Currently, smc/hvc calls are made with parameters set
to zeros. We are using multiple scmi instances within
a VM. We are sharing the same smc-id(func_id) with all
scmi instance. The hypervisor needs a way to distinguish
among hvc calls made from different instances.
This patch series introduces new compatible string which
can be used to pass shmem channel address as parameters
to smc/hvc calls.
---
v5 -> avoid computing page and offset in send function
Link: https://lore.kernel.org/all/[email protected]/
v4 -> split shmem address into 4KB-pages and offset
v3 -> pass shmem channel address as parameter
v2 -> fix the compilation erros on 32bit platform(see below)
Reported-by: kernel test robot <[email protected]>
Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
v1 -> original patches
Nikunj Kela (2):
dt-bindings: firmware: arm,scmi: support for parameter in smc/hvc call
firmware: arm_scmi: Augment SMC/HVC to allow optional parameters
.../bindings/firmware/arm,scmi.yaml | 8 ++++-
drivers/firmware/arm_scmi/driver.c | 1 +
drivers/firmware/arm_scmi/smc.c | 30 ++++++++++++++++++-
3 files changed, 37 insertions(+), 2 deletions(-)
--
2.17.1
This change adds support for passing shmem channel address as parameters
in smc/hvc call. The address is split into 4KB-page and offset.
This is useful when multiple scmi instances are using same smc-id
and firmware needs to distinguish among the instances.
Signed-off-by: Nikunj Kela <[email protected]>
---
drivers/firmware/arm_scmi/driver.c | 1 +
drivers/firmware/arm_scmi/smc.c | 30 +++++++++++++++++++++++++++++-
2 files changed, 30 insertions(+), 1 deletion(-)
diff --git a/drivers/firmware/arm_scmi/driver.c b/drivers/firmware/arm_scmi/driver.c
index e7d97b59963b..b5957cc12fee 100644
--- a/drivers/firmware/arm_scmi/driver.c
+++ b/drivers/firmware/arm_scmi/driver.c
@@ -2914,6 +2914,7 @@ static const struct of_device_id scmi_of_match[] = {
#endif
#ifdef CONFIG_ARM_SCMI_TRANSPORT_SMC
{ .compatible = "arm,scmi-smc", .data = &scmi_smc_desc},
+ { .compatible = "arm,scmi-smc-param", .data = &scmi_smc_desc},
#endif
#ifdef CONFIG_ARM_SCMI_TRANSPORT_VIRTIO
{ .compatible = "arm,scmi-virtio", .data = &scmi_virtio_desc},
diff --git a/drivers/firmware/arm_scmi/smc.c b/drivers/firmware/arm_scmi/smc.c
index 93272e4bbd12..621c37efe3ec 100644
--- a/drivers/firmware/arm_scmi/smc.c
+++ b/drivers/firmware/arm_scmi/smc.c
@@ -20,6 +20,23 @@
#include "common.h"
+/*
+ * The shmem address is split into 4K page and offset.
+ * This is to make sure the parameters fit in 32bit arguments of the
+ * smc/hvc call to keep it uniform across smc32/smc64 conventions.
+ * This however limits the shmem address to 44 bit.
+ *
+ * These optional parameters can be used to distinguish among multiple
+ * scmi instances that are using the same smc-id.
+ * The page parameter is passed in r1/x1/w1 register and the offset parameter
+ * is passed in r2/x2/w2 register.
+ */
+
+#define SHMEM_SIZE (SZ_4K)
+#define SHMEM_SHIFT 12
+#define SHMEM_PAGE(x) (_UL((x) >> SHMEM_SHIFT))
+#define SHMEM_OFFSET(x) ((x) & (SHMEM_SIZE - 1))
+
/**
* struct scmi_smc - Structure representing a SCMI smc transport
*
@@ -30,6 +47,8 @@
* @inflight: Atomic flag to protect access to Tx/Rx shared memory area.
* Used when operating in atomic mode.
* @func_id: smc/hvc call function id
+ * @param_page: 4K page number of the shmem channel
+ * @param_offset: Offset within the 4K page of the shmem channel
*/
struct scmi_smc {
@@ -40,6 +59,8 @@ struct scmi_smc {
#define INFLIGHT_NONE MSG_TOKEN_MAX
atomic_t inflight;
u32 func_id;
+ u32 param_page;
+ u32 param_offset;
};
static irqreturn_t smc_msg_done_isr(int irq, void *data)
@@ -137,6 +158,10 @@ static int smc_chan_setup(struct scmi_chan_info *cinfo, struct device *dev,
if (ret < 0)
return ret;
+ if (of_device_is_compatible(dev->of_node, "arm,scmi-smc-param")) {
+ scmi_info->param_page = SHMEM_PAGE(res.start);
+ scmi_info->param_offset = SHMEM_OFFSET(res.start);
+ }
/*
* If there is an interrupt named "a2p", then the service and
* completion of a message is signaled by an interrupt rather than by
@@ -179,6 +204,8 @@ static int smc_send_message(struct scmi_chan_info *cinfo,
{
struct scmi_smc *scmi_info = cinfo->transport_info;
struct arm_smccc_res res;
+ unsigned long page = scmi_info->param_page;
+ unsigned long offset = scmi_info->param_offset;
/*
* Channel will be released only once response has been
@@ -188,7 +215,8 @@ static int smc_send_message(struct scmi_chan_info *cinfo,
shmem_tx_prepare(scmi_info->shmem, xfer, cinfo);
- arm_smccc_1_1_invoke(scmi_info->func_id, 0, 0, 0, 0, 0, 0, 0, &res);
+ arm_smccc_1_1_invoke(scmi_info->func_id, page, offset, 0, 0, 0, 0, 0,
+ &res);
/* Only SMCCC_RET_NOT_SUPPORTED is valid error code */
if (res.a0) {
--
2.17.1
Currently, smc/hvc calls are made with smc-id only. The parameters are
all set to zeros. This change defines a new compatible string that can
be used to pass shmem address(4KB-page, offset) as two parameters in
SMC/HVC doorbell.
