[PATCH v2 1/9] drm/msm: add msm_gem_get_and_pin_iova_range

[PATCH v2 1/9] drm/msm: add msm_gem_get_and_pin_iova_range

[Jonathan Marek]

This function allows pinning iova to a specific page range (for a6xx GMU).

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/msm_drv.h     |  6 +++++-
 drivers/gpu/drm/msm/msm_gem.c     | 28 +++++++++++++++++++++-------
 drivers/gpu/drm/msm/msm_gem_vma.c |  6 ++++--
 3 files changed, 30 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/msm/msm_drv.h b/drivers/gpu/drm/msm/msm_drv.h
index 194d900a460e..966fd9068c94 100644
--- a/drivers/gpu/drm/msm/msm_drv.h
+++ b/drivers/gpu/drm/msm/msm_drv.h
@@ -236,7 +236,8 @@ int msm_crtc_enable_vblank(struct drm_crtc *crtc);
 void msm_crtc_disable_vblank(struct drm_crtc *crtc);
 
 int msm_gem_init_vma(struct msm_gem_address_space *aspace,
-		struct msm_gem_vma *vma, int npages);
+		struct msm_gem_vma *vma, int npages,
+		u64 range_start, u64 range_end);
 void msm_gem_purge_vma(struct msm_gem_address_space *aspace,
 		struct msm_gem_vma *vma);
 void msm_gem_unmap_vma(struct msm_gem_address_space *aspace,
@@ -276,6 +277,9 @@ vm_fault_t msm_gem_fault(struct vm_fault *vmf);
 uint64_t msm_gem_mmap_offset(struct drm_gem_object *obj);
 int msm_gem_get_iova(struct drm_gem_object *obj,
 		struct msm_gem_address_space *aspace, uint64_t *iova);
+int msm_gem_get_and_pin_iova_range(struct drm_gem_object *obj,
+		struct msm_gem_address_space *aspace, uint64_t *iova,
+		u64 range_start, u64 range_end);
 int msm_gem_get_and_pin_iova(struct drm_gem_object *obj,
 		struct msm_gem_address_space *aspace, uint64_t *iova);
 uint64_t msm_gem_iova(struct drm_gem_object *obj,
diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index 5a6a79fbc9d6..d8f56a34c117 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -389,7 +389,8 @@ put_iova(struct drm_gem_object *obj)
 }
 
 static int msm_gem_get_iova_locked(struct drm_gem_object *obj,
-		struct msm_gem_address_space *aspace, uint64_t *iova)
+		struct msm_gem_address_space *aspace, uint64_t *iova,
+		u64 range_start, u64 range_end)
 {
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
 	struct msm_gem_vma *vma;
@@ -404,7 +405,8 @@ static int msm_gem_get_iova_locked(struct drm_gem_object *obj,
 		if (IS_ERR(vma))
 			return PTR_ERR(vma);
 
-		ret = msm_gem_init_vma(aspace, vma, obj->size >> PAGE_SHIFT);
+		ret = msm_gem_init_vma(aspace, vma, obj->size >> PAGE_SHIFT,
+			range_start, range_end);
 		if (ret) {
 			del_vma(vma);
 			return ret;
@@ -443,9 +445,13 @@ static int msm_gem_pin_iova(struct drm_gem_object *obj,
 			msm_obj->sgt, obj->size >> PAGE_SHIFT);
 }
 
-/* get iova and pin it. Should have a matching put */
-int msm_gem_get_and_pin_iova(struct drm_gem_object *obj,
-		struct msm_gem_address_space *aspace, uint64_t *iova)
+/*
+ * get iova and pin it. Should have a matching put
+ * limits iova to specified range (in pages)
+ */
+int msm_gem_get_and_pin_iova_range(struct drm_gem_object *obj,
+		struct msm_gem_address_space *aspace, uint64_t *iova,
+		u64 range_start, u64 range_end)
 {
 	struct msm_gem_object *msm_obj = to_msm_bo(obj);
 	u64 local;
@@ -453,7 +459,8 @@ int msm_gem_get_and_pin_iova(struct drm_gem_object *obj,
 
 	mutex_lock(&msm_obj->lock);
 
-	ret = msm_gem_get_iova_locked(obj, aspace, &local);
+	ret = msm_gem_get_iova_locked(obj, aspace, &local,
+		range_start, range_end);
 
 	if (!ret)
 		ret = msm_gem_pin_iova(obj, aspace);
@@ -465,6 +472,13 @@ int msm_gem_get_and_pin_iova(struct drm_gem_object *obj,
 	return ret;
 }
 
+/* get iova and pin it. Should have a matching put */
+int msm_gem_get_and_pin_iova(struct drm_gem_object *obj,
+		struct msm_gem_address_space *aspace, uint64_t *iova)
+{
+	return msm_gem_get_and_pin_iova_range(obj, aspace, iova, 0, U64_MAX);
+}
+
 /*
  * Get an iova but don't pin it. Doesn't need a put because iovas are currently
  * valid for the life of the object
@@ -476,7 +490,7 @@ int msm_gem_get_iova(struct drm_gem_object *obj,
 	int ret;
 
 	mutex_lock(&msm_obj->lock);
-	ret = msm_gem_get_iova_locked(obj, aspace, iova);
+	ret = msm_gem_get_iova_locked(obj, aspace, iova, 0, U64_MAX);
 	mutex_unlock(&msm_obj->lock);
 
 	return ret;
diff --git a/drivers/gpu/drm/msm/msm_gem_vma.c b/drivers/gpu/drm/msm/msm_gem_vma.c
index 1af5354bcd46..407b7ab82818 100644
--- a/drivers/gpu/drm/msm/msm_gem_vma.c
+++ b/drivers/gpu/drm/msm/msm_gem_vma.c
@@ -103,7 +103,8 @@ void msm_gem_close_vma(struct msm_gem_address_space *aspace,
 
 /* Initialize a new vma and allocate an iova for it */
 int msm_gem_init_vma(struct msm_gem_address_space *aspace,
-		struct msm_gem_vma *vma, int npages)
+		struct msm_gem_vma *vma, int npages,
+		u64 range_start, u64 range_end)
 {
 	int ret;
 
@@ -111,7 +112,8 @@ int msm_gem_init_vma(struct msm_gem_address_space *aspace,
 		return -EBUSY;
 
 	spin_lock(&aspace->lock);
-	ret = drm_mm_insert_node(&aspace->mm, &vma->node, npages);
+	ret = drm_mm_insert_node_in_range(&aspace->mm, &vma->node, npages, 0,
+		0, range_start, range_end, 0);
 	spin_unlock(&aspace->lock);
 
 	if (ret)
-- 
2.26.1

[PATCH v2 2/9] drm/msm: add internal MSM_BO_MAP_PRIV flag

[Jonathan Marek]

This flag sets IOMMU_PRIV, which is required for some a6xx GMU objects.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/msm_gem.c | 3 +++
 drivers/gpu/drm/msm/msm_gem.h | 1 +
 2 files changed, 4 insertions(+)

diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
index d8f56a34c117..6277fde13df9 100644
--- a/drivers/gpu/drm/msm/msm_gem.c
+++ b/drivers/gpu/drm/msm/msm_gem.c
@@ -428,6 +428,9 @@ static int msm_gem_pin_iova(struct drm_gem_object *obj,
 	if (!(msm_obj->flags & MSM_BO_GPU_READONLY))
 		prot |= IOMMU_WRITE;
 
+	if (msm_obj->flags & MSM_BO_MAP_PRIV)
+		prot |= IOMMU_PRIV;
+
 	WARN_ON(!mutex_is_locked(&msm_obj->lock));
 
 	if (WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED))
diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
index 30584eaf8cc8..972490b14ba5 100644
--- a/drivers/gpu/drm/msm/msm_gem.h
+++ b/drivers/gpu/drm/msm/msm_gem.h
@@ -13,6 +13,7 @@
 
 /* Additional internal-use only BO flags: */
 #define MSM_BO_STOLEN        0x10000000    /* try to use stolen/splash memory */
+#define MSM_BO_MAP_PRIV      0x20000000    /* use IOMMU_PRIV when mapping */
 
 struct msm_gem_address_space {
 	const char *name;
-- 
2.26.1

Re: [Freedreno] [PATCH v2 2/9] drm/msm: add internal MSM_BO_MAP_PRIV flag

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:20PM -0400, Jonathan Marek wrote:
> This flag sets IOMMU_PRIV, which is required for some a6xx GMU objects.

The GMU doesn't _need_  privileged surfaces, but it sure is a good idea.
As a bonus with the APRIV bit on some a6xx targets we can set all of the
global buffers as privileged so people can't go out and overwrite the memstore
any more.

Reviewed-by: Jordan Crouse <jcrouse@codeauorora.org>

> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/msm_gem.c | 3 +++
>  drivers/gpu/drm/msm/msm_gem.h | 1 +
>  2 files changed, 4 insertions(+)
> 
> diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
> index d8f56a34c117..6277fde13df9 100644
> --- a/drivers/gpu/drm/msm/msm_gem.c
> +++ b/drivers/gpu/drm/msm/msm_gem.c
> @@ -428,6 +428,9 @@ static int msm_gem_pin_iova(struct drm_gem_object *obj,
>  	if (!(msm_obj->flags & MSM_BO_GPU_READONLY))
>  		prot |= IOMMU_WRITE;
>  
> +	if (msm_obj->flags & MSM_BO_MAP_PRIV)
> +		prot |= IOMMU_PRIV;
> +
>  	WARN_ON(!mutex_is_locked(&msm_obj->lock));
>  
>  	if (WARN_ON(msm_obj->madv != MSM_MADV_WILLNEED))
> diff --git a/drivers/gpu/drm/msm/msm_gem.h b/drivers/gpu/drm/msm/msm_gem.h
> index 30584eaf8cc8..972490b14ba5 100644
> --- a/drivers/gpu/drm/msm/msm_gem.h
> +++ b/drivers/gpu/drm/msm/msm_gem.h
> @@ -13,6 +13,7 @@
>  
>  /* Additional internal-use only BO flags: */
>  #define MSM_BO_STOLEN        0x10000000    /* try to use stolen/splash memory */
> +#define MSM_BO_MAP_PRIV      0x20000000    /* use IOMMU_PRIV when mapping */
>  
>  struct msm_gem_address_space {
>  	const char *name;
> -- 
> 2.26.1
> 
> _______________________________________________
> Freedreno mailing list
> Freedreno@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/freedreno

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 3/9] drm/msm/a6xx: use msm_gem for GMU memory objects

[Jonathan Marek]

This gives more fine-grained control over how memory is allocated over the
DMA api. In particular, it allows using an address range or pinning to
a fixed address.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 115 ++++++++++++++++++--------
 drivers/gpu/drm/msm/adreno/a6xx_gmu.h |   9 +-
 drivers/gpu/drm/msm/adreno/a6xx_hfi.c |   6 +-
 3 files changed, 88 insertions(+), 42 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index c4e71abbdd53..c9fd0470a321 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -2,14 +2,16 @@
 /* Copyright (c) 2017-2019 The Linux Foundation. All rights reserved. */
 
 #include <linux/clk.h>
-#include <linux/dma-mapping.h>
 #include <linux/interconnect.h>
 #include <linux/pm_domain.h>
 #include <linux/pm_opp.h>
 #include <soc/qcom/cmd-db.h>
+#include <drm/drm_gem.h>
 
 #include "a6xx_gpu.h"
 #include "a6xx_gmu.xml.h"
+#include "msm_gem.h"
+#include "msm_mmu.h"
 
 static void a6xx_gmu_fault(struct a6xx_gmu *gmu)
 {
@@ -620,7 +622,7 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 	gmu_write(gmu, REG_A6XX_GMU_CM3_BOOT_CONFIG, 0x02);
 
 	/* Write the iova of the HFI table */
-	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_ADDR, gmu->hfi->iova);
+	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_ADDR, gmu->hfi.iova);
 	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_INFO, 1);
 
 	gmu_write(gmu, REG_A6XX_GMU_AHB_FENCE_RANGE_0,
@@ -919,34 +921,77 @@ int a6xx_gmu_stop(struct a6xx_gpu *a6xx_gpu)
 	return 0;
 }
 
-static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo)
+static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
 {
-	if (IS_ERR_OR_NULL(bo))
-		return;
+	msm_gem_kernel_put(gmu->hfi.obj, gmu->aspace, false);
+	msm_gem_kernel_put(gmu->debug.obj, gmu->aspace, false);
+
+	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
+	msm_gem_address_space_put(gmu->aspace);
+}
+
+static int a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo,
+		size_t size, u64 iova)
+{
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct drm_device *dev = a6xx_gpu->base.base.dev;
+	uint32_t flags = MSM_BO_WC;
+	u64 range_start, range_end;
+	int ret;
+
+	size = PAGE_ALIGN(size);
+	if (!iova) {
+		/* no fixed address - use GMU's uncached range */
+		range_start = 0x60000000;
+		range_end = 0x80000000;
+	} else {
+		/* range for fixed address */
+		range_start = iova;
+		range_end = iova + size;
+	}
+
+	bo->obj = msm_gem_new(dev, size, flags);
+	if (IS_ERR(bo->obj))
+		return PTR_ERR(bo->obj);
 
-	dma_free_wc(gmu->dev, bo->size, bo->virt, bo->iova);
-	kfree(bo);
+	ret = msm_gem_get_and_pin_iova_range(bo->obj, gmu->aspace, &bo->iova,
+		range_start >> PAGE_SHIFT, range_end >> PAGE_SHIFT);
+	if (ret) {
+		drm_gem_object_put(bo->obj);
+		return ret;
+	}
+
+	bo->virt = msm_gem_get_vaddr(bo->obj);
+	bo->size = size;
+
+	return 0;
 }
 
-static struct a6xx_gmu_bo *a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu,
-		size_t size)
+static int a6xx_gmu_memory_probe(struct a6xx_gmu *gmu)
 {
-	struct a6xx_gmu_bo *bo;
+	struct iommu_domain *domain;
+	int ret;
 
-	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
-	if (!bo)
-		return ERR_PTR(-ENOMEM);
+	domain = iommu_domain_alloc(&platform_bus_type);
+	if (!domain)
+		return -ENODEV;
 
-	bo->size = PAGE_ALIGN(size);
+	domain->geometry.aperture_start = 0x00000000;
+	domain->geometry.aperture_end = 0x7fffffff;
 
-	bo->virt = dma_alloc_wc(gmu->dev, bo->size, &bo->iova, GFP_KERNEL);
+	gmu->aspace = msm_gem_address_space_create(gmu->dev, domain, "gmu");
+	if (IS_ERR(gmu->aspace)) {
+		iommu_domain_free(domain);
+		return PTR_ERR(gmu->aspace);
+	}
 
-	if (!bo->virt) {
-		kfree(bo);
-		return ERR_PTR(-ENOMEM);
+	ret = gmu->aspace->mmu->funcs->attach(gmu->aspace->mmu);
+	if (ret) {
+		msm_gem_address_space_put(gmu->aspace);
+		return ret;
 	}
 
-	return bo;
+	return 0;
 }
 
 /* Return the 'arc-level' for the given frequency */
@@ -1204,7 +1249,7 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 	iounmap(gmu->mmio);
 	gmu->mmio = NULL;
 
-	a6xx_gmu_memory_free(gmu, gmu->hfi);
+	a6xx_gmu_memory_free(gmu);
 
 	free_irq(gmu->gmu_irq, gmu);
 	free_irq(gmu->hfi_irq, gmu);
@@ -1226,15 +1271,7 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 
 	gmu->dev = &pdev->dev;
 
-	/* Pass force_dma false to require the DT to set the dma region */
-	ret = of_dma_configure(gmu->dev, node, false);
-	if (ret)
-		return ret;
-
-	/* Set the mask after the of_dma_configure() */
-	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(31));
-	if (ret)
-		return ret;
+	of_dma_configure(gmu->dev, node, true);
 
 	/* Fow now, don't do anything fancy until we get our feet under us */
 	gmu->idle_level = GMU_IDLE_STATE_ACTIVE;
@@ -1246,20 +1283,26 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	if (ret)
 		goto err_put_device;
 
+	ret = a6xx_gmu_memory_probe(gmu);
+	if (ret)
+		goto err_put_device;
+
 	/* Allocate memory for for the HFI queues */
-	gmu->hfi = a6xx_gmu_memory_alloc(gmu, SZ_16K);
-	if (IS_ERR(gmu->hfi))
+	ret = a6xx_gmu_memory_alloc(gmu, &gmu->hfi, SZ_16K, 0);
+	if (ret)
 		goto err_memory;
 
 	/* Allocate memory for the GMU debug region */
-	gmu->debug = a6xx_gmu_memory_alloc(gmu, SZ_16K);
-	if (IS_ERR(gmu->debug))
+	ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
+	if (ret)
 		goto err_memory;
 
 	/* Map the GMU registers */
 	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
-	if (IS_ERR(gmu->mmio))
+	if (IS_ERR(gmu->mmio)) {
+		ret = PTR_ERR(gmu->mmio);
 		goto err_memory;
+	}
 
