drm/nouveau/gsp: fix potential leak of memory used during acpi init
Subsystems (3)
PATCH 1/1 drm/nouveau/gsp: fix potential leak of memory used during acpi init
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
--- .../drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c | 20 +++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-)
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +14 -6
@@ -719,7 +719,6 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps)union acpi_object argv4 = { .buffer.type =ACPI_TYPE_BUFFER , .buffer.length =4 , - .buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ), }, *obj; caps->status =0xffff ;
@@ -727,17 +726,22 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps)if (!acpi_check_dsm (handle, &NVOP_DSM_GUID ,NVOP_DSM_REV ,BIT_ULL (0x1a )))return ; + argv4.buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ); +if (!argv4.buffer.pointer) +return ; + obj =acpi_evaluate_dsm (handle, &NVOP_DSM_GUID ,NVOP_DSM_REV ,0x1a , &argv4);if (!obj) -return ; + goto done;if (WARN_ON (obj->type !=ACPI_TYPE_BUFFER ) ||WARN_ON (obj->buffer.length !=4 )) -return ; + goto done; caps->status =0 ; caps->optimusCaps = *(u32 *)obj->buffer.pointer; +done:ACPI_FREE (obj);kfree (argv4.buffer.pointer);
@@ -754,24 +758,28 @@ r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt)union acpi_object argv4 = { .buffer.type =ACPI_TYPE_BUFFER , .buffer.length =sizeof (caps), - .buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ), }, *obj; jt->status =0xffff ; + argv4.buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ); +if (!argv4.buffer.pointer) +return ; + obj =acpi_evaluate_dsm (handle, &JT_DSM_GUID ,JT_DSM_REV ,0x1 , &argv4);if (!obj) -return ; + goto done;if (WARN_ON (obj->type !=ACPI_TYPE_BUFFER ) ||WARN_ON (obj->buffer.length !=4 )) -return ; + goto done; jt->status =0 ; jt->jtCaps = *(u32 *)obj->buffer.pointer; jt->jtRevId = (jt->jtCaps &0xfff00000 ) >>20 ; jt->bSBIOSCaps =0 ; +done:ACPI_FREE (obj);kfree (argv4.buffer.pointer);
Signature · 1 line
2.49.0
Quoted 9 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
Reported-by Danilo Krummrich <[email protected]>Fixes 176fdcbddfd2 ("drm/nouveau/gsp/r535: add support for bootingGSP-RM")
Needs to +Cc the stable kernel
But, see below
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
Quoted 11 lines
Signed-off-by Ben Skeggs <[email protected]>---
.../drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c | 20 +++++++++++++---- --1 file changed, 14 insertions(+), 6 deletions(-)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0
This seems to be based on a code move that is not yet in mainline.
Therefore, backporting the bugfix to stable seems difficult. Since that code move is already in drm-misc-next, it would seem that it can only be solved with two distinct patches for stable and for -next.
But this needs to be judged by a maintainer.
Quoted 18 lines
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -719,7 +719,6 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = 4, - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
caps->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0@@ -727,17 +726,22 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) if (!acpi_check_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV,BIT_ULL (0x1a ))) return;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
GFP_KERNEL);
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) · line 732
Quoted 3 lines
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +
This could be done immediately after the creation of argv4. That way it's more difficult to have the leak again if something is inserted later on.
Quoted 44 lines
obj =
acpi_evaluate_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, 0x1a, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
caps->status= 0;
caps->optimusCaps= *(u32 *)obj->buffer.pointer;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -754,24 +758,28 @@ r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = sizeof(caps), - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
jt->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
GFP_KERNEL);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &JT_DSM_GUID, JT_DSM_REV, 0x1, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
Quoted 7 lines
jt->status= 0;
jt->jtCaps= *(u32 *)obj->buffer.pointer;jt->jtRevId= (jt->jtCaps&0xfff00000) >> 20;jt->bSBIOSCaps= 0;
+done:
'done' seems like a bad name considering that the operations are aborted with a WARN_ON above. Better 'abort' or sth like that.
P.
Quoted 3 lines
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);
Danilo Krummrich
maintainerQuoted 8 lines · Philipp Stanner
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
Quoted 5 lines · Ben Skeggs
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0This seems to be based on a code move that is not yet in mainline.
It is, it did land with v6.16-rc1.
Quoted 3 lines
Therefore, backporting the bugfix to stable seems difficult. Since that code move is already in drm-misc-next, it would seem that it can only be solved with two distinct patches for stable and for -next.
drm-misc-fixes is the relevant target branch and given the above, it contains the code move as well.
