Re: [PATCH v2 2/2] platform/x86/amd/pmf: Use ring buffer to store custom BIOS input values

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 11732 bytes --]

On Fri, 7 Nov 2025, Shyam Sundar S K wrote:
> On 11/5/2025 22:16, Ilpo Järvinen wrote:
> > On Wed, 15 Oct 2025, Shyam Sundar S K wrote:
> > 
> >> Custom BIOS input values can be updated by multiple sources, such as power
> >> mode changes and sensor events, each triggering a custom BIOS input event.
> >> When these events occur in rapid succession, new data may overwrite
> >> previous values before they are processed, resulting in lost updates.
> >>
> >> To address this, introduce a fixed-size, power-of-two ring buffer to
> >> capture every custom BIOS input event, storing both the pending request
> >> and its associated input values. Access to the ring buffer is synchronized
> >> using a mutex.
> >>
> >> The previous use of memset() to clear the pending request structure after
> >> each event is removed, as each BIOS input value is now copied into the
> >> buffer as a snapshot. Consumers now process entries directly from the ring
> >> buffer, making explicit clearing of the pending request structure
> >> unnecessary.
> >>
> >> Co-developed-by: Patil Rajesh Reddy <Patil.Reddy@amd.com>
> >> Signed-off-by: Patil Rajesh Reddy <Patil.Reddy@amd.com>
> >> Signed-off-by: Shyam Sundar S K <Shyam-sundar.S-k@amd.com>
> >> ---
> >> v2:
> >>  - Add dev_WARN_ONCE()
> >>  - Change variable name rb_mutex to cbi_mutex
> >>  - Move tail increment logic above pending request check
> >>
> >>  drivers/platform/x86/amd/pmf/acpi.c   | 41 +++++++++++++++++++++++++++
> >>  drivers/platform/x86/amd/pmf/core.c   |  2 ++
> >>  drivers/platform/x86/amd/pmf/pmf.h    | 20 +++++++++++++
> >>  drivers/platform/x86/amd/pmf/spc.c    | 23 +++++++++------
> >>  drivers/platform/x86/amd/pmf/tee-if.c |  2 ++
> >>  5 files changed, 79 insertions(+), 9 deletions(-)
> >>
> >> diff --git a/drivers/platform/x86/amd/pmf/acpi.c b/drivers/platform/x86/amd/pmf/acpi.c
> >> index 13c4fec2c7ef..870a56f1fe07 100644
> >> --- a/drivers/platform/x86/amd/pmf/acpi.c
> >> +++ b/drivers/platform/x86/amd/pmf/acpi.c
> >> @@ -331,6 +331,43 @@ int apmf_get_sbios_requests(struct amd_pmf_dev *pdev, struct apmf_sbios_req *req
> >>  									 req, sizeof(*req));
> >>  }
> >>  
> >> +/* Store custom BIOS inputs data in ring buffer */
> >> +static void amd_pmf_custom_bios_inputs_rb(struct amd_pmf_dev *pmf_dev)
> >> +{
> >> +	struct cbi_ring_buffer *rb = &pmf_dev->cbi_buf;
> >> +	struct bios_input_entry entry = { };
> > 
> > Too generic struct name.
> > 
> >> +	int i;
> >> +
> >> +	guard(mutex)(&pmf_dev->cbi_mutex);
> >> +
> >> +	switch (pmf_dev->cpu_id) {
> >> +	case AMD_CPU_ID_PS:
> >> +		for (i = 0; i < ARRAY_SIZE(custom_bios_inputs_v1); i++)
> >> +			entry.val[i] = pmf_dev->req1.custom_policy[i];
> >> +		entry.preq = pmf_dev->req1.pending_req;
> >> +		break;
> >> +	case PCI_DEVICE_ID_AMD_1AH_M20H_ROOT:
> >> +	case PCI_DEVICE_ID_AMD_1AH_M60H_ROOT:
> >> +		for (i = 0; i < ARRAY_SIZE(custom_bios_inputs); i++)
> > 
> > Add include.