 	/* Get the HFI and GMU interrupts */
 	gmu->hfi_irq = a6xx_gmu_get_irq(gmu, pdev, "hfi", a6xx_hfi_irq);
@@ -1288,11 +1331,11 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	iounmap(gmu->mmio);
 	free_irq(gmu->gmu_irq, gmu);
 	free_irq(gmu->hfi_irq, gmu);
-err_memory:
-	a6xx_gmu_memory_free(gmu, gmu->hfi);
 
 	ret = -ENODEV;
 
+err_memory:
+	a6xx_gmu_memory_free(gmu);
 err_put_device:
 	/* Drop reference taken in of_find_device_by_node */
 	put_device(gmu->dev);
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
index 4af65a36d5ca..cd66a5e1d7e9 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
@@ -10,9 +10,10 @@
 #include "a6xx_hfi.h"
 
 struct a6xx_gmu_bo {
+	struct drm_gem_object *obj;
 	void *virt;
 	size_t size;
-	dma_addr_t iova;
+	u64 iova;
 };
 
 /*
@@ -43,6 +44,8 @@ struct a6xx_gmu_bo {
 struct a6xx_gmu {
 	struct device *dev;
 
+	struct msm_gem_address_space *aspace;
+
 	void * __iomem mmio;
 
 	int hfi_irq;
@@ -52,8 +55,8 @@ struct a6xx_gmu {
 
 	int idle_level;
 
-	struct a6xx_gmu_bo *hfi;
-	struct a6xx_gmu_bo *debug;
+	struct a6xx_gmu_bo hfi;
+	struct a6xx_gmu_bo debug;
 
 	int nr_clocks;
 	struct clk_bulk_data *clocks;
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
index e450e0b97211..b90343d4caf0 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
@@ -176,8 +176,8 @@ static int a6xx_hfi_send_gmu_init(struct a6xx_gmu *gmu, int boot_state)
 {
 	struct a6xx_hfi_msg_gmu_init_cmd msg = { 0 };
 
-	msg.dbg_buffer_addr = (u32) gmu->debug->iova;
-	msg.dbg_buffer_size = (u32) gmu->debug->size;
+	msg.dbg_buffer_addr = (u32) gmu->debug.iova;
+	msg.dbg_buffer_size = (u32) gmu->debug.size;
 	msg.boot_state = boot_state;
 
 	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_INIT, &msg, sizeof(msg),
@@ -385,7 +385,7 @@ static void a6xx_hfi_queue_init(struct a6xx_hfi_queue *queue,
 
 void a6xx_hfi_init(struct a6xx_gmu *gmu)
 {
-	struct a6xx_gmu_bo *hfi = gmu->hfi;
+	struct a6xx_gmu_bo *hfi = &gmu->hfi;
 	struct a6xx_hfi_queue_table_header *table = hfi->virt;
 	struct a6xx_hfi_queue_header *headers = hfi->virt + sizeof(*table);
 	u64 offset;
-- 
2.26.1

Re: [PATCH v2 3/9] drm/msm/a6xx: use msm_gem for GMU memory objects

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:21PM -0400, Jonathan Marek wrote:
> This gives more fine-grained control over how memory is allocated over the
> DMA api. In particular, it allows using an address range or pinning to
> a fixed address.

Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org>

> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.c | 115 ++++++++++++++++++--------
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.h |   9 +-
>  drivers/gpu/drm/msm/adreno/a6xx_hfi.c |   6 +-
>  3 files changed, 88 insertions(+), 42 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> index c4e71abbdd53..c9fd0470a321 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> @@ -2,14 +2,16 @@
>  /* Copyright (c) 2017-2019 The Linux Foundation. All rights reserved. */
>  
>  #include <linux/clk.h>
> -#include <linux/dma-mapping.h>
>  #include <linux/interconnect.h>
>  #include <linux/pm_domain.h>
>  #include <linux/pm_opp.h>
>  #include <soc/qcom/cmd-db.h>
> +#include <drm/drm_gem.h>
>  
>  #include "a6xx_gpu.h"
>  #include "a6xx_gmu.xml.h"
> +#include "msm_gem.h"
> +#include "msm_mmu.h"
>  
>  static void a6xx_gmu_fault(struct a6xx_gmu *gmu)
>  {
> @@ -620,7 +622,7 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  	gmu_write(gmu, REG_A6XX_GMU_CM3_BOOT_CONFIG, 0x02);
>  
>  	/* Write the iova of the HFI table */
> -	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_ADDR, gmu->hfi->iova);
> +	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_ADDR, gmu->hfi.iova);
>  	gmu_write(gmu, REG_A6XX_GMU_HFI_QTBL_INFO, 1);
>  
>  	gmu_write(gmu, REG_A6XX_GMU_AHB_FENCE_RANGE_0,
> @@ -919,34 +921,77 @@ int a6xx_gmu_stop(struct a6xx_gpu *a6xx_gpu)
>  	return 0;
>  }
>  
> -static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo)
> +static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
>  {
> -	if (IS_ERR_OR_NULL(bo))
> -		return;
> +	msm_gem_kernel_put(gmu->hfi.obj, gmu->aspace, false);
> +	msm_gem_kernel_put(gmu->debug.obj, gmu->aspace, false);
> +
> +	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
> +	msm_gem_address_space_put(gmu->aspace);
> +}
> +
> +static int a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo,
> +		size_t size, u64 iova)
> +{
> +	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
> +	struct drm_device *dev = a6xx_gpu->base.base.dev;
> +	uint32_t flags = MSM_BO_WC;
> +	u64 range_start, range_end;
> +	int ret;
> +
> +	size = PAGE_ALIGN(size);
> +	if (!iova) {
> +		/* no fixed address - use GMU's uncached range */
> +		range_start = 0x60000000;
> +		range_end = 0x80000000;
> +	} else {
> +		/* range for fixed address */
> +		range_start = iova;
> +		range_end = iova + size;
> +	}
> +
> +	bo->obj = msm_gem_new(dev, size, flags);
> +	if (IS_ERR(bo->obj))
> +		return PTR_ERR(bo->obj);
>  
> -	dma_free_wc(gmu->dev, bo->size, bo->virt, bo->iova);
> -	kfree(bo);
> +	ret = msm_gem_get_and_pin_iova_range(bo->obj, gmu->aspace, &bo->iova,
> +		range_start >> PAGE_SHIFT, range_end >> PAGE_SHIFT);
> +	if (ret) {
> +		drm_gem_object_put(bo->obj);
> +		return ret;
> +	}
> +
> +	bo->virt = msm_gem_get_vaddr(bo->obj);
> +	bo->size = size;
> +
> +	return 0;
>  }
>  
> -static struct a6xx_gmu_bo *a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu,
> -		size_t size)
> +static int a6xx_gmu_memory_probe(struct a6xx_gmu *gmu)
>  {
> -	struct a6xx_gmu_bo *bo;
> +	struct iommu_domain *domain;
> +	int ret;
>  
> -	bo = kzalloc(sizeof(*bo), GFP_KERNEL);
> -	if (!bo)
> -		return ERR_PTR(-ENOMEM);
> +	domain = iommu_domain_alloc(&platform_bus_type);
> +	if (!domain)
> +		return -ENODEV;
>  
> -	bo->size = PAGE_ALIGN(size);
> +	domain->geometry.aperture_start = 0x00000000;
> +	domain->geometry.aperture_end = 0x7fffffff;
>  
> -	bo->virt = dma_alloc_wc(gmu->dev, bo->size, &bo->iova, GFP_KERNEL);
> +	gmu->aspace = msm_gem_address_space_create(gmu->dev, domain, "gmu");
> +	if (IS_ERR(gmu->aspace)) {
> +		iommu_domain_free(domain);
> +		return PTR_ERR(gmu->aspace);
> +	}
>  
> -	if (!bo->virt) {
> -		kfree(bo);
> -		return ERR_PTR(-ENOMEM);
> +	ret = gmu->aspace->mmu->funcs->attach(gmu->aspace->mmu);
> +	if (ret) {
> +		msm_gem_address_space_put(gmu->aspace);
> +		return ret;
>  	}
>  
> -	return bo;
> +	return 0;
>  }
>  
>  /* Return the 'arc-level' for the given frequency */
> @@ -1204,7 +1249,7 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
>  	iounmap(gmu->mmio);
>  	gmu->mmio = NULL;
>  
> -	a6xx_gmu_memory_free(gmu, gmu->hfi);
> +	a6xx_gmu_memory_free(gmu);
>  
>  	free_irq(gmu->gmu_irq, gmu);
>  	free_irq(gmu->hfi_irq, gmu);
> @@ -1226,15 +1271,7 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  
>  	gmu->dev = &pdev->dev;
>  
> -	/* Pass force_dma false to require the DT to set the dma region */
> -	ret = of_dma_configure(gmu->dev, node, false);
> -	if (ret)
> -		return ret;
> -
> -	/* Set the mask after the of_dma_configure() */
> -	ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(31));
> -	if (ret)
> -		return ret;
> +	of_dma_configure(gmu->dev, node, true);
>  
>  	/* Fow now, don't do anything fancy until we get our feet under us */
>  	gmu->idle_level = GMU_IDLE_STATE_ACTIVE;
> @@ -1246,20 +1283,26 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  	if (ret)
>  		goto err_put_device;
>  
> +	ret = a6xx_gmu_memory_probe(gmu);
> +	if (ret)
> +		goto err_put_device;
> +
>  	/* Allocate memory for for the HFI queues */
> -	gmu->hfi = a6xx_gmu_memory_alloc(gmu, SZ_16K);
> -	if (IS_ERR(gmu->hfi))
> +	ret = a6xx_gmu_memory_alloc(gmu, &gmu->hfi, SZ_16K, 0);
> +	if (ret)
>  		goto err_memory;
>  
>  	/* Allocate memory for the GMU debug region */
> -	gmu->debug = a6xx_gmu_memory_alloc(gmu, SZ_16K);
> -	if (IS_ERR(gmu->debug))
> +	ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
> +	if (ret)
>  		goto err_memory;
>  
>  	/* Map the GMU registers */
>  	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
> -	if (IS_ERR(gmu->mmio))
> +	if (IS_ERR(gmu->mmio)) {
> +		ret = PTR_ERR(gmu->mmio);
>  		goto err_memory;
> +	}
>  
>  	/* Get the HFI and GMU interrupts */
>  	gmu->hfi_irq = a6xx_gmu_get_irq(gmu, pdev, "hfi", a6xx_hfi_irq);
> @@ -1288,11 +1331,11 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  	iounmap(gmu->mmio);
>  	free_irq(gmu->gmu_irq, gmu);
>  	free_irq(gmu->hfi_irq, gmu);
> -err_memory:
> -	a6xx_gmu_memory_free(gmu, gmu->hfi);
>  
>  	ret = -ENODEV;
>  
> +err_memory:
> +	a6xx_gmu_memory_free(gmu);
>  err_put_device:
>  	/* Drop reference taken in of_find_device_by_node */
>  	put_device(gmu->dev);
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> index 4af65a36d5ca..cd66a5e1d7e9 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> @@ -10,9 +10,10 @@
>  #include "a6xx_hfi.h"
>  
>  struct a6xx_gmu_bo {
> +	struct drm_gem_object *obj;
>  	void *virt;
>  	size_t size;
> -	dma_addr_t iova;
> +	u64 iova;
>  };
>  
>  /*
> @@ -43,6 +44,8 @@ struct a6xx_gmu_bo {
>  struct a6xx_gmu {
>  	struct device *dev;
>  
> +	struct msm_gem_address_space *aspace;
> +
>  	void * __iomem mmio;
>  
>  	int hfi_irq;
> @@ -52,8 +55,8 @@ struct a6xx_gmu {
>  
>  	int idle_level;
>  
> -	struct a6xx_gmu_bo *hfi;
> -	struct a6xx_gmu_bo *debug;
> +	struct a6xx_gmu_bo hfi;
> +	struct a6xx_gmu_bo debug;
>  
>  	int nr_clocks;
>  	struct clk_bulk_data *clocks;
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> index e450e0b97211..b90343d4caf0 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> @@ -176,8 +176,8 @@ static int a6xx_hfi_send_gmu_init(struct a6xx_gmu *gmu, int boot_state)
>  {
>  	struct a6xx_hfi_msg_gmu_init_cmd msg = { 0 };
>  
> -	msg.dbg_buffer_addr = (u32) gmu->debug->iova;
> -	msg.dbg_buffer_size = (u32) gmu->debug->size;
> +	msg.dbg_buffer_addr = (u32) gmu->debug.iova;
> +	msg.dbg_buffer_size = (u32) gmu->debug.size;
>  	msg.boot_state = boot_state;
>  
>  	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_INIT, &msg, sizeof(msg),
> @@ -385,7 +385,7 @@ static void a6xx_hfi_queue_init(struct a6xx_hfi_queue *queue,
>  
>  void a6xx_hfi_init(struct a6xx_gmu *gmu)
>  {
> -	struct a6xx_gmu_bo *hfi = gmu->hfi;
> +	struct a6xx_gmu_bo *hfi = &gmu->hfi;
>  	struct a6xx_hfi_queue_table_header *table = hfi->virt;
>  	struct a6xx_hfi_queue_header *headers = hfi->virt + sizeof(*table);
>  	u64 offset;
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 4/9] drm/msm/a6xx: add A640/A650 to gpulist

[Jonathan Marek]

Add Adreno 640 and 650 GPU info to the gpulist.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/adreno_device.c | 24 ++++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.c    |  2 +-
 drivers/gpu/drm/msm/adreno/adreno_gpu.h    | 10 +++++++++
 3 files changed, 35 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/msm/adreno/adreno_device.c b/drivers/gpu/drm/msm/adreno/adreno_device.c
index cb3a6e597d76..1156f72532a4 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_device.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_device.c
@@ -189,6 +189,30 @@ static const struct adreno_info gpulist[] = {
 		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
 		.init = a6xx_gpu_init,
 		.zapfw = "a630_zap.mdt",
+	}, {
+		.rev = ADRENO_REV(6, 4, 0, ANY_ID),
+		.revn = 640,
+		.name = "A640",
+		.fw = {
+			[ADRENO_FW_SQE] = "a630_sqe.fw",
+			[ADRENO_FW_GMU] = "a640_gmu.bin",
+		},
+		.gmem = SZ_1M,
+		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
+		.init = a6xx_gpu_init,
+		.zapfw = "a640_zap.mdt",
+	}, {
+		.rev = ADRENO_REV(6, 5, 0, ANY_ID),
+		.revn = 650,
+		.name = "A650",
+		.fw = {
+			[ADRENO_FW_SQE] = "a650_sqe.fw",
+			[ADRENO_FW_GMU] = "a650_gmu.bin",
+		},
+		.gmem = SZ_1M + SZ_128K,
+		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
+		.init = a6xx_gpu_init,
+		.zapfw = "a650_zap.mdt",
 	},
 };
 
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 1d5c43c22269..a7647eaacc7a 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -197,7 +197,7 @@ int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value)
 		*value = adreno_gpu->gmem;
 		return 0;
 	case MSM_PARAM_GMEM_BASE:
-		*value = 0x100000;
+		*value = !adreno_is_a650(adreno_gpu) ? 0x100000 : 0;
 		return 0;
 	case MSM_PARAM_CHIP_ID:
 		*value = adreno_gpu->rev.patchid |
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index 9ff4e550e7bd..88ae1b2813ef 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -237,6 +237,16 @@ static inline int adreno_is_a630(struct adreno_gpu *gpu)
        return gpu->revn == 630;
 }
 
+static inline int adreno_is_a640(struct adreno_gpu *gpu)
+{
+       return gpu->revn == 640;
+}
+
+static inline int adreno_is_a650(struct adreno_gpu *gpu)
+{
+       return gpu->revn == 650;
+}
+
 int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value);
 const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu,
 		const char *fwname);
-- 
2.26.1

Re: [PATCH v2 4/9] drm/msm/a6xx: add A640/A650 to gpulist

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:22PM -0400, Jonathan Marek wrote:
> Add Adreno 640 and 650 GPU info to the gpulist.
> 

Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org>

> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/adreno_device.c | 24 ++++++++++++++++++++++
>  drivers/gpu/drm/msm/adreno/adreno_gpu.c    |  2 +-
>  drivers/gpu/drm/msm/adreno/adreno_gpu.h    | 10 +++++++++
>  3 files changed, 35 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_device.c b/drivers/gpu/drm/msm/adreno/adreno_device.c
> index cb3a6e597d76..1156f72532a4 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_device.c
> +++ b/drivers/gpu/drm/msm/adreno/adreno_device.c
> @@ -189,6 +189,30 @@ static const struct adreno_info gpulist[] = {
>  		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
>  		.init = a6xx_gpu_init,
>  		.zapfw = "a630_zap.mdt",
> +	}, {
> +		.rev = ADRENO_REV(6, 4, 0, ANY_ID),
> +		.revn = 640,
> +		.name = "A640",
> +		.fw = {
> +			[ADRENO_FW_SQE] = "a630_sqe.fw",
> +			[ADRENO_FW_GMU] = "a640_gmu.bin",
> +		},
> +		.gmem = SZ_1M,
> +		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
> +		.init = a6xx_gpu_init,
> +		.zapfw = "a640_zap.mdt",
> +	}, {
> +		.rev = ADRENO_REV(6, 5, 0, ANY_ID),
> +		.revn = 650,
> +		.name = "A650",
> +		.fw = {
> +			[ADRENO_FW_SQE] = "a650_sqe.fw",
> +			[ADRENO_FW_GMU] = "a650_gmu.bin",
> +		},
> +		.gmem = SZ_1M + SZ_128K,
> +		.inactive_period = DRM_MSM_INACTIVE_PERIOD,
> +		.init = a6xx_gpu_init,
> +		.zapfw = "a650_zap.mdt",
>  	},
>  };
>  
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> index 1d5c43c22269..a7647eaacc7a 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> @@ -197,7 +197,7 @@ int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value)
>  		*value = adreno_gpu->gmem;
>  		return 0;
>  	case MSM_PARAM_GMEM_BASE:
> -		*value = 0x100000;
> +		*value = !adreno_is_a650(adreno_gpu) ? 0x100000 : 0;

This will likely be 0 from here on out. This is okay for now, but we might need
to consider adding the gmem base to the GPU list to avoid some ugly
conditionals.

