[PATCH v7 12/12] drm/xe/nvm: add support for non-posted erase
From: Alexander Usyskin
Date: Wed Mar 26 2025 - 11:46:05 EST
From: "Abliyev, Reuven" <reuven.abliyev@xxxxxxxxx>
Erase command is slow on discrete graphics storage
and may overshot PCI completion timeout.
BMG introduces the ability to have non-posted erase.
Add driver support for non-posted erase with polling
for erase completion.
Signed-off-by: Abliyev, Reuven <reuven.abliyev@xxxxxxxxx>
Signed-off-by: Alexander Usyskin <alexander.usyskin@xxxxxxxxx>
---
drivers/gpu/drm/xe/xe_nvm.c | 25 +++++++++++++++++
drivers/mtd/devices/mtd_intel_dg.c | 43 ++++++++++++++++++++++++++++--
include/linux/intel_dg_nvm_aux.h | 2 ++
3 files changed, 68 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/xe/xe_nvm.c b/drivers/gpu/drm/xe/xe_nvm.c
index 8aec20bc629a..dd91f2e37661 100644
--- a/drivers/gpu/drm/xe/xe_nvm.c
+++ b/drivers/gpu/drm/xe/xe_nvm.c
@@ -14,7 +14,15 @@
#include "xe_sriov.h"
#define GEN12_GUNIT_NVM_BASE 0x00102040
+#define GEN12_DEBUG_NVM_BASE 0x00101018
+
+#define GEN12_CNTL_PROTECTED_NVM_REG 0x0010100C
+
#define GEN12_GUNIT_NVM_SIZE 0x80
+#define GEN12_DEBUG_NVM_SIZE 0x4
+
+#define NVM_NON_POSTED_ERASE_CHICKEN_BIT BIT(13)
+
#define HECI_FW_STATUS_2_NVM_ACCESS_MODE BIT(3)
static const struct intel_dg_nvm_region regions[INTEL_DG_NVM_REGIONS] = {
@@ -28,6 +36,16 @@ static void xe_nvm_release_dev(struct device *dev)
{
}
+static bool xe_nvm_non_posted_erase(struct xe_device *xe)
+{
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+
+ if (xe->info.platform != XE_BATTLEMAGE)
+ return false;
+ return !(xe_mmio_read32(>->mmio, XE_REG(GEN12_CNTL_PROTECTED_NVM_REG)) &
+ NVM_NON_POSTED_ERASE_CHICKEN_BIT);
+}
+
static bool xe_nvm_writable_override(struct xe_device *xe)
{
struct xe_gt *gt = xe_root_mmio_gt(xe);
@@ -85,6 +103,7 @@ void xe_nvm_init(struct xe_device *xe)
nvm = xe->nvm;
nvm->writable_override = xe_nvm_writable_override(xe);
+ nvm->non_posted_erase = xe_nvm_non_posted_erase(xe);
nvm->bar.parent = &pdev->resource[0];
nvm->bar.start = GEN12_GUNIT_NVM_BASE + pdev->resource[0].start;
nvm->bar.end = nvm->bar.start + GEN12_GUNIT_NVM_SIZE - 1;
@@ -92,6 +111,12 @@ void xe_nvm_init(struct xe_device *xe)
nvm->bar.desc = IORES_DESC_NONE;
nvm->regions = regions;
+ nvm->bar2.parent = &pdev->resource[0];
+ nvm->bar2.start = GEN12_DEBUG_NVM_BASE + pdev->resource[0].start;
+ nvm->bar2.end = nvm->bar2.start + GEN12_DEBUG_NVM_SIZE - 1;
+ nvm->bar2.flags = IORESOURCE_MEM;
+ nvm->bar2.desc = IORES_DESC_NONE;
+
aux_dev = &nvm->aux_dev;
aux_dev->name = "nvm";
diff --git a/drivers/mtd/devices/mtd_intel_dg.c b/drivers/mtd/devices/mtd_intel_dg.c
index 9f4bb15a03b8..c898107a588f 100644
--- a/drivers/mtd/devices/mtd_intel_dg.c
+++ b/drivers/mtd/devices/mtd_intel_dg.c
@@ -28,6 +28,9 @@ struct intel_dg_nvm {
struct mtd_info mtd;
struct mutex lock; /* region access lock */
void __iomem *base;
+ void __iomem *base2;
+ bool non_posted_erase;
+
size_t size;
unsigned int nregions;
struct {
@@ -44,6 +47,7 @@ struct intel_dg_nvm {