> > 
> >> +			entry.val[i] = pmf_dev->req.custom_policy[i];
> >> +		entry.preq = pmf_dev->req.pending_req;
> >> +		break;
> >> +	default:
> >> +		return;
> >> +	}
> >> +
> >> +	if (CIRC_SPACE(rb->head, rb->tail, CUSTOM_BIOS_INPUT_RB_SZ) > 0) {
> >> +		rb->data[rb->head] = entry;
> >> +		rb->head = (rb->head + 1) & (CUSTOM_BIOS_INPUT_RB_SZ - 1);
> >> +	} else {
> >> +		/* Rare case: ensures the newest BIOS input value is kept */
> >> +		dev_WARN_ONCE(pmf_dev->dev, 1, "Overwriting BIOS input value, data may be lost\n");
> > 
> > Is this really WARN_ONCE thing? That is, does it indicate a bug in the 
> > kernel code?
> 
> If we hit this case, we might end up in a debug spew and hence put it
> under WARN_ONCE (that way we are not lost in debugging the odd cases,
> if we are under this scenario).

Why doesn't printing just a dev_warn() suffice (it should be just as 
visible in the log provided as the stacktrace, etc. is)? Why you need 
stacktrace, etc. to differentiate?

-- 
 i.

> Not a bug in the kernel code..
> 
> > 
> > Add include.
> > 
> >> +		rb->data[rb->head] = entry;
> >> +		rb->head = (rb->head + 1) & (CUSTOM_BIOS_INPUT_RB_SZ - 1);
> >> +		rb->tail = (rb->tail + 1) & (CUSTOM_BIOS_INPUT_RB_SZ - 1);
> > 
> > Since you're under lock, can you make room separately first and then add
> > the entry on a common path?
> > 
> >> +	}
> >> +}
> >> +
> >>  static void amd_pmf_handle_early_preq(struct amd_pmf_dev *pdev)
> >>  {
> >>  	if (!pdev->cb_flag)
> >> @@ -356,6 +393,8 @@ static void apmf_event_handler_v2(acpi_handle handle, u32 event, void *data)
> >>  	dev_dbg(pmf_dev->dev, "Pending request (preq): 0x%x\n", pmf_dev->req.pending_req);
> >>  
> >>  	amd_pmf_handle_early_preq(pmf_dev);
> >> +
> >> +	amd_pmf_custom_bios_inputs_rb(pmf_dev);
> >>  }
> >>  
> >>  static void apmf_event_handler_v1(acpi_handle handle, u32 event, void *data)
> >> @@ -374,6 +413,8 @@ static void apmf_event_handler_v1(acpi_handle handle, u32 event, void *data)
> >>  	dev_dbg(pmf_dev->dev, "Pending request (preq1): 0x%x\n", pmf_dev->req1.pending_req);
> >>  
> >>  	amd_pmf_handle_early_preq(pmf_dev);
> >> +
> >> +	amd_pmf_custom_bios_inputs_rb(pmf_dev);
> >>  }
> >>  
> >>  static void apmf_event_handler(acpi_handle handle, u32 event, void *data)
> >> diff --git a/drivers/platform/x86/amd/pmf/core.c b/drivers/platform/x86/amd/pmf/core.c
> >> index bc544a4a5266..8b97eba00dd3 100644
> >> --- a/drivers/platform/x86/amd/pmf/core.c
> >> +++ b/drivers/platform/x86/amd/pmf/core.c
> >> @@ -468,6 +468,7 @@ static int amd_pmf_probe(struct platform_device *pdev)
> >>  	mutex_init(&dev->lock);
