From: Ilia Levi <ilia.levi@intel.com>
To: intel-xe@lists.freedesktop.org
Cc: ilia.levi@intel.com, piotr.piorkowski@intel.com,
niranjana.vishwanathapura@intel.com, koby.elbaz@intel.com,
yaron.avizrat@intel.com
Subject: [PATCH v5 3/5] drm/xe: Initial MSI-X support for HW engines
Date: Thu, 28 Nov 2024 14:53:43 +0200 [thread overview]
Message-ID: <20241128125345.1115-4-ilia.levi@intel.com> (raw)
In-Reply-To: <20241128125345.1115-1-ilia.levi@intel.com>
- Configure the HW engines to work with MSI-X
- Program the LRC to use memirq infra (similar to VF)
- CS_INT_VEC field added to the LRC
Bspec: 60342, 72547
Signed-off-by: Ilia Levi <ilia.levi@intel.com>
---
drivers/gpu/drm/xe/regs/xe_engine_regs.h | 3 +++
drivers/gpu/drm/xe/regs/xe_lrc_layout.h | 3 +++
drivers/gpu/drm/xe/xe_exec_queue.c | 4 +++-
drivers/gpu/drm/xe/xe_exec_queue_types.h | 2 ++
drivers/gpu/drm/xe/xe_execlist.c | 10 +++++++---
drivers/gpu/drm/xe/xe_hw_engine.c | 7 +++++--
drivers/gpu/drm/xe/xe_lrc.c | 24 ++++++++++++++++++++----
drivers/gpu/drm/xe/xe_lrc.h | 2 +-
8 files changed, 44 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/xe/regs/xe_engine_regs.h b/drivers/gpu/drm/xe/regs/xe_engine_regs.h
index 7c78496e6213..d86219dedde2 100644
--- a/drivers/gpu/drm/xe/regs/xe_engine_regs.h
+++ b/drivers/gpu/drm/xe/regs/xe_engine_regs.h
@@ -83,6 +83,8 @@
#define RING_IMR(base) XE_REG((base) + 0xa8)
#define RING_INT_STATUS_RPT_PTR(base) XE_REG((base) + 0xac)
+#define CS_INT_VEC(base) XE_REG((base) + 0x1b8)
+
#define RING_EIR(base) XE_REG((base) + 0xb0)
#define RING_EMR(base) XE_REG((base) + 0xb4)
#define RING_ESR(base) XE_REG((base) + 0xb8)
@@ -138,6 +140,7 @@
#define RING_MODE(base) XE_REG((base) + 0x29c)
#define GFX_DISABLE_LEGACY_MODE REG_BIT(3)
+#define GFX_MSIX_INTERRUPT_ENABLE REG_BIT(13)
#define RING_TIMESTAMP(base) XE_REG((base) + 0x358)
diff --git a/drivers/gpu/drm/xe/regs/xe_lrc_layout.h b/drivers/gpu/drm/xe/regs/xe_lrc_layout.h
index 045dfd09db99..57944f90bbf6 100644
--- a/drivers/gpu/drm/xe/regs/xe_lrc_layout.h
+++ b/drivers/gpu/drm/xe/regs/xe_lrc_layout.h
@@ -25,6 +25,9 @@
#define CTX_INT_SRC_REPORT_REG (CTX_LRI_INT_REPORT_PTR + 3)
#define CTX_INT_SRC_REPORT_PTR (CTX_LRI_INT_REPORT_PTR + 4)
+#define CTX_CS_INT_VEC_REG 0x5a
+#define CTX_CS_INT_VEC_DATA (CTX_CS_INT_VEC_REG + 1)
+
#define INDIRECT_CTX_RING_HEAD (0x02 + 1)
#define INDIRECT_CTX_RING_TAIL (0x04 + 1)
#define INDIRECT_CTX_RING_START (0x06 + 1)
diff --git a/drivers/gpu/drm/xe/xe_exec_queue.c b/drivers/gpu/drm/xe/xe_exec_queue.c
index aab9e561153d..9c94be571900 100644
--- a/drivers/gpu/drm/xe/xe_exec_queue.c
+++ b/drivers/gpu/drm/xe/xe_exec_queue.c
@@ -16,6 +16,7 @@
#include "xe_hw_engine_class_sysfs.h"
#include "xe_hw_engine_group.h"
#include "xe_hw_fence.h"
+#include "xe_irq.h"
#include "xe_lrc.h"
#include "xe_macros.h"
#include "xe_migrate.h"
@@ -68,6 +69,7 @@ static struct xe_exec_queue *__xe_exec_queue_alloc(struct xe_device *xe,
q->gt = gt;
q->class = hwe->class;
q->width = width;
+ q->msix_vec = XE_IRQ_DEFAULT_MSIX;
q->logical_mask = logical_mask;
