Compiling Mali Midgard driver for Linux v6.5.0 (Mainline) in order to obtain OpenCL with closed-source libMali for Rock Pi N10
|
6 min read
As Linux kernel versions advance some out-of-tree modules begin to fall into obscurity and obsolescence.
While I appreciate the efforts of LibreELEC team to try to maintain Mali Midgard drivers working the latest kernel supported is 5.16 which is eternity past if you need Mainline Linux.
This is a diffstat specially aimed at Rock Pi N10 users - perhaps it could work for Rock Pi 4, I don't know! - for Linux kernel v6.5.0, latest Mainline as of today.
Just cd to mali-midgard/driver/product/kernel/drivers/gpu/arm/midgard and make.
Make sure you adhere to the exact same compiler you used to build the kernel. For example, I used distcc to build the kernel but forgot to pass CC environment variable as in make -j16 CC="distcc gcc" and that was enough for build to fail.
Mali driver you just built which creates /dev/mali0 node for closed-source libMali needs to be inserted without panfrost loaded. Make sure you modprobe -r panfrost from a real console or via ssh and then insmod mali-midgard/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase.ko which should result in the following lines to your dmesg:
[18631.912063] mali_kbase: loading out-of-tree module taints kernel.
[18631.915590] mali ff9a0000.gpu: Continuing without Mali regulator control
[18631.915613] mali ff9a0000.gpu: Continuing without Mali bus clock
[18631.916570] mali ff9a0000.gpu: GPU identified as 0x0860 r2p0 status 0
[18631.916714] mali ff9a0000.gpu: Protected mode not available
[18631.917414] mali ff9a0000.gpu: Probed as mali0
Make sure /etc/OpenCL/vendors/ has some .icd pointing to the correct libMali.so. I recommend you give dlprimitives and its children pytorch_dlprim a good try with your newly-acquired OpenCL capabilities.
From 65df8d090cfeff8a692008e20f2241f7d325e267 Mon Sep 17 00:00:00 2001
From: Geraldo Nascimento <geraldogabriel@gmail.com>
Date: Fri, 1 Sep 2023 00:50:56 -0300
Subject: [PATCH 1/1] Very hacky, tentative support for Linux v6.5.0 (Mainline)
---
.../kernel/drivers/gpu/arm/midgard/Makefile | 5 +-
.../backend/gpu/mali_kbase_irq_linux.c | 16 +---
.../gpu/arm/midgard/mali_kbase_core_linux.c | 15 +--
.../drivers/gpu/arm/midgard/mali_kbase_defs.h | 4 +-
.../drivers/gpu/arm/midgard/mali_kbase_mem.c | 2 +-
.../gpu/arm/midgard/mali_kbase_mem_linux.c | 38 ++++----
.../gpu/arm/midgard/mali_kbase_mem_pool.c | 2 +-
.../arm/midgard/thirdparty/mali_kbase_mmap.c | 91 ++-----------------
8 files changed, 47 insertions(+), 126 deletions(-)
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/Makefile b/driver/product/kernel/drivers/gpu/arm/midgard/Makefile
index 08b2fa9..22929b6 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/Makefile
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/Makefile
@@ -20,7 +20,8 @@
#
-KDIR ?= /lib/modules/$(shell uname -r)/build
+#KDIR ?= /lib/modules/$(shell uname -r)/build
+KDIR=/usr/src/linux
BUSLOG_PATH_RELATIVE = $(CURDIR)/../../../..