>  		return 0;
>  	case MSM_PARAM_CHIP_ID:
>  		*value = adreno_gpu->rev.patchid |
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> index 9ff4e550e7bd..88ae1b2813ef 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> @@ -237,6 +237,16 @@ static inline int adreno_is_a630(struct adreno_gpu *gpu)
>         return gpu->revn == 630;
>  }
>  
> +static inline int adreno_is_a640(struct adreno_gpu *gpu)
> +{
> +       return gpu->revn == 640;
> +}
> +
> +static inline int adreno_is_a650(struct adreno_gpu *gpu)
> +{
> +       return gpu->revn == 650;
> +}
> +
>  int adreno_get_param(struct msm_gpu *gpu, uint32_t param, uint64_t *value);
>  const struct firmware *adreno_request_fw(struct adreno_gpu *adreno_gpu,
>  		const char *fwname);
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 5/9] drm/msm/a6xx: HFI v2 for A640 and A650

[Jonathan Marek]

Add HFI v2 code paths required by Adreno 640 and 650 GPUs.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c |  66 ++++++++++++---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.h |   7 ++
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c |   6 +-
 drivers/gpu/drm/msm/adreno/a6xx_hfi.c | 117 ++++++++++++++++++++++++--
 drivers/gpu/drm/msm/adreno/a6xx_hfi.h |  50 ++++++++++-
 5 files changed, 222 insertions(+), 24 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index c9fd0470a321..b22a69e2f4b0 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -129,8 +129,6 @@ static void __a6xx_gmu_set_freq(struct a6xx_gmu *gmu, int index)
 	if (ret)
 		dev_err(gmu->dev, "GMU set GPU frequency error: %d\n", ret);
 
-	gmu->freq = gmu->gpu_freqs[index];
-
 	/*
 	 * Eventually we will want to scale the path vote with the frequency but
 	 * for now leave it at max so that the performance is nominal.
@@ -154,7 +152,12 @@ void a6xx_gmu_set_freq(struct msm_gpu *gpu, unsigned long freq)
 
 	gmu->current_perf_index = perf_index;
 
-	__a6xx_gmu_set_freq(gmu, perf_index);
+	if (gmu->legacy)
+		__a6xx_gmu_set_freq(gmu, perf_index);
+	else
+		a6xx_hfi_set_freq(gmu, perf_index);
+
+	gmu->freq = gmu->gpu_freqs[perf_index];
 }
 
 unsigned long a6xx_gmu_get_freq(struct msm_gpu *gpu)
@@ -234,8 +237,13 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 
 	switch (state) {
 	case GMU_OOB_GPU_SET:
-		request = GMU_OOB_GPU_SET_REQUEST;
-		ack = GMU_OOB_GPU_SET_ACK;
+		if (gmu->legacy) {
+			request = GMU_OOB_GPU_SET_REQUEST;
+			ack = GMU_OOB_GPU_SET_ACK;
+		} else {
+			request = GMU_OOB_GPU_SET_REQUEST_NEW;
+			ack = GMU_OOB_GPU_SET_ACK_NEW;
+		}
 		name = "GPU_SET";
 		break;
 	case GMU_OOB_BOOT_SLUMBER:
@@ -274,6 +282,13 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 /* Clear a pending OOB state in the GMU */
 void a6xx_gmu_clear_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
 {
+	if (!gmu->legacy) {
+		WARN_ON(state != GMU_OOB_GPU_SET);
+		gmu_write(gmu, REG_A6XX_GMU_HOST2GMU_INTR_SET,
+			1 << GMU_OOB_GPU_SET_CLEAR_NEW);
+		return;
+	}
+
 	switch (state) {
 	case GMU_OOB_GPU_SET:
 		gmu_write(gmu, REG_A6XX_GMU_HOST2GMU_INTR_SET,
@@ -296,6 +311,9 @@ static int a6xx_sptprac_enable(struct a6xx_gmu *gmu)
 	int ret;
 	u32 val;
 
+	if (!gmu->legacy)
+		return 0;
+
 	gmu_write(gmu, REG_A6XX_GMU_GX_SPTPRAC_POWER_CONTROL, 0x778000);
 
 	ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_SPTPRAC_PWR_CLK_STATUS, val,
@@ -315,6 +333,9 @@ static void a6xx_sptprac_disable(struct a6xx_gmu *gmu)
 	u32 val;
 	int ret;
 
+	if (!gmu->legacy)
+		return;
+
 	/* Make sure retention is on */
 	gmu_rmw(gmu, REG_A6XX_GPU_CC_GX_GDSCR, 0, (1 << 11));
 
@@ -358,6 +379,11 @@ static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
 	if (gmu->idle_level < GMU_IDLE_STATE_SPTP)
 		a6xx_sptprac_disable(gmu);
 
+	if (!gmu->legacy) {
+		ret = a6xx_hfi_send_prep_slumber(gmu);
+		goto out;
+	}
+
 	/* Tell the GMU to get ready to slumber */
 	gmu_write(gmu, REG_A6XX_GMU_BOOT_SLUMBER_OPTION, 1);
 
@@ -373,6 +399,7 @@ static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
 		}
 	}
 
+out:
 	/* Put fence into allow mode */
 	gmu_write(gmu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, 0);
 	return ret;
@@ -642,9 +669,11 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 	if (ret)
 		return ret;
 
-	ret = a6xx_gmu_gfx_rail_on(gmu);
-	if (ret)
-		return ret;
+	if (gmu->legacy) {
+		ret = a6xx_gmu_gfx_rail_on(gmu);
+		if (ret)
+			return ret;
+	}
 
 	/* Enable SPTP_PC if the CPU is responsible for it */
 	if (gmu->idle_level < GMU_IDLE_STATE_SPTP) {
@@ -763,7 +792,10 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
 	enable_irq(gmu->hfi_irq);
 
 	/* Set the GPU to the current freq */
-	__a6xx_gmu_set_freq(gmu, gmu->current_perf_index);
+	if (gmu->legacy)
+		__a6xx_gmu_set_freq(gmu, gmu->current_perf_index);
+	else
+		a6xx_hfi_set_freq(gmu, gmu->current_perf_index);
 
 	/*
 	 * "enable" the GX power domain which won't actually do anything but it
@@ -1262,6 +1294,7 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 
 int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 {
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
 	struct platform_device *pdev = of_find_device_by_node(node);
 	int ret;
@@ -1287,16 +1320,21 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	if (ret)
 		goto err_put_device;
 
+	if (!adreno_is_a640(adreno_gpu) && !adreno_is_a650(adreno_gpu)) {
+		/* HFI v1, has sptprac */
+		gmu->legacy = true;
+
+		/* Allocate memory for the GMU debug region */
+		ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
+		if (ret)
+			goto err_memory;
+	}
+
 	/* Allocate memory for for the HFI queues */
 	ret = a6xx_gmu_memory_alloc(gmu, &gmu->hfi, SZ_16K, 0);
 	if (ret)
 		goto err_memory;
 
-	/* Allocate memory for the GMU debug region */
-	ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
-	if (ret)
-		goto err_memory;
-
 	/* Map the GMU registers */
 	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
 	if (IS_ERR(gmu->mmio)) {
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
index cd66a5e1d7e9..463e2d5f2bb9 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
@@ -79,6 +79,7 @@ struct a6xx_gmu {
 
 	bool initialized;
 	bool hung;
+	bool legacy; /* a618 or a630 */
 };
 
 static inline u32 gmu_read(struct a6xx_gmu *gmu, u32 offset)
@@ -159,10 +160,16 @@ enum a6xx_gmu_oob_state {
 #define GMU_OOB_GPU_SET_ACK	24
 #define GMU_OOB_GPU_SET_CLEAR	24
 
+#define GMU_OOB_GPU_SET_REQUEST_NEW	30
+#define GMU_OOB_GPU_SET_ACK_NEW		31
+#define GMU_OOB_GPU_SET_CLEAR_NEW	31
+
 
 void a6xx_hfi_init(struct a6xx_gmu *gmu);
 int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state);
 void a6xx_hfi_stop(struct a6xx_gmu *gmu);
+int a6xx_hfi_send_prep_slumber(struct a6xx_gmu *gmu);
+int a6xx_hfi_set_freq(struct a6xx_gmu *gmu, int index);
 
 bool a6xx_gmu_gx_is_on(struct a6xx_gmu *gmu);
 bool a6xx_gmu_sptprac_is_on(struct a6xx_gmu *gmu);
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index 68af24150de5..a860d4970e10 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -566,8 +566,10 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
 	 */
 	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
 
-	/* Take the GMU out of its special boot mode */
-	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
+	if (a6xx_gpu->gmu.legacy) {
+		/* Take the GMU out of its special boot mode */
+		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
+	}
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
index b90343d4caf0..f9db69e77121 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
@@ -17,10 +17,14 @@ static const char * const a6xx_hfi_msg_id[] = {
 	HFI_MSG_ID(HFI_H2F_MSG_BW_TABLE),
 	HFI_MSG_ID(HFI_H2F_MSG_PERF_TABLE),
 	HFI_MSG_ID(HFI_H2F_MSG_TEST),
+	HFI_MSG_ID(HFI_H2F_MSG_START),
+	HFI_MSG_ID(HFI_H2F_MSG_CORE_FW_START),
+	HFI_MSG_ID(HFI_H2F_MSG_GX_BW_PERF_VOTE),
+	HFI_MSG_ID(HFI_H2F_MSG_PREPARE_SLUMBER),
 };
 
-static int a6xx_hfi_queue_read(struct a6xx_hfi_queue *queue, u32 *data,
-		u32 dwords)
+static int a6xx_hfi_queue_read(struct a6xx_gmu *gmu,
+	struct a6xx_hfi_queue *queue, u32 *data, u32 dwords)
 {
 	struct a6xx_hfi_queue_header *header = queue->header;
 	u32 i, hdr, index = header->read_index;
@@ -48,6 +52,9 @@ static int a6xx_hfi_queue_read(struct a6xx_hfi_queue *queue, u32 *data,
 		index = (index + 1) % header->size;
 	}
 
+	if (!gmu->legacy)
+		index = ALIGN(index, 4) % header->size;
+
 	header->read_index = index;
 	return HFI_HEADER_SIZE(hdr);
 }
@@ -73,6 +80,12 @@ static int a6xx_hfi_queue_write(struct a6xx_gmu *gmu,
 		index = (index + 1) % header->size;
 	}
 
+	/* Cookify any non used data at the end of the write buffer */
+	if (!gmu->legacy) {
+		for (; index % 4; index = (index + 1) % header->size)
+			queue->data[index] = 0xfafafafa;
+	}
+
 	header->write_index = index;
 	spin_unlock(&queue->lock);
 
@@ -106,7 +119,7 @@ static int a6xx_hfi_wait_for_ack(struct a6xx_gmu *gmu, u32 id, u32 seqnum,
 		struct a6xx_hfi_msg_response resp;
 
 		/* Get the next packet */
-		ret = a6xx_hfi_queue_read(queue, (u32 *) &resp,
+		ret = a6xx_hfi_queue_read(gmu, queue, (u32 *) &resp,
 			sizeof(resp) >> 2);
 
 		/* If the queue is empty our response never made it */
@@ -195,6 +208,28 @@ static int a6xx_hfi_get_fw_version(struct a6xx_gmu *gmu, u32 *version)
 		version, sizeof(*version));
 }
 
+static int a6xx_hfi_send_perf_table_v1(struct a6xx_gmu *gmu)
+{
+	struct a6xx_hfi_msg_perf_table_v1 msg = { 0 };
+	int i;
+
+	msg.num_gpu_levels = gmu->nr_gpu_freqs;
+	msg.num_gmu_levels = gmu->nr_gmu_freqs;
+
+	for (i = 0; i < gmu->nr_gpu_freqs; i++) {
+		msg.gx_votes[i].vote = gmu->gx_arc_votes[i];
+		msg.gx_votes[i].freq = gmu->gpu_freqs[i] / 1000;
+	}
+
+	for (i = 0; i < gmu->nr_gmu_freqs; i++) {
+		msg.cx_votes[i].vote = gmu->cx_arc_votes[i];
+		msg.cx_votes[i].freq = gmu->gmu_freqs[i] / 1000;
+	}
+
+	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_PERF_TABLE, &msg, sizeof(msg),
+		NULL, 0);
+}
+
 static int a6xx_hfi_send_perf_table(struct a6xx_gmu *gmu)
 {
 	struct a6xx_hfi_msg_perf_table msg = { 0 };
@@ -205,6 +240,7 @@ static int a6xx_hfi_send_perf_table(struct a6xx_gmu *gmu)
 
 	for (i = 0; i < gmu->nr_gpu_freqs; i++) {
 		msg.gx_votes[i].vote = gmu->gx_arc_votes[i];
+		msg.gx_votes[i].acd = 0xffffffff;
 		msg.gx_votes[i].freq = gmu->gpu_freqs[i] / 1000;
 	}
 
@@ -306,7 +342,45 @@ static int a6xx_hfi_send_test(struct a6xx_gmu *gmu)
 		NULL, 0);
 }
 
-int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
+int a6xx_hfi_send_start(struct a6xx_gmu *gmu)
+{
+	struct a6xx_hfi_msg_start msg = { 0 };
+
+	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_START, &msg, sizeof(msg),
+		NULL, 0);
+}
+
+int a6xx_hfi_send_core_fw_start(struct a6xx_gmu *gmu)
+{
+	struct a6xx_hfi_msg_core_fw_start msg = { 0 };
+
+	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_CORE_FW_START, &msg,
+		sizeof(msg), NULL, 0);
+}
+
+int a6xx_hfi_set_freq(struct a6xx_gmu *gmu, int index)
+{
+	struct a6xx_hfi_gx_bw_perf_vote_cmd msg = { 0 };
+
+	msg.ack_type = 1; /* blocking */
+	msg.freq = index;
+	msg.bw = 0; /* TODO: bus scaling */
+
+	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_GX_BW_PERF_VOTE, &msg,
+		sizeof(msg), NULL, 0);
+}
+
+int a6xx_hfi_send_prep_slumber(struct a6xx_gmu *gmu)
+{
+	struct a6xx_hfi_prep_slumber_cmd msg = { 0 };
+
+	/* TODO: should freq and bw fields be non-zero ? */
+
+	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_PREPARE_SLUMBER, &msg,
+		sizeof(msg), NULL, 0);
+}
+
+static int a6xx_hfi_start_v1(struct a6xx_gmu *gmu, int boot_state)
 {
 	int ret;
 
@@ -324,7 +398,7 @@ int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
 	 * the GMU firmware
 	 */
 
-	ret = a6xx_hfi_send_perf_table(gmu);
+	ret = a6xx_hfi_send_perf_table_v1(gmu);
 	if (ret)
 		return ret;
 
@@ -341,6 +415,37 @@ int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
 	return 0;
 }
 
+int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
+{
+	int ret;
+
+	if (gmu->legacy)
+		return a6xx_hfi_start_v1(gmu, boot_state);
+
+
+	ret = a6xx_hfi_send_perf_table(gmu);
+	if (ret)
+		return ret;
+
+	ret = a6xx_hfi_send_bw_table(gmu);
+	if (ret)
+		return ret;
+
+	ret = a6xx_hfi_send_core_fw_start(gmu);
+	if (ret)
+		return ret;
+
+	/*
+	 * Downstream driver sends this in its "a6xx_hw_init" equivalent,
+	 * but seems to be no harm in sending it here
+	 */
+	ret = a6xx_hfi_send_start(gmu);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
 void a6xx_hfi_stop(struct a6xx_gmu *gmu)
 {
 	int i;
@@ -415,5 +520,5 @@ void a6xx_hfi_init(struct a6xx_gmu *gmu)
 	/* GMU response queue */
 	offset += SZ_4K;
 	a6xx_hfi_queue_init(&gmu->queues[1], &headers[1], hfi->virt + offset,
-		hfi->iova + offset, 4);
+		hfi->iova + offset, gmu->legacy ? 4 : 1);
 }
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.h b/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
index 60d1319fa44f..2bd670ca42d6 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
@@ -51,7 +51,8 @@ struct a6xx_hfi_queue {
 /* HFI message types */
 
 #define HFI_MSG_CMD 0
-#define HFI_MSG_ACK 2
+#define HFI_MSG_ACK 1
+#define HFI_MSG_ACK_V1 2
 
 #define HFI_F2H_MSG_ACK 126
 
@@ -94,7 +95,13 @@ struct perf_level {
 	u32 freq;
 };
 
-struct a6xx_hfi_msg_perf_table {
+struct perf_gx_level {
+	u32 vote;
+	u32 acd;
+	u32 freq;
+};
+
+struct a6xx_hfi_msg_perf_table_v1 {
 	u32 header;
 	u32 num_gpu_levels;
 	u32 num_gmu_levels;
@@ -103,6 +110,15 @@ struct a6xx_hfi_msg_perf_table {
 	struct perf_level cx_votes[4];
 };
 
+struct a6xx_hfi_msg_perf_table {
+	u32 header;
+	u32 num_gpu_levels;
+	u32 num_gmu_levels;
+
+	struct perf_gx_level gx_votes[16];
+	struct perf_level cx_votes[4];
+};
+
 #define HFI_H2F_MSG_BW_TABLE 3
 
 struct a6xx_hfi_msg_bw_table {
@@ -124,4 +140,34 @@ struct a6xx_hfi_msg_test {
 	u32 header;
 };
 
+#define HFI_H2F_MSG_START 10
+
+struct a6xx_hfi_msg_start {
+	u32 header;
+};
+
+#define HFI_H2F_MSG_CORE_FW_START 14
+
+struct a6xx_hfi_msg_core_fw_start {
+	u32 header;
+	u32 handle;
+};
+
+#define HFI_H2F_MSG_GX_BW_PERF_VOTE 30
+
+struct a6xx_hfi_gx_bw_perf_vote_cmd {
+	u32 header;
+	u32 ack_type;
+	u32 freq;
+	u32 bw;
+};
+
+#define HFI_H2F_MSG_PREPARE_SLUMBER 33
+
+struct a6xx_hfi_prep_slumber_cmd {
+	u32 header;
+	u32 bw;
+	u32 freq;
+};
+
 #endif
-- 
2.26.1

Re: [PATCH v2 5/9] drm/msm/a6xx: HFI v2 for A640 and A650

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:23PM -0400, Jonathan Marek wrote:
> Add HFI v2 code paths required by Adreno 640 and 650 GPUs.

