[v3,14/27] powerpc/powernv/pmem: Add support for Admin commands
diff mbox series

Message ID 20200221032720.33893-15-alastair@au1.ibm.com
State New
Headers show
Series
  • Add support for OpenCAPI Persistent Memory devices
Related show

Commit Message

Alastair D'Silva Feb. 21, 2020, 3:27 a.m. UTC
From: Alastair D'Silva <alastair@d-silva.org>

This patch requests the metadata required to issue admin commands, as well
as some helper functions to construct and check the completion of the
commands.

Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
---
 arch/powerpc/platforms/powernv/pmem/ocxl.c    |  65 ++++++++
 .../platforms/powernv/pmem/ocxl_internal.c    | 153 ++++++++++++++++++
 .../platforms/powernv/pmem/ocxl_internal.h    |  61 +++++++
 3 files changed, 279 insertions(+)

Comments

Andrew Donnellan Feb. 27, 2020, 8:22 a.m. UTC | #1
On 21/2/20 2:27 pm, Alastair D'Silva wrote:
> From: Alastair D'Silva <alastair@d-silva.org>
> 
> This patch requests the metadata required to issue admin commands, as well
> as some helper functions to construct and check the completion of the
> commands.
> 
> Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
> ---
>   arch/powerpc/platforms/powernv/pmem/ocxl.c    |  65 ++++++++
>   .../platforms/powernv/pmem/ocxl_internal.c    | 153 ++++++++++++++++++
>   .../platforms/powernv/pmem/ocxl_internal.h    |  61 +++++++
>   3 files changed, 279 insertions(+)
> 
> diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl.c b/arch/powerpc/platforms/powernv/pmem/ocxl.c
> index 431212c9f0cc..4e782d22605b 100644
> --- a/arch/powerpc/platforms/powernv/pmem/ocxl.c
> +++ b/arch/powerpc/platforms/powernv/pmem/ocxl.c
> @@ -216,6 +216,58 @@ static int register_lpc_mem(struct ocxlpmem *ocxlpmem)
>   	return 0;
>   }
>   
> +/**
> + * extract_command_metadata() - Extract command data from MMIO & save it for further use
> + * @ocxlpmem: the device metadata
> + * @offset: The base address of the command data structures (address of CREQO)
> + * @command_metadata: A pointer to the command metadata to populate
> + * Return: 0 on success, negative on failure
> + */
> +static int extract_command_metadata(struct ocxlpmem *ocxlpmem, u32 offset,
> +					struct command_metadata *command_metadata)
> +{
> +	int rc;
> +	u64 tmp;
> +
> +	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset, OCXL_LITTLE_ENDIAN,
> +				     &tmp);
> +	if (rc)
> +		return rc;
> +
> +	command_metadata->request_offset = tmp >> 32;
> +	command_metadata->response_offset = tmp & 0xFFFFFFFF;
> +
> +	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset + 8, OCXL_LITTLE_ENDIAN,
> +				     &tmp);
> +	if (rc)
> +		return rc;
> +
> +	command_metadata->data_offset = tmp >> 32;
> +	command_metadata->data_size = tmp & 0xFFFFFFFF;
> +
> +	command_metadata->id = 0;
> +
> +	return 0;
> +}
> +
> +/**
> + * setup_command_metadata() - Set up the command metadata
> + * @ocxlpmem: the device metadata
> + */
> +static int setup_command_metadata(struct ocxlpmem *ocxlpmem)
> +{
> +	int rc;
> +
> +	mutex_init(&ocxlpmem->admin_command.lock);
> +
> +	rc = extract_command_metadata(ocxlpmem, GLOBAL_MMIO_ACMA_CREQO,
> +				      &ocxlpmem->admin_command);
> +	if (rc)
> +		return rc;
> +
> +	return 0;
> +}
> +
>   /**
>    * is_usable() - Is a controller usable?
>    * @ocxlpmem: the device metadata
> @@ -456,6 +508,14 @@ static int probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>   	}
>   	ocxlpmem->pdev = pdev;
>   
> +	ocxlpmem->timeouts[ADMIN_COMMAND_ERRLOG] = 2000; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_HEARTBEAT] = 100; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_SMART] = 100; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_DUMP] = 1000; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_STATS] = 100; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_SHUTDOWN] = 1000; // ms
> +	ocxlpmem->timeouts[ADMIN_COMMAND_FW_UPDATE] = 16000; // ms

Why are we keeping these timeouts in a per device struct? I can't see 
anywhere where we change these values.

> +
>   	pci_set_drvdata(pdev, ocxlpmem);
>   
>   	ocxlpmem->ocxl_fn = ocxl_function_open(pdev);
> @@ -501,6 +561,11 @@ static int probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>   		goto err;
>   	}
>   
> +	if (setup_command_metadata(ocxlpmem)) {
> +		dev_err(&pdev->dev, "Could not read OCXL command matada\n");

metadata

Also, "OCXL command metadata" is misleading, this is a pmem specific 
thing, not an OpenCAPI thing, I would prefer just "command metadata".

