drm/xe/pcode: Treat pcode as per-tile rather than per-GT

There's only one instance of the pcode per tile, and for GT-related
accesses both the primary and media GT share the same register
interface.  Since Xe was using per-GT locking, the pcode mutex wasn't
actually protecting everything that it should since concurrent accesses
related to a tile's primary GT and media GT were possible.

Fixes: dd08ebf6 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240829220619.789159-5-matthew.d.roper@intel.com
(cherry picked from commit 3034cc81)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

drivers/gpu/drm/xe/compat-i915-headers/intel_pcode.h

@@ -13,7 +13,7 @@ static inline int
 snb_pcode_write_timeout(struct intel_uncore *uncore, u32 mbox, u32 val,
 			int fast_timeout_us, int slow_timeout_ms)
 {
-	return xe_pcode_write_timeout(__compat_uncore_to_gt(uncore), mbox, val,
+	return xe_pcode_write_timeout(__compat_uncore_to_tile(uncore), mbox, val,
 				      slow_timeout_ms ?: 1);
 }
 
@@ -21,13 +21,13 @@ static inline int
 snb_pcode_write(struct intel_uncore *uncore, u32 mbox, u32 val)
 {
 
-	return xe_pcode_write(__compat_uncore_to_gt(uncore), mbox, val);
+	return xe_pcode_write(__compat_uncore_to_tile(uncore), mbox, val);
 }
 
 static inline int
 snb_pcode_read(struct intel_uncore *uncore, u32 mbox, u32 *val, u32 *val1)
 {
-	return xe_pcode_read(__compat_uncore_to_gt(uncore), mbox, val, val1);
+	return xe_pcode_read(__compat_uncore_to_tile(uncore), mbox, val, val1);
 }
 
 static inline int
@@ -35,7 +35,7 @@ skl_pcode_request(struct intel_uncore *uncore, u32 mbox,
 		  u32 request, u32 reply_mask, u32 reply,
 		  int timeout_base_ms)
 {
-	return xe_pcode_request(__compat_uncore_to_gt(uncore), mbox, request, reply_mask, reply,
+	return xe_pcode_request(__compat_uncore_to_tile(uncore), mbox, request, reply_mask, reply,
 				timeout_base_ms);
 }
 

drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h

@@ -17,6 +17,13 @@ static inline struct xe_gt *__compat_uncore_to_gt(struct intel_uncore *uncore)
 	return xe_root_mmio_gt(xe);
 }
 
+static inline struct xe_tile *__compat_uncore_to_tile(struct intel_uncore *uncore)
+{
+	struct xe_device *xe = container_of(uncore, struct xe_device, uncore);
+
+	return xe_device_get_root_tile(xe);
+}
+
 static inline u32 intel_uncore_read(struct intel_uncore *uncore,
 				    i915_reg_t i915_reg)
 {

drivers/gpu/drm/xe/xe_device_types.h

@@ -203,6 +203,12 @@ struct xe_tile {
 		} vf;
 	} sriov;
 
+	/** @pcode: tile's PCODE */
+	struct {
+		/** @pcode.lock: protecting tile's PCODE mailbox data */
+		struct mutex lock;
+	} pcode;
+
 	/** @migrate: Migration helper for vram blits and clearing */
 	struct xe_migrate *migrate;
 

drivers/gpu/drm/xe/xe_gt.c

@@ -47,7 +47,6 @@
 #include "xe_migrate.h"
 #include "xe_mmio.h"
 #include "xe_pat.h"
-#include "xe_pcode.h"
 #include "xe_pm.h"
 #include "xe_mocs.h"
 #include "xe_reg_sr.h"
@@ -387,7 +386,6 @@ int xe_gt_init_early(struct xe_gt *gt)
 	xe_tuning_process_gt(gt);
 
 	xe_force_wake_init_gt(gt, gt_to_fw(gt));
-	xe_pcode_init(gt);
 	spin_lock_init(&gt->global_invl_lock);
 
 	return 0;

drivers/gpu/drm/xe/xe_gt_types.h

@@ -310,12 +310,6 @@ struct xe_gt {
 	/** @eclass: per hardware engine class interface on the GT */
 	struct xe_hw_engine_class_intf  eclass[XE_ENGINE_CLASS_MAX];
 
