Tentative fix for NVIDIA 470.103.01 driver for Linux 5.18-rc1+

nvidia-470xx-fix-linux-5.18.patch

From 0ec02378bcc0d35d713fd93ba42c3848bc8b1077 Mon Sep 17 00:00:00 2001
From: Joan Bruguera <[email protected]>
Date: Sat, 12 Mar 2022 08:37:32 +0100
Subject: [PATCH 1/2] Buildfix NVIDIA 470.103.01 for Linux 5.18-rc1 (part 1)

---
 common/inc/nv-linux.h       | 7 +++++++
 nvidia-drm/nvidia-drm-drv.c | 4 ++++
 nvidia-drm/nvidia-drm-gem.c | 6 ++++++
 3 files changed, 17 insertions(+)

diff --git a/common/inc/nv-linux.h b/common/inc/nv-linux.h
index 82e19d6..58e66c7 100644
--- a/common/inc/nv-linux.h
+++ b/common/inc/nv-linux.h
@@ -11,6 +11,8 @@
 #ifndef _NV_LINUX_H_
 #define _NV_LINUX_H_
 
+#warning Do not buy NVIDIA for your next GPU!
+#include <linux/version.h>
 #include "nvstatus.h"
 #include "nv.h"
 #include "nv-ioctl-numa.h"
@@ -958,7 +960,12 @@ static inline pgprot_t nv_adjust_pgprot(pgprot_t vm_prot, NvU32 extra)
      * When AMD memory encryption is enabled, device memory mappings with the
      * C-bit set read as 0xFF, so ensure the bit is cleared for user mappings.
      */
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0) && defined(_ASM_X86_PGTABLE_H))
+    // Rel. commit "x86/coco: Add API to handle encryption mask" (Kirill A. Shutemov, 22 Feb 2022)
+    prot = __pgprot(__sme_clr(pgprot_val(prot)));
+#else
     prot = pgprot_decrypted(prot);
+#endif
 #endif
     return prot;
 }
diff --git a/nvidia-drm/nvidia-drm-drv.c b/nvidia-drm/nvidia-drm-drv.c
index d7067a4..f57cbc8 100644
--- a/nvidia-drm/nvidia-drm-drv.c
+++ b/nvidia-drm/nvidia-drm-drv.c
@@ -20,6 +20,7 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
+#include <linux/version.h>
 #include "nvidia-drm-conftest.h" /* NV_DRM_AVAILABLE and NV_DRM_DRM_GEM_H_PRESENT */
 
 #include "nvidia-drm-priv.h"
@@ -254,7 +255,10 @@ nv_drm_init_mode_config(struct nv_drm_device *nv_dev,
 #if defined(NV_DRM_FORMAT_MODIFIERS_PRESENT)
     /* Allow clients to define framebuffer layouts using DRM format modifiers */
 
+#if (LINUX_VERSION_CODE < KERNEL_VERSION(5, 18, 0))
+    // Rel. commit "drm: remove allow_fb_modifiers" (Tomohito Esaki, 28 Jan 2022)
     dev->mode_config.allow_fb_modifiers = true;
+#endif
 #endif
 
     /* Initialize output polling support */
diff --git a/nvidia-drm/nvidia-drm-gem.c b/nvidia-drm/nvidia-drm-gem.c
index 82de645..8cd3f76 100644
--- a/nvidia-drm/nvidia-drm-gem.c
+++ b/nvidia-drm/nvidia-drm-gem.c
@@ -49,6 +49,12 @@
 
 #include "nv-mm.h"
 
+#include <linux/version.h>
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(5, 18, 0))
+// Rel. commit "dma-buf-map: Rename to iosys-map" (Lucas De Marchi, 4 Feb 2022)
+#define dma_buf_map iosys_map
+#endif
+
 void nv_drm_gem_free(struct drm_gem_object *gem)
 {
     struct nv_drm_gem_object *nv_gem = to_nv_gem_object(gem);
-- 
2.35.1


