summaryrefslogtreecommitdiff
path: root/drivers/misc/habanalabs
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/misc/habanalabs')
-rw-r--r--drivers/misc/habanalabs/common/device.c11
-rw-r--r--drivers/misc/habanalabs/common/firmware_if.c5
-rw-r--r--drivers/misc/habanalabs/common/habanalabs.h1
-rw-r--r--drivers/misc/habanalabs/common/habanalabs_ioctl.c2
-rw-r--r--drivers/misc/habanalabs/common/memory.c10
-rw-r--r--drivers/misc/habanalabs/common/mmu.c6
-rw-r--r--drivers/misc/habanalabs/common/mmu_v1.c12
-rw-r--r--drivers/misc/habanalabs/gaudi/gaudi.c3
-rw-r--r--drivers/misc/habanalabs/goya/goya.c3
9 files changed, 43 insertions, 10 deletions
diff --git a/drivers/misc/habanalabs/common/device.c b/drivers/misc/habanalabs/common/device.c
index 1456eabf9601..69d04eca767f 100644
--- a/drivers/misc/habanalabs/common/device.c
+++ b/drivers/misc/habanalabs/common/device.c
@@ -1037,7 +1037,7 @@ kill_processes:
if (hard_reset) {
/* Release kernel context */
- if (hl_ctx_put(hdev->kernel_ctx) == 1)
+ if (hdev->kernel_ctx && hl_ctx_put(hdev->kernel_ctx) == 1)
hdev->kernel_ctx = NULL;
hl_vm_fini(hdev);
hl_mmu_fini(hdev);
@@ -1487,6 +1487,15 @@ void hl_device_fini(struct hl_device *hdev)
}
}
+ /* Disable PCI access from device F/W so it won't send us additional
+ * interrupts. We disable MSI/MSI-X at the halt_engines function and we
+ * can't have the F/W sending us interrupts after that. We need to
+ * disable the access here because if the device is marked disable, the
+ * message won't be send. Also, in case of heartbeat, the device CPU is
+ * marked as disable so this message won't be sent
+ */
+ hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS);
+
/* Mark device as disabled */
hdev->disabled = true;
diff --git a/drivers/misc/habanalabs/common/firmware_if.c b/drivers/misc/habanalabs/common/firmware_if.c
index 20f77f58edef..c9a12980218a 100644
--- a/drivers/misc/habanalabs/common/firmware_if.c
+++ b/drivers/misc/habanalabs/common/firmware_if.c
@@ -402,6 +402,10 @@ int hl_fw_cpucp_pci_counters_get(struct hl_device *hdev,
}
counters->rx_throughput = result;
+ memset(&pkt, 0, sizeof(pkt));
+ pkt.ctl = cpu_to_le32(CPUCP_PACKET_PCIE_THROUGHPUT_GET <<
+ CPUCP_PKT_CTL_OPCODE_SHIFT);
+
/* Fetch PCI tx counter */
pkt.index = cpu_to_le32(cpucp_pcie_throughput_tx);
rc = hdev->asic_funcs->send_cpu_message(hdev, (u32 *) &pkt, sizeof(pkt),
@@ -414,6 +418,7 @@ int hl_fw_cpucp_pci_counters_get(struct hl_device *hdev,
counters->tx_throughput = result;
/* Fetch PCI replay counter */
+ memset(&pkt, 0, sizeof(pkt));
pkt.ctl = cpu_to_le32(CPUCP_PACKET_PCIE_REPLAY_CNT_GET <<
CPUCP_PKT_CTL_OPCODE_SHIFT);
diff --git a/drivers/misc/habanalabs/common/habanalabs.h b/drivers/misc/habanalabs/common/habanalabs.h
index e0d7f5fbaa5c..60e16dc4bcac 100644
--- a/drivers/misc/habanalabs/common/habanalabs.h
+++ b/drivers/misc/habanalabs/common/habanalabs.h
@@ -2182,6 +2182,7 @@ void hl_mmu_v1_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu);
int hl_mmu_va_to_pa(struct hl_ctx *ctx, u64 virt_addr, u64 *phys_addr);
int hl_mmu_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr,
struct hl_mmu_hop_info *hops);
+bool hl_is_dram_va(struct hl_device *hdev, u64 virt_addr);
int hl_fw_load_fw_to_device(struct hl_device *hdev, const char *fw_name,
void __iomem *dst, u32 src_offset, u32 size);
diff --git a/drivers/misc/habanalabs/common/habanalabs_ioctl.c b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
index 12efbd9d2e3a..d25892d61ec9 100644
--- a/drivers/misc/habanalabs/common/habanalabs_ioctl.c
+++ b/drivers/misc/habanalabs/common/habanalabs_ioctl.c
@@ -133,6 +133,8 @@ static int hw_idle(struct hl_device *hdev, struct hl_info_args *args)
hw_idle.is_idle = hdev->asic_funcs->is_device_idle(hdev,
&hw_idle.busy_engines_mask_ext, NULL);
+ hw_idle.busy_engines_mask =
+ lower_32_bits(hw_idle.busy_engines_mask_ext);
return copy_to_user(out, &hw_idle,
min((size_t) max_size, sizeof(hw_idle))) ? -EFAULT : 0;
diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c
