-nor.c
>>> index 196b52f..e4145cd 100644
>>> --- a/drivers/mtd/spi-nor/spi-nor.c
>>> +++ b/drivers/mtd/spi-nor/spi-nor.c
>>> @@ -986,6 +986,11 @@ static const struct flash_info spi_nor_ids[] = {
>>> SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>>> SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>>> },
>>> + {
>>> + "gd25q256", INFO(0xc84019, 0, 64 * 1024, 512,
>>> + SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>>> + SPI_NOR_4B_OPCODES | SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>>> + },
>>> /* Intel/Numonyx -- xxxs33b */
>>> { "160s33b", INFO(0x898911, 0, 64 * 1024, 32, 0) },
>>> @@ -2365,6 +2370,7 @@ static int spi_nor_init_params(struct spi_nor
>>> *nor,
>>> SNOR_HWCAPS_PP_QUAD)) {
>>> switch (JEDEC_MFR(info)) {
>>> case SNOR_MFR_MACRONIX:
>>> + case SNOR_MFR_GIGADEVICE:
>>> params->quad_enable = macronix_quad_enable;
>> Here, you've have changed the Quad Enable requirement for *all*
>> Gigadevice memories with Quad SPI capabilities.
>>
>> However, I'm reading the GD25Q128 datasheet and it claims that the QE
>> bit is BIT(1) of the Status Register 2. Hence some
>> spansion*_quad_enable() should be used, as before your patch.
>>
>> Then, still according to the datasheet, the GD25Q128 memory is compliant
>> with the JESD216 specification (minor 0) but neither with rev A (minor
>> 5) nor rev B (minor 6).
>> So its Basic Flash Parameter Table is limited to 9 DWORDs instead of 16
>> DWORDs, hence doesn't provide the Quad Enable requirements. It means
>> that the SFDP tables would not help to select the right _quad_enable()
>> function by overriding the choice made by the switch() statement above.
>>
>> tl;dr
>> This chunk would introduce a regression with some already supported
>> Gigadevice memories. So I reject this patch, sorry.
>
> After check some other Gigadevice memories, I found it's true as you
> mentioned.
> Some memories use S9 as the QE bit, but some use S6.
> Do you have some ideas for this case? Add a check for the full
> jedec_id or encode
> the QE bit in the flash_info?
> I am a new bee in the flash failed, very appreciate for your advice.
Historically spi-nor.c assumed that the procedure to set the QE bit
could be chosen based on the manufacturer ID only, hence without needing
to check the whole JEDEC ID. Obviouly this is no longer true for
Gigadevice SPI NOR memories...
So you need a solution which is backward compatible with the existing
code so you patch would not introduce any regression for other SPI NOR
memories.
Then, I suggest you add a .quad_enable member in 'struct flash_info'.
#define USE_CLSR BIT(14) /* use CLSR command */
+
+ int (*quad_enable)(struct spi_nor *nor);
};
#define JEDEC_MFR(info) ((info)->id[0]
Also in spi_nor_init_params():
/* Select the procedure to set the Quad Enable bit. */
if (params->hwcaps.mask & (SNOR_HWCAPS_READ_QUAD |
SNOR_HWCAPS_PP_QUAD)) {
switch (JEDEC_MFR(info)) {
case SNOR_MFR_MACRONIX:
params->quad_enable = macronix_quad_enable;
break;
case SNOR_MFR_MICRON:
break;
default:
/* Kept only for backward compatibility purpose. */
params->quad_enable = spansion_quad_enable;
break;
}
}
+
+ if (info->quad_enable)
+ params->quad_enable = info->quad_enable;
/* Override the parameters with data read from SFDP tables. */
nor->addr_width = 0;
nor->mtd.erasesize = 0;
if ((info->flags & (SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ)) &&
!(info->flags & SPI_NOR_SKIP_SFDP)) {
struct spi_nor_flash_parameter sfdp_params;
memcpy(&sfdp_params, params, sizeof(sfdp_params));
if (spi_nor_parse_sfdp(nor, &sfdp_params)) {
nor->addr_width = 0;
nor->mtd.erasesize = 0;
} else {
memcpy(params, &sfdp_params, sizeof(*params));
}
}
And finally when adding the gigadevice entries in the spi_nor_ids[]
array, don't forget to initialize the new member:
+ {
+ "gd25q256", .quad_enable = macronix_quad_enable, + INFO([...])
+ },
>>
>> Best regards,
>>
>> Cyrille
>>
>>> break;
>>>
>>
>>
>>
>
>
>
Hi Cyrille:
On 2017年08月23日 15:46, Cyrille Pitchen wrote:
> Hi Andy,
>
> Le 16/08/2017 à 05:40, Andy Yan a écrit :
>> Hi Cyrille:
>>
>>
>> On 2017年08月16日 00:04, Cyrille Pitchen wrote:
>>> Hi Andy,
>>>
>>> Le 25/07/2017 à 12:12, Andy Yan a écrit :
>>>> Add support for GD25Q256, a 32MiB SPI Nor
>>>> flash from Gigadevice.
>>>>
>>>> Signed-off-by: Andy Yan <[email protected]>
>>>>
>>>> ---
>>>>
>>>> Changes in v4:
>>>> - add SPI_NOR_HAS_LOCK and SPI_NOR_HAS_TB
>>> Between v3 and v4, I see that you've also changed the procedure to the
>>> the Quad Enable bit on all Gigadevice memories with QSPI capabilities.
>>> This is not a detail and should have been reported here.
>> Sorry, I will keep this in mind.