From 6b934e829fb942d1735e2371320ddbec089c6900 Mon Sep 17 00:00:00 2001
From: Joan Bruguera <[email protected]>
Date: Mon, 4 Apr 2022 19:59:51 +0200
Subject: [PATCH 2/2] Buildfix NVIDIA 470.103.01 for Linux 5.18-rc1 (part 2)

See also: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d56baf6efaf14e2910610216c581ca71d6940012
          https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7968778914e53788a01c2dee2692cab157de9ac0
---
 nvidia-uvm/uvm_gpu.c    |  8 ++++----
 nvidia/linux_nvswitch.c | 18 +++++++++---------
 nvidia/nv.c             | 18 +++++++++---------
 3 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/nvidia-uvm/uvm_gpu.c b/nvidia-uvm/uvm_gpu.c
index f58c00b..c9607a8 100644
--- a/nvidia-uvm/uvm_gpu.c
+++ b/nvidia-uvm/uvm_gpu.c
@@ -3059,16 +3059,16 @@ out:
 
 NV_STATUS uvm_gpu_map_cpu_pages(uvm_gpu_t *gpu, struct page *page, size_t size, NvU64 *dma_addr_out)
 {
-    NvU64 dma_addr = pci_map_page(gpu->parent->pci_dev, page, 0, size, PCI_DMA_BIDIRECTIONAL);
+    NvU64 dma_addr = dma_map_page(&gpu->parent->pci_dev->dev, page, 0, size, DMA_BIDIRECTIONAL);
 
     UVM_ASSERT(PAGE_ALIGNED(size));
 
-    if (pci_dma_mapping_error(gpu->parent->pci_dev, dma_addr))
+    if (dma_mapping_error(&gpu->parent->pci_dev->dev, dma_addr))
         return NV_ERR_OPERATING_SYSTEM;
 
     if (dma_addr < gpu->parent->dma_addressable_start ||
         dma_addr + size - 1 > gpu->parent->dma_addressable_limit) {
-        pci_unmap_page(gpu->parent->pci_dev, dma_addr, size, PCI_DMA_BIDIRECTIONAL);
+        dma_unmap_page(&gpu->parent->pci_dev->dev, dma_addr, size, DMA_BIDIRECTIONAL);
         UVM_ERR_PRINT_RL("PCI mapped range [0x%llx, 0x%llx) not in the addressable range [0x%llx, 0x%llx), GPU %s\n",
                          dma_addr,
                          dma_addr + (NvU64)size,
@@ -3102,7 +3102,7 @@ void uvm_gpu_unmap_cpu_pages(uvm_gpu_t *gpu, NvU64 dma_address, size_t size)
     if (gpu->parent->npu)
         dma_address = nv_expand_nvlink_addr(dma_address);
     dma_address += gpu->parent->dma_addressable_start;
-    pci_unmap_page(gpu->parent->pci_dev, dma_address, size, PCI_DMA_BIDIRECTIONAL);
+    dma_unmap_page(&gpu->parent->pci_dev->dev, dma_address, size, DMA_BIDIRECTIONAL);
     atomic64_sub(size, &gpu->parent->mapped_cpu_pages_size);
 }
 
diff --git a/nvidia/linux_nvswitch.c b/nvidia/linux_nvswitch.c
index ddfc53c..714b9eb 100644
--- a/nvidia/linux_nvswitch.c
+++ b/nvidia/linux_nvswitch.c
@@ -2139,11 +2139,11 @@ _nvswitch_to_pci_dma_direction
 )
 {
     if (direction == NVSWITCH_DMA_DIR_TO_SYSMEM)
-        return PCI_DMA_FROMDEVICE;
+        return DMA_FROM_DEVICE;
     else if (direction == NVSWITCH_DMA_DIR_FROM_SYSMEM)
-        return PCI_DMA_TODEVICE;
+        return DMA_TO_DEVICE;
     else
-        return PCI_DMA_BIDIRECTIONAL;
+        return DMA_BIDIRECTIONAL;
 }
 
 NvlStatus
@@ -2164,9 +2164,9 @@ nvswitch_os_map_dma_region
 
     dma_dir = _nvswitch_to_pci_dma_direction(direction);
 