> +		goto err;
> +	}
> +
>   	elapsed = 0;
>   	timeout = ocxlpmem->readiness_timeout + ocxlpmem->memory_available_timeout;
>   	while (!is_usable(ocxlpmem, false)) {
> diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> index 617ca943b1b8..583f48023025 100644
> --- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> +++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> @@ -17,3 +17,156 @@ int ocxlpmem_chi(const struct ocxlpmem *ocxlpmem, u64 *chi)
>   
>   	return 0;
>   }
> +
> +#define COMMAND_REQUEST_SIZE (8 * sizeof(u64))
> +static int scm_command_request(const struct ocxlpmem *ocxlpmem,
> +			       struct command_metadata *cmd, u8 op_code)
> +{
> +	u64 val = op_code;
> +	int rc;
> +	u8 i;
> +
> +	cmd->op_code = op_code;
> +	cmd->id++;
> +
> +	val |= ((u64)cmd->id) << 16;
> +
> +	rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu, cmd->request_offset,
> +				      OCXL_LITTLE_ENDIAN, val);
> +	if (rc)
> +		return rc;
> +
> +	for (i = sizeof(u64); i < COMMAND_REQUEST_SIZE; i += sizeof(u64)) {
> +		rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu,
> +					      cmd->request_offset + i,
> +					      OCXL_LITTLE_ENDIAN, 0);
> +		if (rc)
> +			return rc;
> +	}
> +
> +	return 0;
> +}
> +
> +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code)
> +{
> +	u64 val;
> +	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_CHI,
> +					 OCXL_LITTLE_ENDIAN, &val);
> +	if (rc)
> +		return rc;

Ignoring the value here expected, you're just trying to verify that you 
don't see an error on the read?

> +
> +	return scm_command_request(ocxlpmem, &ocxlpmem->admin_command, op_code);
> +}
> +
> +static int command_response(const struct ocxlpmem *ocxlpmem,
> +			    const struct command_metadata *cmd)
> +{
> +	u64 val;
> +	u16 id;
> +	u8 status;
> +	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu,
> +					 cmd->response_offset,
> +					 OCXL_LITTLE_ENDIAN, &val);
> +	if (rc)
> +		return rc;
> +
> +	status = val & 0xff;
> +	id = (val >> 16) & 0xffff;
> +
> +	if (id != cmd->id) {
> +		dev_warn(&ocxlpmem->dev,
> +			 "Expected response for command %d, but received response for command %d instead.\n",
> +			 cmd->id, id);

If this happens I imagine something's gone pretty wrong - this should 
probably be a dev_err()? And perhaps we want to make sure we return an 
error code rather than whatever status code we get from the MMIO?

> +	}
> +
> +	return status;
> +}
> +
> +int admin_response(const struct ocxlpmem *ocxlpmem)
> +{
> +	return command_response(ocxlpmem, &ocxlpmem->admin_command);
> +}
> +
> +
> +int admin_command_execute(const struct ocxlpmem *ocxlpmem)
> +{
> +	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_HCI,
> +				      OCXL_LITTLE_ENDIAN, GLOBAL_MMIO_HCI_ACRW);
> +}
> +
> +static bool admin_command_complete(const struct ocxlpmem *ocxlpmem)
> +{
> +	u64 val = 0;
> +
> +	int rc = ocxlpmem_chi(ocxlpmem, &val);
> +
> +	WARN_ON(rc);
> +
> +	return (val & GLOBAL_MMIO_CHI_ACRA) != 0;
> +}
> +
> +int admin_command_complete_timeout(const struct ocxlpmem *ocxlpmem,
> +				   int command)
> +{
> +	u32 timeout = ocxlpmem->timeouts[command];
> +	// 32 is the next power of 2 greater than the 20ms minimum for msleep
> +#define TIMEOUT_SLEEP_MILLIS 32
> +	timeout /= TIMEOUT_SLEEP_MILLIS;
> +	if (!timeout)
> +		timeout = DEFAULT_TIMEOUT / TIMEOUT_SLEEP_MILLIS;
> +
> +	while (timeout-- > 0) {
> +		if (admin_command_complete(ocxlpmem))
> +			return 0;
> +		msleep(TIMEOUT_SLEEP_MILLIS);
> +	}

I think the more traditional way to implement timeouts is something more 
like:

   unsigned long timeout = jiffies + msecs_to_jiffies(<timeout period>);
   do {
     <check>
     <sleep>
   } while (time_before(jiffies, timeout));

> +
> +	if (admin_command_complete(ocxlpmem))
> +		return 0;
> +
> +	return -EBUSY;
> +}
> +
> +int admin_response_handled(const struct ocxlpmem *ocxlpmem)
> +{
> +	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_CHIC,
> +				      OCXL_LITTLE_ENDIAN, GLOBAL_MMIO_CHI_ACRA);
> +}

This looks wrong? My reading of the spec is that you're meant to *clear* 
ACRA upon completion of handing, this looks like it's setting ACRA to 1.

