GSI DMA provides specific TREs(Transfer ring element) namely Lock and
Unlock TRE. It provides mutually exclusive access to I2C controller from
any of the processor(Apps,ADSP). Lock prevents other subsystems from
concurrently performing DMA transfers and avoids disturbance to data path.
Basically for shared I2C usecase, lock the SE(Serial Engine) for one of
the processor, complete the transfer, unlock the SE.
Apply Lock TRE for the first transfer of shared SE and Apply Unlock
TRE for the last transfer.
Also change MAX_TRE macro to 5 from 3 because of the two additional TREs.
Signed-off-by: Mukesh Kumar Savaliya <quic_msavaliy@quicinc.com>
---
drivers/dma/qcom/gpi.c | 37 +++++++++++++++++++++++++++++++-
include/linux/dma/qcom-gpi-dma.h | 6 ++++++
2 files changed, 42 insertions(+), 1 deletion(-)
diff --git a/drivers/dma/qcom/gpi.c b/drivers/dma/qcom/gpi.c
index 52a7c8f2498f..c9e71c576680 100644
--- a/drivers/dma/qcom/gpi.c
+++ b/drivers/dma/qcom/gpi.c
@@ -2,6 +2,7 @@
/*
* Copyright (c) 2017-2020, The Linux Foundation. All rights reserved.
* Copyright (c) 2020, Linaro Limited
+ * Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved.
*/
#include <dt-bindings/dma/qcom-gpi.h>
@@ -65,6 +66,14 @@
/* DMA TRE */
#define TRE_DMA_LEN GENMASK(23, 0)
+/* Lock TRE */
+#define TRE_LOCK BIT(0)
+#define TRE_MINOR_TYPE GENMASK(19, 16)
+#define TRE_MAJOR_TYPE GENMASK(23, 20)
+
+/* Unlock TRE */
+#define TRE_I2C_UNLOCK BIT(8)
+
/* Register offsets from gpi-top */
#define GPII_n_CH_k_CNTXT_0_OFFS(n, k) (0x20000 + (0x4000 * (n)) + (0x80 * (k)))
#define GPII_n_CH_k_CNTXT_0_EL_SIZE GENMASK(31, 24)
@@ -516,7 +525,7 @@ struct gpii {
bool ieob_set;
};
-#define MAX_TRE 3
+#define MAX_TRE 5
struct gpi_desc {
struct virt_dma_desc vd;
@@ -1637,6 +1646,19 @@ static int gpi_create_i2c_tre(struct gchan *chan, struct gpi_desc *desc,
struct gpi_tre *tre;
unsigned int i;
+ /* create lock tre for first tranfser */
+ if (i2c->shared_se && i2c->first_msg) {
+ tre = &desc->tre[tre_idx];
+ tre_idx++;
+
+ tre->dword[0] = 0;
+ tre->dword[1] = 0;
+ tre->dword[2] = 0;
+ tre->dword[3] = u32_encode_bits(1, TRE_LOCK);
+ tre->dword[3] |= u32_encode_bits(0, TRE_MINOR_TYPE);
+ tre->dword[3] |= u32_encode_bits(3, TRE_MAJOR_TYPE);
+ }
+
/* first create config tre if applicable */
if (i2c->set_config) {
tre = &desc->tre[tre_idx];
@@ -1695,6 +1717,19 @@ static int gpi_create_i2c_tre(struct gchan *chan, struct gpi_desc *desc,
tre->dword[3] |= u32_encode_bits(1, TRE_FLAGS_IEOT);
}
+ /* Unlock tre for last transfer */
+ if (i2c->shared_se && i2c->last_msg && i2c->op != I2C_READ) {
+ tre = &desc->tre[tre_idx];
+ tre_idx++;
+
+ tre->dword[0] = 0;
+ tre->dword[1] = 0;
+ tre->dword[2] = 0;
+ tre->dword[3] = u32_encode_bits(1, TRE_I2C_UNLOCK);
+ tre->dword[3] |= u32_encode_bits(1, TRE_MINOR_TYPE);
+ tre->dword[3] |= u32_encode_bits(3, TRE_MAJOR_TYPE);
+ }
+
for (i = 0; i < tre_idx; i++)
dev_dbg(dev, "TRE:%d %x:%x:%x:%x\n", i, desc->tre[i].dword[0],
desc->tre[i].dword[1], desc->tre[i].dword[2], desc->tre[i].dword[3]);
diff --git a/include/linux/dma/qcom-gpi-dma.h b/include/linux/dma/qcom-gpi-dma.h
index 6680dd1a43c6..8589c711afae 100644
--- a/include/linux/dma/qcom-gpi-dma.h
+++ b/include/linux/dma/qcom-gpi-dma.h
@@ -65,6 +65,9 @@ enum i2c_op {
* @rx_len: receive length for buffer
* @op: i2c cmd
* @muli-msg: is part of multi i2c r-w msgs
+ * @shared_se: bus is shared between subsystems
+ * @bool first_msg: use it for tracking multimessage xfer
+ * @bool last_msg: use it for tracking multimessage xfer
*/
struct gpi_i2c_config {
u8 set_config;
@@ -78,6 +81,9 @@ struct gpi_i2c_config {
u32 rx_len;
enum i2c_op op;
bool multi_msg;
+ bool shared_se;
+ bool first_msg;
+ bool last_msg;
};
#endif /* QCOM_GPI_DMA_H */
--
2.25.1On 13.11.2024 5:14 PM, Mukesh Kumar Savaliya wrote:
> GSI DMA provides specific TREs(Transfer ring element) namely Lock and
> Unlock TRE. It provides mutually exclusive access to I2C controller from
> any of the processor(Apps,ADSP). Lock prevents other subsystems from
> concurrently performing DMA transfers and avoids disturbance to data path.
