drm/xe/xe_guc: Expand guc_g2g_alloc error hanlding for future change
Subsystems (2)
PATCH 1/1 drm/xe/xe_guc: Expand guc_g2g_alloc error hanlding for future change
On today's driver, xe_device_get_gt(xe, 0); can never return NULL. Hardware-wise there's always at least one tile, and every tilie has a primary GT. If something went wrong during init of tile or GT and we couldn't create/initialize the structures, then we already aborted the device probe immediately and we'll never get further on to places in the code that would be chasing a NULL pointer.
However, there's currently ongoing work to allow the primary GT to be disabled via configfs for debugging purposes. Once that lands, it will be possible for this query to return a NULL pointer. Expand the error handling code for guc_g2g_alloc to manage this in the future.
--- drivers/gpu/drm/xe/xe_guc.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/gpu/drm/xe/xe_guc.c +5 -1
@@ -467,9 +467,13 @@ static int guc_g2g_alloc(struct xe_guc *guc)if (gt->info.id !=0 ) {struct xe_gt *root_gt =xe_device_get_gt (xe,0 ); -struct xe_guc *root_guc = &root_gt->uc.guc; +struct xe_guc *root_guc;struct xe_bo *bo; +if (!root_gt) +return -ENODEV ; + + root_guc = &root_gt->uc.guc; bo =xe_bo_get (root_guc->g2g.bo);if (!bo)return -ENODEV ;
Signature · 1 line
2.43.0
Quoted 11 lines · Jonathan Cavitt
On today's driver,
xe_device_get_gt(xe, 0); can never return NULL. Hardware-wise there's always at least one tile, and every tilie has a primary GT. If something went wrong during init of tile or GT and we couldn't create/initialize the structures, then we already aborted the device probe immediately and we'll never get further on to places in the code that would be chasing a NULL pointer.However, there's currently ongoing work to allow the primary GT to be disabled via configfs for debugging purposes. Once that lands, it will be possible for this query to return a NULL pointer. Expand the error handling code for
guc_g2g_allocto manage this in the future.
Rather than blindly adding checks to any low-level code that calls xe_device_get_gt() (and then raising errors that we hope will do something reasonable when propagated up the call stack), I think we should be putting more thought into what these functions are and how they get called. Functions like guc_g2g_alloc() are something that get called deep within the callstack of GuC-to-GuC communication. We should be thinking about what the purpose/use of this code is. If GuC-to-GuC always requires communication with the first tile's primary GT, then G2G itself should just be disabled at the higher level so that we never call down to this part of the code (e.g., in xe_guc_g2g_wanted that already has other "is this possible?" checks). Or if there's other feasible use cases (e.g., a media GuC on one tile talking to a media GuC on another tile), then we should be changing the code to remove the hardcoded use of "gt0" so that those use cases work.
Matt
Quoted 30 lines
Suggested-by Michal Wajdeczko <[email protected]>Suggested-by Matt Roper <[email protected]>Signed-off-by Jonathan Cavitt <[email protected]>--- drivers/gpu/drm/xe/xe_guc.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)drivers/gpu/drm/xe/xe_guc.c +5
-1@@ -467,9 +467,13 @@ static int guc_g2g_alloc(struct xe_guc *guc)if (gt->info.id !=0 ) {struct xe_gt *root_gt =xe_device_get_gt (xe,0 ); -struct xe_guc *root_guc = &root_gt->uc.guc; +struct xe_guc *root_guc;struct xe_bo *bo; +if (!root_gt) +return -ENODEV ; + + root_guc = &root_gt->uc.guc; bo =xe_bo_get (root_guc->g2g.bo);if (!bo)return -ENODEV ;Signature · 1 line
2.43.0
Signature · 4 lines
Matt Roper Graphics Software Engineer Linux GPU Platform Enablement Intel Corporation
Quoted 28 lines · Matt Roper
Quoted 11 lines · Jonathan Cavitt
On today's driver,
xe_device_get_gt(xe, 0); can never return NULL. Hardware-wise there's always at least one tile, and every tilie has a primary GT. If something went wrong during init of tile or GT and we couldn't create/initialize the structures, then we already aborted the device probe immediately and we'll never get further on to places in the code that would be chasing a NULL pointer.However, there's currently ongoing work to allow the primary GT to be disabled via configfs for debugging purposes. Once that lands, it will be possible for this query to return a NULL pointer. Expand the error handling code for
guc_g2g_allocto manage this in the future.Rather than blindly adding checks to any low-level code that calls
xe_device_get_gt()(and then raising errors that we hope will do something reasonable when propagated up the call stack), I think we should be putting more thought into what these functions are and how they get called. Functions likeguc_g2g_alloc()are something that get called deep within the callstack of GuC-to-GuC communication. We should be thinking about what the purpose/use of this code is. If GuC-to-GuC always requires communication with the first tile's primary GT, then G2G itself should just be disabled at the higher level so that we never call down to this part of the code (e.g., inxe_guc_g2g_wantedthat already has other "is this possible?" checks). Or if there's other feasible use cases (e.g., a media GuC on one tile talking to a media GuC on another tile), then we should be changing the code to remove the hardcoded use of "gt0" so that those use cases work.
Quoted 1 line
Matt
The use of gt0 is purely for convenience here and the code could be reworked to not use it.
A single object is allocated for all G2G buffers and to make things simple the object is owned by gt0, so the logic (simplified) is as follows:
if (gt_id == 0) {
allocate buffer
} else {
fetch gt0
get gtg buffer from gt0
}
We could change the ownership to the first found GuC instead, or move the object outside of the GuC structure entirely and into xe_device.
Daniele
Quoted 30 lines
Quoted 30 lines
Suggested-by Michal Wajdeczko <[email protected]>Suggested-by Matt Roper <[email protected]>Signed-off-by Jonathan Cavitt <[email protected]>--- drivers/gpu/drm/xe/xe_guc.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)drivers/gpu/drm/xe/xe_guc.c +5
-1@@ -467,9 +467,13 @@ static int guc_g2g_alloc(struct xe_guc *guc)if (gt->info.id !=0 ) {struct xe_gt *root_gt =xe_device_get_gt (xe,0 ); -struct xe_guc *root_guc = &root_gt->uc.guc; +struct xe_guc *root_guc;struct xe_bo *bo; +if (!root_gt) +return -ENODEV ; + + root_guc = &root_gt->uc.guc; bo =xe_bo_get (root_guc->g2g.bo);if (!bo)return -ENODEV ;Signature · 1 line
2.43.0