KBASE_PATH_RELATIVE = $(CURDIR)
@@ -32,7 +33,7 @@ endif
# we get the symbols from modules using KBUILD_EXTRA_SYMBOLS to prevent warnings about unknown functions
all:
- $(MAKE) -C $(KDIR) M=$(CURDIR) EXTRA_CFLAGS="-I$(CURDIR)/../../../../include -I$(CURDIR)/../../../../tests/include $(SCONS_CFLAGS)" $(SCONS_CONFIGS) KBUILD_EXTRA_SYMBOLS="$(EXTRA_SYMBOLS)" modules
+ $(MAKE) -C $(KDIR) M=$(CURDIR) EXTRA_CFLAGS="-I$(CURDIR)/../../../../include" KBUILD_EXTRA_SYMBOLS="$(EXTRA_SYMBOLS)" CONFIG_MALI_MIDGARD=m CONFIG_MALI_PLATFORM_NAME=rk modules
clean:
$(MAKE) -C $(KDIR) M=$(CURDIR) clean
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_irq_linux.c b/driver/product/kernel/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_irq_linux.c
index dd0279a..5b4b405 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_irq_linux.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/backend/gpu/mali_kbase_irq_linux.c
@@ -426,8 +426,8 @@ int kbase_install_interrupts(struct kbase_device *kbdev)
u32 i;
for (i = 0; i < nr; i++) {
- err = request_irq(kbdev->irqs[i].irq, kbase_handler_table[i],
- kbdev->irqs[i].flags | IRQF_SHARED,
+ err = devm_request_irq(kbdev->dev, kbdev->irqs[i].irq, kbase_handler_table[i],
+ 0,
dev_name(kbdev->dev),
kbase_tag(kbdev, i));
if (err) {
@@ -436,17 +436,11 @@ int kbase_install_interrupts(struct kbase_device *kbdev)
#ifdef CONFIG_SPARSE_IRQ
dev_err(kbdev->dev, "You have CONFIG_SPARSE_IRQ support enabled - is the interrupt number correct for this configuration?\n");
#endif /* CONFIG_SPARSE_IRQ */
- goto release;
+ return err;
}
}
return 0;
-
- release:
- while (i-- > 0)
- free_irq(kbdev->irqs[i].irq, kbase_tag(kbdev, i));
-
- return err;
}
void kbase_release_interrupts(struct kbase_device *kbdev)
@@ -454,10 +448,10 @@ void kbase_release_interrupts(struct kbase_device *kbdev)
u32 nr = ARRAY_SIZE(kbase_handler_table);
u32 i;
- for (i = 0; i < nr; i++) {
+ /*for (i = 0; i < nr; i++) {
if (kbdev->irqs[i].irq)
free_irq(kbdev->irqs[i].irq, kbase_tag(kbdev, i));
- }
+ }*/
}
void kbase_synchronize_irqs(struct kbase_device *kbdev)
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_core_linux.c b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_core_linux.c
index 58e7335..ee0d6fe 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_core_linux.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_core_linux.c
@@ -888,16 +888,16 @@ static int assign_irqs(struct platform_device *pdev)
/* 3 IRQ resources */
for (i = 0; i < 3; i++) {
- struct resource *irq_res;
+ int irq_res;
int irqtag;
- irq_res = platform_get_resource(pdev, IORESOURCE_IRQ, i);
- if (!irq_res) {
+ irq_res = platform_get_irq(pdev, i);
+ if (irq_res < 0) {
dev_err(kbdev->dev, "No IRQ resource at index %d\n", i);
return -ENOENT;
}
-#ifdef CONFIG_OF
+/*#ifdef CONFIG_OF
if (!strncmp(irq_res->name, "job", 4)) {
irqtag = JOB_IRQ_TAG;
} else if (!strncmp(irq_res->name, "mmu", 4)) {
@@ -911,9 +911,10 @@ static int assign_irqs(struct platform_device *pdev)
}
#else
irqtag = i;
-#endif /* CONFIG_OF */
- kbdev->irqs[irqtag].irq = irq_res->start;
- kbdev->irqs[irqtag].flags = irq_res->flags & IRQF_TRIGGER_MASK;
+#endif*/ /* CONFIG_OF */
+ irqtag = i;
+ kbdev->irqs[irqtag].irq = irq_res;
+ kbdev->irqs[irqtag].flags = 0;
}
return 0;
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_defs.h b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_defs.h
index 1836485..922dc70 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_defs.h
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_defs.h
@@ -1419,8 +1419,8 @@ struct kbase_device {
void __iomem *reg;
struct {
- int irq;
- int flags;
+ unsigned int irq;
+ unsigned long flags;
} irqs[3];
struct clk *clock;
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem.c b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem.c
index 04015e1..9415c97 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem.c
@@ -3401,7 +3401,7 @@ static int kbase_jd_user_buf_map(struct kbase_context *kctx,
address,
alloc->imported.user_buf.nr_pages,
reg->flags & KBASE_REG_GPU_WR ? FOLL_WRITE : 0,
- pages, NULL, NULL);
+ pages, NULL);
#endif
if (pinned_pages <= 0)
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_linux.c b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_linux.c
index 248fa12..58bf3be 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_linux.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_linux.c
@@ -640,7 +640,7 @@ int kbase_mem_evictable_init(struct kbase_context *kctx)
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 1, 0)
kctx->reclaim.batch = 0;
#endif
- register_shrinker(&kctx->reclaim);
+ register_shrinker(&kctx->reclaim, "mali_kbase");
return 0;
}
@@ -1113,7 +1113,7 @@ static struct kbase_va_region *kbase_mem_from_user_buffer(
#else