This is useful when multiple scmi instances are used with common smc-id.
Signed-off-by: Nikunj Kela <[email protected]>
Reviewed-by: Rob Herring <[email protected]>
---
Documentation/devicetree/bindings/firmware/arm,scmi.yaml | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/Documentation/devicetree/bindings/firmware/arm,scmi.yaml b/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
index 5824c43e9893..ad776911f990 100644
--- a/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
+++ b/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
@@ -34,6 +34,10 @@ properties:
- description: SCMI compliant firmware with ARM SMC/HVC transport
items:
- const: arm,scmi-smc
+ - description: SCMI compliant firmware with ARM SMC/HVC transport
+ with shmem address(4KB-page, offset) as parameters
+ items:
+ - const: arm,scmi-smc-param
- description: SCMI compliant firmware with SCMI Virtio transport.
The virtio transport only supports a single device.
items:
@@ -299,7 +303,9 @@ else:
properties:
compatible:
contains:
- const: arm,scmi-smc
+ enum:
+ - arm,scmi-smc
+ - arm,scmi-smc-param
then:
required:
- arm,smc-id
--
2.17.1
On 5/6/23 11:24, Nikunj Kela wrote:
> Currently, smc/hvc calls are made with smc-id only. The parameters are
> all set to zeros. This change defines a new compatible string that can
> be used to pass shmem address(4KB-page, offset) as two parameters in
> SMC/HVC doorbell.
>
> This is useful when multiple scmi instances are used with common smc-id.
>
> Signed-off-by: Nikunj Kela <[email protected]>
> Reviewed-by: Rob Herring <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
--
Florian
On 5/6/23 11:24, Nikunj Kela wrote:
> This change adds support for passing shmem channel address as parameters
> in smc/hvc call. The address is split into 4KB-page and offset.
> This is useful when multiple scmi instances are using same smc-id
> and firmware needs to distinguish among the instances.
>
> Signed-off-by: Nikunj Kela <[email protected]>
Reviewed-by: Florian Fainelli <[email protected]>
--
Florian
On Sat, May 06, 2023 at 11:24:26AM -0700, Nikunj Kela wrote:
> Currently, smc/hvc calls are made with parameters set
> to zeros. We are using multiple scmi instances within
> a VM. We are sharing the same smc-id(func_id) with all
> scmi instance. The hypervisor needs a way to distinguish
> among hvc calls made from different instances.
>
> This patch series introduces new compatible string which
> can be used to pass shmem channel address as parameters
> to smc/hvc calls.
>
> ---
> v5 -> avoid computing page and offset in send function
> Link: https://lore.kernel.org/all/[email protected]/
>
Thanks for the updated patch. I will pick this up for v6.5 and update
once queued.
--
Regards,
Sudeep
On Sat, 06 May 2023 11:24:26 -0700, Nikunj Kela wrote:
> Currently, smc/hvc calls are made with parameters set
> to zeros. We are using multiple scmi instances within
> a VM. We are sharing the same smc-id(func_id) with all
> scmi instance. The hypervisor needs a way to distinguish
> among hvc calls made from different instances.
>
> This patch series introduces new compatible string which
> can be used to pass shmem channel address as parameters
> to smc/hvc calls.
>
> [...]
Applied to sudeep.holla/linux (for-next/scmi/updates), thanks!
[1/2] dt-bindings: firmware: arm,scmi: support for parameter in smc/hvc call
https://git.kernel.org/sudeep.holla/c/8f9d530cffc1
[2/2] firmware: arm_scmi: Augment SMC/HVC to allow optional parameters
https://git.kernel.org/sudeep.holla/c/5f2ea10a808a
--
Regards,
Sudeep
On 6/30/2023 2:44 AM, Sudeep Holla wrote:
> On Wed, Jun 28, 2023 at 01:38:32PM -0700, Nikunj Kela wrote:
>> On 5/31/2023 5:26 AM, Sudeep Holla wrote:
>>> On Sat, 06 May 2023 11:24:26 -0700, Nikunj Kela wrote:
>>>> Currently, smc/hvc calls are made with parameters set
>>>> to zeros. We are using multiple scmi instances within
>>>> a VM. We are sharing the same smc-id(func_id) with all
>>>> scmi instance. The hypervisor needs a way to distinguish
>>>> among hvc calls made from different instances.
>>>>
>>>> This patch series introduces new compatible string which
>>>> can be used to pass shmem channel address as parameters
>>>> to smc/hvc calls.
>>>>
>>>> [...]
>>> Applied to sudeep.holla/linux (for-next/scmi/updates), thanks!
>> Hi Sudeep, our hypervisor team is evaluating other scmi transport
>> options(including new qcom specific transport) along with this one so there
>> is a possibility that we might not use this solution. If you think this
>> patch is not useful to others, you can hold off its merge. Sorry about the
>> last minute inconvenience.
> Firstly, not sure why you took this in private. I am going to reply on the
> list and from now on I will never trust or rush to take any Qualcomm changes.
> Nothing personal but I have to follow that to keep it simple. It is in
> Linux tree yesterday or today and yes it is too late.
I understand your frustration and trust me I am even more disappointed
by the decision by our hypervisor team and that too this late. My
efforts on this series are all wasted. If you like me to post a revert
of this patch series, I will be happy to do that. Please don't let this
one example reflect whole Qualcomm culture. Again, extremely sorry about
all the efforts you put in.
>