> +
> +void warn_status(const struct ocxlpmem *ocxlpmem, const char *message,
> +		     u8 status)
> +{
> +	const char *text = "Unknown";
> +
> +	switch (status) {
> +	case STATUS_SUCCESS:
> +		text = "Success";
> +		break;
> +
> +	case STATUS_MEM_UNAVAILABLE:
> +		text = "Persistent memory unavailable";
> +		break;
> +
> +	case STATUS_BAD_OPCODE:
> +		text = "Bad opcode";
> +		break;
> +
> +	case STATUS_BAD_REQUEST_PARM:
> +		text = "Bad request parameter";
> +		break;
> +
> +	case STATUS_BAD_DATA_PARM:
> +		text = "Bad data parameter";
> +		break;
> +
> +	case STATUS_DEBUG_BLOCKED:
> +		text = "Debug action blocked";
> +		break;
> +
> +	case STATUS_FAIL:
> +		text = "Failed";
> +		break;
> +	}
> +
> +	dev_warn(&ocxlpmem->dev, "%s: %s (%x)\n", message, text, status);
> +}
> diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> index ba0301533d00..2fef68c71271 100644
> --- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> +++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> @@ -7,6 +7,7 @@
>   #include <linux/mm.h>
>   
>   #define LABEL_AREA_SIZE	(1UL << PA_SECTION_SHIFT)
> +#define DEFAULT_TIMEOUT 100
>   
>   #define GLOBAL_MMIO_CHI		0x000
>   #define GLOBAL_MMIO_CHIC	0x008
> @@ -80,6 +81,16 @@
>   #define STATUS_FW_ARG_INVALID	0x51
>   #define STATUS_FW_INVALID	0x52
>   
> +struct command_metadata {
> +	u32 request_offset;
> +	u32 response_offset;
> +	u32 data_offset;
> +	u32 data_size;
> +	struct mutex lock;
> +	u16 id;
> +	u8 op_code;
> +};
> +
>   struct ocxlpmem_function0 {
>   	struct pci_dev *pdev;
>   	struct ocxl_fn *ocxl_fn;
> @@ -95,9 +106,11 @@ struct ocxlpmem {
>   	struct ocxl_afu *ocxl_afu;
>   	struct ocxl_context *ocxl_context;
>   	void *metadata_addr;
> +	struct command_metadata admin_command;
>   	struct resource pmem_res;
>   	struct nd_region *nd_region;
>   	char fw_version[8+1];
> +	u32 timeouts[ADMIN_COMMAND_MAX+1];
>   
>   	u32 max_controller_dump_size;
>   	u16 scm_revision; // major/minor
> @@ -122,3 +135,51 @@ struct ocxlpmem {
>    * Returns 0 on success, negative on error
>    */
>   int ocxlpmem_chi(const struct ocxlpmem *ocxlpmem, u64 *chi);
> +
> +/**
> + * admin_command_request() - Issue an admin command request
> + * @ocxlpmem: the device metadata
> + * @op_code: The op-code for the command
> + *
> + * Returns an identifier for the command, or negative on error
> + */
> +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code);
> +
> +/**
> + * admin_response() - Validate an admin response
> + * @ocxlpmem: the device metadata
> + * Returns the status code of the command, or negative on error
> + */
> +int admin_response(const struct ocxlpmem *ocxlpmem);
> +
> +/**
> + * admin_command_execute() - Notify the controller to start processing a pending admin command
> + * @ocxlpmem: the device metadata
> + * Returns 0 on success, negative on error
> + */
> +int admin_command_execute(const struct ocxlpmem *ocxlpmem);
> +
> +/**
> + * admin_command_complete_timeout() - Wait for an admin command to finish executing
> + * @ocxlpmem: the device metadata
> + * @command: the admin command to wait for completion (determines the timeout)
> + * Returns 0 on success, -EBUSY on timeout
> + */
> +int admin_command_complete_timeout(const struct ocxlpmem *ocxlpmem,
> +				   int command);
> +
> +/**
> + * admin_response_handled() - Notify the controller that the admin response has been handled
> + * @ocxlpmem: the device metadata
> + * Returns 0 on success, negative on failure
> + */
> +int admin_response_handled(const struct ocxlpmem *ocxlpmem);
> +
> +/**
> + * warn_status() - Emit a kernel warning showing a command status.
> + * @ocxlpmem: the device metadata
> + * @message: A message to accompany the warning
> + * @status: The command status
> + */
> +void warn_status(const struct ocxlpmem *ocxlpmem, const char *message,
> +		 u8 status);
>
Andrew Donnellan Feb. 27, 2020, 8:27 a.m. UTC | #2
On 27/2/20 7:22 pm, Andrew Donnellan wrote:
>> +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code)
>> +{
>> +    u64 val;
>> +    int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, 
>> GLOBAL_MMIO_CHI,
>> +                     OCXL_LITTLE_ENDIAN, &val);
>> +    if (rc)
>> +        return rc;
> 
> Ignoring the value here expected, you're just trying to verify that you 
> don't see an error on the read?

I see that in the next patch, in ns_command_request() you check that 
NSCRA is 1 - did you mean to check that ACRA = 1 here?
Dan Williams Feb. 27, 2020, 5:01 p.m. UTC | #3
On Thu, Feb 20, 2020 at 7:28 PM Alastair D'Silva <alastair@au1.ibm.com> wrote:
>
> From: Alastair D'Silva <alastair@d-silva.org>
>
> This patch requests the metadata required to issue admin commands, as well
> as some helper functions to construct and check the completion of the
> commands.