> >>  	mutex_init(&dev->update_mutex);
> >>  	mutex_init(&dev->cb_mutex);
> >> +	mutex_init(&dev->cbi_mutex);
> >>  
> >>  	apmf_acpi_init(dev);
> >>  	platform_set_drvdata(pdev, dev);
> >> @@ -494,6 +495,7 @@ static void amd_pmf_remove(struct platform_device *pdev)
> >>  	mutex_destroy(&dev->lock);
> >>  	mutex_destroy(&dev->update_mutex);
> >>  	mutex_destroy(&dev->cb_mutex);
> >> +	mutex_destroy(&dev->cbi_mutex);
> >>  }
> >>  
> >>  static const struct attribute_group *amd_pmf_driver_groups[] = {
> >> diff --git a/drivers/platform/x86/amd/pmf/pmf.h b/drivers/platform/x86/amd/pmf/pmf.h
> >> index 2145df4128cd..6cb1e228c48e 100644
> >> --- a/drivers/platform/x86/amd/pmf/pmf.h
> >> +++ b/drivers/platform/x86/amd/pmf/pmf.h
> >> @@ -12,6 +12,7 @@
> >>  #define PMF_H
> >>  
> >>  #include <linux/acpi.h>
> >> +#include <linux/circ_buf.h>
> >>  #include <linux/input.h>
> >>  #include <linux/platform_device.h>
> >>  #include <linux/platform_profile.h>
> >> @@ -120,6 +121,7 @@ struct cookie_header {
> >>  #define APTS_MAX_STATES		16
> >>  #define CUSTOM_BIOS_INPUT_BITS	GENMASK(16, 7)
> >>  #define BIOS_INPUTS_MAX		10
> >> +#define CUSTOM_BIOS_INPUT_RB_SZ	64	/* Must be power of two for CIRC_* macros */
> > 
> > This is not really size but there are n entries. The total size will be 
> > something else. I'd call it NUM or ENTRIES, or something like that. 
> > Perhaps CUSTOM_BIOS_INPUT_RING_ENTRIES but that is a bit to the long side 
> > of things?
> > 
> >>  typedef void (*apmf_event_handler_t)(acpi_handle handle, u32 event, void *data);
> >>  
> >> @@ -359,6 +361,22 @@ struct pmf_bios_inputs_prev {
> >>  	u32 custom_bios_inputs[BIOS_INPUTS_MAX];
> >>  };
> >>  
> >> +/**
> >> + * struct bios_input_entry - Snapshot of custom BIOS input event
> >> + * @val: Array of custom BIOS input values
> >> + * @preq: Pending request value associated with this event
> >> + */
> >> +struct bios_input_entry {
> >> +	u32 val[BIOS_INPUTS_MAX];
> >> +	u32 preq;
> >> +};
> >> +
> >> +struct cbi_ring_buffer {
> >> +	struct bios_input_entry data[CUSTOM_BIOS_INPUT_RB_SZ];
> >> +	int head;
> >> +	int tail;
> >> +};
> >> +
> >>  struct amd_pmf_dev {
> >>  	void __iomem *regbase;
> >>  	void __iomem *smu_virt_addr;
> >> @@ -407,6 +425,8 @@ struct amd_pmf_dev {
> >>  	struct apmf_sbios_req_v1 req1;
> >>  	struct pmf_bios_inputs_prev cb_prev; /* To preserve custom BIOS inputs */
> >>  	bool cb_flag;			     /* To handle first custom BIOS input */
> >> +	struct cbi_ring_buffer cbi_buf;
> >> +	struct mutex cbi_mutex;		     /* Protects ring buffer access */
> > 
> > Include missing?
> 
> yes..