#define NVM_VALSIG_REG 0x00000010
#define NVM_ADDRESS_REG 0x00000040
#define NVM_REGION_ID_REG 0x00000044
+#define NVM_DEBUG_REG 0x00000000
/*
* [15:0]-Erase size = 0x0010 4K 0x0080 32K 0x0100 64K
* [23:16]-Reserved
@@ -75,6 +79,9 @@ struct intel_dg_nvm {
#define NVM_FREG_ADDR_SHIFT 12
#define NVM_FREG_MIN_REGION_SIZE 0xFFF
+#define NVM_NON_POSTED_ERASE_DONE BIT(23)
+#define NVM_NON_POSTED_ERASE_DONE_ITER 3000
+
static inline void idg_nvm_set_region_id(struct intel_dg_nvm *nvm, u8 region)
{
iowrite32((u32)region, nvm->base + NVM_REGION_ID_REG);
@@ -370,11 +377,30 @@ idg_erase(struct intel_dg_nvm *nvm, u8 region, loff_t from, u64 len, u64 *fail_a
{
u64 i;
const u32 block = 0x10;
+ u32 reg;
+ u32 iter = 0;
void __iomem *base = nvm->base;
+ void __iomem *base2 = nvm->base2;
for (i = 0; i < len; i += SZ_4K) {
iowrite32(from + i, base + NVM_ADDRESS_REG);
iowrite32(region << 24 | block, base + NVM_ERASE_REG);
+ if (nvm->non_posted_erase) {
+ /* Wait for Erase Done */
+ reg = ioread32(base2 + NVM_DEBUG_REG);
+ while (!(reg & NVM_NON_POSTED_ERASE_DONE) &&
+ ++iter < NVM_NON_POSTED_ERASE_DONE_ITER) {
+ msleep(10);
+ reg = ioread32(base2 + NVM_DEBUG_REG);
+ }
+ if (reg & NVM_NON_POSTED_ERASE_DONE) {
+ /* Clear Erase Done */
+ iowrite32(reg, base2 + NVM_DEBUG_REG);
+ } else {
+ *fail_addr = from + i;
+ return -ETIME;
+ }
+ }
/* Since the writes are via sguint
* we cannot do back to back erases.
*/
@@ -383,7 +409,8 @@ idg_erase(struct intel_dg_nvm *nvm, u8 region, loff_t from, u64 len, u64 *fail_a
return len;
}
-static int intel_dg_nvm_init(struct intel_dg_nvm *nvm, struct device *device)
+static int intel_dg_nvm_init(struct intel_dg_nvm *nvm, struct device *device,
+ bool non_posted_erase)
{
int ret;
unsigned int i, n;
@@ -443,7 +470,10 @@ static int intel_dg_nvm_init(struct intel_dg_nvm *nvm, struct device *device)
n++;
}
+ nvm->non_posted_erase = non_posted_erase;
+
dev_dbg(device, "Registered %d regions\n", n);
+ dev_dbg(device, "Non posted erase %d\n", nvm->non_posted_erase);
/* Need to add 1 to the amount of memory
* so it is reported as an even block
@@ -778,7 +808,16 @@ static int intel_dg_mtd_probe(struct auxiliary_device *aux_dev,
goto err;
}
- ret = intel_dg_nvm_init(nvm, device);
+ if (invm->non_posted_erase) {
+ nvm->base2 = devm_ioremap_resource(device, &invm->bar2);
+ if (IS_ERR(nvm->base2)) {
+ dev_err(device, "base2 mmio not mapped\n");
+ ret = PTR_ERR(nvm->base2);
+ goto err;
+ }
+ }
+
+ ret = intel_dg_nvm_init(nvm, device, invm->non_posted_erase);
if (ret < 0) {
dev_err(device, "cannot initialize nvm %d\n", ret);
goto err;
diff --git a/include/linux/intel_dg_nvm_aux.h b/include/linux/intel_dg_nvm_aux.h
index 68df634c994c..bee25dfc6982 100644
--- a/include/linux/intel_dg_nvm_aux.h
+++ b/include/linux/intel_dg_nvm_aux.h
@@ -17,7 +17,9 @@ struct intel_dg_nvm_region {
struct intel_dg_nvm_dev {
struct auxiliary_device aux_dev;
bool writable_override;
+ bool non_posted_erase;
struct resource bar;
+ struct resource bar2;
const struct intel_dg_nvm_region *regions;
};
--
2.43.0