-	/** @pcode: GT's PCODE */
-	struct {
-		/** @pcode.lock: protecting GT's PCODE mailbox data */
-		struct mutex lock;
-	} pcode;
-
 	/** @sysfs: sysfs' kobj used by xe_gt_sysfs */
 	struct kobject *sysfs;
 

drivers/gpu/drm/xe/xe_guc_pc.c

@@ -915,7 +915,7 @@ static void pc_init_pcode_freq(struct xe_guc_pc *pc)
 	u32 min = DIV_ROUND_CLOSEST(pc->rpn_freq, GT_FREQUENCY_MULTIPLIER);
 	u32 max = DIV_ROUND_CLOSEST(pc->rp0_freq, GT_FREQUENCY_MULTIPLIER);
 
-	XE_WARN_ON(xe_pcode_init_min_freq_table(pc_to_gt(pc), min, max));
+	XE_WARN_ON(xe_pcode_init_min_freq_table(gt_to_tile(pc_to_gt(pc)), min, max));
 }
 
 static int pc_init_freqs(struct xe_guc_pc *pc)

drivers/gpu/drm/xe/xe_hwmon.c

@@ -441,14 +441,14 @@ static int xe_hwmon_pcode_read_i1(struct xe_gt *gt, u32 *uval)
 	if (gt_to_xe(gt)->info.platform == XE_DG2)
 		return -ENXIO;
 
-	return xe_pcode_read(gt, PCODE_MBOX(PCODE_POWER_SETUP,
+	return xe_pcode_read(gt_to_tile(gt), PCODE_MBOX(PCODE_POWER_SETUP,
 			     POWER_SETUP_SUBCOMMAND_READ_I1, 0),
 			     uval, NULL);
 }
 
 static int xe_hwmon_pcode_write_i1(struct xe_gt *gt, u32 uval)
 {
-	return xe_pcode_write(gt, PCODE_MBOX(PCODE_POWER_SETUP,
+	return xe_pcode_write(gt_to_tile(gt), PCODE_MBOX(PCODE_POWER_SETUP,
 			      POWER_SETUP_SUBCOMMAND_WRITE_I1, 0),
 			      (uval & POWER_SETUP_I1_DATA_MASK));
 }

drivers/gpu/drm/xe/xe_pcode.c

@@ -12,7 +12,6 @@
 
 #include "xe_assert.h"
 #include "xe_device.h"
-#include "xe_gt.h"
 #include "xe_mmio.h"
 #include "xe_pcode_api.h"
 
@@ -30,7 +29,7 @@
  * - PCODE for display operations
  */
 
-static int pcode_mailbox_status(struct xe_gt *gt)
+static int pcode_mailbox_status(struct xe_tile *tile)
 {
 	u32 err;
 	static const struct pcode_err_decode err_decode[] = {
@@ -45,9 +44,9 @@ static int pcode_mailbox_status(struct xe_gt *gt)
 		[PCODE_ERROR_MASK] = {-EPROTO, "Unknown"},
 	};
 
-	err = xe_mmio_read32(gt, PCODE_MAILBOX) & PCODE_ERROR_MASK;
+	err = xe_mmio_read32(tile->primary_gt, PCODE_MAILBOX) & PCODE_ERROR_MASK;
 	if (err) {
-		drm_err(&gt_to_xe(gt)->drm, "PCODE Mailbox failed: %d %s", err,
+		drm_err(&tile_to_xe(tile)->drm, "PCODE Mailbox failed: %d %s", err,
 			err_decode[err].str ?: "Unknown");
 		return err_decode[err].errno ?: -EPROTO;
 	}
@@ -55,84 +54,85 @@ static int pcode_mailbox_status(struct xe_gt *gt)
 	return 0;
 }
 