I still feel like there is a chance for optimization here but the perfect is the
enemy of the good and that would be a good follow on for the future.

Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org>

> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.c |  66 ++++++++++++---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.h |   7 ++
>  drivers/gpu/drm/msm/adreno/a6xx_gpu.c |   6 +-
>  drivers/gpu/drm/msm/adreno/a6xx_hfi.c | 117 ++++++++++++++++++++++++--
>  drivers/gpu/drm/msm/adreno/a6xx_hfi.h |  50 ++++++++++-
>  5 files changed, 222 insertions(+), 24 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> index c9fd0470a321..b22a69e2f4b0 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> @@ -129,8 +129,6 @@ static void __a6xx_gmu_set_freq(struct a6xx_gmu *gmu, int index)
>  	if (ret)
>  		dev_err(gmu->dev, "GMU set GPU frequency error: %d\n", ret);
>  
> -	gmu->freq = gmu->gpu_freqs[index];
> -
>  	/*
>  	 * Eventually we will want to scale the path vote with the frequency but
>  	 * for now leave it at max so that the performance is nominal.
> @@ -154,7 +152,12 @@ void a6xx_gmu_set_freq(struct msm_gpu *gpu, unsigned long freq)
>  
>  	gmu->current_perf_index = perf_index;
>  
> -	__a6xx_gmu_set_freq(gmu, perf_index);
> +	if (gmu->legacy)
> +		__a6xx_gmu_set_freq(gmu, perf_index);
> +	else
> +		a6xx_hfi_set_freq(gmu, perf_index);
> +
> +	gmu->freq = gmu->gpu_freqs[perf_index];
>  }
>  
>  unsigned long a6xx_gmu_get_freq(struct msm_gpu *gpu)
> @@ -234,8 +237,13 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
>  
>  	switch (state) {
>  	case GMU_OOB_GPU_SET:
> -		request = GMU_OOB_GPU_SET_REQUEST;
> -		ack = GMU_OOB_GPU_SET_ACK;
> +		if (gmu->legacy) {
> +			request = GMU_OOB_GPU_SET_REQUEST;
> +			ack = GMU_OOB_GPU_SET_ACK;
> +		} else {
> +			request = GMU_OOB_GPU_SET_REQUEST_NEW;
> +			ack = GMU_OOB_GPU_SET_ACK_NEW;
> +		}
>  		name = "GPU_SET";
>  		break;
>  	case GMU_OOB_BOOT_SLUMBER:
> @@ -274,6 +282,13 @@ int a6xx_gmu_set_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
>  /* Clear a pending OOB state in the GMU */
>  void a6xx_gmu_clear_oob(struct a6xx_gmu *gmu, enum a6xx_gmu_oob_state state)
>  {
> +	if (!gmu->legacy) {
> +		WARN_ON(state != GMU_OOB_GPU_SET);
> +		gmu_write(gmu, REG_A6XX_GMU_HOST2GMU_INTR_SET,
> +			1 << GMU_OOB_GPU_SET_CLEAR_NEW);
> +		return;
> +	}
> +
>  	switch (state) {
>  	case GMU_OOB_GPU_SET:
>  		gmu_write(gmu, REG_A6XX_GMU_HOST2GMU_INTR_SET,
> @@ -296,6 +311,9 @@ static int a6xx_sptprac_enable(struct a6xx_gmu *gmu)
>  	int ret;
>  	u32 val;
>  
> +	if (!gmu->legacy)
> +		return 0;
> +
>  	gmu_write(gmu, REG_A6XX_GMU_GX_SPTPRAC_POWER_CONTROL, 0x778000);
>  
>  	ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_SPTPRAC_PWR_CLK_STATUS, val,
> @@ -315,6 +333,9 @@ static void a6xx_sptprac_disable(struct a6xx_gmu *gmu)
>  	u32 val;
>  	int ret;
>  
> +	if (!gmu->legacy)
> +		return;
> +
>  	/* Make sure retention is on */
>  	gmu_rmw(gmu, REG_A6XX_GPU_CC_GX_GDSCR, 0, (1 << 11));
>  
> @@ -358,6 +379,11 @@ static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
>  	if (gmu->idle_level < GMU_IDLE_STATE_SPTP)
>  		a6xx_sptprac_disable(gmu);
>  
> +	if (!gmu->legacy) {
> +		ret = a6xx_hfi_send_prep_slumber(gmu);
> +		goto out;
> +	}
> +
>  	/* Tell the GMU to get ready to slumber */
>  	gmu_write(gmu, REG_A6XX_GMU_BOOT_SLUMBER_OPTION, 1);
>  
> @@ -373,6 +399,7 @@ static int a6xx_gmu_notify_slumber(struct a6xx_gmu *gmu)
>  		}
>  	}
>  
> +out:
>  	/* Put fence into allow mode */
>  	gmu_write(gmu, REG_A6XX_GMU_AO_AHB_FENCE_CTRL, 0);
>  	return ret;
> @@ -642,9 +669,11 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  	if (ret)
>  		return ret;
>  
> -	ret = a6xx_gmu_gfx_rail_on(gmu);
> -	if (ret)
> -		return ret;
> +	if (gmu->legacy) {
> +		ret = a6xx_gmu_gfx_rail_on(gmu);
> +		if (ret)
> +			return ret;
> +	}
>  
>  	/* Enable SPTP_PC if the CPU is responsible for it */
>  	if (gmu->idle_level < GMU_IDLE_STATE_SPTP) {
> @@ -763,7 +792,10 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
>  	enable_irq(gmu->hfi_irq);
>  
>  	/* Set the GPU to the current freq */
> -	__a6xx_gmu_set_freq(gmu, gmu->current_perf_index);
> +	if (gmu->legacy)
> +		__a6xx_gmu_set_freq(gmu, gmu->current_perf_index);
> +	else
> +		a6xx_hfi_set_freq(gmu, gmu->current_perf_index);
>  
>  	/*
>  	 * "enable" the GX power domain which won't actually do anything but it
> @@ -1262,6 +1294,7 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
>  
>  int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  {
> +	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
>  	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
>  	struct platform_device *pdev = of_find_device_by_node(node);
>  	int ret;
> @@ -1287,16 +1320,21 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  	if (ret)
>  		goto err_put_device;
>  
> +	if (!adreno_is_a640(adreno_gpu) && !adreno_is_a650(adreno_gpu)) {
> +		/* HFI v1, has sptprac */
> +		gmu->legacy = true;
> +
> +		/* Allocate memory for the GMU debug region */
> +		ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
> +		if (ret)
> +			goto err_memory;
> +	}
> +
>  	/* Allocate memory for for the HFI queues */
>  	ret = a6xx_gmu_memory_alloc(gmu, &gmu->hfi, SZ_16K, 0);
>  	if (ret)
>  		goto err_memory;
>  
> -	/* Allocate memory for the GMU debug region */
> -	ret = a6xx_gmu_memory_alloc(gmu, &gmu->debug, SZ_16K, 0);
> -	if (ret)
> -		goto err_memory;
> -
>  	/* Map the GMU registers */
>  	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
>  	if (IS_ERR(gmu->mmio)) {
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> index cd66a5e1d7e9..463e2d5f2bb9 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> @@ -79,6 +79,7 @@ struct a6xx_gmu {
>  
>  	bool initialized;
>  	bool hung;
> +	bool legacy; /* a618 or a630 */
>  };
>  
>  static inline u32 gmu_read(struct a6xx_gmu *gmu, u32 offset)
> @@ -159,10 +160,16 @@ enum a6xx_gmu_oob_state {
>  #define GMU_OOB_GPU_SET_ACK	24
>  #define GMU_OOB_GPU_SET_CLEAR	24
>  
> +#define GMU_OOB_GPU_SET_REQUEST_NEW	30
> +#define GMU_OOB_GPU_SET_ACK_NEW		31
> +#define GMU_OOB_GPU_SET_CLEAR_NEW	31
> +
>  
>  void a6xx_hfi_init(struct a6xx_gmu *gmu);
>  int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state);
>  void a6xx_hfi_stop(struct a6xx_gmu *gmu);
> +int a6xx_hfi_send_prep_slumber(struct a6xx_gmu *gmu);
> +int a6xx_hfi_set_freq(struct a6xx_gmu *gmu, int index);
>  
>  bool a6xx_gmu_gx_is_on(struct a6xx_gmu *gmu);
>  bool a6xx_gmu_sptprac_is_on(struct a6xx_gmu *gmu);
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> index 68af24150de5..a860d4970e10 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
> @@ -566,8 +566,10 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
>  	 */
>  	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_GPU_SET);
>  
> -	/* Take the GMU out of its special boot mode */
> -	a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
> +	if (a6xx_gpu->gmu.legacy) {
> +		/* Take the GMU out of its special boot mode */
> +		a6xx_gmu_clear_oob(&a6xx_gpu->gmu, GMU_OOB_BOOT_SLUMBER);
> +	}
>  
>  	return ret;
>  }
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> index b90343d4caf0..f9db69e77121 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.c
> @@ -17,10 +17,14 @@ static const char * const a6xx_hfi_msg_id[] = {
>  	HFI_MSG_ID(HFI_H2F_MSG_BW_TABLE),
>  	HFI_MSG_ID(HFI_H2F_MSG_PERF_TABLE),
>  	HFI_MSG_ID(HFI_H2F_MSG_TEST),
> +	HFI_MSG_ID(HFI_H2F_MSG_START),
> +	HFI_MSG_ID(HFI_H2F_MSG_CORE_FW_START),
> +	HFI_MSG_ID(HFI_H2F_MSG_GX_BW_PERF_VOTE),
> +	HFI_MSG_ID(HFI_H2F_MSG_PREPARE_SLUMBER),
>  };
>  
> -static int a6xx_hfi_queue_read(struct a6xx_hfi_queue *queue, u32 *data,
> -		u32 dwords)
> +static int a6xx_hfi_queue_read(struct a6xx_gmu *gmu,
> +	struct a6xx_hfi_queue *queue, u32 *data, u32 dwords)
>  {
>  	struct a6xx_hfi_queue_header *header = queue->header;
>  	u32 i, hdr, index = header->read_index;
> @@ -48,6 +52,9 @@ static int a6xx_hfi_queue_read(struct a6xx_hfi_queue *queue, u32 *data,
>  		index = (index + 1) % header->size;
>  	}
>  
> +	if (!gmu->legacy)
> +		index = ALIGN(index, 4) % header->size;
> +
>  	header->read_index = index;
>  	return HFI_HEADER_SIZE(hdr);
>  }
> @@ -73,6 +80,12 @@ static int a6xx_hfi_queue_write(struct a6xx_gmu *gmu,
>  		index = (index + 1) % header->size;
>  	}
>  
> +	/* Cookify any non used data at the end of the write buffer */
> +	if (!gmu->legacy) {
> +		for (; index % 4; index = (index + 1) % header->size)
> +			queue->data[index] = 0xfafafafa;
> +	}
> +
>  	header->write_index = index;
>  	spin_unlock(&queue->lock);
>  
> @@ -106,7 +119,7 @@ static int a6xx_hfi_wait_for_ack(struct a6xx_gmu *gmu, u32 id, u32 seqnum,
>  		struct a6xx_hfi_msg_response resp;
>  
>  		/* Get the next packet */
> -		ret = a6xx_hfi_queue_read(queue, (u32 *) &resp,
> +		ret = a6xx_hfi_queue_read(gmu, queue, (u32 *) &resp,
>  			sizeof(resp) >> 2);
>  
>  		/* If the queue is empty our response never made it */
> @@ -195,6 +208,28 @@ static int a6xx_hfi_get_fw_version(struct a6xx_gmu *gmu, u32 *version)
>  		version, sizeof(*version));
>  }
>  
> +static int a6xx_hfi_send_perf_table_v1(struct a6xx_gmu *gmu)
> +{
> +	struct a6xx_hfi_msg_perf_table_v1 msg = { 0 };
> +	int i;
> +
> +	msg.num_gpu_levels = gmu->nr_gpu_freqs;
> +	msg.num_gmu_levels = gmu->nr_gmu_freqs;
> +
> +	for (i = 0; i < gmu->nr_gpu_freqs; i++) {
> +		msg.gx_votes[i].vote = gmu->gx_arc_votes[i];
> +		msg.gx_votes[i].freq = gmu->gpu_freqs[i] / 1000;
> +	}
> +
> +	for (i = 0; i < gmu->nr_gmu_freqs; i++) {
> +		msg.cx_votes[i].vote = gmu->cx_arc_votes[i];
> +		msg.cx_votes[i].freq = gmu->gmu_freqs[i] / 1000;
> +	}
> +
> +	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_PERF_TABLE, &msg, sizeof(msg),
> +		NULL, 0);
> +}
> +
>  static int a6xx_hfi_send_perf_table(struct a6xx_gmu *gmu)
>  {
>  	struct a6xx_hfi_msg_perf_table msg = { 0 };
> @@ -205,6 +240,7 @@ static int a6xx_hfi_send_perf_table(struct a6xx_gmu *gmu)
>  
>  	for (i = 0; i < gmu->nr_gpu_freqs; i++) {
>  		msg.gx_votes[i].vote = gmu->gx_arc_votes[i];
> +		msg.gx_votes[i].acd = 0xffffffff;
>  		msg.gx_votes[i].freq = gmu->gpu_freqs[i] / 1000;
>  	}
>  
> @@ -306,7 +342,45 @@ static int a6xx_hfi_send_test(struct a6xx_gmu *gmu)
>  		NULL, 0);
>  }
>  
> -int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
> +int a6xx_hfi_send_start(struct a6xx_gmu *gmu)
> +{
> +	struct a6xx_hfi_msg_start msg = { 0 };
> +
> +	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_START, &msg, sizeof(msg),
> +		NULL, 0);
> +}
> +
> +int a6xx_hfi_send_core_fw_start(struct a6xx_gmu *gmu)
> +{
> +	struct a6xx_hfi_msg_core_fw_start msg = { 0 };
> +
> +	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_CORE_FW_START, &msg,
> +		sizeof(msg), NULL, 0);
> +}
> +
> +int a6xx_hfi_set_freq(struct a6xx_gmu *gmu, int index)
> +{
> +	struct a6xx_hfi_gx_bw_perf_vote_cmd msg = { 0 };
> +
> +	msg.ack_type = 1; /* blocking */
> +	msg.freq = index;
> +	msg.bw = 0; /* TODO: bus scaling */
> +
> +	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_GX_BW_PERF_VOTE, &msg,
> +		sizeof(msg), NULL, 0);
> +}
> +
> +int a6xx_hfi_send_prep_slumber(struct a6xx_gmu *gmu)
> +{
> +	struct a6xx_hfi_prep_slumber_cmd msg = { 0 };
> +
> +	/* TODO: should freq and bw fields be non-zero ? */
> +
> +	return a6xx_hfi_send_msg(gmu, HFI_H2F_MSG_PREPARE_SLUMBER, &msg,
> +		sizeof(msg), NULL, 0);
> +}
> +
> +static int a6xx_hfi_start_v1(struct a6xx_gmu *gmu, int boot_state)
>  {
>  	int ret;
>  
> @@ -324,7 +398,7 @@ int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
>  	 * the GMU firmware
>  	 */
>  
> -	ret = a6xx_hfi_send_perf_table(gmu);
> +	ret = a6xx_hfi_send_perf_table_v1(gmu);
>  	if (ret)
>  		return ret;
>  
> @@ -341,6 +415,37 @@ int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
>  	return 0;
>  }
>  
> +int a6xx_hfi_start(struct a6xx_gmu *gmu, int boot_state)
> +{
> +	int ret;
> +
> +	if (gmu->legacy)
> +		return a6xx_hfi_start_v1(gmu, boot_state);
> +
> +
> +	ret = a6xx_hfi_send_perf_table(gmu);
> +	if (ret)
> +		return ret;
> +
> +	ret = a6xx_hfi_send_bw_table(gmu);
> +	if (ret)
> +		return ret;
> +
> +	ret = a6xx_hfi_send_core_fw_start(gmu);
> +	if (ret)
> +		return ret;
> +
> +	/*
> +	 * Downstream driver sends this in its "a6xx_hw_init" equivalent,
> +	 * but seems to be no harm in sending it here
> +	 */
> +	ret = a6xx_hfi_send_start(gmu);
> +	if (ret)
> +		return ret;
> +
> +	return 0;
> +}
> +
>  void a6xx_hfi_stop(struct a6xx_gmu *gmu)
>  {
>  	int i;
> @@ -415,5 +520,5 @@ void a6xx_hfi_init(struct a6xx_gmu *gmu)
>  	/* GMU response queue */
>  	offset += SZ_4K;
>  	a6xx_hfi_queue_init(&gmu->queues[1], &headers[1], hfi->virt + offset,
> -		hfi->iova + offset, 4);
> +		hfi->iova + offset, gmu->legacy ? 4 : 1);
>  }
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_hfi.h b/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
> index 60d1319fa44f..2bd670ca42d6 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_hfi.h
> @@ -51,7 +51,8 @@ struct a6xx_hfi_queue {
>  /* HFI message types */
>  
>  #define HFI_MSG_CMD 0
> -#define HFI_MSG_ACK 2
> +#define HFI_MSG_ACK 1
> +#define HFI_MSG_ACK_V1 2
>  
>  #define HFI_F2H_MSG_ACK 126
>  
> @@ -94,7 +95,13 @@ struct perf_level {
>  	u32 freq;
>  };
>  
> -struct a6xx_hfi_msg_perf_table {
> +struct perf_gx_level {
> +	u32 vote;
> +	u32 acd;
> +	u32 freq;
> +};
> +
> +struct a6xx_hfi_msg_perf_table_v1 {
>  	u32 header;
>  	u32 num_gpu_levels;
>  	u32 num_gmu_levels;
> @@ -103,6 +110,15 @@ struct a6xx_hfi_msg_perf_table {
>  	struct perf_level cx_votes[4];
>  };
>  
> +struct a6xx_hfi_msg_perf_table {
> +	u32 header;
> +	u32 num_gpu_levels;
> +	u32 num_gmu_levels;
> +
> +	struct perf_gx_level gx_votes[16];
> +	struct perf_level cx_votes[4];
> +};
> +
>  #define HFI_H2F_MSG_BW_TABLE 3
>  
>  struct a6xx_hfi_msg_bw_table {
> @@ -124,4 +140,34 @@ struct a6xx_hfi_msg_test {
>  	u32 header;
>  };
>  
> +#define HFI_H2F_MSG_START 10
> +
> +struct a6xx_hfi_msg_start {
> +	u32 header;
> +};
> +
> +#define HFI_H2F_MSG_CORE_FW_START 14
> +
> +struct a6xx_hfi_msg_core_fw_start {
> +	u32 header;
> +	u32 handle;
> +};
> +
> +#define HFI_H2F_MSG_GX_BW_PERF_VOTE 30
> +
> +struct a6xx_hfi_gx_bw_perf_vote_cmd {
> +	u32 header;
> +	u32 ack_type;
> +	u32 freq;
> +	u32 bw;
> +};
> +
> +#define HFI_H2F_MSG_PREPARE_SLUMBER 33
> +
> +struct a6xx_hfi_prep_slumber_cmd {
> +	u32 header;
> +	u32 bw;
> +	u32 freq;
> +};
> +
>  #endif
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 6/9] drm/msm/a6xx: A640/A650 GMU firmware path

