From 7cb779a6867fea00b4209bcf6de2f178a743247d Mon Sep 17 00:00:00 2001 From: Stanislav Fomichev Date: Mon, 11 Sep 2023 12:47:30 -0700 Subject: bpf: Clarify error expectations from bpf_clone_redirect Commit 151e887d8ff9 ("veth: Fixing transmit return status for dropped packets") exposed the fact that bpf_clone_redirect is capable of returning raw NET_XMIT_XXX return codes. This is in the conflict with its UAPI doc which says the following: "0 on success, or a negative error in case of failure." Update the UAPI to reflect the fact that bpf_clone_redirect can return positive error numbers, but don't explicitly define their meaning. Reported-by: Daniel Borkmann Signed-off-by: Stanislav Fomichev Signed-off-by: Daniel Borkmann Link: https://lore.kernel.org/bpf/20230911194731.286342-1-sdf@google.com --- include/uapi/linux/bpf.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'include/uapi') diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index 8790b3962e4b..0448700890f7 100644 --- a/include/uapi/linux/bpf.h +++ b/include/uapi/linux/bpf.h @@ -1962,7 +1962,9 @@ union bpf_attr { * performed again, if the helper is used in combination with * direct packet access. * Return - * 0 on success, or a negative error in case of failure. + * 0 on success, or a negative error in case of failure. Positive + * error indicates a potential drop or congestion in the target + * device. The particular positive error codes are not defined. * * u64 bpf_get_current_pid_tgid(void) * Description -- cgit v1.2.3 From 531108ec5b5cd45ec6272a6115e73275baef7d22 Mon Sep 17 00:00:00 2001 From: Alexey Dobriyan Date: Tue, 12 Sep 2023 19:23:21 +0300 Subject: uapi: stddef.h: Fix header guard location The #endif for the header guard wasn't at the end of the header. This was harmless since the define that escaped was already testing for its own redefinition. Regardless, move the #endif to the correct place. Signed-off-by: Alexey Dobriyan Fixes: c8248faf3ca2 ("Compiler Attributes: counted_by: Adjust name and identifier expansion") Link: https://lore.kernel.org/r/b1f5081e-339d-421d-81b2-cbb94e1f6f5f@p183 Co-developed-by: Kees Cook Signed-off-by: Kees Cook --- include/uapi/linux/stddef.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'include/uapi') diff --git a/include/uapi/linux/stddef.h b/include/uapi/linux/stddef.h index 7c3fc3980881..c027b2070d79 100644 --- a/include/uapi/linux/stddef.h +++ b/include/uapi/linux/stddef.h @@ -44,8 +44,9 @@ struct { } __empty_ ## NAME; \ TYPE NAME[]; \ } -#endif #ifndef __counted_by #define __counted_by(m) #endif + +#endif /* _UAPI_LINUX_STDDEF_H */ -- cgit v1.2.3 From 32a4ec211d4164e667d9d0b807fadf02053cd2e9 Mon Sep 17 00:00:00 2001 From: Alexey Dobriyan Date: Tue, 12 Sep 2023 19:22:24 +0300 Subject: uapi: stddef.h: Fix __DECLARE_FLEX_ARRAY for C++ __DECLARE_FLEX_ARRAY(T, member) macro expands to struct { struct {} __empty_member; T member[]; }; which is subtly wrong in C++ because sizeof(struct{}) is 1 not 0, changing UAPI structures layouts. This can be fixed by expanding to T member[]; Now g++ doesn't like "T member[]" either, throwing errors on the following code: struct S { union { T1 member1[]; T2 member2[]; }; }; or struct S { T member[]; }; Use "T member[0];" which seems to work and does the right thing wrt structure layout. Signed-off-by: Alexey Dobriyan Fixes: 3080ea5553cc ("stddef: Introduce DECLARE_FLEX_ARRAY() helper") Link: https://lore.kernel.org/r/97242381-f1ec-4a4a-9472-1a464f575657@p183 