-static int __pcode_mailbox_rw(struct xe_gt *gt, u32 mbox, u32 *data0, u32 *data1,
+static int __pcode_mailbox_rw(struct xe_tile *tile, u32 mbox, u32 *data0, u32 *data1,
 			      unsigned int timeout_ms, bool return_data,
 			      bool atomic)
 {
+	struct xe_gt *mmio = tile->primary_gt;
 	int err;
 
-	if (gt_to_xe(gt)->info.skip_pcode)
+	if (tile_to_xe(tile)->info.skip_pcode)
 		return 0;
 
-	if ((xe_mmio_read32(gt, PCODE_MAILBOX) & PCODE_READY) != 0)
+	if ((xe_mmio_read32(mmio, PCODE_MAILBOX) & PCODE_READY) != 0)
 		return -EAGAIN;
 
-	xe_mmio_write32(gt, PCODE_DATA0, *data0);
-	xe_mmio_write32(gt, PCODE_DATA1, data1 ? *data1 : 0);
-	xe_mmio_write32(gt, PCODE_MAILBOX, PCODE_READY | mbox);
+	xe_mmio_write32(mmio, PCODE_DATA0, *data0);
+	xe_mmio_write32(mmio, PCODE_DATA1, data1 ? *data1 : 0);
+	xe_mmio_write32(mmio, PCODE_MAILBOX, PCODE_READY | mbox);
 
-	err = xe_mmio_wait32(gt, PCODE_MAILBOX, PCODE_READY, 0,
+	err = xe_mmio_wait32(mmio, PCODE_MAILBOX, PCODE_READY, 0,
 			     timeout_ms * USEC_PER_MSEC, NULL, atomic);
 	if (err)
 		return err;
 
 	if (return_data) {
-		*data0 = xe_mmio_read32(gt, PCODE_DATA0);
+		*data0 = xe_mmio_read32(mmio, PCODE_DATA0);
 		if (data1)
-			*data1 = xe_mmio_read32(gt, PCODE_DATA1);
+			*data1 = xe_mmio_read32(mmio, PCODE_DATA1);
 	}
 
-	return pcode_mailbox_status(gt);
+	return pcode_mailbox_status(tile);
 }
 
-static int pcode_mailbox_rw(struct xe_gt *gt, u32 mbox, u32 *data0, u32 *data1,
+static int pcode_mailbox_rw(struct xe_tile *tile, u32 mbox, u32 *data0, u32 *data1,
 			    unsigned int timeout_ms, bool return_data,
 			    bool atomic)
 {
-	if (gt_to_xe(gt)->info.skip_pcode)
+	if (tile_to_xe(tile)->info.skip_pcode)
 		return 0;
 
-	lockdep_assert_held(&gt->pcode.lock);
+	lockdep_assert_held(&tile->pcode.lock);
 
-	return __pcode_mailbox_rw(gt, mbox, data0, data1, timeout_ms, return_data, atomic);
+	return __pcode_mailbox_rw(tile, mbox, data0, data1, timeout_ms, return_data, atomic);
 }
 
-int xe_pcode_write_timeout(struct xe_gt *gt, u32 mbox, u32 data, int timeout)
+int xe_pcode_write_timeout(struct xe_tile *tile, u32 mbox, u32 data, int timeout)
 {
 	int err;
 
-	mutex_lock(&gt->pcode.lock);
-	err = pcode_mailbox_rw(gt, mbox, &data, NULL, timeout, false, false);
-	mutex_unlock(&gt->pcode.lock);
+	mutex_lock(&tile->pcode.lock);
+	err = pcode_mailbox_rw(tile, mbox, &data, NULL, timeout, false, false);
+	mutex_unlock(&tile->pcode.lock);
 
 	return err;
 }
 
-int xe_pcode_read(struct xe_gt *gt, u32 mbox, u32 *val, u32 *val1)
+int xe_pcode_read(struct xe_tile *tile, u32 mbox, u32 *val, u32 *val1)
 {
 	int err;
 