[Jonathan Marek]

Newer GPUs have different GMU firmware path.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 135 +++++++++++++++++++---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.h     |  11 ++
 drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h |   6 +
 3 files changed, 136 insertions(+), 16 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index b22a69e2f4b0..4aef5fe985d6 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -571,6 +571,8 @@ static void a6xx_gmu_power_config(struct a6xx_gmu *gmu)
 {
 	/* Disable GMU WB/RB buffer */
 	gmu_write(gmu, REG_A6XX_GMU_SYS_BUS_CONFIG, 0x1);
+	gmu_write(gmu, REG_A6XX_GMU_ICACHE_CONFIG, 0x1);
+	gmu_write(gmu, REG_A6XX_GMU_DCACHE_CONFIG, 0x1);
 
 	gmu_write(gmu, REG_A6XX_GMU_PWR_COL_INTER_FRAME_CTRL, 0x9c40400);
 
@@ -600,14 +602,91 @@ static void a6xx_gmu_power_config(struct a6xx_gmu *gmu)
 		A6XX_GMU_RPMH_CTRL_GFX_VOTE_ENABLE);
 }
 
+static int in_range(u32 addr, u32 start, u32 size)
+{
+	return addr >= start && addr < start + size;
+}
+
+static int a6xx_gmu_fw_load(struct a6xx_gmu *gmu)
+{
+	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
+	const struct firmware *fw_image = adreno_gpu->fw[ADRENO_FW_GMU];
+	const struct block_header {
+		u32 addr;
+		u32 size;
+		u32 type;
+		u32 value;
+		u32 data[];
+	} *blk;
+	struct a6xx_gmu_bo *mem;
+	u32 reg;
+
+	u32 itcm_base = 0x00000000;
+	u32 dtcm_base = 0x00040000;
+
+	if (adreno_is_a650(adreno_gpu))
+		dtcm_base = 0x10004000;
+
+	if (gmu->legacy) {
+		/* Sanity check the size of the firmware that was loaded */
+		if (fw_image->size > 0x8000) {
+			DRM_DEV_ERROR(gmu->dev,
+				"GMU firmware is bigger than the available region\n");
+			return -EINVAL;
+		}
+
+		gmu_write_bulk(gmu, REG_A6XX_GMU_CM3_ITCM_START,
+			       (u32*) fw_image->data, fw_image->size >> 2);
+		return 0;
+	}
+
+
+	for (blk = (const struct block_header *) fw_image->data;
+	     (const u8*) blk < fw_image->data + fw_image->size;
+	     blk = (const struct block_header *) &blk->data[blk->size >> 2]) {
+		if (blk->size == 0)
+			continue;
+
+		reg = 0;
+		mem = NULL;
+
+		if (in_range(blk->addr, itcm_base, SZ_16K))
+			reg = REG_A6XX_GMU_CM3_ITCM_START + ((blk->addr - itcm_base) >> 2);
+		else if (in_range(blk->addr, dtcm_base, SZ_16K))
+			reg = REG_A6XX_GMU_CM3_DTCM_START + ((blk->addr - dtcm_base) >> 2);
+		else if (in_range(blk->addr, gmu->icache.iova, gmu->icache.size))
+			mem = &gmu->icache;
+		else if (in_range(blk->addr, gmu->dcache.iova, gmu->dcache.size))
+			mem = &gmu->dcache;
+		else if (in_range(blk->addr, gmu->dummy.iova, gmu->dummy.size))
+			mem = &gmu->dummy;
+		else {
+			DRM_DEV_ERROR(gmu->dev,
+				"failed to match fw block (addr=%.8x size=%d data[0]=%.8x)\n",
+				blk->addr, blk->size, blk->data[0]);
+		}
+
+		if (reg)
+			gmu_write_bulk(gmu, reg, blk->data, blk->size >> 2);
+
+		if (mem)
+			memcpy(mem->virt + blk->addr - mem->iova, blk->data, blk->size);
+	}
+
+	return 0;
+}
+
 static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 {
 	static bool rpmh_init;
 	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
 	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
-	int i, ret;
+	int ret;
 	u32 chipid;
-	u32 *image;
+
+	if (adreno_is_a650(adreno_gpu))
+		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF, 1);
 
 	if (state == GMU_WARM_BOOT) {
 		ret = a6xx_rpmh_start(gmu);
@@ -618,13 +697,6 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 			"GMU firmware is not loaded\n"))
 			return -ENOENT;
 
-		/* Sanity check the size of the firmware that was loaded */
-		if (adreno_gpu->fw[ADRENO_FW_GMU]->size > 0x8000) {
-			DRM_DEV_ERROR(gmu->dev,
-				"GMU firmware is bigger than the available region\n");
-			return -EINVAL;
-		}
-
 		/* Turn on register retention */
 		gmu_write(gmu, REG_A6XX_GMU_GENERAL_7, 1);
 
@@ -638,11 +710,9 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 				return ret;
 		}
 
-		image = (u32 *) adreno_gpu->fw[ADRENO_FW_GMU]->data;
-
-		for (i = 0; i < adreno_gpu->fw[ADRENO_FW_GMU]->size >> 2; i++)
-			gmu_write(gmu, REG_A6XX_GMU_CM3_ITCM_START + i,
-				image[i]);
+		ret = a6xx_gmu_fw_load(gmu);
+		if (ret)
+			return ret;
 	}
 
 	gmu_write(gmu, REG_A6XX_GMU_CM3_FW_INIT_RESULT, 0);
@@ -769,12 +839,20 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
 	/* Enable the GMU interrupt */
 	gmu_write(gmu, REG_A6XX_GMU_AO_HOST_INTERRUPT_CLR, ~0);
 	gmu_write(gmu, REG_A6XX_GMU_AO_HOST_INTERRUPT_MASK, ~A6XX_GMU_IRQ_MASK);
+
 	enable_irq(gmu->gmu_irq);
 
 	/* Check to see if we are doing a cold or warm boot */
 	status = gmu_read(gmu, REG_A6XX_GMU_GENERAL_7) == 1 ?
 		GMU_WARM_BOOT : GMU_COLD_BOOT;
 
+	/*
+	 * Warm boot path does not work on newer GPUs
+	 * Presumably this is because icache/dcache regions must be restored
+	 */
+	if (!gmu->legacy)
+		status = GMU_COLD_BOOT;
+
 	ret = a6xx_gmu_fw_start(gmu, status);
 	if (ret)
 		goto out;
@@ -957,6 +1035,9 @@ static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
 {
 	msm_gem_kernel_put(gmu->hfi.obj, gmu->aspace, false);
 	msm_gem_kernel_put(gmu->debug.obj, gmu->aspace, false);
+	msm_gem_kernel_put(gmu->icache.obj, gmu->aspace, false);
+	msm_gem_kernel_put(gmu->dcache.obj, gmu->aspace, false);
+	msm_gem_kernel_put(gmu->dummy.obj, gmu->aspace, false);
 
 	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
 	msm_gem_address_space_put(gmu->aspace);
@@ -974,12 +1055,14 @@ static int a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo,
 	size = PAGE_ALIGN(size);
 	if (!iova) {
 		/* no fixed address - use GMU's uncached range */
-		range_start = 0x60000000;
+		range_start = 0x60000000 + PAGE_SIZE; /* skip dummy page */
 		range_end = 0x80000000;
 	} else {
 		/* range for fixed address */
 		range_start = iova;
 		range_end = iova + size;
+		/* use IOMMU_PRIV for icache/dcache */
+		flags |= MSM_BO_MAP_PRIV;
 	}
 
 	bo->obj = msm_gem_new(dev, size, flags);
@@ -1320,7 +1403,27 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	if (ret)
 		goto err_put_device;
 
-	if (!adreno_is_a640(adreno_gpu) && !adreno_is_a650(adreno_gpu)) {
+	/* Allocate memory for the GMU dummy page */
+	ret = a6xx_gmu_memory_alloc(gmu, &gmu->dummy, SZ_4K, 0x60000000);
+	if (ret)
+		goto err_memory;
+
+	if (adreno_is_a650(adreno_gpu)) {
+		ret = a6xx_gmu_memory_alloc(gmu, &gmu->icache,
+			SZ_16M - SZ_16K, 0x04000);
+		if (ret)
+			goto err_memory;
+	} else if (adreno_is_a640(adreno_gpu)) {
+		ret = a6xx_gmu_memory_alloc(gmu, &gmu->icache,
+			SZ_256K - SZ_16K, 0x04000);
+		if (ret)
+			goto err_memory;
+
+		ret = a6xx_gmu_memory_alloc(gmu, &gmu->dcache,
+			SZ_256K - SZ_16K, 0x44000);
+		if (ret)
+			goto err_memory;
+	} else {
 		/* HFI v1, has sptprac */
 		gmu->legacy = true;
 
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
index 463e2d5f2bb9..526911010cb1 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
@@ -57,6 +57,9 @@ struct a6xx_gmu {
 
 	struct a6xx_gmu_bo hfi;
 	struct a6xx_gmu_bo debug;
+	struct a6xx_gmu_bo icache;
+	struct a6xx_gmu_bo dcache;
+	struct a6xx_gmu_bo dummy;
 
 	int nr_clocks;
 	struct clk_bulk_data *clocks;
@@ -92,6 +95,14 @@ static inline void gmu_write(struct a6xx_gmu *gmu, u32 offset, u32 value)
 	return msm_writel(value, gmu->mmio + (offset << 2));
 }
 
+static inline void
+gmu_write_bulk(struct a6xx_gmu *gmu, u32 offset, const u32 *data, u32 size)
+{
+	int i;
+	for (i = 0; i < size; i++)
+		gmu_write(gmu, offset + i, data[i]);
+}
+
 static inline void gmu_rmw(struct a6xx_gmu *gmu, u32 reg, u32 mask, u32 or)
 {
 	u32 val = gmu_read(gmu, reg);
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
index 1cc1c135236b..eb2cd41dae6e 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
@@ -101,6 +101,10 @@ static inline uint32_t A6XX_HFI_IRQ_OOB_MASK(uint32_t val)
 
 #define REG_A6XX_GMU_DCVS_RETURN				0x000023ff
 
+#define REG_A6XX_GMU_ICACHE_CONFIG				0x00004c00
+
+#define REG_A6XX_GMU_DCACHE_CONFIG				0x00004c01
+
 #define REG_A6XX_GMU_SYS_BUS_CONFIG				0x00004c0f
 
 #define REG_A6XX_GMU_CM3_SYSRESET				0x00005000
@@ -199,6 +203,8 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
 
 #define REG_A6XX_GPU_GMU_CX_GMU_RPMH_POWER_STATE		0x000050ec
 
+#define REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF			0x000050f0
+
 #define REG_A6XX_GMU_BOOT_KMD_LM_HANDSHAKE			0x000051f0
 
 #define REG_A6XX_GMU_LLM_GLM_SLEEP_CTRL				0x00005157
-- 
2.26.1

Re: [PATCH v2 6/9] drm/msm/a6xx: A640/A650 GMU firmware path

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:24PM -0400, Jonathan Marek wrote:
> Newer GPUs have different GMU firmware path.
> 
> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 135 +++++++++++++++++++---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.h     |  11 ++
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h |   6 +
>  3 files changed, 136 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> index b22a69e2f4b0..4aef5fe985d6 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> @@ -571,6 +571,8 @@ static void a6xx_gmu_power_config(struct a6xx_gmu *gmu)
>  {
>  	/* Disable GMU WB/RB buffer */
>  	gmu_write(gmu, REG_A6XX_GMU_SYS_BUS_CONFIG, 0x1);
> +	gmu_write(gmu, REG_A6XX_GMU_ICACHE_CONFIG, 0x1);
> +	gmu_write(gmu, REG_A6XX_GMU_DCACHE_CONFIG, 0x1);
>  
>  	gmu_write(gmu, REG_A6XX_GMU_PWR_COL_INTER_FRAME_CTRL, 0x9c40400);
>  
> @@ -600,14 +602,91 @@ static void a6xx_gmu_power_config(struct a6xx_gmu *gmu)
>  		A6XX_GMU_RPMH_CTRL_GFX_VOTE_ENABLE);
>  }
>  
> +static int in_range(u32 addr, u32 start, u32 size)
> +{
> +	return addr >= start && addr < start + size;
> +}

I can't believe they don't have a clever macro in include/kernel.h for range
checking.

> +static int a6xx_gmu_fw_load(struct a6xx_gmu *gmu)
> +{
> +	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
> +	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
> +	const struct firmware *fw_image = adreno_gpu->fw[ADRENO_FW_GMU];
> +	const struct block_header {
> +		u32 addr;
> +		u32 size;
> +		u32 type;
> +		u32 value;
> +		u32 data[];
> +	} *blk;
> +	struct a6xx_gmu_bo *mem;
> +	u32 reg;
> +
> +	u32 itcm_base = 0x00000000;
> +	u32 dtcm_base = 0x00040000;
> +
> +	if (adreno_is_a650(adreno_gpu))
> +		dtcm_base = 0x10004000;
> +

Yet another one of those spots that might grow an ugly conditional.