However, you're right that this fix won't apply to anything before v6.16-rc1. Given that, it makes sense to leave a note below the '---' line that this fix won't apply before v6.16-rc1 and that a backported patch will be sent to stable once this one hit Linus' tree.
Quoted 19 lines
But this needs to be judged by a maintainer.
Quoted 17 lines
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -719,7 +719,6 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = 4, - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
caps->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0@@ -727,17 +726,22 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) if (!acpi_check_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV,BIT_ULL (0x1a ))) return;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
Quoted 8 lines
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) · line 732
Quoted 4 lines
GFP_KERNEL);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +This could be done immediately after the creation of argv4. That way it's more difficult to have the leak again if something is inserted later on.
I think the idea was to avoid a potential unwind path after acpi_check_dsm().
Quoted 46 lines
Quoted 46 lines
obj =
acpi_evaluate_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, 0x1a, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
caps->status= 0;
caps->optimusCaps= *(u32 *)obj->buffer.pointer;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -754,24 +758,28 @@ r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = sizeof(caps), - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
jt->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
GFP_KERNEL);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &JT_DSM_GUID, JT_DSM_REV, 0x1, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
jt->status= 0;
Quoted 8 lines
Quoted 5 lines
jt->jtCaps= *(u32 *)obj->buffer.pointer;jt->jtRevId= (jt->jtCaps&0xfff00000) >> 20;jt->bSBIOSCaps= 0;
+done:'done' seems like a bad name considering that the operations are aborted with a
WARN_ONabove. Better 'abort' or sth like that.
I think some neutral name is fine, since we also enter this code path when everything went well, maybe 'out_free' or just 'free'?
Quoted 6 lines
P.
Quoted 3 lines
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);
Danilo Krummrich
maintainerQuoted 16 lines · Danilo Krummrich
Quoted 8 lines · Philipp Stanner
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c
Quoted 5 lines · Ben Skeggs
drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0This seems to be based on a code move that is not yet in mainline.
It is, it did land with v6.16-rc1.
Quoted 3 lines
Therefore, backporting the bugfix to stable seems difficult. Since that code move is already in drm-misc-next, it would seem that it can only be solved with two distinct patches for stable and for -next.
Quoted 7 lines
drm-misc-fixes is the relevant target branch and given the above, it contains the code move as well.
However, you're right that this fix won't apply to anything before v6.16-rc1. Given that, it makes sense to leave a note below the '---' line that this fix won't apply before v6.16-rc1 and that a backported patch will be sent to stable once this one hit Linus' tree.
Nevermind, I think this file wasn't split up and is correctly considered as moved by git, so I think the patch should apply as is.
Quoted 5 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
If you've got a kmemleak trace, it would also be good to put it here in the commit message so that we can distinguish this bug from potential other leaks.
P.
Quoted 90 lines
Reported-by Danilo Krummrich <[email protected]>Fixes 176fdcbddfd2 ("drm/nouveau/gsp/r535: add support for bootingGSP-RM")
Signed-off-by Ben Skeggs <[email protected]>---
.../drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c | 20 +++++++++++++---- --1 file changed, 14 insertions(+), 6 deletions(-)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -719,7 +719,6 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = 4, - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
caps->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0@@ -727,17 +726,22 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) if (!acpi_check_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV,BIT_ULL (0x1a ))) return;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
GFP_KERNEL);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, 0x1a, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
caps->status= 0;
caps->optimusCaps= *(u32 *)obj->buffer.pointer;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -754,24 +758,28 @@ r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = sizeof(caps), - .buffer.pointer =kmalloc (argv4.buffer.length, GFP_KERNEL), }, *obj;
jt->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-0+ argv4.buffer.pointer = kmalloc(argv4.buffer.length,
GFP_KERNEL);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +3
-0+if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &JT_DSM_GUID, JT_DSM_REV, 0x1, &argv4);
if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
jt->status= 0;
jt->jtCaps= *(u32 *)obj->buffer.pointer;jt->jtRevId= (jt->jtCaps&0xfff00000) >> 20;jt->bSBIOSCaps= 0;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);
Danilo Krummrich
maintainerQuoted 10 lines · Philipp Stanner
Quoted 5 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
If you've got a kmemleak trace, it would also be good to put it here in the commit message so that we can distinguish this bug from potential other leaks.