-    *dma_handle = (NvU64)pci_map_single(pdev, cpu_addr, size, dma_dir);
+    *dma_handle = (NvU64) dma_map_single(&pdev->dev, cpu_addr, size, dma_dir);
 
-    if (pci_dma_mapping_error(pdev, *dma_handle))
+    if (dma_mapping_error(&pdev->dev, *dma_handle))
     {
         pr_err("nvidia-nvswitch: unable to create PCI DMA mapping\n");
         return -NVL_ERR_GENERIC;
@@ -2193,7 +2193,7 @@ nvswitch_os_unmap_dma_region
 
     dma_dir = _nvswitch_to_pci_dma_direction(direction);
 
-    pci_unmap_single(pdev, dma_handle, size, dma_dir);
+    dma_unmap_single(&pdev->dev, dma_handle, size, dma_dir);
 
     return NVL_SUCCESS;
 }
@@ -2210,7 +2210,7 @@ nvswitch_os_set_dma_mask
     if (!pdev)
         return -NVL_BAD_ARGS;
 
-    if (pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_addr_width)))
+    if (dma_set_mask(&pdev->dev, DMA_BIT_MASK(dma_addr_width)))
         return -NVL_ERR_GENERIC;
 
     return NVL_SUCCESS;
@@ -2233,7 +2233,7 @@ nvswitch_os_sync_dma_region_for_cpu
 
     dma_dir = _nvswitch_to_pci_dma_direction(direction);
 
-    pci_dma_sync_single_for_cpu(pdev, dma_handle, size, dma_dir);
+    dma_sync_single_for_cpu(&pdev->dev, dma_handle, size, dma_dir);
 
     return NVL_SUCCESS;
 }
@@ -2255,7 +2255,7 @@ nvswitch_os_sync_dma_region_for_device
 
     dma_dir = _nvswitch_to_pci_dma_direction(direction);
 
-    pci_dma_sync_single_for_device(pdev, dma_handle, size, dma_dir);
+    dma_sync_single_for_device(&pdev->dev, dma_handle, size, dma_dir);
 
     return NVL_SUCCESS;
 }
diff --git a/nvidia/nv.c b/nvidia/nv.c
index 92a7dc6..5065158 100644
--- a/nvidia/nv.c
+++ b/nvidia/nv.c
@@ -2821,13 +2821,13 @@ nv_set_dma_address_size(
      */
     if (!nvl->tce_bypass_enabled)
     {
-        pci_set_dma_mask(nvl->pci_dev, new_mask);
+        dma_set_mask(&nvl->pci_dev->dev, new_mask);
         /* Certain kernels have a bug which causes pci_set_consistent_dma_mask
          * to call GPL sme_active symbol, this bug has already been fixed in a
          * minor release update but detect the failure scenario here to prevent
          * an installation regression */
 #if !NV_IS_EXPORT_SYMBOL_GPL_sme_active
-        pci_set_consistent_dma_mask(nvl->pci_dev, new_mask);
+        dma_set_coherent_mask(&nvl->pci_dev->dev, new_mask);
 #endif
     }
 }
@@ -4524,19 +4524,19 @@ NvU64 NV_API_CALL nv_get_dma_start_address(
      * as the starting address for all DMA mappings.
      */
     saved_dma_mask = pci_dev->dma_mask;
-    if (pci_set_dma_mask(pci_dev, DMA_BIT_MASK(64)) != 0)
+    if (dma_set_mask(&pci_dev->dev, DMA_BIT_MASK(64)) != 0)
     {
         goto done;
     }
 
-    dma_addr = pci_map_single(pci_dev, NULL, 1, DMA_BIDIRECTIONAL);
-    if (pci_dma_mapping_error(pci_dev, dma_addr))
+    dma_addr = dma_map_single(&pci_dev->dev, NULL, 1, DMA_BIDIRECTIONAL);
+    if (dma_mapping_error(&pci_dev->dev, dma_addr))
     {
-        pci_set_dma_mask(pci_dev, saved_dma_mask);
+        dma_set_mask(&pci_dev->dev, saved_dma_mask);
         goto done;
     }
 