What are the admin commands? Any pointer to a spec? Why does Linux
need to support these commands?
Alastair D'Silva Feb. 27, 2020, 11:51 p.m. UTC | #4
On Thu, 2020-02-27 at 19:22 +1100, Andrew Donnellan wrote:
> On 21/2/20 2:27 pm, Alastair D'Silva wrote:
> > From: Alastair D'Silva <alastair@d-silva.org>
> > 
> > This patch requests the metadata required to issue admin commands,
> > as well
> > as some helper functions to construct and check the completion of
> > the
> > commands.
> > 
> > Signed-off-by: Alastair D'Silva <alastair@d-silva.org>
> > ---
> >   arch/powerpc/platforms/powernv/pmem/ocxl.c    |  65 ++++++++
> >   .../platforms/powernv/pmem/ocxl_internal.c    | 153
> > ++++++++++++++++++
> >   .../platforms/powernv/pmem/ocxl_internal.h    |  61 +++++++
> >   3 files changed, 279 insertions(+)
> > 
> > diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl.c
> > b/arch/powerpc/platforms/powernv/pmem/ocxl.c
> > index 431212c9f0cc..4e782d22605b 100644
> > --- a/arch/powerpc/platforms/powernv/pmem/ocxl.c
> > +++ b/arch/powerpc/platforms/powernv/pmem/ocxl.c
> > @@ -216,6 +216,58 @@ static int register_lpc_mem(struct ocxlpmem
> > *ocxlpmem)
> >   	return 0;
> >   }
> >   
> > +/**
> > + * extract_command_metadata() - Extract command data from MMIO &
> > save it for further use
> > + * @ocxlpmem: the device metadata
> > + * @offset: The base address of the command data structures
> > (address of CREQO)
> > + * @command_metadata: A pointer to the command metadata to
> > populate
> > + * Return: 0 on success, negative on failure
> > + */
> > +static int extract_command_metadata(struct ocxlpmem *ocxlpmem, u32
> > offset,
> > +					struct command_metadata
> > *command_metadata)
> > +{
> > +	int rc;
> > +	u64 tmp;
> > +
> > +	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset,
> > OCXL_LITTLE_ENDIAN,
> > +				     &tmp);
> > +	if (rc)
> > +		return rc;
> > +
> > +	command_metadata->request_offset = tmp >> 32;
> > +	command_metadata->response_offset = tmp & 0xFFFFFFFF;
> > +
> > +	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset + 8,
> > OCXL_LITTLE_ENDIAN,
> > +				     &tmp);
> > +	if (rc)
> > +		return rc;
> > +
> > +	command_metadata->data_offset = tmp >> 32;
> > +	command_metadata->data_size = tmp & 0xFFFFFFFF;
> > +
> > +	command_metadata->id = 0;
> > +
> > +	return 0;
> > +}
> > +
> > +/**
> > + * setup_command_metadata() - Set up the command metadata
> > + * @ocxlpmem: the device metadata
> > + */
> > +static int setup_command_metadata(struct ocxlpmem *ocxlpmem)
> > +{
> > +	int rc;
> > +
> > +	mutex_init(&ocxlpmem->admin_command.lock);
> > +
> > +	rc = extract_command_metadata(ocxlpmem, GLOBAL_MMIO_ACMA_CREQO,
> > +				      &ocxlpmem->admin_command);
> > +	if (rc)
> > +		return rc;
> > +
> > +	return 0;
> > +}
> > +
> >   /**
> >    * is_usable() - Is a controller usable?
> >    * @ocxlpmem: the device metadata
> > @@ -456,6 +508,14 @@ static int probe(struct pci_dev *pdev, const
> > struct pci_device_id *ent)
> >   	}
> >   	ocxlpmem->pdev = pdev;
> >   
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_ERRLOG] = 2000; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_HEARTBEAT] = 100; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_SMART] = 100; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_DUMP] = 1000; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_STATS] = 100; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_SHUTDOWN] = 1000; // ms
> > +	ocxlpmem->timeouts[ADMIN_COMMAND_FW_UPDATE] = 16000; // ms
> 
> Why are we keeping these timeouts in a per device struct? I can't
> see 
> anywhere where we change these values.
> 

These are overwritten in a later patch, which I've missed! thanks for
pointing this out.

These initial values will be overwritten by card specific timeouts.

> > +
> >   	pci_set_drvdata(pdev, ocxlpmem);
> >   
> >   	ocxlpmem->ocxl_fn = ocxl_function_open(pdev);
> > @@ -501,6 +561,11 @@ static int probe(struct pci_dev *pdev, const
> > struct pci_device_id *ent)
> >   		goto err;
> >   	}
> >   
> > +	if (setup_command_metadata(ocxlpmem)) {
> > +		dev_err(&pdev->dev, "Could not read OCXL command
> > matada\n");
> 
> metadata

Wow, not sure how that happened.

> 
> Also, "OCXL command metadata" is misleading, this is a pmem specific 
> thing, not an OpenCAPI thing, I would prefer just "command metadata".
> 

Ok

> > +		goto err;
> > +	}
> > +
> >   	elapsed = 0;
> >   	timeout = ocxlpmem->readiness_timeout + ocxlpmem-
> > >memory_available_timeout;
> >   	while (!is_usable(ocxlpmem, false)) {
> > diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> > b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> > index 617ca943b1b8..583f48023025 100644
> > --- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> > +++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
> > @@ -17,3 +17,156 @@ int ocxlpmem_chi(const struct ocxlpmem
> > *ocxlpmem, u64 *chi)
> >   
> >   	return 0;
> >   }
> > +
> > +#define COMMAND_REQUEST_SIZE (8 * sizeof(u64))
> > +static int scm_command_request(const struct ocxlpmem *ocxlpmem,
> > +			       struct command_metadata *cmd, u8
> > op_code)
> > +{
> > +	u64 val = op_code;
> > +	int rc;
> > +	u8 i;
> > +
> > +	cmd->op_code = op_code;
> > +	cmd->id++;
> > +
> > +	val |= ((u64)cmd->id) << 16;
> > +
> > +	rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu, cmd-
> > >request_offset,
> > +				      OCXL_LITTLE_ENDIAN, val);
> > +	if (rc)
> > +		return rc;
> > +
> > +	for (i = sizeof(u64); i < COMMAND_REQUEST_SIZE; i +=
> > sizeof(u64)) {
> > +		rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu,
> > +					      cmd->request_offset + i,
> > +					      OCXL_LITTLE_ENDIAN, 0);
> > +		if (rc)
> > +			return rc;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code)
> > +{
> > +	u64 val;
> > +	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu,
> > GLOBAL_MMIO_CHI,
> > +					 OCXL_LITTLE_ENDIAN, &val);
> > +	if (rc)
> > +		return rc;
> 
> Ignoring the value here expected, you're just trying to verify that
> you 
> don't see an error on the read?
> 

This was some vestigial code that should be removed.