> +	if (gmu->legacy) {
> +		/* Sanity check the size of the firmware that was loaded */
> +		if (fw_image->size > 0x8000) {
> +			DRM_DEV_ERROR(gmu->dev,
> +				"GMU firmware is bigger than the available region\n");
> +			return -EINVAL;
> +		}
> +
> +		gmu_write_bulk(gmu, REG_A6XX_GMU_CM3_ITCM_START,
> +			       (u32*) fw_image->data, fw_image->size >> 2);
> +		return 0;
> +	}
> +
> +
> +	for (blk = (const struct block_header *) fw_image->data;
> +	     (const u8*) blk < fw_image->data + fw_image->size;
> +	     blk = (const struct block_header *) &blk->data[blk->size >> 2]) {
> +		if (blk->size == 0)
> +			continue;
> +
> +		reg = 0;
> +		mem = NULL;
> +
> +		if (in_range(blk->addr, itcm_base, SZ_16K))
> +			reg = REG_A6XX_GMU_CM3_ITCM_START + ((blk->addr - itcm_base) >> 2);

Personally I would just gmu_write_bulk() here just to avoid the extra registers.

> +		else if (in_range(blk->addr, dtcm_base, SZ_16K))
> +			reg = REG_A6XX_GMU_CM3_DTCM_START + ((blk->addr - dtcm_base) >> 2);

> +		else if (in_range(blk->addr, gmu->icache.iova, gmu->icache.size))
> +			mem = &gmu->icache;

I would mempcpy here. 

> +		else if (in_range(blk->addr, gmu->dcache.iova, gmu->dcache.size))

> +			mem = &gmu->dcache;
> +		else if (in_range(blk->addr, gmu->dummy.iova, gmu->dummy.size))
> +			mem = &gmu->dummy;
> +		else {
> +			DRM_DEV_ERROR(gmu->dev,
> +				"failed to match fw block (addr=%.8x size=%d data[0]=%.8x)\n",
> +				blk->addr, blk->size, blk->data[0]);
> +		}
> +
> +		if (reg)
> +			gmu_write_bulk(gmu, reg, blk->data, blk->size >> 2);
> +
> +		if (mem)
> +			memcpy(mem->virt + blk->addr - mem->iova, blk->data, blk->size);
> +	}
> +
> +	return 0;
> +}
> +
>  static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  {
>  	static bool rpmh_init;
>  	struct a6xx_gpu *a6xx_gpu = container_of(gmu, struct a6xx_gpu, gmu);
>  	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
> -	int i, ret;
> +	int ret;
>  	u32 chipid;
> -	u32 *image;
> +
> +	if (adreno_is_a650(adreno_gpu))
> +		gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF, 1);
>  
>  	if (state == GMU_WARM_BOOT) {
>  		ret = a6xx_rpmh_start(gmu);
> @@ -618,13 +697,6 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  			"GMU firmware is not loaded\n"))
>  			return -ENOENT;
>  
> -		/* Sanity check the size of the firmware that was loaded */
> -		if (adreno_gpu->fw[ADRENO_FW_GMU]->size > 0x8000) {
> -			DRM_DEV_ERROR(gmu->dev,
> -				"GMU firmware is bigger than the available region\n");
> -			return -EINVAL;
> -		}
> -
>  		/* Turn on register retention */
>  		gmu_write(gmu, REG_A6XX_GMU_GENERAL_7, 1);
>  
> @@ -638,11 +710,9 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  				return ret;
>  		}
>  
> -		image = (u32 *) adreno_gpu->fw[ADRENO_FW_GMU]->data;
> -
> -		for (i = 0; i < adreno_gpu->fw[ADRENO_FW_GMU]->size >> 2; i++)
> -			gmu_write(gmu, REG_A6XX_GMU_CM3_ITCM_START + i,
> -				image[i]);
> +		ret = a6xx_gmu_fw_load(gmu);
> +		if (ret)
> +			return ret;
>  	}
>  
>  	gmu_write(gmu, REG_A6XX_GMU_CM3_FW_INIT_RESULT, 0);
> @@ -769,12 +839,20 @@ int a6xx_gmu_resume(struct a6xx_gpu *a6xx_gpu)
>  	/* Enable the GMU interrupt */
>  	gmu_write(gmu, REG_A6XX_GMU_AO_HOST_INTERRUPT_CLR, ~0);
>  	gmu_write(gmu, REG_A6XX_GMU_AO_HOST_INTERRUPT_MASK, ~A6XX_GMU_IRQ_MASK);
> +

Extra whitespace change.

>  	enable_irq(gmu->gmu_irq);
>  
>  	/* Check to see if we are doing a cold or warm boot */
>  	status = gmu_read(gmu, REG_A6XX_GMU_GENERAL_7) == 1 ?
>  		GMU_WARM_BOOT : GMU_COLD_BOOT;
>  
> +	/*
> +	 * Warm boot path does not work on newer GPUs
> +	 * Presumably this is because icache/dcache regions must be restored
> +	 */
> +	if (!gmu->legacy)
> +		status = GMU_COLD_BOOT;
> +

To be honest, I'm not 100% sure WARM_BOOT even makes sense on a630. Downstream
4.19 has no paths that actually use WARM_BOOT. It might be worth a shot to
try removing it in a future patch.

>  	ret = a6xx_gmu_fw_start(gmu, status);
>  	if (ret)
>  		goto out;
> @@ -957,6 +1035,9 @@ static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
>  {
>  	msm_gem_kernel_put(gmu->hfi.obj, gmu->aspace, false);
>  	msm_gem_kernel_put(gmu->debug.obj, gmu->aspace, false);
> +	msm_gem_kernel_put(gmu->icache.obj, gmu->aspace, false);
> +	msm_gem_kernel_put(gmu->dcache.obj, gmu->aspace, false);
> +	msm_gem_kernel_put(gmu->dummy.obj, gmu->aspace, false);
>  
>  	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
>  	msm_gem_address_space_put(gmu->aspace);
> @@ -974,12 +1055,14 @@ static int a6xx_gmu_memory_alloc(struct a6xx_gmu *gmu, struct a6xx_gmu_bo *bo,
>  	size = PAGE_ALIGN(size);
>  	if (!iova) {
>  		/* no fixed address - use GMU's uncached range */
> -		range_start = 0x60000000;
> +		range_start = 0x60000000 + PAGE_SIZE; /* skip dummy page */
>  		range_end = 0x80000000;
>  	} else {
>  		/* range for fixed address */
>  		range_start = iova;
>  		range_end = iova + size;
> +		/* use IOMMU_PRIV for icache/dcache */
> +		flags |= MSM_BO_MAP_PRIV;
>  	}
>  
>  	bo->obj = msm_gem_new(dev, size, flags);
> @@ -1320,7 +1403,27 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  	if (ret)
>  		goto err_put_device;
>  
> -	if (!adreno_is_a640(adreno_gpu) && !adreno_is_a650(adreno_gpu)) {
> +	/* Allocate memory for the GMU dummy page */
> +	ret = a6xx_gmu_memory_alloc(gmu, &gmu->dummy, SZ_4K, 0x60000000);
> +	if (ret)
> +		goto err_memory;
> +
> +	if (adreno_is_a650(adreno_gpu)) {
> +		ret = a6xx_gmu_memory_alloc(gmu, &gmu->icache,
> +			SZ_16M - SZ_16K, 0x04000);
> +		if (ret)
> +			goto err_memory;
> +	} else if (adreno_is_a640(adreno_gpu)) {
> +		ret = a6xx_gmu_memory_alloc(gmu, &gmu->icache,
> +			SZ_256K - SZ_16K, 0x04000);
> +		if (ret)
> +			goto err_memory;
> +
> +		ret = a6xx_gmu_memory_alloc(gmu, &gmu->dcache,
> +			SZ_256K - SZ_16K, 0x44000);
> +		if (ret)
> +			goto err_memory;
> +	} else {
>  		/* HFI v1, has sptprac */
>  		gmu->legacy = true;
>  
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> index 463e2d5f2bb9..526911010cb1 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> @@ -57,6 +57,9 @@ struct a6xx_gmu {
>  
>  	struct a6xx_gmu_bo hfi;
>  	struct a6xx_gmu_bo debug;
> +	struct a6xx_gmu_bo icache;
> +	struct a6xx_gmu_bo dcache;
> +	struct a6xx_gmu_bo dummy;
>  
>  	int nr_clocks;
>  	struct clk_bulk_data *clocks;
> @@ -92,6 +95,14 @@ static inline void gmu_write(struct a6xx_gmu *gmu, u32 offset, u32 value)
>  	return msm_writel(value, gmu->mmio + (offset << 2));
>  }
>  
> +static inline void
> +gmu_write_bulk(struct a6xx_gmu *gmu, u32 offset, const u32 *data, u32 size)
> +{
> +	int i;
> +	for (i = 0; i < size; i++)
> +		gmu_write(gmu, offset + i, data[i]);
> +}

mempcy_toio() might be helpful here instead.

> +
>  static inline void gmu_rmw(struct a6xx_gmu *gmu, u32 reg, u32 mask, u32 or)
>  {
>  	u32 val = gmu_read(gmu, reg);
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> index 1cc1c135236b..eb2cd41dae6e 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> @@ -101,6 +101,10 @@ static inline uint32_t A6XX_HFI_IRQ_OOB_MASK(uint32_t val)
>  
>  #define REG_A6XX_GMU_DCVS_RETURN				0x000023ff
>  
> +#define REG_A6XX_GMU_ICACHE_CONFIG				0x00004c00
> +
> +#define REG_A6XX_GMU_DCACHE_CONFIG				0x00004c01
> +
>  #define REG_A6XX_GMU_SYS_BUS_CONFIG				0x00004c0f
>  
>  #define REG_A6XX_GMU_CM3_SYSRESET				0x00005000
> @@ -199,6 +203,8 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
>  
>  #define REG_A6XX_GPU_GMU_CX_GMU_RPMH_POWER_STATE		0x000050ec
>  
> +#define REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF			0x000050f0
> +
>  #define REG_A6XX_GMU_BOOT_KMD_LM_HANDSHAKE			0x000051f0
>  
>  #define REG_A6XX_GMU_LLM_GLM_SLEEP_CTRL				0x00005157
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

RE: [PATCH v2 6/9] drm/msm/a6xx: A640/A650 GMU firmware path

[Ruhl, Michael J]

>-----Original Message-----
>From: Jonathan Marek <jonathan@marek.ca>
>Sent: Tuesday, April 21, 2020 7:41 PM
>To: freedreno@lists.freedesktop.org
>Cc: Rob Clark <robdclark@gmail.com>; Sean Paul <sean@poorly.run>; David
>Airlie <airlied@linux.ie>; Daniel Vetter <daniel@ffwll.ch>; Jordan Crouse
><jcrouse@codeaurora.org>; Sharat Masetty <smasetty@codeaurora.org>;
>Ruhl, Michael J <michael.j.ruhl@intel.com>; open list:DRM DRIVER FOR MSM
>ADRENO GPU <linux-arm-msm@vger.kernel.org>; open list:DRM DRIVER FOR
>MSM ADRENO GPU <dri-devel@lists.freedesktop.org>; open list <linux-
>kernel@vger.kernel.org>
>Subject: [PATCH v2 6/9] drm/msm/a6xx: A640/A650 GMU firmware path
>
>Newer GPUs have different GMU firmware path.
>
>Signed-off-by: Jonathan Marek <jonathan@marek.ca>
>---
> drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 135 +++++++++++++++++++-
>--
> drivers/gpu/drm/msm/adreno/a6xx_gmu.h     |  11 ++
> drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h |   6 +
> 3 files changed, 136 insertions(+), 16 deletions(-)
>
>diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
>b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
>index b22a69e2f4b0..4aef5fe985d6 100644
>--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
>+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
>@@ -571,6 +571,8 @@ static void a6xx_gmu_power_config(struct a6xx_gmu
>*gmu)
> {
> 	/* Disable GMU WB/RB buffer */
> 	gmu_write(gmu, REG_A6XX_GMU_SYS_BUS_CONFIG, 0x1);
>+	gmu_write(gmu, REG_A6XX_GMU_ICACHE_CONFIG, 0x1);
>+	gmu_write(gmu, REG_A6XX_GMU_DCACHE_CONFIG, 0x1);
>
> 	gmu_write(gmu, REG_A6XX_GMU_PWR_COL_INTER_FRAME_CTRL,
>0x9c40400);
>
>@@ -600,14 +602,91 @@ static void a6xx_gmu_power_config(struct
>a6xx_gmu *gmu)
> 		A6XX_GMU_RPMH_CTRL_GFX_VOTE_ENABLE);
> }
>
>+static int in_range(u32 addr, u32 start, u32 size)
>+{
>+	return addr >= start && addr < start + size;
>+}

Minor nit:

should this return a bool?

M

[PATCH v2 7/9] drm/msm/a6xx: update pdc/rscc GMU registers for A640/A650

[Jonathan Marek]

Update the gmu_pdc registers for A640 and A650.

Some of the RSCC registers on A650 are in a separate region.

Note this also changes the address of these registers:

RSCC_TCS1_DRV0_STATUS
RSCC_TCS2_DRV0_STATUS
RSCC_TCS3_DRV0_STATUS

Based on the values in msm-4.14 and msm-4.19 kernels.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 90 ++++++++++++++---------
 drivers/gpu/drm/msm/adreno/a6xx_gmu.h     | 10 +++
 drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h | 38 +++++-----
 3 files changed, 85 insertions(+), 53 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index 4aef5fe985d6..60ed5360fd8a 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -421,7 +421,7 @@ static int a6xx_rpmh_start(struct a6xx_gmu *gmu)
 		return ret;
 	}
 
-	ret = gmu_poll_timeout(gmu, REG_A6XX_RSCC_SEQ_BUSY_DRV0, val,
+	ret = gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_SEQ_BUSY_DRV0, val,
 		!val, 100, 10000);
 
 	if (ret) {
@@ -447,7 +447,7 @@ static void a6xx_rpmh_stop(struct a6xx_gmu *gmu)
 
 	gmu_write(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ, 1);
 
-	ret = gmu_poll_timeout(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0,
+	ret = gmu_poll_timeout_rscc(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0,
 		val, val & (1 << 16), 100, 10000);
 	if (ret)
 		DRM_DEV_ERROR(gmu->dev, "Unable to power off the GPU RSC\n");
@@ -470,32 +470,48 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
 	struct platform_device *pdev = to_platform_device(gmu->dev);
 	void __iomem *pdcptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc");
 	void __iomem *seqptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc_seq");
+	uint32_t pdc_address_offset;
 
 	if (!pdcptr || !seqptr)
 		goto err;
 
+	if (adreno_is_a618(adreno_gpu) || adreno_is_a640(adreno_gpu))
+		pdc_address_offset = 0x30090;
+	else if (adreno_is_a650(adreno_gpu))
+		pdc_address_offset = 0x300a0;
+	else
+		pdc_address_offset = 0x30080;
+
 	/* Disable SDE clock gating */
-	gmu_write(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0, BIT(24));
+	gmu_write_rscc(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0, BIT(24));
 
 	/* Setup RSC PDC handshake for sleep and wakeup */
-	gmu_write(gmu, REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0, 1);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 2, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 2, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 4, 0x80000000);
-	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 4, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_OVERRIDE_START_ADDR, 0);
-	gmu_write(gmu, REG_A6XX_RSCC_PDC_SEQ_START_ADDR, 0x4520);
-	gmu_write(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_LO, 0x4510);
-	gmu_write(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_HI, 0x4514);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0, 1);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 2, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 2, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 4, 0x80000000);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 4, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_OVERRIDE_START_ADDR, 0);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_SEQ_START_ADDR, 0x4520);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_LO, 0x4510);
+	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_HI, 0x4514);
 
 	/* Load RSC sequencer uCode for sleep and wakeup */
-	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xa7a506a0);
-	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xa1e6a6e7);
-	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e081e1);
-	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xe9a982e2);
-	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020e8a8);
+	if (adreno_is_a650(adreno_gpu)) {
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xeaaae5a0);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xe1a1ebab);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e0a581);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xecac82e2);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020edad);
+	} else {
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xa7a506a0);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xa1e6a6e7);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e081e1);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xe9a982e2);
+		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020e8a8);
+	}
 
 	/* Load PDC sequencer uCode for power up and power down sequence */
 	pdc_write(seqptr, REG_A6XX_PDC_GPU_SEQ_MEM_0, 0xfebea1e1);
@@ -516,10 +532,7 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_DATA + 4, 0x0);
 
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_MSGID + 8, 0x10108);
-	if (adreno_is_a618(adreno_gpu))
-		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, 0x30090);
-	else
-		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, 0x30080);
+	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, pdc_address_offset);
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_DATA + 8, 0x0);
 
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD_ENABLE_BANK, 7);
@@ -531,17 +544,12 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
 
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_MSGID + 4, 0x10108);
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 4, 0x30000);
-	if (adreno_is_a618(adreno_gpu))
+	if (adreno_is_a618(adreno_gpu) || adreno_is_a650(adreno_gpu))
 		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 4, 0x2);
 	else
 		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 4, 0x3);
-
-
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_MSGID + 8, 0x10108);
-	if (adreno_is_a618(adreno_gpu))
-		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, 0x30090);
-	else
-		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, 0x30080);
+	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, pdc_address_offset);
 	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 8, 0x3);
 
 	/* Setup GPU PDC */
@@ -784,13 +792,13 @@ static void a6xx_gmu_rpmh_off(struct a6xx_gmu *gmu)
 	u32 val;
 