unreferenced object 0xffff8ed5029bfb28 (size 8):
comm "(udev-worker)", pid 464, jiffies 4294672444 hex dump (first 8 bytes):
7c b4 d4 79 01 59 36 6c |..y.Y6l
backtrace (crc 4461fdb7):
__kmalloc_cache_noprof+0x31b/0x410
r535_gsp_acpi_jt+0x7c/0x110 [nouveau]
r535_gsp_set_system_info+0x153/0x390 [nouveau]
r535_gsp_oneinit+0xa4d/0xf50 [nouveau]
tu102_gsp_oneinit+0x124/0x440 [nouveau]
nvkm_subdev_oneinit_+0x3f/0x90 [nouveau]
nvkm_subdev_init_+0x33/0xa0 [nouveau]
nvkm_subdev_init+0x46/0x60 [nouveau]
nvkm_device_init+0x167/0x1f0 [nouveau]
nvkm_udevice_init+0x4b/0x70 [nouveau]
nvkm_object_init+0x3a/0x110 [nouveau]
nvkm_ioctl_new+0x13a/0x200 [nouveau]
nvkm_ioctl+0x9f/0x140 [nouveau]
nvif_object_ctor+0x11a/0x1a0 [nouveau]
nvif_device_ctor+0x2a/0x80 [nouveau]
nouveau_drm_device_new+0x157/0x2e0 [nouveau]
unreferenced object 0xffff8ed502a37580 (size 32):
comm "(udev-worker)", pid 464, jiffies 4294672444
hex dump (first 32 bytes):
01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace (crc f1da05aa):
__kmalloc_noprof+0x3ac/0x500
acpi_ut_initialize_buffer+0x67/0xc0
acpi_evaluate_object+0x272/0x2c0
acpi_evaluate_dsm+0xb4/0x120
r535_gsp_acpi_jt+0xa3/0x110 [nouveau]
r535_gsp_set_system_info+0x153/0x390 [nouveau]
r535_gsp_oneinit+0xa4d/0xf50 [nouveau]
tu102_gsp_oneinit+0x124/0x440 [nouveau]
nvkm_subdev_oneinit_+0x3f/0x90 [nouveau]
nvkm_subdev_init_+0x33/0xa0 [nouveau]
nvkm_subdev_init+0x46/0x60 [nouveau]
nvkm_device_init+0x167/0x1f0 [nouveau]
nvkm_udevice_init+0x4b/0x70 [nouveau]
nvkm_object_init+0x3a/0x110 [nouveau]
nvkm_ioctl_new+0x13a/0x200 [nouveau]
nvkm_ioctl+0x9f/0x140 [nouveau]
unreferenced object 0xffff8ed5029bf1c0 (size 8):
comm "(udev-worker)", pid 464, jiffies 4294672446
hex dump (first 8 bytes):
cc bb d4 79 01 59 3c 84 ...y.Y<.
backtrace (crc 30e1d939):
__kmalloc_cache_noprof+0x31b/0x410
r535_gsp_acpi_caps+0x7e/0x120 [nouveau]
r535_gsp_set_system_info+0x162/0x390 [nouveau]
r535_gsp_oneinit+0xa4d/0xf50 [nouveau]
tu102_gsp_oneinit+0x124/0x440 [nouveau]
nvkm_subdev_oneinit_+0x3f/0x90 [nouveau]
nvkm_subdev_init_+0x33/0xa0 [nouveau]
nvkm_subdev_init+0x46/0x60 [nouveau]
nvkm_device_init+0x167/0x1f0 [nouveau]
nvkm_udevice_init+0x4b/0x70 [nouveau]
nvkm_object_init+0x3a/0x110 [nouveau]
nvkm_ioctl_new+0x13a/0x200 [nouveau]
nvkm_ioctl+0x9f/0x140 [nouveau]
nvif_object_ctor+0x11a/0x1a0 [nouveau]
nvif_device_ctor+0x2a/0x80 [nouveau]
nouveau_drm_device_new+0x157/0x2e0 [nouveau]Quoted 60 lines · Danilo Krummrich
Quoted 10 lines · Philipp Stanner
Quoted 5 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
If you've got a kmemleak trace, it would also be good to put it here in the commit message so that we can distinguish this bug from potential other leaks.