> 
> > 
> >>  };
> >>  
> >>  struct apmf_sps_prop_granular_v2 {
> >> diff --git a/drivers/platform/x86/amd/pmf/spc.c b/drivers/platform/x86/amd/pmf/spc.c
> >> index 85192c7536b8..669680ce580a 100644
> >> --- a/drivers/platform/x86/amd/pmf/spc.c
> >> +++ b/drivers/platform/x86/amd/pmf/spc.c
> >> @@ -150,12 +150,23 @@ static void amd_pmf_update_bios_inputs(struct amd_pmf_dev *pdev, u32 pending_req
> >>  static void amd_pmf_get_custom_bios_inputs(struct amd_pmf_dev *pdev,
> >>  					   struct ta_pmf_enact_table *in)
> >>  {
> >> +	struct cbi_ring_buffer *rb = &pdev->cbi_buf;
> >> +	struct bios_input_entry entry = { };
> >>  	unsigned int i;
> >>  
> >> +	guard(mutex)(&pdev->cbi_mutex);
> >> +
> >>  	for (i = 0; i < ARRAY_SIZE(custom_bios_inputs); i++)
> >>  		amd_pmf_set_ta_custom_bios_input(in, i, pdev->cb_prev.custom_bios_inputs[i]);
> >>  
> >> -	if (!(pdev->req.pending_req || pdev->req1.pending_req))
> >> +	if (CIRC_CNT(rb->head, rb->tail, CUSTOM_BIOS_INPUT_RB_SZ) == 0)
> > 
> > Include missing.
> 
> circ_buf.h is included in pmf.h since acpi.c and spc.c are using it.
> 
> > 
> > As a general comment unrelated to this patch, these files to be missing 
> > other headers too.
> 
> Yeah. I get your point. Let me add all the relavent missing headers
> and respin a new version.
> 
> ack to other comments in the patch.
> 
> Thanks,
> Shyam
> > 
> >> +		return;	/* return if ring buffer is empty */
> >> +
> >> +	entry = rb->data[rb->tail];
> >> +	rb->tail = (rb->tail + 1) & (CUSTOM_BIOS_INPUT_RB_SZ - 1);
> >> +
> >> +	/* If no active custom BIOS input pending request, do not consume further work */
> >> +	if (!entry.preq)
> >>  		return;
> >>  
> >>  	if (!pdev->smart_pc_enabled)
> >> @@ -165,20 +176,14 @@ static void amd_pmf_get_custom_bios_inputs(struct amd_pmf_dev *pdev,
> >>  	case PMF_IF_V1:
> >>  		if (!is_apmf_bios_input_notifications_supported(pdev))
> >>  			return;
> >> -		amd_pmf_update_bios_inputs(pdev, pdev->req1.pending_req, custom_bios_inputs_v1,
> >> -					   pdev->req1.custom_policy, in);
> >> +		amd_pmf_update_bios_inputs(pdev, entry.preq, custom_bios_inputs_v1, entry.val, in);
> >>  		break;
> >>  	case PMF_IF_V2:
> >> -		amd_pmf_update_bios_inputs(pdev, pdev->req.pending_req, custom_bios_inputs,
> >> -					   pdev->req.custom_policy, in);
> >> +		amd_pmf_update_bios_inputs(pdev, entry.preq, custom_bios_inputs, entry.val, in);
> > 
> > Could we pass the entry instead now?
> > 
> >>  		break;
> >>  	default:
> >>  		break;
> >>  	}
> >> -
> >> -	/* Clear pending requests after handling */
> >> -	memset(&pdev->req, 0, sizeof(pdev->req));
> >> -	memset(&pdev->req1, 0, sizeof(pdev->req1));
> >>  }
> >>  
> >>  static void amd_pmf_get_c0_residency(u16 *core_res, size_t size, struct ta_pmf_enact_table *in)
> >> diff --git a/drivers/platform/x86/amd/pmf/tee-if.c b/drivers/platform/x86/amd/pmf/tee-if.c
> >> index 6e8116bef4f6..add742e33e1e 100644
> >> --- a/drivers/platform/x86/amd/pmf/tee-if.c
> >> +++ b/drivers/platform/x86/amd/pmf/tee-if.c
> >> @@ -579,6 +579,8 @@ int amd_pmf_init_smart_pc(struct amd_pmf_dev *dev)
> >>  		status = ret == TA_PMF_TYPE_SUCCESS;
> >>  		if (status) {
> >>  			dev->cb_flag = true;
> >> +			dev->cbi_buf.head = 0;
> >> +			dev->cbi_buf.tail = 0;
> >>  			break;
> >>  		}
> >>  		amd_pmf_tee_deinit(dev);
> >>
> > 
>