 	/* Make sure there are no outstanding RPMh votes */
-	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS0_DRV0_STATUS, val,
+	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS0_DRV0_STATUS, val,
 		(val & 1), 100, 10000);
-	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS1_DRV0_STATUS, val,
+	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS1_DRV0_STATUS, val,
 		(val & 1), 100, 10000);
-	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS2_DRV0_STATUS, val,
+	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS2_DRV0_STATUS, val,
 		(val & 1), 100, 10000);
-	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS3_DRV0_STATUS, val,
+	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS3_DRV0_STATUS, val,
 		(val & 1), 100, 1000);
 }
 
@@ -1349,6 +1357,7 @@ static int a6xx_gmu_get_irq(struct a6xx_gmu *gmu, struct platform_device *pdev,
 
 void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 {
+	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
 	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
 
 	if (!gmu->initialized)
@@ -1362,7 +1371,10 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
 	}
 
 	iounmap(gmu->mmio);
+	if (adreno_is_a650(adreno_gpu))
+		iounmap(gmu->rscc);
 	gmu->mmio = NULL;
+	gmu->rscc = NULL;
 
 	a6xx_gmu_memory_free(gmu);
 
@@ -1445,6 +1457,14 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 		goto err_memory;
 	}
 
+	if (adreno_is_a650(adreno_gpu)) {
+		gmu->rscc = a6xx_gmu_get_mmio(pdev, "rscc");
+		if (IS_ERR(gmu->rscc))
+			goto err_mmio;
+	} else {
+		gmu->rscc = gmu->mmio + 0x23000;
+	}
+
 	/* Get the HFI and GMU interrupts */
 	gmu->hfi_irq = a6xx_gmu_get_irq(gmu, pdev, "hfi", a6xx_hfi_irq);
 	gmu->gmu_irq = a6xx_gmu_get_irq(gmu, pdev, "gmu", a6xx_gmu_irq);
@@ -1470,6 +1490,8 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 
 err_mmio:
 	iounmap(gmu->mmio);
+	if (adreno_is_a650(adreno_gpu))
+		iounmap(gmu->rscc);
 	free_irq(gmu->gmu_irq, gmu);
 	free_irq(gmu->hfi_irq, gmu);
 
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
index 526911010cb1..aed02b2b7659 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
@@ -47,6 +47,7 @@ struct a6xx_gmu {
 	struct msm_gem_address_space *aspace;
 
 	void * __iomem mmio;
+	void * __iomem rscc;
 
 	int hfi_irq;
 	int gmu_irq;
@@ -126,6 +127,15 @@ static inline u64 gmu_read64(struct a6xx_gmu *gmu, u32 lo, u32 hi)
 	readl_poll_timeout((gmu)->mmio + ((addr) << 2), val, cond, \
 		interval, timeout)
 
+static inline void gmu_write_rscc(struct a6xx_gmu *gmu, u32 offset, u32 value)
+{
+	return msm_writel(value, gmu->rscc + (offset << 2));
+}
+
+#define gmu_poll_timeout_rscc(gmu, addr, val, cond, interval, timeout) \
+	readl_poll_timeout((gmu)->rscc + ((addr) << 2), val, cond, \
+		interval, timeout)
+
 /*
  * These are the available OOB (out of band requests) to the GMU where "out of
  * band" means that the CPU talks to the GMU directly and not through HFI.
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
index eb2cd41dae6e..b4357ea550ec 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
@@ -336,8 +336,6 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
 
 #define REG_A6XX_GMU_AO_SPARE_CNTL				0x00009316
 
-#define REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0			0x00008c04
-
 #define REG_A6XX_GMU_RSCC_CONTROL_REQ				0x00009307
 
 #define REG_A6XX_GMU_RSCC_CONTROL_ACK				0x00009308
@@ -350,39 +348,41 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
 
 #define REG_A6XX_GPU_CC_GX_DOMAIN_MISC				0x00009d42
 
-#define REG_A6XX_RSCC_PDC_SEQ_START_ADDR			0x00008c08
+#define REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0			0x00000004
+
+#define REG_A6XX_RSCC_PDC_SEQ_START_ADDR			0x00000008
 
-#define REG_A6XX_RSCC_PDC_MATCH_VALUE_LO			0x00008c09
+#define REG_A6XX_RSCC_PDC_MATCH_VALUE_LO			0x00000009
 
-#define REG_A6XX_RSCC_PDC_MATCH_VALUE_HI			0x00008c0a
+#define REG_A6XX_RSCC_PDC_MATCH_VALUE_HI			0x0000000a
 
-#define REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0				0x00008c0b
+#define REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0				0x0000000b
 
-#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR			0x00008c0d
+#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR			0x0000000d
 
-#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA			0x00008c0e
+#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA			0x0000000e
 
-#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_L_DRV0		0x00008c82
+#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_L_DRV0		0x00000082
 
-#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_H_DRV0		0x00008c83
+#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_H_DRV0		0x00000083
 
-#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_EN_DRV0			0x00008c89
+#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_EN_DRV0			0x00000089
 
-#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_OUTPUT_DRV0		0x00008c8c
+#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_OUTPUT_DRV0		0x0000008c
 
-#define REG_A6XX_RSCC_OVERRIDE_START_ADDR			0x00008d00
+#define REG_A6XX_RSCC_OVERRIDE_START_ADDR			0x00000100
 
-#define REG_A6XX_RSCC_SEQ_BUSY_DRV0				0x00008d01
+#define REG_A6XX_RSCC_SEQ_BUSY_DRV0				0x00000101
 
-#define REG_A6XX_RSCC_SEQ_MEM_0_DRV0				0x00008d80
+#define REG_A6XX_RSCC_SEQ_MEM_0_DRV0				0x00000180
 
-#define REG_A6XX_RSCC_TCS0_DRV0_STATUS				0x00008f46
+#define REG_A6XX_RSCC_TCS0_DRV0_STATUS				0x00000346
 
-#define REG_A6XX_RSCC_TCS1_DRV0_STATUS				0x000090ae
+#define REG_A6XX_RSCC_TCS1_DRV0_STATUS				0x000003ee
 
-#define REG_A6XX_RSCC_TCS2_DRV0_STATUS				0x00009216
+#define REG_A6XX_RSCC_TCS2_DRV0_STATUS				0x00000496
 
-#define REG_A6XX_RSCC_TCS3_DRV0_STATUS				0x0000937e
+#define REG_A6XX_RSCC_TCS3_DRV0_STATUS				0x0000053e
 
 
 #endif /* A6XX_GMU_XML */
-- 
2.26.1

Re: [PATCH v2 7/9] drm/msm/a6xx: update pdc/rscc GMU registers for A640/A650

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:25PM -0400, Jonathan Marek wrote:
> Update the gmu_pdc registers for A640 and A650.
> 
> Some of the RSCC registers on A650 are in a separate region.
> 
> Note this also changes the address of these registers:
> 
> RSCC_TCS1_DRV0_STATUS
> RSCC_TCS2_DRV0_STATUS
> RSCC_TCS3_DRV0_STATUS
> 
> Based on the values in msm-4.14 and msm-4.19 kernels.
> 
> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 90 ++++++++++++++---------
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.h     | 10 +++
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h | 38 +++++-----
>  3 files changed, 85 insertions(+), 53 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> index 4aef5fe985d6..60ed5360fd8a 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> @@ -421,7 +421,7 @@ static int a6xx_rpmh_start(struct a6xx_gmu *gmu)
>  		return ret;
>  	}
>  
> -	ret = gmu_poll_timeout(gmu, REG_A6XX_RSCC_SEQ_BUSY_DRV0, val,
> +	ret = gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_SEQ_BUSY_DRV0, val,
>  		!val, 100, 10000);
>  
>  	if (ret) {
> @@ -447,7 +447,7 @@ static void a6xx_rpmh_stop(struct a6xx_gmu *gmu)
>  
>  	gmu_write(gmu, REG_A6XX_GMU_RSCC_CONTROL_REQ, 1);
>  
> -	ret = gmu_poll_timeout(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0,
> +	ret = gmu_poll_timeout_rscc(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0,
>  		val, val & (1 << 16), 100, 10000);
>  	if (ret)
>  		DRM_DEV_ERROR(gmu->dev, "Unable to power off the GPU RSC\n");
> @@ -470,32 +470,48 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
>  	struct platform_device *pdev = to_platform_device(gmu->dev);
>  	void __iomem *pdcptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc");
>  	void __iomem *seqptr = a6xx_gmu_get_mmio(pdev, "gmu_pdc_seq");
> +	uint32_t pdc_address_offset;
>  
>  	if (!pdcptr || !seqptr)
>  		goto err;
>  
> +	if (adreno_is_a618(adreno_gpu) || adreno_is_a640(adreno_gpu))
> +		pdc_address_offset = 0x30090;
> +	else if (adreno_is_a650(adreno_gpu))
> +		pdc_address_offset = 0x300a0;
> +	else
> +		pdc_address_offset = 0x30080;

This might be another one of those things that we should consider adding to the
gpu list.

> +
>  	/* Disable SDE clock gating */
> -	gmu_write(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0, BIT(24));
> +	gmu_write_rscc(gmu, REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0, BIT(24));
>  
>  	/* Setup RSC PDC handshake for sleep and wakeup */
> -	gmu_write(gmu, REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0, 1);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 2, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 2, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 4, 0x80000000);
> -	gmu_write(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 4, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_OVERRIDE_START_ADDR, 0);
> -	gmu_write(gmu, REG_A6XX_RSCC_PDC_SEQ_START_ADDR, 0x4520);
> -	gmu_write(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_LO, 0x4510);
> -	gmu_write(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_HI, 0x4514);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0, 1);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 2, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 2, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA + 4, 0x80000000);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR + 4, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_OVERRIDE_START_ADDR, 0);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_SEQ_START_ADDR, 0x4520);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_LO, 0x4510);
> +	gmu_write_rscc(gmu, REG_A6XX_RSCC_PDC_MATCH_VALUE_HI, 0x4514);
>  
>  	/* Load RSC sequencer uCode for sleep and wakeup */
> -	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xa7a506a0);
> -	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xa1e6a6e7);
> -	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e081e1);
> -	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xe9a982e2);
> -	gmu_write(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020e8a8);
> +	if (adreno_is_a650(adreno_gpu)) {
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xeaaae5a0);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xe1a1ebab);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e0a581);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xecac82e2);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020edad);
> +	} else {
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0, 0xa7a506a0);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 1, 0xa1e6a6e7);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 2, 0xa2e081e1);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 3, 0xe9a982e2);
> +		gmu_write_rscc(gmu, REG_A6XX_RSCC_SEQ_MEM_0_DRV0 + 4, 0x0020e8a8);
> +	}
>  
>  	/* Load PDC sequencer uCode for power up and power down sequence */
>  	pdc_write(seqptr, REG_A6XX_PDC_GPU_SEQ_MEM_0, 0xfebea1e1);
> @@ -516,10 +532,7 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_DATA + 4, 0x0);
>  
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_MSGID + 8, 0x10108);
> -	if (adreno_is_a618(adreno_gpu))
> -		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, 0x30090);
> -	else
> -		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, 0x30080);
> +	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_ADDR + 8, pdc_address_offset);
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS1_CMD0_DATA + 8, 0x0);
>  
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD_ENABLE_BANK, 7);
> @@ -531,17 +544,12 @@ static void a6xx_gmu_rpmh_init(struct a6xx_gmu *gmu)
>  
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_MSGID + 4, 0x10108);
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 4, 0x30000);
> -	if (adreno_is_a618(adreno_gpu))
> +	if (adreno_is_a618(adreno_gpu) || adreno_is_a650(adreno_gpu))
>  		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 4, 0x2);
>  	else
>  		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 4, 0x3);
> -
> -
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_MSGID + 8, 0x10108);
> -	if (adreno_is_a618(adreno_gpu))
> -		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, 0x30090);
> -	else
> -		pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, 0x30080);
> +	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_ADDR + 8, pdc_address_offset);
>  	pdc_write(pdcptr, REG_A6XX_PDC_GPU_TCS3_CMD0_DATA + 8, 0x3);
>  
>  	/* Setup GPU PDC */
> @@ -784,13 +792,13 @@ static void a6xx_gmu_rpmh_off(struct a6xx_gmu *gmu)
>  	u32 val;
>  
>  	/* Make sure there are no outstanding RPMh votes */
> -	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS0_DRV0_STATUS, val,
> +	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS0_DRV0_STATUS, val,
>  		(val & 1), 100, 10000);
> -	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS1_DRV0_STATUS, val,
> +	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS1_DRV0_STATUS, val,
>  		(val & 1), 100, 10000);
> -	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS2_DRV0_STATUS, val,
> +	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS2_DRV0_STATUS, val,
>  		(val & 1), 100, 10000);
> -	gmu_poll_timeout(gmu, REG_A6XX_RSCC_TCS3_DRV0_STATUS, val,
> +	gmu_poll_timeout_rscc(gmu, REG_A6XX_RSCC_TCS3_DRV0_STATUS, val,
>  		(val & 1), 100, 1000);
>  }
>  
> @@ -1349,6 +1357,7 @@ static int a6xx_gmu_get_irq(struct a6xx_gmu *gmu, struct platform_device *pdev,
>  
>  void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
>  {
> +	struct adreno_gpu *adreno_gpu = &a6xx_gpu->base;
>  	struct a6xx_gmu *gmu = &a6xx_gpu->gmu;
>  
>  	if (!gmu->initialized)
> @@ -1362,7 +1371,10 @@ void a6xx_gmu_remove(struct a6xx_gpu *a6xx_gpu)
>  	}
>  
>  	iounmap(gmu->mmio);
> +	if (adreno_is_a650(adreno_gpu))
> +		iounmap(gmu->rscc);

Another transition point. Every thing after a650 will use the rscc region. It
might be better to do a resource check here instead:

if (platform_get_resource_byname(pdev, "rscc", IORESOURCE_MEM))
   iounmap(gmu->rscc);

It would be really nice to use devm_ but I'm skeptical of how well devres works
with sub-platform devices. Sean Paul has fought with that before and it takes a
combination of platform_device_put() and some good luck. I've mentioned to Rob
before that we might need to think of chaining components together and make GMU
a top level platform-device, but I think the only way we can do that today is to
make GMU a peer of adreno and I'm not sure if that is something we want to do.

Other than that, I really like reusing the gmu->rscc pointer for both types.
That is clever.


>  	gmu->mmio = NULL;
> +	gmu->rscc = NULL;
>  
>  	a6xx_gmu_memory_free(gmu);
>  
> @@ -1445,6 +1457,14 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  		goto err_memory;
>  	}
>  
> +	if (adreno_is_a650(adreno_gpu)) {
> +		gmu->rscc = a6xx_gmu_get_mmio(pdev, "rscc");
> +		if (IS_ERR(gmu->rscc))
> +			goto err_mmio;
> +	} else {
> +		gmu->rscc = gmu->mmio + 0x23000;
> +	}
> +
>  	/* Get the HFI and GMU interrupts */
>  	gmu->hfi_irq = a6xx_gmu_get_irq(gmu, pdev, "hfi", a6xx_hfi_irq);
>  	gmu->gmu_irq = a6xx_gmu_get_irq(gmu, pdev, "gmu", a6xx_gmu_irq);
> @@ -1470,6 +1490,8 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  
>  err_mmio:
>  	iounmap(gmu->mmio);
> +	if (adreno_is_a650(adreno_gpu))
> +		iounmap(gmu->rscc);

Same concern as above.