> > +
> > +	return scm_command_request(ocxlpmem, &ocxlpmem->admin_command,
> > op_code);
> > +}
> > +
> > +static int command_response(const struct ocxlpmem *ocxlpmem,
> > +			    const struct command_metadata *cmd)
> > +{
> > +	u64 val;
> > +	u16 id;
> > +	u8 status;
> > +	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu,
> > +					 cmd->response_offset,
> > +					 OCXL_LITTLE_ENDIAN, &val);
> > +	if (rc)
> > +		return rc;
> > +
> > +	status = val & 0xff;
> > +	id = (val >> 16) & 0xffff;
> > +
> > +	if (id != cmd->id) {
> > +		dev_warn(&ocxlpmem->dev,
> > +			 "Expected response for command %d, but
> > received response for command %d instead.\n",
> > +			 cmd->id, id);
> 
> If this happens I imagine something's gone pretty wrong - this
> should 
> probably be a dev_err()? And perhaps we want to make sure we return
> an 
> error code rather than whatever status code we get from the MMIO?
> 

Ok

> > +	}
> > +
> > +	return status;
> > +}
> > +
> > +int admin_response(const struct ocxlpmem *ocxlpmem)
> > +{
> > +	return command_response(ocxlpmem, &ocxlpmem->admin_command);
> > +}
> > +
> > +
> > +int admin_command_execute(const struct ocxlpmem *ocxlpmem)
> > +{
> > +	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu,
> > GLOBAL_MMIO_HCI,
> > +				      OCXL_LITTLE_ENDIAN,
> > GLOBAL_MMIO_HCI_ACRW);
> > +}
> > +
> > +static bool admin_command_complete(const struct ocxlpmem
> > *ocxlpmem)
> > +{
> > +	u64 val = 0;
> > +
> > +	int rc = ocxlpmem_chi(ocxlpmem, &val);
> > +
> > +	WARN_ON(rc);
> > +
> > +	return (val & GLOBAL_MMIO_CHI_ACRA) != 0;
> > +}
> > +
> > +int admin_command_complete_timeout(const struct ocxlpmem
> > *ocxlpmem,
> > +				   int command)
> > +{
> > +	u32 timeout = ocxlpmem->timeouts[command];
> > +	// 32 is the next power of 2 greater than the 20ms minimum for
> > msleep
> > +#define TIMEOUT_SLEEP_MILLIS 32
> > +	timeout /= TIMEOUT_SLEEP_MILLIS;
> > +	if (!timeout)
> > +		timeout = DEFAULT_TIMEOUT / TIMEOUT_SLEEP_MILLIS;
> > +
> > +	while (timeout-- > 0) {
> > +		if (admin_command_complete(ocxlpmem))
> > +			return 0;
> > +		msleep(TIMEOUT_SLEEP_MILLIS);
> > +	}
> 
> I think the more traditional way to implement timeouts is something
> more 
> like:
> 
>    unsigned long timeout = jiffies + msecs_to_jiffies(<timeout
> period>);
>    do {
>      <check>
>      <sleep>
>    } while (time_before(jiffies, timeout));
> 

ok

> > +
> > +	if (admin_command_complete(ocxlpmem))
> > +		return 0;
> > +
> > +	return -EBUSY;
> > +}
> > +
> > +int admin_response_handled(const struct ocxlpmem *ocxlpmem)
> > +{
> > +	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu,
> > GLOBAL_MMIO_CHIC,
> > +				      OCXL_LITTLE_ENDIAN,
> > GLOBAL_MMIO_CHI_ACRA);
> > +}
> 
> This looks wrong? My reading of the spec is that you're meant to
> *clear* 
> ACRA upon completion of handing, this looks like it's setting ACRA to
> 1.
> 

Writing a 1 to the CHIC register clears the respective bit in the CHI
register. I'll add a comment.