q->fence_irq = >->fence_irq[hwe->class];
q->ring_ops = gt->ring_ops[hwe->class];
@@ -117,7 +119,7 @@ static int __xe_exec_queue_init(struct xe_exec_queue *q)
}
for (i = 0; i < q->width; ++i) {
- q->lrc[i] = xe_lrc_create(q->hwe, q->vm, SZ_16K);
+ q->lrc[i] = xe_lrc_create(q->hwe, q->vm, SZ_16K, q->msix_vec);
if (IS_ERR(q->lrc[i])) {
err = PTR_ERR(q->lrc[i]);
goto err_unlock;
diff --git a/drivers/gpu/drm/xe/xe_exec_queue_types.h b/drivers/gpu/drm/xe/xe_exec_queue_types.h
index 1158b6062a6c..eec8f9935a58 100644
--- a/drivers/gpu/drm/xe/xe_exec_queue_types.h
+++ b/drivers/gpu/drm/xe/xe_exec_queue_types.h
@@ -63,6 +63,8 @@ struct xe_exec_queue {
char name[MAX_FENCE_NAME_LEN];
/** @width: width (number BB submitted per exec) of this exec queue */
u16 width;
+ /** @msix_vec: MSI-X vector (for platforms that support it) */
+ u16 msix_vec;
/** @fence_irq: fence IRQ used to signal job completion */
struct xe_hw_fence_irq *fence_irq;
diff --git a/drivers/gpu/drm/xe/xe_execlist.c b/drivers/gpu/drm/xe/xe_execlist.c
index a8c416a48812..5ef96deaa881 100644
--- a/drivers/gpu/drm/xe/xe_execlist.c
+++ b/drivers/gpu/drm/xe/xe_execlist.c
@@ -17,6 +17,7 @@
#include "xe_exec_queue.h"
#include "xe_gt.h"
#include "xe_hw_fence.h"
+#include "xe_irq.h"
#include "xe_lrc.h"
#include "xe_macros.h"
#include "xe_mmio.h"
@@ -47,6 +48,7 @@ static void __start_lrc(struct xe_hw_engine *hwe, struct xe_lrc *lrc,
struct xe_mmio *mmio = >->mmio;
struct xe_device *xe = gt_to_xe(gt);
u64 lrc_desc;
+ u32 ring_mode = _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE);
lrc_desc = xe_lrc_descriptor(lrc);
@@ -80,8 +82,10 @@ static void __start_lrc(struct xe_hw_engine *hwe, struct xe_lrc *lrc,
xe_mmio_write32(mmio, RING_HWS_PGA(hwe->mmio_base),
xe_bo_ggtt_addr(hwe->hwsp));
xe_mmio_read32(mmio, RING_HWS_PGA(hwe->mmio_base));
- xe_mmio_write32(mmio, RING_MODE(hwe->mmio_base),
- _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE));
+
+ if (xe_device_has_msix(gt_to_xe(hwe->gt)))
+ ring_mode |= _MASKED_BIT_ENABLE(GFX_MSIX_INTERRUPT_ENABLE);
+ xe_mmio_write32(mmio, RING_MODE(hwe->mmio_base), ring_mode);
xe_mmio_write32(mmio, RING_EXECLIST_SQ_CONTENTS_LO(hwe->mmio_base),
lower_32_bits(lrc_desc));
@@ -265,7 +269,7 @@ struct xe_execlist_port *xe_execlist_port_create(struct xe_device *xe,
port->hwe = hwe;
- port->lrc = xe_lrc_create(hwe, NULL, SZ_16K);
+ port->lrc = xe_lrc_create(hwe, NULL, SZ_16K, XE_IRQ_DEFAULT_MSIX);
if (IS_ERR(port->lrc)) {
err = PTR_ERR(port->lrc);
goto err;
diff --git a/drivers/gpu/drm/xe/xe_hw_engine.c b/drivers/gpu/drm/xe/xe_hw_engine.c
index c4b0dc3be39c..0a1baaedff38 100644
--- a/drivers/gpu/drm/xe/xe_hw_engine.c
+++ b/drivers/gpu/drm/xe/xe_hw_engine.c
@@ -324,6 +324,7 @@ void xe_hw_engine_enable_ring(struct xe_hw_engine *hwe)
{
u32 ccs_mask =
xe_hw_engine_mask_per_class(hwe->gt, XE_ENGINE_CLASS_COMPUTE);
+ u32 ring_mode = _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE);
if (hwe->class == XE_ENGINE_CLASS_COMPUTE && ccs_mask)
xe_mmio_write32(&hwe->gt->mmio, RCU_MODE,
@@ -332,8 +333,10 @@ void xe_hw_engine_enable_ring(struct xe_hw_engine *hwe)
