From: Amber Lin <Amber.Lin@amd.com>
To: <amd-gfx@lists.freedesktop.org>
Cc: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>,
Felix Kuehling <felix.kuehling@amd.com>
Subject: [PATCH 3/3] drm/amdkfd: Add support for more per-process flag
Date: Thu, 6 Mar 2025 10:55:55 -0500 [thread overview]
Message-ID: <20250306155555.4687-3-Amber.Lin@amd.com> (raw)
In-Reply-To: <20250306155555.4687-1-Amber.Lin@amd.com>
From: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Add support for more per-process flags starting with option to configure
MFMA precision for gfx 9.5
v2: Change flag name to KFD_PROC_FLAG_MFMA_HIGH_PRECISION
Remove unused else condition
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 3 ++-
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 6 ++++--
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h | 6 ++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c | 6 ++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c | 6 ++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c | 6 ++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c | 6 ++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c | 11 +++++++++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c | 6 ++++--
include/uapi/linux/kfd_ioctl.h | 5 ++++-
10 files changed, 43 insertions(+), 18 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 8c2e92378b49..1e9dd00620bf 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -606,7 +606,8 @@ static int kfd_ioctl_set_memory_policy(struct file *filep,
default_policy,
alternate_policy,
(void __user *)args->alternate_aperture_base,
- args->alternate_aperture_size))
+ args->alternate_aperture_size,
+ args->misc_process_flag))
err = -EINVAL;
out:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index d23c6a358d34..2afcc1b4856a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2596,7 +2596,8 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
bool retval = true;
@@ -2611,7 +2612,8 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm,
default_policy,
alternate_policy,
alternate_aperture_base,
- alternate_aperture_size);
+ alternate_aperture_size,
+ misc_process_properties);
if (retval)
goto out;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index 7146e227e2c1..122eb745e9c4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -174,7 +174,8 @@ struct device_queue_manager_ops {
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
int (*process_termination)(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
@@ -210,7 +211,8 @@ struct device_queue_manager_asic_ops {
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
void (*init_sdma_vm)(struct device_queue_manager *dqm,
struct queue *q,
struct qcm_process_device *qpd);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
index 32bedef912b3..0508ef5a41d7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c
@@ -40,7 +40,8 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
static int update_qpd_cik(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
static void init_sdma_vm(struct device_queue_manager *dqm,
@@ -88,7 +89,8 @@ static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
uint32_t default_mtype;
uint32_t ape1_mtype;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
index b5f5f141353b..ba6e3d747ccd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
@@ -36,7 +36,8 @@ static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
void device_queue_manager_init_v10(
struct device_queue_manager_asic_ops *asic_ops)
@@ -61,7 +62,8 @@ static bool set_cache_memory_policy_v10(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
index f436878d0d62..8b447d04558f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v11.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
void device_queue_manager_init_v11(
struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v11(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
index 62ca1c8fcbaf..3550da3a46f9 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v12.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
void device_queue_manager_init_v12(
struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v12(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
qpd->sh_mem_config = (SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) |
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
index c734eb9b505f..4635077aa905 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c
@@ -35,7 +35,8 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
void device_queue_manager_init_v9(
struct device_queue_manager_asic_ops *asic_ops)
@@ -60,7 +61,8 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
qpd->sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
@@ -73,6 +75,11 @@ static bool set_cache_memory_policy_v9(struct device_queue_manager *dqm,
KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 5, 0))
qpd->sh_mem_config |= (1 << SH_MEM_CONFIG__F8_MODE__SHIFT);
+ if (KFD_GC_VERSION(dqm->dev->kfd) == IP_VERSION(9, 5, 0)) {
+ if (misc_process_properties & KFD_PROC_FLAG_MFMA_HIGH_PRECISION)
+ qpd->sh_mem_config |= 1 << SH_MEM_CONFIG__PRECISION_MODE__SHIFT;
+ }
+
qpd->sh_mem_ape1_limit = 0;
qpd->sh_mem_ape1_base = 0;
qpd->sh_mem_bases = compute_sh_mem_bases_64bit(qpd_to_pdd(qpd));
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
index 320518f41890..dad83356e976 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c
@@ -40,7 +40,8 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size);
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties);
static int update_qpd_vi(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
static void init_sdma_vm(struct device_queue_manager *dqm,
@@ -89,7 +90,8 @@ static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm,
enum cache_policy default_policy,
enum cache_policy alternate_policy,
void __user *alternate_aperture_base,
- uint64_t alternate_aperture_size)
+ uint64_t alternate_aperture_size,
+ u32 misc_process_properties)
{
uint32_t default_mtype;
uint32_t ape1_mtype;
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index b0160b09987c..e7b4b1551529 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -150,6 +150,9 @@ struct kfd_dbg_device_info_entry {
#define KFD_IOC_CACHE_POLICY_COHERENT 0
#define KFD_IOC_CACHE_POLICY_NONCOHERENT 1
+/* Misc. per process flags */
+#define KFD_PROC_FLAG_MFMA_HIGH_PRECISION (1 << 0)
+
struct kfd_ioctl_set_memory_policy_args {
__u64 alternate_aperture_base; /* to KFD */
__u64 alternate_aperture_size; /* to KFD */
@@ -157,7 +160,7 @@ struct kfd_ioctl_set_memory_policy_args {
__u32 gpu_id; /* to KFD */
__u32 default_policy; /* to KFD */
__u32 alternate_policy; /* to KFD */
- __u32 pad;
+ __u32 misc_process_flag; /* to KFD */
};
/*
--
2.34.1
prev parent reply other threads:[~2025-03-06 15:56 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-03-06 15:55 [PATCH 1/3] drm/amdkfd: Set per-process flags only once cik/vi Amber Lin
2025-03-06 15:55 ` [PATCH 2/3] drm/amdkfd: Set per-process flags only once for gfx9/10/11/12 Amber Lin
2025-03-06 15:55 ` Amber Lin [this message]
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=20250306155555.4687-3-Amber.Lin@amd.com \
--to=amber.lin@amd.com \
--cc=Harish.Kasiviswanathan@amd.com \
--cc=amd-gfx@lists.freedesktop.org \
--cc=felix.kuehling@amd.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.