-    pci_unmap_single(pci_dev, dma_addr, 1, DMA_BIDIRECTIONAL);
+    dma_unmap_single(&pci_dev->dev, dma_addr, 1, DMA_BIDIRECTIONAL);
 
     /*
      * From IBM: "For IODA2, native DMA bypass or KVM TCE-based implementation
@@ -4568,7 +4568,7 @@ NvU64 NV_API_CALL nv_get_dma_start_address(
          */
         nv_printf(NV_DBG_WARNINGS,
             "NVRM: DMA window limited by platform\n");
-        pci_set_dma_mask(pci_dev, saved_dma_mask);
+        dma_set_mask(&pci_dev->dev, saved_dma_mask);
         goto done;
     }
     else if ((dma_addr & saved_dma_mask) != 0)
@@ -4587,7 +4587,7 @@ NvU64 NV_API_CALL nv_get_dma_start_address(
              */
             nv_printf(NV_DBG_WARNINGS,
                 "NVRM: DMA window limited by memory size\n");
-            pci_set_dma_mask(pci_dev, saved_dma_mask);
+            dma_set_mask(&pci_dev->dev, saved_dma_mask);
             goto done;
         }
     }
-- 
2.35.1


From 689ede6109d1b8aef914809f290e75632c338b9f Mon Sep 17 00:00:00 2001
From: Joan Bruguera <[email protected]>
Date: Sat, 9 Apr 2022 19:42:24 +0200
Subject: [PATCH] Buildfix NVIDIA 470.103.01 for Linux 5.18-rc2

acpi_bus_get_device was removed and its trivial replacement acpi_fetch_acpi_dev
is GPL-exported, so just disable ACPI support as I don't want to deal with it.
---
 nvidia/nv-acpi.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/nvidia/nv-acpi.c b/nvidia/nv-acpi.c
index faf8b44..22a9152 100644
--- a/nvidia/nv-acpi.c
+++ b/nvidia/nv-acpi.c
@@ -16,7 +16,8 @@
 
 #include <linux/acpi.h>
 
-#if defined(NV_LINUX_ACPI_EVENTS_SUPPORTED)
+#include <linux/version.h>
+#if defined(NV_LINUX_ACPI_EVENTS_SUPPORTED) && (LINUX_VERSION_CODE < KERNEL_VERSION(5, 18, 0))
 static NV_STATUS   nv_acpi_extract_integer (const union acpi_object *, void *, NvU32, NvU32 *);
 static NV_STATUS   nv_acpi_extract_buffer  (const union acpi_object *, void *, NvU32, NvU32 *);
 static NV_STATUS   nv_acpi_extract_package (const union acpi_object *, void *, NvU32, NvU32 *);
-- 
2.35.1

I've made some tweaks to your code for nvidia-510.60x and it works a treat. Thanks very much for your work.

it can work on nvidia 510.68?

I haven't got nvidia 510.68. Is that a beta version?

Is the last nvidia versión, downloadnable from website https://www.nvidia.it/Download/driverResults.aspx/187177/it

I can't with dkms, i take error

I don't use dkms with nvidia. I've never found it satisfactory. I can't promise but show me the exact error you get and I'll see if there's anything I can do.

I use 5.18-rc2 with last nvidia driver buy It can load dkms when install driver.
It possibile without dkms?

I don't have any NVIDIA hardware other than the 470xx one so I can't test it over 510xx, but at least 510.60.02 builds fine after patching, so it may work (the nvidia-uvm.c part seems missing from the newer driver so you can ignore that - but make sure dkms is not failing because of it).

If you're having trouble with dkms you can try to use this manual steps:
https://gist.github.com/joanbm/144a965c36fc1dc0d1f1b9be3438a368?permalink_comment_id=3914496#gistcomment-3914496

I have patched no-compat32 nvidia folder and now i had can install driver and has install dkms for kernel module then Xorg no load

without the kernel module the xorg server could not be started and therefore I stayed in a terminal interface, I used your patch and it worked, great job, thank you very much for the work done