xe_hw_engine_mmio_write32(hwe, RING_HWSTAM(0), ~0x0);
xe_hw_engine_mmio_write32(hwe, RING_HWS_PGA(0),
xe_bo_ggtt_addr(hwe->hwsp));
- xe_hw_engine_mmio_write32(hwe, RING_MODE(0),
- _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE));
+
+ if (xe_device_has_msix(gt_to_xe(hwe->gt)))
+ ring_mode |= _MASKED_BIT_ENABLE(GFX_MSIX_INTERRUPT_ENABLE);
+ xe_hw_engine_mmio_write32(hwe, RING_MODE(0), ring_mode);
xe_hw_engine_mmio_write32(hwe, RING_MI_MODE(0),
_MASKED_BIT_DISABLE(STOP_RING));
xe_hw_engine_mmio_read32(hwe, RING_MI_MODE(0));
diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
index 22e58c6e2a35..bbb9ffbf6367 100644
--- a/drivers/gpu/drm/xe/xe_lrc.c
+++ b/drivers/gpu/drm/xe/xe_lrc.c
@@ -584,6 +584,7 @@ static void set_memory_based_intr(u32 *regs, struct xe_hw_engine *hwe)
{
struct xe_memirq *memirq = >_to_tile(hwe->gt)->memirq;
struct xe_device *xe = gt_to_xe(hwe->gt);
+ u8 num_regs;
if (!xe_device_uses_memirq(xe))
return;
@@ -593,12 +594,18 @@ static void set_memory_based_intr(u32 *regs, struct xe_hw_engine *hwe)
regs[CTX_INT_MASK_ENABLE_REG] = RING_IMR(0).addr;
regs[CTX_INT_MASK_ENABLE_PTR] = xe_memirq_enable_ptr(memirq);
- regs[CTX_LRI_INT_REPORT_PTR] = MI_LOAD_REGISTER_IMM | MI_LRI_NUM_REGS(2) |
+ num_regs = xe_device_has_msix(xe) ? 3 : 2;
+ regs[CTX_LRI_INT_REPORT_PTR] = MI_LOAD_REGISTER_IMM | MI_LRI_NUM_REGS(num_regs) |
MI_LRI_LRM_CS_MMIO | MI_LRI_FORCE_POSTED;
regs[CTX_INT_STATUS_REPORT_REG] = RING_INT_STATUS_RPT_PTR(0).addr;
regs[CTX_INT_STATUS_REPORT_PTR] = xe_memirq_status_ptr(memirq, hwe);
regs[CTX_INT_SRC_REPORT_REG] = RING_INT_SRC_RPT_PTR(0).addr;
regs[CTX_INT_SRC_REPORT_PTR] = xe_memirq_source_ptr(memirq, hwe);
+
+ if (xe_device_has_msix(xe)) {
+ regs[CTX_CS_INT_VEC_REG] = CS_INT_VEC(0).addr;
+ /* CTX_CS_INT_VEC_DATA will be set in xe_lrc_init */
+ }
}
static int lrc_ring_mi_mode(struct xe_hw_engine *hwe)
@@ -876,7 +883,7 @@ static void xe_lrc_finish(struct xe_lrc *lrc)
#define PVC_CTX_ACC_CTR_THOLD (0x2a + 1)
static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
- struct xe_vm *vm, u32 ring_size)
+ struct xe_vm *vm, u32 ring_size, u16 msix_vec)
{
struct xe_gt *gt = hwe->gt;
struct xe_tile *tile = gt_to_tile(gt);
@@ -945,6 +952,14 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
xe_drm_client_add_bo(vm->xef->client, lrc->bo);
}
+ if (xe_device_has_msix(xe)) {
+ xe_lrc_write_ctx_reg(lrc, CTX_INT_STATUS_REPORT_PTR,
+ xe_memirq_status_ptr(&tile->memirq, hwe));
+ xe_lrc_write_ctx_reg(lrc, CTX_INT_SRC_REPORT_PTR,
+ xe_memirq_source_ptr(&tile->memirq, hwe));
+ xe_lrc_write_ctx_reg(lrc, CTX_CS_INT_VEC_DATA, msix_vec << 16 | msix_vec);
+ }
+
if (xe_gt_has_indirect_ring_state(gt)) {
xe_lrc_write_ctx_reg(lrc, CTX_INDIRECT_RING_STATE,
__xe_lrc_indirect_ring_ggtt_addr(lrc));
@@ -1005,6 +1020,7 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
* @hwe: Hardware Engine
* @vm: The VM (address space)
* @ring_size: LRC ring size
+ * @msix_vec: MSI-X interrupt vector (for platforms that support it)
*
* Allocate and initialize the Logical Ring Context (LRC).