> > +
> > +void warn_status(const struct ocxlpmem *ocxlpmem, const char
> > *message,
> > +		     u8 status)
> > +{
> > +	const char *text = "Unknown";
> > +
> > +	switch (status) {
> > +	case STATUS_SUCCESS:
> > +		text = "Success";
> > +		break;
> > +
> > +	case STATUS_MEM_UNAVAILABLE:
> > +		text = "Persistent memory unavailable";
> > +		break;
> > +
> > +	case STATUS_BAD_OPCODE:
> > +		text = "Bad opcode";
> > +		break;
> > +
> > +	case STATUS_BAD_REQUEST_PARM:
> > +		text = "Bad request parameter";
> > +		break;
> > +
> > +	case STATUS_BAD_DATA_PARM:
> > +		text = "Bad data parameter";
> > +		break;
> > +
> > +	case STATUS_DEBUG_BLOCKED:
> > +		text = "Debug action blocked";
> > +		break;
> > +
> > +	case STATUS_FAIL:
> > +		text = "Failed";
> > +		break;
> > +	}
> > +
> > +	dev_warn(&ocxlpmem->dev, "%s: %s (%x)\n", message, text,
> > status);
> > +}
> > diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> > b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> > index ba0301533d00..2fef68c71271 100644
> > --- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> > +++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
> > @@ -7,6 +7,7 @@
> >   #include <linux/mm.h>
> >   
> >   #define LABEL_AREA_SIZE	(1UL << PA_SECTION_SHIFT)
> > +#define DEFAULT_TIMEOUT 100
> >   
> >   #define GLOBAL_MMIO_CHI		0x000
> >   #define GLOBAL_MMIO_CHIC	0x008
> > @@ -80,6 +81,16 @@
> >   #define STATUS_FW_ARG_INVALID	0x51
> >   #define STATUS_FW_INVALID	0x52
> >   
> > +struct command_metadata {
> > +	u32 request_offset;
> > +	u32 response_offset;
> > +	u32 data_offset;
> > +	u32 data_size;
> > +	struct mutex lock;
> > +	u16 id;
> > +	u8 op_code;
> > +};
> > +
> >   struct ocxlpmem_function0 {
> >   	struct pci_dev *pdev;
> >   	struct ocxl_fn *ocxl_fn;
> > @@ -95,9 +106,11 @@ struct ocxlpmem {
> >   	struct ocxl_afu *ocxl_afu;
> >   	struct ocxl_context *ocxl_context;
> >   	void *metadata_addr;
> > +	struct command_metadata admin_command;
> >   	struct resource pmem_res;
> >   	struct nd_region *nd_region;
> >   	char fw_version[8+1];
> > +	u32 timeouts[ADMIN_COMMAND_MAX+1];
> >   
> >   	u32 max_controller_dump_size;
> >   	u16 scm_revision; // major/minor
> > @@ -122,3 +135,51 @@ struct ocxlpmem {
> >    * Returns 0 on success, negative on error
> >    */
> >   int ocxlpmem_chi(const struct ocxlpmem *ocxlpmem, u64 *chi);
> > +
> > +/**
> > + * admin_command_request() - Issue an admin command request
> > + * @ocxlpmem: the device metadata
> > + * @op_code: The op-code for the command
> > + *
> > + * Returns an identifier for the command, or negative on error
> > + */
> > +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code);
> > +
> > +/**
> > + * admin_response() - Validate an admin response
> > + * @ocxlpmem: the device metadata
> > + * Returns the status code of the command, or negative on error
> > + */
> > +int admin_response(const struct ocxlpmem *ocxlpmem);
> > +
> > +/**
> > + * admin_command_execute() - Notify the controller to start
> > processing a pending admin command
> > + * @ocxlpmem: the device metadata
> > + * Returns 0 on success, negative on error
> > + */
> > +int admin_command_execute(const struct ocxlpmem *ocxlpmem);
> > +
> > +/**
> > + * admin_command_complete_timeout() - Wait for an admin command to
> > finish executing
> > + * @ocxlpmem: the device metadata
> > + * @command: the admin command to wait for completion (determines
> > the timeout)
> > + * Returns 0 on success, -EBUSY on timeout
> > + */
> > +int admin_command_complete_timeout(const struct ocxlpmem
> > *ocxlpmem,
> > +				   int command);
> > +
> > +/**
> > + * admin_response_handled() - Notify the controller that the admin
> > response has been handled
> > + * @ocxlpmem: the device metadata
> > + * Returns 0 on success, negative on failure
> > + */
> > +int admin_response_handled(const struct ocxlpmem *ocxlpmem);
> > +
> > +/**
> > + * warn_status() - Emit a kernel warning showing a command status.
> > + * @ocxlpmem: the device metadata
> > + * @message: A message to accompany the warning
> > + * @status: The command status
> > + */
> > +void warn_status(const struct ocxlpmem *ocxlpmem, const char
> > *message,
> > +		 u8 status);
> >
Alastair D'Silva Feb. 27, 2020, 11:54 p.m. UTC | #5
On Thu, 2020-02-27 at 19:27 +1100, Andrew Donnellan wrote:
> On 27/2/20 7:22 pm, Andrew Donnellan wrote:
> > > +int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code)
> > > +{
> > > +    u64 val;
> > > +    int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, 
> > > GLOBAL_MMIO_CHI,
> > > +                     OCXL_LITTLE_ENDIAN, &val);
> > > +    if (rc)
> > > +        return rc;
> > 
> > Ignoring the value here expected, you're just trying to verify that
> > you 
> > don't see an error on the read?
> 
> I see that in the next patch, in ns_command_request() you check that 
> NSCRA is 1 - did you mean to check that ACRA = 1 here?
> 
> 

I was in one version, but that was causing problems in startup since
there was successful prior command to assert ACRA.

I should remove the NSCRA check too.
Alastair D'Silva Feb. 27, 2020, 11:57 p.m. UTC | #6
On Thu, 2020-02-27 at 09:01 -0800, Dan Williams wrote:
> On Thu, Feb 20, 2020 at 7:28 PM Alastair D'Silva <
> alastair@au1.ibm.com> wrote:
> > From: Alastair D'Silva <alastair@d-silva.org>
> > 
> > This patch requests the metadata required to issue admin commands,
> > as well
> > as some helper functions to construct and check the completion of
> > the
> > commands.
> 
> What are the admin commands? Any pointer to a spec? Why does Linux
> need to support these commands?


I'll flesh these out for the next spin, thanks.