>  	free_irq(gmu->gmu_irq, gmu);
>  	free_irq(gmu->hfi_irq, gmu);
>  
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> index 526911010cb1..aed02b2b7659 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> @@ -47,6 +47,7 @@ struct a6xx_gmu {
>  	struct msm_gem_address_space *aspace;
>  
>  	void * __iomem mmio;
> +	void * __iomem rscc;
>  
>  	int hfi_irq;
>  	int gmu_irq;
> @@ -126,6 +127,15 @@ static inline u64 gmu_read64(struct a6xx_gmu *gmu, u32 lo, u32 hi)
>  	readl_poll_timeout((gmu)->mmio + ((addr) << 2), val, cond, \
>  		interval, timeout)
>  
> +static inline void gmu_write_rscc(struct a6xx_gmu *gmu, u32 offset, u32 value)
> +{
> +	return msm_writel(value, gmu->rscc + (offset << 2));
> +}
> +
> +#define gmu_poll_timeout_rscc(gmu, addr, val, cond, interval, timeout) \
> +	readl_poll_timeout((gmu)->rscc + ((addr) << 2), val, cond, \
> +		interval, timeout)
> +
>  /*
>   * These are the available OOB (out of band requests) to the GMU where "out of
>   * band" means that the CPU talks to the GMU directly and not through HFI.
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> index eb2cd41dae6e..b4357ea550ec 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> @@ -336,8 +336,6 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
>  
>  #define REG_A6XX_GMU_AO_SPARE_CNTL				0x00009316
>  
> -#define REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0			0x00008c04
> -
>  #define REG_A6XX_GMU_RSCC_CONTROL_REQ				0x00009307
>  
>  #define REG_A6XX_GMU_RSCC_CONTROL_ACK				0x00009308
> @@ -350,39 +348,41 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
>  
>  #define REG_A6XX_GPU_CC_GX_DOMAIN_MISC				0x00009d42
>  
> -#define REG_A6XX_RSCC_PDC_SEQ_START_ADDR			0x00008c08
> +#define REG_A6XX_GPU_RSCC_RSC_STATUS0_DRV0			0x00000004
> +
> +#define REG_A6XX_RSCC_PDC_SEQ_START_ADDR			0x00000008
>  
> -#define REG_A6XX_RSCC_PDC_MATCH_VALUE_LO			0x00008c09
> +#define REG_A6XX_RSCC_PDC_MATCH_VALUE_LO			0x00000009
>  
> -#define REG_A6XX_RSCC_PDC_MATCH_VALUE_HI			0x00008c0a
> +#define REG_A6XX_RSCC_PDC_MATCH_VALUE_HI			0x0000000a
>  
> -#define REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0				0x00008c0b
> +#define REG_A6XX_RSCC_PDC_SLAVE_ID_DRV0				0x0000000b
>  
> -#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR			0x00008c0d
> +#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_ADDR			0x0000000d
>  
> -#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA			0x00008c0e
> +#define REG_A6XX_RSCC_HIDDEN_TCS_CMD0_DATA			0x0000000e
>  
> -#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_L_DRV0		0x00008c82
> +#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_L_DRV0		0x00000082
>  
> -#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_H_DRV0		0x00008c83
> +#define REG_A6XX_RSCC_TIMESTAMP_UNIT0_TIMESTAMP_H_DRV0		0x00000083
>  
> -#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_EN_DRV0			0x00008c89
> +#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_EN_DRV0			0x00000089
>  
> -#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_OUTPUT_DRV0		0x00008c8c
> +#define REG_A6XX_RSCC_TIMESTAMP_UNIT1_OUTPUT_DRV0		0x0000008c
>  
> -#define REG_A6XX_RSCC_OVERRIDE_START_ADDR			0x00008d00
> +#define REG_A6XX_RSCC_OVERRIDE_START_ADDR			0x00000100
>  
> -#define REG_A6XX_RSCC_SEQ_BUSY_DRV0				0x00008d01
> +#define REG_A6XX_RSCC_SEQ_BUSY_DRV0				0x00000101
>  
> -#define REG_A6XX_RSCC_SEQ_MEM_0_DRV0				0x00008d80
> +#define REG_A6XX_RSCC_SEQ_MEM_0_DRV0				0x00000180
>  
> -#define REG_A6XX_RSCC_TCS0_DRV0_STATUS				0x00008f46
> +#define REG_A6XX_RSCC_TCS0_DRV0_STATUS				0x00000346
>  
> -#define REG_A6XX_RSCC_TCS1_DRV0_STATUS				0x000090ae
> +#define REG_A6XX_RSCC_TCS1_DRV0_STATUS				0x000003ee
>  
> -#define REG_A6XX_RSCC_TCS2_DRV0_STATUS				0x00009216
> +#define REG_A6XX_RSCC_TCS2_DRV0_STATUS				0x00000496
>  
> -#define REG_A6XX_RSCC_TCS3_DRV0_STATUS				0x0000937e
> +#define REG_A6XX_RSCC_TCS3_DRV0_STATUS				0x0000053e
>  
>  
>  #endif /* A6XX_GMU_XML */
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 8/9] drm/msm/a6xx: enable GMU log

[Jonathan Marek]

This is required for a650 to work.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 15 +++++++++++++++
 drivers/gpu/drm/msm/adreno/a6xx_gmu.h     |  1 +
 drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h |  4 ++++
 3 files changed, 20 insertions(+)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
index 60ed5360fd8a..f24e436daafe 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
@@ -201,6 +201,12 @@ static int a6xx_gmu_start(struct a6xx_gmu *gmu)
 	u32 val;
 
 	gmu_write(gmu, REG_A6XX_GMU_CM3_SYSRESET, 1);
+
+	/* Set the log wptr index
+	 * note: downstream saves the value in poweroff and restores it here
+	 */
+	gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_RESP, 0);
+
 	gmu_write(gmu, REG_A6XX_GMU_CM3_SYSRESET, 0);
 
 	ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_CM3_FW_INIT_RESULT, val,
@@ -740,6 +746,9 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
 
 	gmu_write(gmu, REG_A6XX_GMU_HFI_SFR_ADDR, chipid);
 
+	gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_MSG,
+		  gmu->log.iova | (gmu->log.size / SZ_4K - 1));
+
 	/* Set up the lowest idle level on the GMU */
 	a6xx_gmu_power_config(gmu);
 
@@ -1046,6 +1055,7 @@ static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
 	msm_gem_kernel_put(gmu->icache.obj, gmu->aspace, false);
 	msm_gem_kernel_put(gmu->dcache.obj, gmu->aspace, false);
 	msm_gem_kernel_put(gmu->dummy.obj, gmu->aspace, false);
+	msm_gem_kernel_put(gmu->log.obj, gmu->aspace, false);
 
 	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
 	msm_gem_address_space_put(gmu->aspace);
@@ -1450,6 +1460,11 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
 	if (ret)
 		goto err_memory;
 
+	/* Allocate memory for the GMU log region */
+	ret = a6xx_gmu_memory_alloc(gmu, &gmu->log, SZ_4K, 0);
+	if (ret)
+		goto err_memory;
+
 	/* Map the GMU registers */
 	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
 	if (IS_ERR(gmu->mmio)) {
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
index aed02b2b7659..be65720f32cf 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
@@ -61,6 +61,7 @@ struct a6xx_gmu {
 	struct a6xx_gmu_bo icache;
 	struct a6xx_gmu_bo dcache;
 	struct a6xx_gmu_bo dummy;
+	struct a6xx_gmu_bo log;
 
 	int nr_clocks;
 	struct clk_bulk_data *clocks;
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
index b4357ea550ec..176ae94d9fe6 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
@@ -205,6 +205,10 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
 
 #define REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF			0x000050f0
 
+#define REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_MSG			0x00005100
+
+#define REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_RESP			0x00005101
+
 #define REG_A6XX_GMU_BOOT_KMD_LM_HANDSHAKE			0x000051f0
 
 #define REG_A6XX_GMU_LLM_GLM_SLEEP_CTRL				0x00005157
-- 
2.26.1

Re: [PATCH v2 8/9] drm/msm/a6xx: enable GMU log

[Jordan Crouse]

On Tue, Apr 21, 2020 at 07:41:26PM -0400, Jonathan Marek wrote:
> This is required for a650 to work.

Reviewed-by: Jordan Crouse <jcrouse@codeaurora.org>

> Signed-off-by: Jonathan Marek <jonathan@marek.ca>
> ---
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.c     | 15 +++++++++++++++
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.h     |  1 +
>  drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h |  4 ++++
>  3 files changed, 20 insertions(+)
> 
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> index 60ed5360fd8a..f24e436daafe 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.c
> @@ -201,6 +201,12 @@ static int a6xx_gmu_start(struct a6xx_gmu *gmu)
>  	u32 val;
>  
>  	gmu_write(gmu, REG_A6XX_GMU_CM3_SYSRESET, 1);
> +
> +	/* Set the log wptr index
> +	 * note: downstream saves the value in poweroff and restores it here
> +	 */
> +	gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_RESP, 0);
> +
>  	gmu_write(gmu, REG_A6XX_GMU_CM3_SYSRESET, 0);
>  
>  	ret = gmu_poll_timeout(gmu, REG_A6XX_GMU_CM3_FW_INIT_RESULT, val,
> @@ -740,6 +746,9 @@ static int a6xx_gmu_fw_start(struct a6xx_gmu *gmu, unsigned int state)
>  
>  	gmu_write(gmu, REG_A6XX_GMU_HFI_SFR_ADDR, chipid);
>  
> +	gmu_write(gmu, REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_MSG,
> +		  gmu->log.iova | (gmu->log.size / SZ_4K - 1));
> +
>  	/* Set up the lowest idle level on the GMU */
>  	a6xx_gmu_power_config(gmu);
>  
> @@ -1046,6 +1055,7 @@ static void a6xx_gmu_memory_free(struct a6xx_gmu *gmu)
>  	msm_gem_kernel_put(gmu->icache.obj, gmu->aspace, false);
>  	msm_gem_kernel_put(gmu->dcache.obj, gmu->aspace, false);
>  	msm_gem_kernel_put(gmu->dummy.obj, gmu->aspace, false);
> +	msm_gem_kernel_put(gmu->log.obj, gmu->aspace, false);
>  
>  	gmu->aspace->mmu->funcs->detach(gmu->aspace->mmu);
>  	msm_gem_address_space_put(gmu->aspace);
> @@ -1450,6 +1460,11 @@ int a6xx_gmu_init(struct a6xx_gpu *a6xx_gpu, struct device_node *node)
>  	if (ret)
>  		goto err_memory;
>  
> +	/* Allocate memory for the GMU log region */
> +	ret = a6xx_gmu_memory_alloc(gmu, &gmu->log, SZ_4K, 0);
> +	if (ret)
> +		goto err_memory;
> +
>  	/* Map the GMU registers */
>  	gmu->mmio = a6xx_gmu_get_mmio(pdev, "gmu");
>  	if (IS_ERR(gmu->mmio)) {
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> index aed02b2b7659..be65720f32cf 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.h
> @@ -61,6 +61,7 @@ struct a6xx_gmu {
>  	struct a6xx_gmu_bo icache;
>  	struct a6xx_gmu_bo dcache;
>  	struct a6xx_gmu_bo dummy;
> +	struct a6xx_gmu_bo log;
>  
>  	int nr_clocks;
>  	struct clk_bulk_data *clocks;
> diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> index b4357ea550ec..176ae94d9fe6 100644
> --- a/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> +++ b/drivers/gpu/drm/msm/adreno/a6xx_gmu.xml.h
> @@ -205,6 +205,10 @@ static inline uint32_t A6XX_GMU_GPU_NAP_CTRL_SID(uint32_t val)
>  
>  #define REG_A6XX_GPU_GMU_CX_GMU_CX_FAL_INTF			0x000050f0
>  
> +#define REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_MSG			0x00005100
> +
> +#define REG_A6XX_GPU_GMU_CX_GMU_PWR_COL_CP_RESP			0x00005101
> +
>  #define REG_A6XX_GMU_BOOT_KMD_LM_HANDSHAKE			0x000051f0
>  
>  #define REG_A6XX_GMU_LLM_GLM_SLEEP_CTRL				0x00005157
> -- 
> 2.26.1
> 

-- 
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH v2 9/9] drm/msm/a6xx: update a6xx_hw_init for A640 and A650

[Jonathan Marek]

Adreno 640 and 650 GPUs need some registers set differently.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
---
 drivers/gpu/drm/msm/adreno/a6xx.xml.h | 14 +++++++
 drivers/gpu/drm/msm/adreno/a6xx_gpu.c | 56 ++++++++++++++++++++++-----
 2 files changed, 61 insertions(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a6xx.xml.h b/drivers/gpu/drm/msm/adreno/a6xx.xml.h
index ed78fee2a262..47840b73cdda 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx.xml.h
+++ b/drivers/gpu/drm/msm/adreno/a6xx.xml.h
@@ -1047,6 +1047,8 @@ enum a6xx_tex_type {
 
 #define REG_A6XX_CP_MISC_CNTL					0x00000840
 
+#define REG_A6XX_CP_APRIV_CNTL					0x00000844
+
 #define REG_A6XX_CP_ROQ_THRESHOLDS_1				0x000008c1
 
 #define REG_A6XX_CP_ROQ_THRESHOLDS_2				0x000008c2
@@ -1764,6 +1766,8 @@ static inline uint32_t A6XX_CP_PROTECT_REG_MASK_LEN(uint32_t val)
 
 #define REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL			0x00000010
 
+#define REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL			0x00000011
+
 #define REG_A6XX_RBBM_INTERFACE_HANG_INT_CNTL			0x0000001f
 
 #define REG_A6XX_RBBM_INT_CLEAR_CMD				0x00000037
@@ -2418,6 +2422,16 @@ static inline uint32_t A6XX_UCHE_CLIENT_PF_PERFSEL(uint32_t val)
 
 #define REG_A6XX_TPL1_NC_MODE_CNTL				0x0000b604
 
+#define REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0			0x0000b608
+
+#define REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1			0x0000b609
+
+#define REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2			0x0000b60a
+
+#define REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3			0x0000b60b
+
+#define REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4			0x0000b60c
+
 #define REG_A6XX_TPL1_PERFCTR_TP_SEL_0				0x0000b610
 
 #define REG_A6XX_TPL1_PERFCTR_TP_SEL_1				0x0000b611
diff --git a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
index a860d4970e10..e1eb34fa3a99 100644
--- a/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a6xx_gpu.c
@@ -414,7 +414,17 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
 		a6xx_set_hwcg(gpu, true);
 
 	/* VBIF/GBIF start*/
-	gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
+	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu)) {
+		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE0, 0x00071620);
+		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE1, 0x00071620);
+		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE2, 0x00071620);
+		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
+		gpu_write(gpu, REG_A6XX_GBIF_QSB_SIDE3, 0x00071620);
+		gpu_write(gpu, REG_A6XX_RBBM_GBIF_CLIENT_QOS_CNTL, 0x3);
+	} else {
+		gpu_write(gpu, REG_A6XX_RBBM_VBIF_CLIENT_QOS_CNTL, 0x3);
+	}
+
 	if (adreno_is_a630(adreno_gpu))
 		gpu_write(gpu, REG_A6XX_VBIF_GATE_OFF_WRREQ_EN, 0x00000009);
 
@@ -429,25 +439,35 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_LO, 0xfffff000);
 	gpu_write(gpu, REG_A6XX_UCHE_WRITE_THRU_BASE_HI, 0x0001ffff);
 
-	/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
-	gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN_LO,
-		REG_A6XX_UCHE_GMEM_RANGE_MIN_HI, 0x00100000);
+	if (!adreno_is_a650(adreno_gpu)) {
+		/* Set the GMEM VA range [0x100000:0x100000 + gpu->gmem - 1] */
+		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MIN_LO,
+			REG_A6XX_UCHE_GMEM_RANGE_MIN_HI, 0x00100000);
 
-	gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX_LO,
-		REG_A6XX_UCHE_GMEM_RANGE_MAX_HI,
-		0x00100000 + adreno_gpu->gmem - 1);
+		gpu_write64(gpu, REG_A6XX_UCHE_GMEM_RANGE_MAX_LO,
+			REG_A6XX_UCHE_GMEM_RANGE_MAX_HI,
+			0x00100000 + adreno_gpu->gmem - 1);
+	}
 
 	gpu_write(gpu, REG_A6XX_UCHE_FILTER_CNTL, 0x804);
 	gpu_write(gpu, REG_A6XX_UCHE_CACHE_WAYS, 0x4);
 
-	gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
+	if (adreno_is_a640(adreno_gpu) || adreno_is_a650(adreno_gpu))
+		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x02000140);
+	else
+		gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_2, 0x010000c0);
 	gpu_write(gpu, REG_A6XX_CP_ROQ_THRESHOLDS_1, 0x8040362c);
 
 	/* Setting the mem pool size */
 	gpu_write(gpu, REG_A6XX_CP_MEM_POOL_SIZE, 128);
 
 	/* Setting the primFifo thresholds default values */
-	gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, (0x300 << 11));
+	if (adreno_is_a650(adreno_gpu))
+		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00300000);
+	else if (adreno_is_a640(adreno_gpu))
+		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, 0x00200000);
+	else
+		gpu_write(gpu, REG_A6XX_PC_DBG_ECO_CNTL, (0x300 << 11));
 
 	/* Set the AHB default slave response to "ERROR" */
 	gpu_write(gpu, REG_A6XX_CP_AHB_CNTL, 0x1);
@@ -471,6 +491,19 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
 
 	gpu_write(gpu, REG_A6XX_UCHE_CLIENT_PF, 1);
 
+	/* Set weights for bicubic filtering */
+	if (adreno_is_a650(adreno_gpu)) {
+		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_0, 0);
+		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_1,
+			0x3fe05ff4);
+		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_2,
+			0x3fa0ebee);
+		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_3,
+			0x3f5193ed);
+		gpu_write(gpu, REG_A6XX_TPL1_BICUBIC_WEIGHTS_TABLE_4,
+			0x3f0243f0);
+	}
+
 	/* Protect registers from the CP */
 	gpu_write(gpu, REG_A6XX_CP_PROTECT_CNTL, 0x00000003);
 
@@ -508,6 +541,11 @@ static int a6xx_hw_init(struct msm_gpu *gpu)
 			A6XX_PROTECT_RDONLY(0x980, 0x4));
 	gpu_write(gpu, REG_A6XX_CP_PROTECT(25), A6XX_PROTECT_RW(0xa630, 0x0));
 
+	if (adreno_is_a650(adreno_gpu)) {
+		gpu_write(gpu, REG_A6XX_CP_APRIV_CNTL,
+			(1 << 6) | (1 << 5) | (1 << 3) | (1 << 2) | (1 << 1));
+	}
+
 	/* Enable interrupts */
 	gpu_write(gpu, REG_A6XX_RBBM_INT_0_MASK, A6XX_INT_MASK);
 
-- 
2.26.1