Signed-off-by: Kees Cook --- include/uapi/linux/stddef.h | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'include/uapi') diff --git a/include/uapi/linux/stddef.h b/include/uapi/linux/stddef.h index c027b2070d79..5c6c4269f7ef 100644 --- a/include/uapi/linux/stddef.h +++ b/include/uapi/linux/stddef.h @@ -29,6 +29,11 @@ struct TAG { MEMBERS } ATTRS NAME; \ } +#ifdef __cplusplus +/* sizeof(struct{}) is 1 in C++, not 0, can't use C version of the macro. */ +#define __DECLARE_FLEX_ARRAY(T, member) \ + T member[0] +#else /** * __DECLARE_FLEX_ARRAY() - Declare a flexible array usable in a union * @@ -44,6 +49,7 @@ struct { } __empty_ ## NAME; \ TYPE NAME[]; \ } +#endif #ifndef __counted_by #define __counted_by(m) -- cgit v1.2.3 From d59e75eef52d89201aaf5342a3ac23ddf3e9b112 Mon Sep 17 00:00:00 2001 From: Danilo Krummrich Date: Mon, 2 Oct 2023 15:46:48 +0200 Subject: drm/nouveau: exec: report max pushs through getparam Report the maximum number of IBs that can be pushed with a single DRM_IOCTL_NOUVEAU_EXEC through DRM_IOCTL_NOUVEAU_GETPARAM. While the maximum number of IBs per ring might vary between chipsets, the kernel will make sure that userspace can only push a fraction of the maximum number of IBs per ring per job, such that we avoid a situation where there's only a single job occupying the ring, which could potentially lead to the ring run dry. Using DRM_IOCTL_NOUVEAU_GETPARAM to report the maximum number of IBs that can be pushed with a single DRM_IOCTL_NOUVEAU_EXEC implies that all channels of a given device have the same ring size. Reviewed-by: Dave Airlie Reviewed-by: Lyude Paul Acked-by: Faith Ekstrand Signed-off-by: Danilo Krummrich Link: https://patchwork.freedesktop.org/patch/msgid/20231002135008.10651-3-dakr@redhat.com --- drivers/gpu/drm/nouveau/nouveau_abi16.c | 21 +++++++++++++++++++++ drivers/gpu/drm/nouveau/nouveau_chan.c | 2 +- drivers/gpu/drm/nouveau/nouveau_dma.h | 3 +++ drivers/gpu/drm/nouveau/nouveau_exec.c | 7 ++++--- drivers/gpu/drm/nouveau/nouveau_exec.h | 10 ++++++++++ include/uapi/drm/nouveau_drm.h | 10 ++++++++++ 6 files changed, 49 insertions(+), 4 deletions(-) (limited to 'include/uapi') diff --git a/drivers/gpu/drm/nouveau/nouveau_abi16.c b/drivers/gpu/drm/nouveau/nouveau_abi16.c index 30afbec9e3b1..2edd7bb13fae 100644 --- a/drivers/gpu/drm/nouveau/nouveau_abi16.c +++ b/drivers/gpu/drm/nouveau/nouveau_abi16.c @@ -31,6 +31,7 @@ #include "nouveau_drv.h" #include "nouveau_dma.h" +#include "nouveau_exec.h" #include "nouveau_gem.h" #include "nouveau_chan.h" #include "nouveau_abi16.h" @@ -183,6 +184,20 @@ nouveau_abi16_fini(struct nouveau_abi16 *abi16) cli->abi16 = NULL; } +static inline int +getparam_dma_ib_max(struct nvif_device *device) +{ + const struct nvif_mclass dmas[] = { + { NV03_CHANNEL_DMA, 0 }, + { NV10_CHANNEL_DMA, 0 }, + { NV17_CHANNEL_DMA, 0 }, + { NV40_CHANNEL_DMA, 0 }, + {} + }; + + return nvif_mclass(&device->object, dmas) < 0 ? NV50_DMA_IB_MAX : 0; +} + int nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS) { @@ -247,6 +262,12 @@ nouveau_abi16_ioctl_getparam(ABI16_IOCTL_ARGS) case NOUVEAU_GETPARAM_GRAPH_UNITS: getparam->value = nvkm_gr_units(gr); break; + case NOUVEAU_GETPARAM_EXEC_PUSH_MAX: { + int ib_max = getparam_dma_ib_max(device); + + getparam->value = nouveau_exec_push_max_from_ib_max(ib_max); + break; + } default: NV_PRINTK(dbg, cli, "unknown parameter %lld\n", getparam->param); return -EINVAL; diff --git a/drivers/gpu/drm/nouveau/nouveau_chan.c