unreferenced object
0xffff8ed5029bfb28(size 8):
comm "(udev-worker)", pid 464, jiffies 4294672444 hex dump (first 8 bytes):
7c b4 d4 79 01 59 36 6c |..y.Y6l backtrace (crc4461fdb7):__kmalloc_cache_noprof+0x31b/0x410 r535_gsp_acpi_jt+0x7c/0x110 [nouveau] r535_gsp_set_system_info+0x153/0x390 [nouveau] r535_gsp_oneinit+0xa4d/0xf50 [nouveau] tu102_gsp_oneinit+0x124/0x440 [nouveau] nvkm_subdev_oneinit_+0x3f/0x90 [nouveau] nvkm_subdev_init_+0x33/0xa0 [nouveau] nvkm_subdev_init+0x46/0x60 [nouveau] nvkm_device_init+0x167/0x1f0 [nouveau] nvkm_udevice_init+0x4b/0x70 [nouveau] nvkm_object_init+0x3a/0x110 [nouveau] nvkm_ioctl_new+0x13a/0x200 [nouveau] nvkm_ioctl+0x9f/0x140 [nouveau] nvif_object_ctor+0x11a/0x1a0 [nouveau] nvif_device_ctor+0x2a/0x80 [nouveau] nouveau_drm_device_new+0x157/0x2e0 [nouveau]unreferenced object
0xffff8ed502a37580(size 32):
comm "(udev-worker)", pid 464, jiffies 4294672444
hex dump (first 32 bytes):
01 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ backtrace (crcf1da05aa):__kmalloc_noprof+0x3ac/0x500 acpi_ut_initialize_buffer+0x67/0xc0 acpi_evaluate_object+0x272/0x2c0 acpi_evaluate_dsm+0xb4/0x120 r535_gsp_acpi_jt+0xa3/0x110 [nouveau] r535_gsp_set_system_info+0x153/0x390 [nouveau] r535_gsp_oneinit+0xa4d/0xf50 [nouveau] tu102_gsp_oneinit+0x124/0x440 [nouveau] nvkm_subdev_oneinit_+0x3f/0x90 [nouveau] nvkm_subdev_init_+0x33/0xa0 [nouveau] nvkm_subdev_init+0x46/0x60 [nouveau] nvkm_device_init+0x167/0x1f0 [nouveau] nvkm_udevice_init+0x4b/0x70 [nouveau] nvkm_object_init+0x3a/0x110 [nouveau] nvkm_ioctl_new+0x13a/0x200 [nouveau] nvkm_ioctl+0x9f/0x140 [nouveau]unreferenced object
0xffff8ed5029bf1c0(size 8):
comm "(udev-worker)", pid 464, jiffies 4294672446 hex dump (first 8 bytes):
cc bb d4 79 01 59 3c 84 ...y.Y<. backtrace (crc30e1d939):
Quoted 17 lines
__kmalloc_cache_noprof+0x31b/0x410 r535_gsp_acpi_caps+0x7e/0x120 [nouveau] r535_gsp_set_system_info+0x162/0x390 [nouveau] r535_gsp_oneinit+0xa4d/0xf50 [nouveau] tu102_gsp_oneinit+0x124/0x440 [nouveau] nvkm_subdev_oneinit_+0x3f/0x90 [nouveau] nvkm_subdev_init_+0x33/0xa0 [nouveau] nvkm_subdev_init+0x46/0x60 [nouveau] nvkm_device_init+0x167/0x1f0 [nouveau] nvkm_udevice_init+0x4b/0x70 [nouveau] nvkm_object_init+0x3a/0x110 [nouveau] nvkm_ioctl_new+0x13a/0x200 [nouveau] nvkm_ioctl+0x9f/0x140 [nouveau] nvif_object_ctor+0x11a/0x1a0 [nouveau] nvif_device_ctor+0x2a/0x80 [nouveau] nouveau_drm_device_new+0x157/0x2e0 [nouveau]
Hm, interesting. That's not the memory leak we have observed. I've got this trace for leaks of similar size:
unreferenced object 0xff110001518f9348 (size 8):
comm "kworker/0:4", pid 141, jiffies 4294719397
hex dump (first 8 bytes):
98 93 36 c3 c1 b9 80 3e ..6....>
backtrace (crc a04f7c42):
kmalloc_trace+0x28a/0x380
r535_gsp_rpc_set_system_info+0x7e8/0x17f0 [nouveau]
r535_gsp_oneinit+0x1ab0/0x2820 [nouveau]
nvkm_subdev_oneinit_.part.0+0x91/0x1a0 [nouveau]
nvkm_subdev_init_+0xfe/0x2b0 [nouveau]
nvkm_subdev_init+0xa7/0xc0 [nouveau]
nvkm_device_init+0x347/0x540 [nouveau]
nvkm_udevice_init+0x97/0xf0 [nouveau]
nvkm_object_init+0xc3/0x3e0 [nouveau]
nvkm_ioctl_new+0x36c/0x6b0 [nouveau]
nvkm_ioctl+0x1c0/0x4a0 [nouveau]
nvif_object_ctor+0x3c9/0x740 [nouveau]
nvif_device_ctor+0x2e/0x100 [nouveau]
nouveau_drm_device_new+0x367/0x910 [nouveau]
nouveau_drm_probe+0x10c/0x450 [nouveau]
local_pci_probe+0xe8/0x1a0