Patch
diff mbox series

diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl.c b/arch/powerpc/platforms/powernv/pmem/ocxl.c
index 431212c9f0cc..4e782d22605b 100644
--- a/arch/powerpc/platforms/powernv/pmem/ocxl.c
+++ b/arch/powerpc/platforms/powernv/pmem/ocxl.c
@@ -216,6 +216,58 @@  static int register_lpc_mem(struct ocxlpmem *ocxlpmem)
 	return 0;
 }
 
+/**
+ * extract_command_metadata() - Extract command data from MMIO & save it for further use
+ * @ocxlpmem: the device metadata
+ * @offset: The base address of the command data structures (address of CREQO)
+ * @command_metadata: A pointer to the command metadata to populate
+ * Return: 0 on success, negative on failure
+ */
+static int extract_command_metadata(struct ocxlpmem *ocxlpmem, u32 offset,
+					struct command_metadata *command_metadata)
+{
+	int rc;
+	u64 tmp;
+
+	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset, OCXL_LITTLE_ENDIAN,
+				     &tmp);
+	if (rc)
+		return rc;
+
+	command_metadata->request_offset = tmp >> 32;
+	command_metadata->response_offset = tmp & 0xFFFFFFFF;
+
+	rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, offset + 8, OCXL_LITTLE_ENDIAN,
+				     &tmp);
+	if (rc)
+		return rc;
+
+	command_metadata->data_offset = tmp >> 32;
+	command_metadata->data_size = tmp & 0xFFFFFFFF;
+
+	command_metadata->id = 0;
+
+	return 0;
+}
+
+/**
+ * setup_command_metadata() - Set up the command metadata
+ * @ocxlpmem: the device metadata
+ */
+static int setup_command_metadata(struct ocxlpmem *ocxlpmem)
+{
+	int rc;
+
+	mutex_init(&ocxlpmem->admin_command.lock);
+
+	rc = extract_command_metadata(ocxlpmem, GLOBAL_MMIO_ACMA_CREQO,
+				      &ocxlpmem->admin_command);
+	if (rc)
+		return rc;
+
+	return 0;
+}
+
 /**
  * is_usable() - Is a controller usable?
  * @ocxlpmem: the device metadata
@@ -456,6 +508,14 @@  static int probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	}
 	ocxlpmem->pdev = pdev;
 
+	ocxlpmem->timeouts[ADMIN_COMMAND_ERRLOG] = 2000; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_HEARTBEAT] = 100; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_SMART] = 100; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_DUMP] = 1000; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_CONTROLLER_STATS] = 100; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_SHUTDOWN] = 1000; // ms
+	ocxlpmem->timeouts[ADMIN_COMMAND_FW_UPDATE] = 16000; // ms
+
 	pci_set_drvdata(pdev, ocxlpmem);
 
 	ocxlpmem->ocxl_fn = ocxl_function_open(pdev);
@@ -501,6 +561,11 @@  static int probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 		goto err;
 	}
 
+	if (setup_command_metadata(ocxlpmem)) {
+		dev_err(&pdev->dev, "Could not read OCXL command matada\n");
+		goto err;
+	}
+
 	elapsed = 0;
 	timeout = ocxlpmem->readiness_timeout + ocxlpmem->memory_available_timeout;
 	while (!is_usable(ocxlpmem, false)) {
diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
index 617ca943b1b8..583f48023025 100644
--- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
+++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.c
@@ -17,3 +17,156 @@  int ocxlpmem_chi(const struct ocxlpmem *ocxlpmem, u64 *chi)
 
 	return 0;
 }
+
+#define COMMAND_REQUEST_SIZE (8 * sizeof(u64))
+static int scm_command_request(const struct ocxlpmem *ocxlpmem,
+			       struct command_metadata *cmd, u8 op_code)
+{
+	u64 val = op_code;
+	int rc;
+	u8 i;
+
+	cmd->op_code = op_code;
+	cmd->id++;
+
+	val |= ((u64)cmd->id) << 16;
+
+	rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu, cmd->request_offset,
+				      OCXL_LITTLE_ENDIAN, val);
+	if (rc)
+		return rc;
+
+	for (i = sizeof(u64); i < COMMAND_REQUEST_SIZE; i += sizeof(u64)) {
+		rc = ocxl_global_mmio_write64(ocxlpmem->ocxl_afu,
+					      cmd->request_offset + i,
+					      OCXL_LITTLE_ENDIAN, 0);
+		if (rc)
+			return rc;
+	}
+
+	return 0;
+}
+
+int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code)
+{
+	u64 val;
+	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_CHI,
+					 OCXL_LITTLE_ENDIAN, &val);
+	if (rc)
+		return rc;
+
+	return scm_command_request(ocxlpmem, &ocxlpmem->admin_command, op_code);
+}
+
+static int command_response(const struct ocxlpmem *ocxlpmem,
+			    const struct command_metadata *cmd)
+{
+	u64 val;
+	u16 id;
+	u8 status;
+	int rc = ocxl_global_mmio_read64(ocxlpmem->ocxl_afu,
+					 cmd->response_offset,
+					 OCXL_LITTLE_ENDIAN, &val);
+	if (rc)
+		return rc;
+
+	status = val & 0xff;
+	id = (val >> 16) & 0xffff;
+
+	if (id != cmd->id) {
+		dev_warn(&ocxlpmem->dev,
+			 "Expected response for command %d, but received response for command %d instead.\n",
+			 cmd->id, id);
+	}
+
+	return status;
+}
+
+int admin_response(const struct ocxlpmem *ocxlpmem)
+{
+	return command_response(ocxlpmem, &ocxlpmem->admin_command);
+}
+
+
+int admin_command_execute(const struct ocxlpmem *ocxlpmem)
+{
+	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_HCI,
+				      OCXL_LITTLE_ENDIAN, GLOBAL_MMIO_HCI_ACRW);
+}
+
+static bool admin_command_complete(const struct ocxlpmem *ocxlpmem)
+{
+	u64 val = 0;
+
+	int rc = ocxlpmem_chi(ocxlpmem, &val);
+
+	WARN_ON(rc);
+
+	return (val & GLOBAL_MMIO_CHI_ACRA) != 0;
+}
+
+int admin_command_complete_timeout(const struct ocxlpmem *ocxlpmem,
+				   int command)
+{
+	u32 timeout = ocxlpmem->timeouts[command];
+	// 32 is the next power of 2 greater than the 20ms minimum for msleep
+#define TIMEOUT_SLEEP_MILLIS 32
+	timeout /= TIMEOUT_SLEEP_MILLIS;
+	if (!timeout)
+		timeout = DEFAULT_TIMEOUT / TIMEOUT_SLEEP_MILLIS;
+
+	while (timeout-- > 0) {
+		if (admin_command_complete(ocxlpmem))
+			return 0;
+		msleep(TIMEOUT_SLEEP_MILLIS);
+	}
+
+	if (admin_command_complete(ocxlpmem))
+		return 0;
+
+	return -EBUSY;
+}
+
+int admin_response_handled(const struct ocxlpmem *ocxlpmem)
+{
+	return ocxl_global_mmio_set64(ocxlpmem->ocxl_afu, GLOBAL_MMIO_CHIC,
+				      OCXL_LITTLE_ENDIAN, GLOBAL_MMIO_CHI_ACRA);
+}
+
+void warn_status(const struct ocxlpmem *ocxlpmem, const char *message,
+		     u8 status)
+{
+	const char *text = "Unknown";
+
+	switch (status) {
+	case STATUS_SUCCESS:
+		text = "Success";
+		break;
+
+	case STATUS_MEM_UNAVAILABLE:
+		text = "Persistent memory unavailable";
+		break;
+
+	case STATUS_BAD_OPCODE:
+		text = "Bad opcode";
+		break;
+
+	case STATUS_BAD_REQUEST_PARM:
+		text = "Bad request parameter";
+		break;
+
+	case STATUS_BAD_DATA_PARM:
+		text = "Bad data parameter";
+		break;
+
+	case STATUS_DEBUG_BLOCKED:
+		text = "Debug action blocked";
+		break;
+
+	case STATUS_FAIL:
+		text = "Failed";
+		break;
+	}
+
+	dev_warn(&ocxlpmem->dev, "%s: %s (%x)\n", message, text, status);
+}
diff --git a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
index ba0301533d00..2fef68c71271 100644
--- a/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
+++ b/arch/powerpc/platforms/powernv/pmem/ocxl_internal.h
@@ -7,6 +7,7 @@ 
 #include <linux/mm.h>
 