> Basically for shared I2C usecase, lock the SE(Serial Engine) for one of
> the processor, complete the transfer, unlock the SE.
>
> Apply Lock TRE for the first transfer of shared SE and Apply Unlock
> TRE for the last transfer.
>
> Also change MAX_TRE macro to 5 from 3 because of the two additional TREs.
>
> Signed-off-by: Mukesh Kumar Savaliya <quic_msavaliy@quicinc.com>
> ---
> drivers/dma/qcom/gpi.c | 37 +++++++++++++++++++++++++++++++-
> include/linux/dma/qcom-gpi-dma.h | 6 ++++++
> 2 files changed, 42 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/dma/qcom/gpi.c b/drivers/dma/qcom/gpi.c
> index 52a7c8f2498f..c9e71c576680 100644
> --- a/drivers/dma/qcom/gpi.c
> +++ b/drivers/dma/qcom/gpi.c
> @@ -2,6 +2,7 @@
> /*
> * Copyright (c) 2017-2020, The Linux Foundation. All rights reserved.
> * Copyright (c) 2020, Linaro Limited
> + * Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved.
> */
>
> #include <dt-bindings/dma/qcom-gpi.h>
> @@ -65,6 +66,14 @@
> /* DMA TRE */
> #define TRE_DMA_LEN GENMASK(23, 0)
>
> +/* Lock TRE */
> +#define TRE_LOCK BIT(0)
> +#define TRE_MINOR_TYPE GENMASK(19, 16)
> +#define TRE_MAJOR_TYPE GENMASK(23, 20)
> +
> +/* Unlock TRE */
> +#define TRE_I2C_UNLOCK BIT(8)
So the lock is generic.. I'd then expect the unlock to be generic, too?
> +
> /* Register offsets from gpi-top */
> #define GPII_n_CH_k_CNTXT_0_OFFS(n, k) (0x20000 + (0x4000 * (n)) + (0x80 * (k)))
> #define GPII_n_CH_k_CNTXT_0_EL_SIZE GENMASK(31, 24)
> @@ -516,7 +525,7 @@ struct gpii {
> bool ieob_set;
> };
>
> -#define MAX_TRE 3
> +#define MAX_TRE 5
>
> struct gpi_desc {
> struct virt_dma_desc vd;
> @@ -1637,6 +1646,19 @@ static int gpi_create_i2c_tre(struct gchan *chan, struct gpi_desc *desc,
> struct gpi_tre *tre;
> unsigned int i;
>
> + /* create lock tre for first tranfser */
> + if (i2c->shared_se && i2c->first_msg) {
Does the first/last logic handle errors well? i.e. what if we
have >= 3 transfers and:
1) the first transfer succeeds but the last doesn't
2) the first transfer succeeds, the second one doesn't and the lock
is submitted again
3) the unlock never suceeds
Konrad
Thanks Konrad for the review !
On 11/16/2024 12:53 AM, Konrad Dybcio wrote:
> On 13.11.2024 5:14 PM, Mukesh Kumar Savaliya wrote:
>> GSI DMA provides specific TREs(Transfer ring element) namely Lock and
>> Unlock TRE. It provides mutually exclusive access to I2C controller from
>> any of the processor(Apps,ADSP). Lock prevents other subsystems from
>> concurrently performing DMA transfers and avoids disturbance to data path.
>> Basically for shared I2C usecase, lock the SE(Serial Engine) for one of
>> the processor, complete the transfer, unlock the SE.
>>
>> Apply Lock TRE for the first transfer of shared SE and Apply Unlock
>> TRE for the last transfer.
>>
>> Also change MAX_TRE macro to 5 from 3 because of the two additional TREs.