index cbe9da4e0211..5d4fbdcaefe3 100644
--- a/drivers/misc/habanalabs/common/memory.c
+++ b/drivers/misc/habanalabs/common/memory.c
@@ -886,8 +886,10 @@ static void unmap_phys_pg_pack(struct hl_ctx *ctx, u64 vaddr,
{
struct hl_device *hdev = ctx->hdev;
u64 next_vaddr, i;
+ bool is_host_addr;
u32 page_size;
+ is_host_addr = !hl_is_dram_va(hdev, vaddr);
page_size = phys_pg_pack->page_size;
next_vaddr = vaddr;
@@ -900,9 +902,13 @@ static void unmap_phys_pg_pack(struct hl_ctx *ctx, u64 vaddr,
/*
* unmapping on Palladium can be really long, so avoid a CPU
* soft lockup bug by sleeping a little between unmapping pages
+ *
+ * In addition, when unmapping host memory we pass through
+ * the Linux kernel to unpin the pages and that takes a long
+ * time. Therefore, sleep every 32K pages to avoid soft lockup
*/
- if (hdev->pldm)
- usleep_range(500, 1000);
+ if (hdev->pldm || (is_host_addr && (i & 0x7FFF) == 0))
+ usleep_range(50, 200);
}
}
diff --git a/drivers/misc/habanalabs/common/mmu.c b/drivers/misc/habanalabs/common/mmu.c
index 33ae953d3a36..28a4638741d8 100644
--- a/drivers/misc/habanalabs/common/mmu.c
+++ b/drivers/misc/habanalabs/common/mmu.c
@@ -9,7 +9,7 @@
#include "habanalabs.h"
-static bool is_dram_va(struct hl_device *hdev, u64 virt_addr)
+bool hl_is_dram_va(struct hl_device *hdev, u64 virt_addr)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
@@ -156,7 +156,7 @@ int hl_mmu_unmap_page(struct hl_ctx *ctx, u64 virt_addr, u32 page_size,
if (!hdev->mmu_enable)
return 0;
- is_dram_addr = is_dram_va(hdev, virt_addr);
+ is_dram_addr = hl_is_dram_va(hdev, virt_addr);
if (is_dram_addr)
mmu_prop = &prop->dmmu;
@@ -236,7 +236,7 @@ int hl_mmu_map_page(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr,
if (!hdev->mmu_enable)
return 0;
- is_dram_addr = is_dram_va(hdev, virt_addr);
+ is_dram_addr = hl_is_dram_va(hdev, virt_addr);
if (is_dram_addr)
mmu_prop = &prop->dmmu;
diff --git a/drivers/misc/habanalabs/common/mmu_v1.c b/drivers/misc/habanalabs/common/mmu_v1.c
index 2ce6ea89d4fa..06d8a44dd5d4 100644
--- a/drivers/misc/habanalabs/common/mmu_v1.c
+++ b/drivers/misc/habanalabs/common/mmu_v1.c
@@ -467,8 +467,16 @@ static void hl_mmu_v1_fini(struct hl_device *hdev)
{
/* MMU H/W fini was already done in device hw_fini() */
- kvfree(hdev->mmu_priv.dr.mmu_shadow_hop0);
- gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool);
+ if (!ZERO_OR_NULL_PTR(hdev->mmu_priv.hr.mmu_shadow_hop0)) {
+ kvfree(hdev->mmu_priv.dr.mmu_shadow_hop0);
+ gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool);
+ }
+
+ /* Make sure that if we arrive here again without init was called we
+ * won't cause kernel panic. This can happen for example if we fail
+ * during hard reset code at certain points
+ */
+ hdev->mmu_priv.dr.mmu_shadow_hop0 = NULL;
}
/**
diff --git a/drivers/misc/habanalabs/gaudi/gaudi.c b/drivers/misc/habanalabs/gaudi/gaudi.c
index 8c09e4466af8..b328ddaa64ee 100644
--- a/drivers/misc/habanalabs/gaudi/gaudi.c
+++ b/drivers/misc/habanalabs/gaudi/gaudi.c
@@ -4002,7 +4002,8 @@ static int gaudi_cb_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP |
VM_DONTCOPY | VM_NORESERVE;
- rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr, dma_addr, size);
+ rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr,
+ (dma_addr - HOST_PHYS_BASE), size);
if (rc)
dev_err(hdev->dev, "dma_mmap_coherent error %d", rc);
diff --git a/drivers/misc/habanalabs/goya/goya.c b/drivers/misc/habanalabs/goya/goya.c
index b8b4aa636b7c..63679a747d2c 100644
--- a/drivers/misc/habanalabs/goya/goya.c
+++ b/drivers/misc/habanalabs/goya/goya.c
@@ -2719,7 +2719,8 @@ static int goya_cb_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP |
VM_DONTCOPY | VM_NORESERVE;
- rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr, dma_addr, size);
+ rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr,
+ (dma_addr - HOST_PHYS_BASE), size);
if (rc)
dev_err(hdev->dev, "dma_mmap_coherent error %d", rc);
generated by cgit v1.2.3 (git 2.39.0) at 2024-12-17 00:29:52 +0300