b/drivers/gpu/drm/nouveau/nouveau_chan.c index 68e9b08a4803..7c97b2886807 100644 --- a/drivers/gpu/drm/nouveau/nouveau_chan.c +++ b/drivers/gpu/drm/nouveau/nouveau_chan.c @@ -454,7 +454,7 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart) chan->user_get = 0x44; chan->user_get_hi = 0x60; chan->dma.ib_base = 0x10000 / 4; - chan->dma.ib_max = (0x02000 / 8) - 1; + chan->dma.ib_max = NV50_DMA_IB_MAX; chan->dma.ib_put = 0; chan->dma.ib_free = chan->dma.ib_max - chan->dma.ib_put; chan->dma.max = chan->dma.ib_base; diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.h b/drivers/gpu/drm/nouveau/nouveau_dma.h index 1744d95b233e..c52cda82353e 100644 --- a/drivers/gpu/drm/nouveau/nouveau_dma.h +++ b/drivers/gpu/drm/nouveau/nouveau_dma.h @@ -49,6 +49,9 @@ void nv50_dma_push(struct nouveau_channel *, u64 addr, u32 length, /* Maximum push buffer size. */ #define NV50_DMA_PUSH_MAX_LENGTH 0x7fffff +/* Maximum IBs per ring. */ +#define NV50_DMA_IB_MAX ((0x02000 / 8) - 1) + /* Object handles - for stuff that's doesn't use handle == oclass. */ enum { NvDmaFB = 0x80000002, diff --git a/drivers/gpu/drm/nouveau/nouveau_exec.c b/drivers/gpu/drm/nouveau/nouveau_exec.c index 5dda94e1318c..c1837ba95fb5 100644 --- a/drivers/gpu/drm/nouveau/nouveau_exec.c +++ b/drivers/gpu/drm/nouveau/nouveau_exec.c @@ -379,7 +379,7 @@ nouveau_exec_ioctl_exec(struct drm_device *dev, struct nouveau_channel *chan = NULL; struct nouveau_exec_job_args args = {}; struct drm_nouveau_exec *req = data; - int ret = 0; + int push_max, ret = 0; if (unlikely(!abi16)) return -ENOMEM; @@ -404,9 +404,10 @@ nouveau_exec_ioctl_exec(struct drm_device *dev, if (!chan->dma.ib_max) return nouveau_abi16_put(abi16, -ENOSYS); - if (unlikely(req->push_count > NOUVEAU_GEM_MAX_PUSH)) { + push_max = nouveau_exec_push_max_from_ib_max(chan->dma.ib_max); + if (unlikely(req->push_count > push_max)) { NV_PRINTK(err, cli, "pushbuf push count exceeds limit: %d max %d\n", - req->push_count, NOUVEAU_GEM_MAX_PUSH); + req->push_count, push_max); return nouveau_abi16_put(abi16, -EINVAL); } diff --git a/drivers/gpu/drm/nouveau/nouveau_exec.h b/drivers/gpu/drm/nouveau/nouveau_exec.h index 778cacd90f65..5488d337bcc0 100644 --- a/drivers/gpu/drm/nouveau/nouveau_exec.h +++ b/drivers/gpu/drm/nouveau/nouveau_exec.h @@ -51,4 +51,14 @@ int nouveau_exec_job_init(struct nouveau_exec_job **job, int nouveau_exec_ioctl_exec(struct drm_device *dev, void *data, struct drm_file *file_priv); +static inline unsigned int +nouveau_exec_push_max_from_ib_max(int ib_max) +{ + /* Limit the number of IBs per job to half the size of the ring in order + * to avoid the ring running dry between submissions and preserve one + * more slot for the job's HW fence. + */ + return ib_max > 1 ? ib_max / 2 - 1 : 0; +} + #endif diff --git a/include/uapi/drm/nouveau_drm.h b/include/uapi/drm/nouveau_drm.h index 8d7402c13e56..eaf9f248619f 100644 --- a/include/uapi/drm/nouveau_drm.h +++ b/include/uapi/drm/nouveau_drm.h @@ -44,6 +44,16 @@ extern "C" { #define NOUVEAU_GETPARAM_PTIMER_TIME 14 #define NOUVEAU_GETPARAM_HAS_BO_USAGE 15 #define NOUVEAU_GETPARAM_HAS_PAGEFLIP 16 + +/** + * @NOUVEAU_GETPARAM_EXEC_PUSH_MAX + * + * Query the maximum amount of IBs that can be pushed through a single + * &drm_nouveau_exec structure and hence a single &DRM_IOCTL_NOUVEAU_EXEC + * ioctl(). + */ +#define NOUVEAU_GETPARAM_EXEC_PUSH_MAX 17 + struct drm_nouveau_getparam { __u64 param; __u64 value; -- cgit v1.2.3