-	mutex_lock(&gt->pcode.lock);
-	err = pcode_mailbox_rw(gt, mbox, val, val1, 1, true, false);
-	mutex_unlock(&gt->pcode.lock);
+	mutex_lock(&tile->pcode.lock);
+	err = pcode_mailbox_rw(tile, mbox, val, val1, 1, true, false);
+	mutex_unlock(&tile->pcode.lock);
 
 	return err;
 }
 
-static int pcode_try_request(struct xe_gt *gt, u32 mbox,
+static int pcode_try_request(struct xe_tile *tile, u32 mbox,
 			     u32 request, u32 reply_mask, u32 reply,
 			     u32 *status, bool atomic, int timeout_us, bool locked)
 {
 	int slept, wait = 10;
 
-	xe_gt_assert(gt, timeout_us > 0);
+	xe_tile_assert(tile, timeout_us > 0);
 
 	for (slept = 0; slept < timeout_us; slept += wait) {
 		if (locked)
-			*status = pcode_mailbox_rw(gt, mbox, &request, NULL, 1, true,
+			*status = pcode_mailbox_rw(tile, mbox, &request, NULL, 1, true,
 						   atomic);
 		else
-			*status = __pcode_mailbox_rw(gt, mbox, &request, NULL, 1, true,
+			*status = __pcode_mailbox_rw(tile, mbox, &request, NULL, 1, true,
 						     atomic);
 		if ((*status == 0) && ((request & reply_mask) == reply))
 			return 0;
@@ -149,7 +149,7 @@ static int pcode_try_request(struct xe_gt *gt, u32 mbox,
 
 /**
  * xe_pcode_request - send PCODE request until acknowledgment
- * @gt: gt
+ * @tile: tile
  * @mbox: PCODE mailbox ID the request is targeted for
  * @request: request ID
  * @reply_mask: mask used to check for request acknowledgment
@@ -166,17 +166,17 @@ static int pcode_try_request(struct xe_gt *gt, u32 mbox,
  * Returns 0 on success, %-ETIMEDOUT in case of a timeout, <0 in case of some
  * other error as reported by PCODE.
  */
-int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
+int xe_pcode_request(struct xe_tile *tile, u32 mbox, u32 request,
 		     u32 reply_mask, u32 reply, int timeout_base_ms)
 {
 	u32 status;
 	int ret;
 
-	xe_gt_assert(gt, timeout_base_ms <= 3);
+	xe_tile_assert(tile, timeout_base_ms <= 3);
 
-	mutex_lock(&gt->pcode.lock);
+	mutex_lock(&tile->pcode.lock);
 
-	ret = pcode_try_request(gt, mbox, request, reply_mask, reply, &status,
+	ret = pcode_try_request(tile, mbox, request, reply_mask, reply, &status,
 				false, timeout_base_ms * 1000, true);
 	if (!ret)
 		goto out;
@@ -191,20 +191,20 @@ int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
 	 * requests, and for any quirks of the PCODE firmware that delays
 	 * the request completion.
 	 */
-	drm_err(&gt_to_xe(gt)->drm,
+	drm_err(&tile_to_xe(tile)->drm,
 		"PCODE timeout, retrying with preemption disabled\n");
 	preempt_disable();
-	ret = pcode_try_request(gt, mbox, request, reply_mask, reply, &status,
+	ret = pcode_try_request(tile, mbox, request, reply_mask, reply, &status,
 				true, 50 * 1000, true);
 	preempt_enable();
 