*
@@ -1012,7 +1028,7 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
* upon failure.
*/
struct xe_lrc *xe_lrc_create(struct xe_hw_engine *hwe, struct xe_vm *vm,
- u32 ring_size)
+ u32 ring_size, u16 msix_vec)
{
struct xe_lrc *lrc;
int err;
@@ -1021,7 +1037,7 @@ struct xe_lrc *xe_lrc_create(struct xe_hw_engine *hwe, struct xe_vm *vm,
if (!lrc)
return ERR_PTR(-ENOMEM);
- err = xe_lrc_init(lrc, hwe, vm, ring_size);
+ err = xe_lrc_init(lrc, hwe, vm, ring_size, msix_vec);
if (err) {
kfree(lrc);
return ERR_PTR(err);
diff --git a/drivers/gpu/drm/xe/xe_lrc.h b/drivers/gpu/drm/xe/xe_lrc.h
index b459dcab8787..4206e6a8b50a 100644
--- a/drivers/gpu/drm/xe/xe_lrc.h
+++ b/drivers/gpu/drm/xe/xe_lrc.h
@@ -42,7 +42,7 @@ struct xe_lrc_snapshot {
#define LRC_PPHWSP_SCRATCH_ADDR (0x34 * 4)
struct xe_lrc *xe_lrc_create(struct xe_hw_engine *hwe, struct xe_vm *vm,
- u32 ring_size);
+ u32 ring_size, u16 msix_vec);
void xe_lrc_destroy(struct kref *ref);
/**
--
2.43.2
next prev parent reply other threads:[~2024-11-28 12:54 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-11-28 12:53 [PATCH v5 0/5] MSI-X support Ilia Levi
2024-11-28 12:53 ` [PATCH v5 1/5] drm/xe: Make irq enabled flag atomic Ilia Levi
2024-12-02 18:32 ` Piotr Piórkowski
2024-12-03 11:47 ` Levi, Ilia
2024-12-05 17:24 ` Rodrigo Vivi
2024-11-28 12:53 ` [PATCH v5 2/5] drm/xe/irq: Separate MSI and MSI-X flows Ilia Levi
2024-12-02 18:59 ` Piotr Piórkowski
2024-12-03 9:04 ` Levi, Ilia
2024-12-05 8:35 ` Piotr Piórkowski
2024-11-28 12:53 ` Ilia Levi [this message]
2024-12-02 19:15 ` [PATCH v5 3/5] drm/xe: Initial MSI-X support for HW engines Piotr Piórkowski
2024-11-28 12:53 ` [PATCH v5 4/5] drm/xe/irq: Manage MSI-X interrupts allocation Ilia Levi
2024-12-05 9:07 ` Piotr Piórkowski
2024-11-28 12:53 ` [PATCH v5 5/5] drm/xe/uapi: Support requesting unique MSI-X for exec queue Ilia Levi
2024-12-05 9:00 ` Piotr Piórkowski
2024-11-28 13:01 ` ✓ CI.Patch_applied: success for MSI-X support Patchwork
2024-11-28 13:01 ` ✓ CI.checkpatch: " Patchwork
2024-11-28 13:03 ` ✓ CI.KUnit: " Patchwork
2024-11-28 13:20 ` ✓ CI.Build: " Patchwork
2024-11-28 13:23 ` ✓ CI.Hooks: " Patchwork
2024-11-28 13:24 ` ✓ CI.checksparse: " Patchwork
2024-11-28 13:42 ` ✓ Xe.CI.BAT: " Patchwork
2024-11-28 15:39 ` ✗ Xe.CI.Full: failure " Patchwork
2024-12-13 7:25 ` [PATCH v6 0/4] " Ilia Levi
2024-12-13 7:25 ` [PATCH v6 1/4] drm/xe/irq: Separate MSI and MSI-X flows Ilia Levi
2024-12-13 7:25 ` [PATCH v6 2/4] drm/xe: Initial MSI-X support for HW engines Ilia Levi
2024-12-13 7:25 ` [PATCH v6 3/4] drm/xe/irq: Manage MSI-X interrupts allocation Ilia Levi
2024-12-13 7:25 ` [PATCH v6 4/4] drm/xe/uapi: Support requesting unique MSI-X for exec queue Ilia Levi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20241128125345.1115-4-ilia.levi@intel.com \
--to=ilia.levi@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=koby.elbaz@intel.com \
--cc=niranjana.vishwanathapura@intel.com \
--cc=piotr.piorkowski@intel.com \
--cc=yaron.avizrat@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.