>>
>> Signed-off-by: Mukesh Kumar Savaliya <quic_msavaliy@quicinc.com>
>> ---
>> drivers/dma/qcom/gpi.c | 37 +++++++++++++++++++++++++++++++-
>> include/linux/dma/qcom-gpi-dma.h | 6 ++++++
>> 2 files changed, 42 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/dma/qcom/gpi.c b/drivers/dma/qcom/gpi.c
>> index 52a7c8f2498f..c9e71c576680 100644
>> --- a/drivers/dma/qcom/gpi.c
>> +++ b/drivers/dma/qcom/gpi.c
>> @@ -2,6 +2,7 @@
>> /*
>> * Copyright (c) 2017-2020, The Linux Foundation. All rights reserved.
>> * Copyright (c) 2020, Linaro Limited
>> + * Copyright (c) 2024 Qualcomm Innovation Center, Inc. All rights reserved.
>> */
>>
>> #include <dt-bindings/dma/qcom-gpi.h>
>> @@ -65,6 +66,14 @@
>> /* DMA TRE */
>> #define TRE_DMA_LEN GENMASK(23, 0)
>>
>> +/* Lock TRE */
>> +#define TRE_LOCK BIT(0)
>> +#define TRE_MINOR_TYPE GENMASK(19, 16)
>> +#define TRE_MAJOR_TYPE GENMASK(23, 20)
>> +
>> +/* Unlock TRE */
>> +#define TRE_I2C_UNLOCK BIT(8)
>
> So the lock is generic.. I'd then expect the unlock to be generic, too?
Absolutely, renamed it for generic as TRE_UNLOCK.
>
>> +
>> /* Register offsets from gpi-top */
>> #define GPII_n_CH_k_CNTXT_0_OFFS(n, k) (0x20000 + (0x4000 * (n)) + (0x80 * (k)))
>> #define GPII_n_CH_k_CNTXT_0_EL_SIZE GENMASK(31, 24)
>> @@ -516,7 +525,7 @@ struct gpii {
>> bool ieob_set;
>> };
>>
>> -#define MAX_TRE 3
>> +#define MAX_TRE 5
>>
>> struct gpi_desc {
>> struct virt_dma_desc vd;
>> @@ -1637,6 +1646,19 @@ static int gpi_create_i2c_tre(struct gchan *chan, struct gpi_desc *desc,
>> struct gpi_tre *tre;
>> unsigned int i;
>>
>> + /* create lock tre for first tranfser */
>> + if (i2c->shared_se && i2c->first_msg) {
>
> Does the first/last logic handle errors well? i.e. what if we
> have >= 3 transfers and:
>
> 1) the first transfer succeeds but the last doesn't
> 2) the first transfer succeeds, the second one doesn't and the lock
> is submitted again
> 3) the unlock never suceeds
>
geni_i2c_gpi_xfer() takes care of any of the error. Upon error, it does
dma_engine_terminate_sync() which resets all the pipes. Internal
downstream also has same implementation.
> Konrad
On 18.11.2024 6:46 AM, Mukesh Kumar Savaliya wrote:
> Thanks Konrad for the review !
>
> On 11/16/2024 12:53 AM, Konrad Dybcio wrote:
>> On 13.11.2024 5:14 PM, Mukesh Kumar Savaliya wrote:
>>> GSI DMA provides specific TREs(Transfer ring element) namely Lock and
>>> Unlock TRE. It provides mutually exclusive access to I2C controller from
>>> any of the processor(Apps,ADSP). Lock prevents other subsystems from
>>> concurrently performing DMA transfers and avoids disturbance to data path.
>>> Basically for shared I2C usecase, lock the SE(Serial Engine) for one of
>>> the processor, complete the transfer, unlock the SE.
>>>
>>> Apply Lock TRE for the first transfer of shared SE and Apply Unlock
>>> TRE for the last transfer.
>>>
>>> Also change MAX_TRE macro to 5 from 3 because of the two additional TREs.
>>>
>>> Signed-off-by: Mukesh Kumar Savaliya <quic_msavaliy@quicinc.com>
>>> ---
[...]
>>> + /* create lock tre for first tranfser */
>>> + if (i2c->shared_se && i2c->first_msg) {
>>
>> Does the first/last logic handle errors well? i.e. what if we
>> have >= 3 transfers and:
>>
>> 1) the first transfer succeeds but the last doesn't
>> 2) the first transfer succeeds, the second one doesn't and the lock
>> is submitted again
>> 3) the unlock never suceeds
>>
> geni_i2c_gpi_xfer() takes care of any of the error. Upon error, it does dma_engine_terminate_sync() which resets all the pipes. Internal downstream also has same implementation.
Okay, this sounds reassuring.
Since the TRE would be locked to APSS, I'm guessing we don't ever need
to worry about gpi_terminate_all() being executed halfway through a
non-APSS transaction?
Konrad
© 2016 - 2024 Red Hat, Inc.