faulted_pages = get_user_pages(address, *va_pages,
reg->flags & KBASE_REG_GPU_WR ? FOLL_WRITE : 0,
- pages, NULL);
+ pages);
#endif
up_read(¤t->mm->mmap_lock);
@@ -1827,11 +1827,7 @@ static int kbase_cpu_mmap(struct kbase_context *kctx,
* See MIDBASE-1057
*/
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0))
- vma->vm_flags |= VM_DONTCOPY | VM_DONTDUMP | VM_DONTEXPAND | VM_IO;
-#else
- vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED | VM_IO;
-#endif
+ vm_flags_set(vma, VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP | VM_IO);
vma->vm_ops = &kbase_vm_ops;
vma->vm_private_data = map;
@@ -1883,7 +1879,7 @@ static int kbase_cpu_mmap(struct kbase_context *kctx,
if (!kaddr) {
unsigned long addr = vma->vm_start + aligned_offset;
- vma->vm_flags |= VM_PFNMAP;
+ vm_flags_set(vma, VM_PFNMAP);
for (i = 0; i < nr_pages; i++) {
phys_addr_t phys;
@@ -1898,7 +1894,7 @@ static int kbase_cpu_mmap(struct kbase_context *kctx,
} else {
WARN_ON(aligned_offset);
/* MIXEDMAP so we can vfree the kaddr early and not track it after map time */
- vma->vm_flags |= VM_MIXEDMAP;
+ vm_flags_set(vma, VM_MIXEDMAP);
/* vmalloc remaping is easy... */
err = remap_vmalloc_range(vma, kaddr, 0);
WARN_ON(err);
@@ -2079,9 +2075,9 @@ int kbase_mmap(struct file *file, struct vm_area_struct *vma)
dev_dbg(dev, "kbase_mmap\n");
if (!(vma->vm_flags & VM_READ))
- vma->vm_flags &= ~VM_MAYREAD;
+ vm_flags_clear(vma, VM_MAYREAD);
if (!(vma->vm_flags & VM_WRITE))
- vma->vm_flags &= ~VM_MAYWRITE;
+ vm_flags_clear(vma, VM_MAYWRITE);
if (0 == nr_pages) {
err = -EINVAL;
@@ -2420,14 +2416,16 @@ void kbasep_os_process_page_usage_update(struct kbase_context *kctx, int pages)
#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 5, 0))
add_mm_counter(mm, MM_FILEPAGES, pages);
#else
- atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ //atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ percpu_counter_add(&mm->rss_stat[MM_FILEPAGES], pages);
#endif
#else
spin_lock(&mm->page_table_lock);
#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 5, 0))
add_mm_counter(mm, MM_FILEPAGES, pages);
#else
- atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ //atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ percpu_counter_add(&mm->rss_stat[MM_FILEPAGES], pages);
#endif
spin_unlock(&mm->page_table_lock);
#endif
@@ -2456,14 +2454,16 @@ static void kbasep_os_process_page_usage_drain(struct kbase_context *kctx)
#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 5, 0))
add_mm_counter(mm, MM_FILEPAGES, -pages);
#else
- atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ //atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ percpu_counter_add(&mm->rss_stat[MM_FILEPAGES], -pages);
#endif
#else
spin_lock(&mm->page_table_lock);
#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 5, 0))
add_mm_counter(mm, MM_FILEPAGES, -pages);
#else
- atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ //atomic_long_add(pages, &mm->rss_stat.count[MM_FILEPAGES]);
+ percpu_counter_add(&mm->rss_stat[MM_FILEPAGES], -pages);
#endif
spin_unlock(&mm->page_table_lock);
#endif
@@ -2495,12 +2495,8 @@ static int kbase_tracking_page_setup(struct kbase_context *kctx, struct vm_area_
spin_unlock(&kctx->mm_update_lock);
/* no real access */
- vma->vm_flags &= ~(VM_READ | VM_MAYREAD | VM_WRITE | VM_MAYWRITE | VM_EXEC | VM_MAYEXEC);
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3, 7, 0))
- vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP | VM_IO;
-#else
- vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND | VM_RESERVED | VM_IO;
-#endif
+ vm_flags_clear(vma, VM_READ | VM_MAYREAD | VM_WRITE | VM_MAYWRITE | VM_EXEC | VM_MAYEXEC);
+ vm_flags_set(vma, VM_DONTCOPY | VM_DONTEXPAND | VM_DONTDUMP | VM_IO);
vma->vm_ops = &kbase_vm_special_ops;
vma->vm_private_data = kctx;
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_pool.c b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_pool.c
index 0f91be1..26578f2 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_pool.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/mali_kbase_mem_pool.c
@@ -397,7 +397,7 @@ int kbase_mem_pool_init(struct kbase_mem_pool *pool,
#if LINUX_VERSION_CODE >= KERNEL_VERSION(3, 1, 0)
pool->reclaim.batch = 0;
#endif
- register_shrinker(&pool->reclaim);
+ register_shrinker(&pool->reclaim, "mali_kbase_pool");
pool_dbg(pool, "initialized\n");
diff --git a/driver/product/kernel/drivers/gpu/arm/midgard/thirdparty/mali_kbase_mmap.c b/driver/product/kernel/drivers/gpu/arm/midgard/thirdparty/mali_kbase_mmap.c
index 3aab51a..eaed808 100644
--- a/driver/product/kernel/drivers/gpu/arm/midgard/thirdparty/mali_kbase_mmap.c
+++ b/driver/product/kernel/drivers/gpu/arm/midgard/thirdparty/mali_kbase_mmap.c
@@ -29,6 +29,7 @@
* Boston, MA 02110-1301, USA.