 out:
-	mutex_unlock(&gt->pcode.lock);
+	mutex_unlock(&tile->pcode.lock);
 	return status ? status : ret;
 }
 /**
  * xe_pcode_init_min_freq_table - Initialize PCODE's QOS frequency table
- * @gt: gt instance
+ * @tile: tile instance
  * @min_gt_freq: Minimal (RPn) GT frequency in units of 50MHz.
  * @max_gt_freq: Maximal (RP0) GT frequency in units of 50MHz.
  *
@@ -227,30 +227,30 @@ int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
  * - -EACCES, "PCODE Rejected"
  * - -EPROTO, "Unknown"
  */
-int xe_pcode_init_min_freq_table(struct xe_gt *gt, u32 min_gt_freq,
+int xe_pcode_init_min_freq_table(struct xe_tile *tile, u32 min_gt_freq,
 				 u32 max_gt_freq)
 {
 	int ret;
 	u32 freq;
 
-	if (!gt_to_xe(gt)->info.has_llc)
+	if (!tile_to_xe(tile)->info.has_llc)
 		return 0;
 
 	if (max_gt_freq <= min_gt_freq)
 		return -EINVAL;
 
-	mutex_lock(&gt->pcode.lock);
+	mutex_lock(&tile->pcode.lock);
 	for (freq = min_gt_freq; freq <= max_gt_freq; freq++) {
 		u32 data = freq << PCODE_FREQ_RING_RATIO_SHIFT | freq;
 
-		ret = pcode_mailbox_rw(gt, PCODE_WRITE_MIN_FREQ_TABLE,
+		ret = pcode_mailbox_rw(tile, PCODE_WRITE_MIN_FREQ_TABLE,
 				       &data, NULL, 1, false, false);
 		if (ret)
 			goto unlock;
 	}
 
 unlock:
-	mutex_unlock(&gt->pcode.lock);
+	mutex_unlock(&tile->pcode.lock);
 	return ret;
 }
 
@@ -270,7 +270,7 @@ int xe_pcode_init_min_freq_table(struct xe_gt *gt, u32 min_gt_freq,
 int xe_pcode_ready(struct xe_device *xe, bool locked)
 {
 	u32 status, request = DGFX_GET_INIT_STATUS;
-	struct xe_gt *gt = xe_root_mmio_gt(xe);
+	struct xe_tile *tile = xe_device_get_root_tile(xe);
 	int timeout_us = 180000000; /* 3 min */
 	int ret;
 
@@ -281,15 +281,15 @@ int xe_pcode_ready(struct xe_device *xe, bool locked)
 		return 0;
 
 	if (locked)
-		mutex_lock(&gt->pcode.lock);
+		mutex_lock(&tile->pcode.lock);
 
-	ret = pcode_try_request(gt, DGFX_PCODE_STATUS, request,
+	ret = pcode_try_request(tile, DGFX_PCODE_STATUS, request,
 				DGFX_INIT_STATUS_COMPLETE,
 				DGFX_INIT_STATUS_COMPLETE,
 				&status, false, timeout_us, locked);
 
 	if (locked)
-		mutex_unlock(&gt->pcode.lock);
+		mutex_unlock(&tile->pcode.lock);
 
 	if (ret)
 		drm_err(&xe->drm,
@@ -300,14 +300,14 @@ int xe_pcode_ready(struct xe_device *xe, bool locked)
 
 /**
  * xe_pcode_init: initialize components of PCODE
- * @gt: gt instance
+ * @tile: tile instance
  *
  * This function initializes the xe_pcode component.
  * To be called once only during probe.
  */
-void xe_pcode_init(struct xe_gt *gt)
+void xe_pcode_init(struct xe_tile *tile)
 {
-	drmm_mutex_init(&gt_to_xe(gt)->drm, &gt->pcode.lock);
+	drmm_mutex_init(&tile_to_xe(tile)->drm, &tile->pcode.lock);
 }
 
 /**

drivers/gpu/drm/xe/xe_pcode.h

@@ -7,21 +7,21 @@
 #define _XE_PCODE_H_
 
 #include <linux/types.h>
-struct xe_gt;
+struct xe_tile;
 struct xe_device;
 