>>>> Changes in v3:
>>>> - rebase on top of spi-nor tree
>>>> - add SPI_NOR_4B_OPCODES flag
>>>>
>>>> Changes in v2:
>>>> - drop one line unnecessary modification
>>>>
>>>> drivers/mtd/spi-nor/spi-nor.c | 6 ++++++
>>>> 1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/drivers/mtd/spi-nor/spi-nor.c
>>>> b/drivers/mtd/spi-nor/spi-nor.c
>>>> index 196b52f..e4145cd 100644
>>>> --- a/drivers/mtd/spi-nor/spi-nor.c
>>>> +++ b/drivers/mtd/spi-nor/spi-nor.c
>>>> @@ -986,6 +986,11 @@ static const struct flash_info spi_nor_ids[] = {
>>>> SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>>>> SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>>>> },
>>>> + {
>>>> + "gd25q256", INFO(0xc84019, 0, 64 * 1024, 512,
>>>> + SECT_4K | SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ |
>>>> + SPI_NOR_4B_OPCODES | SPI_NOR_HAS_LOCK | SPI_NOR_HAS_TB)
>>>> + },
>>>> /* Intel/Numonyx -- xxxs33b */
>>>> { "160s33b", INFO(0x898911, 0, 64 * 1024, 32, 0) },
>>>> @@ -2365,6 +2370,7 @@ static int spi_nor_init_params(struct spi_nor
>>>> *nor,
>>>> SNOR_HWCAPS_PP_QUAD)) {
>>>> switch (JEDEC_MFR(info)) {
>>>> case SNOR_MFR_MACRONIX:
>>>> + case SNOR_MFR_GIGADEVICE:
>>>> params->quad_enable = macronix_quad_enable;
>>> Here, you've have changed the Quad Enable requirement for *all*
>>> Gigadevice memories with Quad SPI capabilities.
>>>
>>> However, I'm reading the GD25Q128 datasheet and it claims that the QE
>>> bit is BIT(1) of the Status Register 2. Hence some
>>> spansion*_quad_enable() should be used, as before your patch.
>>>
>>> Then, still according to the datasheet, the GD25Q128 memory is compliant
>>> with the JESD216 specification (minor 0) but neither with rev A (minor
>>> 5) nor rev B (minor 6).
>>> So its Basic Flash Parameter Table is limited to 9 DWORDs instead of 16
>>> DWORDs, hence doesn't provide the Quad Enable requirements. It means
>>> that the SFDP tables would not help to select the right _quad_enable()
>>> function by overriding the choice made by the switch() statement above.
>>>
>>> tl;dr
>>> This chunk would introduce a regression with some already supported
>>> Gigadevice memories. So I reject this patch, sorry.
>> After check some other Gigadevice memories, I found it's true as you
>> mentioned.
>> Some memories use S9 as the QE bit, but some use S6.
>> Do you have some ideas for this case? Add a check for the full
>> jedec_id or encode
>> the QE bit in the flash_info?
>> I am a new bee in the flash failed, very appreciate for your advice.
> Historically spi-nor.c assumed that the procedure to set the QE bit
> could be chosen based on the manufacturer ID only, hence without needing
> to check the whole JEDEC ID. Obviouly this is no longer true for
> Gigadevice SPI NOR memories...
>
> So you need a solution which is backward compatible with the existing
> code so you patch would not introduce any regression for other SPI NOR
> memories.
>
> Then, I suggest you add a .quad_enable member in 'struct flash_info'.
>
> #define USE_CLSR BIT(14) /* use CLSR command */
> +
> + int (*quad_enable)(struct spi_nor *nor);
> };
>
> #define JEDEC_MFR(info) ((info)->id[0]
>
>
>
> Also in spi_nor_init_params():
>
> /* Select the procedure to set the Quad Enable bit. */
> if (params->hwcaps.mask & (SNOR_HWCAPS_READ_QUAD |
> SNOR_HWCAPS_PP_QUAD)) {
> switch (JEDEC_MFR(info)) {
> case SNOR_MFR_MACRONIX:
> params->quad_enable = macronix_quad_enable;
> break;
>
> case SNOR_MFR_MICRON:
> break;
>
> default:
> /* Kept only for backward compatibility purpose. */
> params->quad_enable = spansion_quad_enable;
> break;
> }
> }
> +
> + if (info->quad_enable)
> + params->quad_enable = info->quad_enable;
>
> /* Override the parameters with data read from SFDP tables. */
> nor->addr_width = 0;
> nor->mtd.erasesize = 0;
> if ((info->flags & (SPI_NOR_DUAL_READ | SPI_NOR_QUAD_READ)) &&
> !(info->flags & SPI_NOR_SKIP_SFDP)) {
> struct spi_nor_flash_parameter sfdp_params;
>
> memcpy(&sfdp_params, params, sizeof(sfdp_params));
> if (spi_nor_parse_sfdp(nor, &sfdp_params)) {
> nor->addr_width = 0;
> nor->mtd.erasesize = 0;
> } else {
> memcpy(params, &sfdp_params, sizeof(*params));
> }
> }
>
>
>
> And finally when adding the gigadevice entries in the spi_nor_ids[]
> array, don't forget to initialize the new member:
>
> + {
> + "gd25q256", .quad_enable = macronix_quad_enable, + INFO([...])
> + },
Thanks very much for your guidance. I will try a new patch very soon.
>>> Best regards,
>>>
>>> Cyrille
>>>
>>>> break;
>>>>
>>>
>>>
>>
>>
>
>
>