*/
+#include <linux/maple_tree.h>
#include "linux/mman.h"
#include "../mali_kbase.h"
@@ -153,97 +154,25 @@ static unsigned long kbase_unmapped_area_topdown(struct vm_unmapped_area_info
*info, bool is_shader_code, bool is_same_4gb_page)
{
struct mm_struct *mm = current->mm;
- struct vm_area_struct *vma;
- unsigned long length, low_limit, high_limit, gap_start, gap_end;
+ //struct vm_area_struct *vma;
+ //unsigned long length, low_limit, high_limit, gap_start, gap_end;
+ unsigned long length, gap_start, gap_end;
+ MA_STATE(mas, &mm->mm_mt, 0, 0);
/* Adjust search length to account for worst case alignment overhead */
length = info->length + info->align_mask;
if (length < info->length)
return -ENOMEM;
- /*
- * Adjust search limits by the desired length.
- * See implementation comment at top of unmapped_area().
- */
- gap_end = info->high_limit;
- if (gap_end < length)
+ if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1,
+ length))
return -ENOMEM;
- high_limit = gap_end - length;
- if (info->low_limit > high_limit)
- return -ENOMEM;
- low_limit = info->low_limit + length;
-
- /* Check highest gap, which does not precede any rbtree node */
- gap_start = mm->highest_vm_end;
- if (gap_start <= high_limit) {
- if (align_and_check(&gap_end, gap_start, info,
+ gap_end = mas.last + 1;
+ gap_start = mas.index;
+ if (align_and_check(&gap_end, gap_start, info,
is_shader_code, is_same_4gb_page))
return gap_end;
- }
-
- /* Check if rbtree root looks promising */
- if (RB_EMPTY_ROOT(&mm->mm_rb))
- return -ENOMEM;
- vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
- if (vma->rb_subtree_gap < length)
- return -ENOMEM;
-
- while (true) {
- /* Visit right subtree if it looks promising */
- gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
- if (gap_start <= high_limit && vma->vm_rb.rb_right) {
- struct vm_area_struct *right =
- rb_entry(vma->vm_rb.rb_right,
- struct vm_area_struct, vm_rb);
- if (right->rb_subtree_gap >= length) {
- vma = right;
- continue;
- }
- }
-
-check_current:
- /* Check if current node has a suitable gap */
- gap_end = vma->vm_start;
- if (gap_end < low_limit)
- return -ENOMEM;
- if (gap_start <= high_limit && gap_end - gap_start >= length) {
- /* We found a suitable gap. Clip it with the original
- * high_limit. */
- if (gap_end > info->high_limit)
- gap_end = info->high_limit;
-
- if (align_and_check(&gap_end, gap_start, info,
- is_shader_code, is_same_4gb_page))
- return gap_end;
- }
-
- /* Visit left subtree if it looks promising */
- if (vma->vm_rb.rb_left) {
- struct vm_area_struct *left =
- rb_entry(vma->vm_rb.rb_left,
- struct vm_area_struct, vm_rb);
- if (left->rb_subtree_gap >= length) {
- vma = left;
- continue;
- }
- }
-
- /* Go back up the rbtree to find next candidate node */
- while (true) {
- struct rb_node *prev = &vma->vm_rb;
-
- if (!rb_parent(prev))
- return -ENOMEM;
- vma = rb_entry(rb_parent(prev),
- struct vm_area_struct, vm_rb);
- if (prev == vma->vm_rb.rb_right) {
- gap_start = vma->vm_prev ?
- vma->vm_prev->vm_end : 0;
- goto check_current;
- }
- }
- }
return -ENOMEM;
}
--
2.39.0