-void xe_pcode_init(struct xe_gt *gt);
+void xe_pcode_init(struct xe_tile *tile);
 int xe_pcode_probe_early(struct xe_device *xe);
 int xe_pcode_ready(struct xe_device *xe, bool locked);
-int xe_pcode_init_min_freq_table(struct xe_gt *gt, u32 min_gt_freq,
+int xe_pcode_init_min_freq_table(struct xe_tile *tile, u32 min_gt_freq,
 				 u32 max_gt_freq);
-int xe_pcode_read(struct xe_gt *gt, u32 mbox, u32 *val, u32 *val1);
-int xe_pcode_write_timeout(struct xe_gt *gt, u32 mbox, u32 val,
+int xe_pcode_read(struct xe_tile *tile, u32 mbox, u32 *val, u32 *val1);
+int xe_pcode_write_timeout(struct xe_tile *tile, u32 mbox, u32 val,
 			   int timeout_ms);
-#define xe_pcode_write(gt, mbox, val) \
-	xe_pcode_write_timeout(gt, mbox, val, 1)
+#define xe_pcode_write(tile, mbox, val) \
+	xe_pcode_write_timeout(tile, mbox, val, 1)
 
-int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
+int xe_pcode_request(struct xe_tile *tile, u32 mbox, u32 request,
 		     u32 reply_mask, u32 reply, int timeout_ms);
 
 #define PCODE_MBOX(mbcmd, param1, param2)\

drivers/gpu/drm/xe/xe_tile.c

@@ -9,6 +9,7 @@
 #include "xe_ggtt.h"
 #include "xe_gt.h"
 #include "xe_migrate.h"
+#include "xe_pcode.h"
 #include "xe_sa.h"
 #include "xe_tile.h"
 #include "xe_tile_sysfs.h"
@@ -124,6 +125,8 @@ int xe_tile_init_early(struct xe_tile *tile, struct xe_device *xe, u8 id)
 	if (IS_ERR(tile->primary_gt))
 		return PTR_ERR(tile->primary_gt);
 
+	xe_pcode_init(tile);
+
 	return 0;
 }
 

drivers/gpu/drm/xe/xe_vram_freq.c

@@ -34,7 +34,6 @@ static ssize_t max_freq_show(struct device *dev, struct device_attribute *attr,
 			     char *buf)
 {
 	struct xe_tile *tile = dev_to_tile(dev);
-	struct xe_gt *gt = tile->primary_gt;
 	u32 val, mbox;
 	int err;
 
@@ -42,7 +41,7 @@ static ssize_t max_freq_show(struct device *dev, struct device_attribute *attr,
 		| REG_FIELD_PREP(PCODE_MB_PARAM1, PCODE_MBOX_FC_SC_READ_FUSED_P0)
 		| REG_FIELD_PREP(PCODE_MB_PARAM2, PCODE_MBOX_DOMAIN_HBM);
 
-	err = xe_pcode_read(gt, mbox, &val, NULL);
+	err = xe_pcode_read(tile, mbox, &val, NULL);
 	if (err)
 		return err;
 
@@ -57,7 +56,6 @@ static ssize_t min_freq_show(struct device *dev, struct device_attribute *attr,
 			     char *buf)
 {
 	struct xe_tile *tile = dev_to_tile(dev);
-	struct xe_gt *gt = tile->primary_gt;
 	u32 val, mbox;
 	int err;
 
@@ -65,7 +63,7 @@ static ssize_t min_freq_show(struct device *dev, struct device_attribute *attr,
 		| REG_FIELD_PREP(PCODE_MB_PARAM1, PCODE_MBOX_FC_SC_READ_FUSED_PN)
 		| REG_FIELD_PREP(PCODE_MB_PARAM2, PCODE_MBOX_DOMAIN_HBM);
 
-	err = xe_pcode_read(gt, mbox, &val, NULL);
+	err = xe_pcode_read(tile, mbox, &val, NULL);
 	if (err)
 		return err;