unreferenced object 0xff11000142e1bf50 (size 8):
comm "kworker/0:4", pid 141, jiffies 4294719397
hex dump (first 8 bytes):
88 b2 58 d0 d2 d7 ac 26 ..X....&
backtrace (crc 8a38119a):
kmalloc_trace+0x28a/0x380
r535_gsp_rpc_set_system_info+0xa8f/0x17f0 [nouveau]
r535_gsp_oneinit+0x1ab0/0x2820 [nouveau]
nvkm_subdev_oneinit_.part.0+0x91/0x1a0 [nouveau]
nvkm_subdev_init_+0xfe/0x2b0 [nouveau]
nvkm_subdev_init+0xa7/0xc0 [nouveau]
nvkm_device_init+0x347/0x540 [nouveau]
nvkm_udevice_init+0x97/0xf0 [nouveau]
nvkm_object_init+0xc3/0x3e0 [nouveau]
nvkm_ioctl_new+0x36c/0x6b0 [nouveau]
nvkm_ioctl+0x1c0/0x4a0 [nouveau]
nvif_object_ctor+0x3c9/0x740 [nouveau]
nvif_device_ctor+0x2e/0x100 [nouveau]
nouveau_drm_device_new+0x367/0x910 [nouveau]
nouveau_drm_probe+0x10c/0x450 [nouveau]Ben, could those be related?
P.
Quoted 8 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
Reported-by Danilo Krummrich <[email protected]>Fixes 176fdcbddfd2 ("drm/nouveau/gsp/r535: add support for booting GSP-RM")Signed-off-by Ben Skeggs <[email protected]>
Applied to drm-misc-fixes, thanks!
Quoted 8 lines · Ben Skeggs
If any of the ACPI calls fail, memory allocated for the input buffer would be leaked. Fix failure paths to free allocated memory.
Also add checks to ensure the allocations succeeded in the first place.
Reported-by Danilo Krummrich <[email protected]>Fixes 176fdcbddfd2 ("drm/nouveau/gsp/r535: add support for booting GSP-RM")Signed-off-by Ben Skeggs <[email protected]>
This
P.
Quoted 73 lines
---
.../drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c | 20 +++++++++++++------1 file changed, 14 insertions(+), 6 deletions(-)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -719,7 +719,6 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = 4, - .buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ), }, *obj;
caps->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-0@@ -727,17 +726,22 @@ r535_gsp_acpi_caps(acpi_handle handle, CAPS_METHOD_DATA *caps) if (!acpi_check_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, BIT_ULL(0x1a))) return;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +4
-0+ argv4.buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ); +if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &NVOP_DSM_GUID, NVOP_DSM_REV, 0x1a, &argv4); if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
caps->status= 0;
caps->optimusCaps= *(u32 *)obj->buffer.pointer;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +0
-1@@ -754,24 +758,28 @@ r535_gsp_acpi_jt(acpi_handle handle, JT_METHOD_DATA *jt) union acpi_object argv4 = { .buffer.type = ACPI_TYPE_BUFFER, .buffer.length = sizeof(caps), - .buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ), }, *obj;
jt->status=0xffff;drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +4
-0+ argv4.buffer.pointer =kmalloc (argv4.buffer.length,GFP_KERNEL ); +if (!argv4.buffer.pointer) +return ; +obj =
acpi_evaluate_dsm(handle, &JT_DSM_GUID, JT_DSM_REV, 0x1, &argv4); if (!obj)drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
if (WARN_ON(obj->type != ACPI_TYPE_BUFFER)||WARN_ON(obj->buffer.length != 4))drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/gsp.c +1
-1-return ; + goto done;
jt->status= 0;
jt->jtCaps= *(u32 *)obj->buffer.pointer;jt->jtRevId= (jt->jtCaps&0xfff00000) >> 20;jt->bSBIOSCaps= 0;
+done:
ACPI_FREE(obj);
kfree(argv4.buffer.pointer);