 #define LABEL_AREA_SIZE	(1UL << PA_SECTION_SHIFT)
+#define DEFAULT_TIMEOUT 100
 
 #define GLOBAL_MMIO_CHI		0x000
 #define GLOBAL_MMIO_CHIC	0x008
@@ -80,6 +81,16 @@ 
 #define STATUS_FW_ARG_INVALID	0x51
 #define STATUS_FW_INVALID	0x52
 
+struct command_metadata {
+	u32 request_offset;
+	u32 response_offset;
+	u32 data_offset;
+	u32 data_size;
+	struct mutex lock;
+	u16 id;
+	u8 op_code;
+};
+
 struct ocxlpmem_function0 {
 	struct pci_dev *pdev;
 	struct ocxl_fn *ocxl_fn;
@@ -95,9 +106,11 @@  struct ocxlpmem {
 	struct ocxl_afu *ocxl_afu;
 	struct ocxl_context *ocxl_context;
 	void *metadata_addr;
+	struct command_metadata admin_command;
 	struct resource pmem_res;
 	struct nd_region *nd_region;
 	char fw_version[8+1];
+	u32 timeouts[ADMIN_COMMAND_MAX+1];
 
 	u32 max_controller_dump_size;
 	u16 scm_revision; // major/minor
@@ -122,3 +135,51 @@  struct ocxlpmem {
  * Returns 0 on success, negative on error
  */
 int ocxlpmem_chi(const struct ocxlpmem *ocxlpmem, u64 *chi);
+
+/**
+ * admin_command_request() - Issue an admin command request
+ * @ocxlpmem: the device metadata
+ * @op_code: The op-code for the command
+ *
+ * Returns an identifier for the command, or negative on error
+ */
+int admin_command_request(struct ocxlpmem *ocxlpmem, u8 op_code);
+
+/**
+ * admin_response() - Validate an admin response
+ * @ocxlpmem: the device metadata
+ * Returns the status code of the command, or negative on error
+ */
+int admin_response(const struct ocxlpmem *ocxlpmem);
+
+/**
+ * admin_command_execute() - Notify the controller to start processing a pending admin command
+ * @ocxlpmem: the device metadata
+ * Returns 0 on success, negative on error
+ */
+int admin_command_execute(const struct ocxlpmem *ocxlpmem);
+
+/**
+ * admin_command_complete_timeout() - Wait for an admin command to finish executing
+ * @ocxlpmem: the device metadata
+ * @command: the admin command to wait for completion (determines the timeout)
+ * Returns 0 on success, -EBUSY on timeout
+ */
+int admin_command_complete_timeout(const struct ocxlpmem *ocxlpmem,
+				   int command);
+
+/**
+ * admin_response_handled() - Notify the controller that the admin response has been handled
+ * @ocxlpmem: the device metadata
+ * Returns 0 on success, negative on failure
+ */
+int admin_response_handled(const struct ocxlpmem *ocxlpmem);
+
+/**
+ * warn_status() - Emit a kernel warning showing a command status.
+ * @ocxlpmem: the device metadata
+ * @message: A message to accompany the warning
+ * @status: The command status
+ */
+void warn_status(const struct ocxlpmem *ocxlpmem, const char *message,
+		 u8 status);