August 2024 - Linux-stable-mirror

[PATCH net] net: mana: Fix RX buf alloc_size alignment and atomic op panic

by Haiyang Zhang

The MANA driver's RX buffer alloc_size is passed into napi_build_skb() to create SKB. skb_shinfo(skb) is located at the end of skb, and its alignment is affected by the alloc_size passed into napi_build_skb(). The size needs to be aligned properly for better performance and atomic operations. Otherwise, on ARM64 CPU, for certain MTU settings like 4000, atomic operations may panic on the skb_shinfo(skb)->dataref due to alignment fault. To fix this bug, add proper alignment to the alloc_size calculation. Sample panic info: [ 253.298819] Unable to handle kernel paging request at virtual address ffff000129ba5cce [ 253.300900] Mem abort info: [ 253.301760] ESR = 0x0000000096000021 [ 253.302825] EC = 0x25: DABT (current EL), IL = 32 bits [ 253.304268] SET = 0, FnV = 0 [ 253.305172] EA = 0, S1PTW = 0 [ 253.306103] FSC = 0x21: alignment fault Call trace: __skb_clone+0xfc/0x198 skb_clone+0x78/0xe0 raw6_local_deliver+0xfc/0x228 ip6_protocol_deliver_rcu+0x80/0x500 ip6_input_finish+0x48/0x80 ip6_input+0x48/0xc0 ip6_sublist_rcv_finish+0x50/0x78 ip6_sublist_rcv+0x1cc/0x2b8 ipv6_list_rcv+0x100/0x150 __netif_receive_skb_list_core+0x180/0x220 netif_receive_skb_list_internal+0x198/0x2a8 __napi_poll+0x138/0x250 net_rx_action+0x148/0x330 handle_softirqs+0x12c/0x3a0 Cc: stable(a)vger.kernel.org Fixes: 80f6215b450e ("net: mana: Add support for jumbo frame") Signed-off-by: Haiyang Zhang <haiyangz(a)microsoft.com> --- drivers/net/ethernet/microsoft/mana/mana_en.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/microsoft/mana/mana_en.c b/drivers/net/ethernet/microsoft/mana/mana_en.c index d2f07e179e86..ae717d06e66f 100644 --- a/drivers/net/ethernet/microsoft/mana/mana_en.c +++ b/drivers/net/ethernet/microsoft/mana/mana_en.c @@ -599,7 +599,11 @@ static void mana_get_rxbuf_cfg(int mtu, u32 *datasize, u32 *alloc_size, else *headroom = XDP_PACKET_HEADROOM; - *alloc_size = mtu + MANA_RXBUF_PAD + *headroom; + *alloc_size = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD + *headroom); + + /* Using page pool in this case, so alloc_size is PAGE_SIZE */ + if (*alloc_size < PAGE_SIZE) + *alloc_size = PAGE_SIZE; *datasize = mtu + ETH_HLEN; } -- 2.34.1

8 months, 2 weeks

3
2
0 0

FAILED: patch "[PATCH] kcov: properly check for softirq context" failed to apply to 5.10-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.10-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y git checkout FETCH_HEAD git cherry-pick -x 7d4df2dad312f270d62fecb0e5c8b086c6d7dcfc # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081226-usual-undocked-fd62@gregkh' --subject-prefix 'PATCH 5.10.y' HEAD^.. Possible dependencies: 7d4df2dad312 ("kcov: properly check for softirq context") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 7d4df2dad312f270d62fecb0e5c8b086c6d7dcfc Mon Sep 17 00:00:00 2001 From: Andrey Konovalov <andreyknvl(a)gmail.com> Date: Mon, 29 Jul 2024 04:21:58 +0200 Subject: [PATCH] kcov: properly check for softirq context When collecting coverage from softirqs, KCOV uses in_serving_softirq() to check whether the code is running in the softirq context. Unfortunately, in_serving_softirq() is > 0 even when the code is running in the hardirq or NMI context for hardirqs and NMIs that happened during a softirq. As a result, if a softirq handler contains a remote coverage collection section and a hardirq with another remote coverage collection section happens during handling the softirq, KCOV incorrectly detects a nested softirq coverate collection section and prints a WARNING, as reported by syzbot. This issue was exposed by commit a7f3813e589f ("usb: gadget: dummy_hcd: Switch to hrtimer transfer scheduler"), which switched dummy_hcd to using hrtimer and made the timer's callback be executed in the hardirq context. Change the related checks in KCOV to account for this behavior of in_serving_softirq() and make KCOV ignore remote coverage collection sections in the hardirq and NMI contexts. This prevents the WARNING printed by syzbot but does not fix the inability of KCOV to collect coverage from the __usb_hcd_giveback_urb when dummy_hcd is in use (caused by a7f3813e589f); a separate patch is required for that. Link: https://lkml.kernel.org/r/20240729022158.92059-1-andrey.konovalov@linux.dev Fixes: 5ff3b30ab57d ("kcov: collect coverage from interrupts") Signed-off-by: Andrey Konovalov <andreyknvl(a)gmail.com> Reported-by: syzbot+2388cdaeb6b10f0c13ac(a)syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=2388cdaeb6b10f0c13ac Acked-by: Marco Elver <elver(a)google.com> Cc: Alan Stern <stern(a)rowland.harvard.edu> Cc: Aleksandr Nogikh <nogikh(a)google.com> Cc: Alexander Potapenko <glider(a)google.com> Cc: Dmitry Vyukov <dvyukov(a)google.com> Cc: Greg Kroah-Hartman <gregkh(a)linuxfoundation.org> Cc: Marcello Sylvester Bauer <sylv(a)sylv.io> Cc: <stable(a)vger.kernel.org> Signed-off-by: Andrew Morton <akpm(a)linux-foundation.org> diff --git a/kernel/kcov.c b/kernel/kcov.c index f0a69d402066..274b6b7c718d 100644 --- a/kernel/kcov.c +++ b/kernel/kcov.c @@ -161,6 +161,15 @@ static void kcov_remote_area_put(struct kcov_remote_area *area, kmsan_unpoison_memory(&area->list, sizeof(area->list)); } +/* + * Unlike in_serving_softirq(), this function returns false when called during + * a hardirq or an NMI that happened in the softirq context. + */ +static inline bool in_softirq_really(void) +{ + return in_serving_softirq() && !in_hardirq() && !in_nmi(); +} + static notrace bool check_kcov_mode(enum kcov_mode needed_mode, struct task_struct *t) { unsigned int mode; @@ -170,7 +179,7 @@ static notrace bool check_kcov_mode(enum kcov_mode needed_mode, struct task_stru * so we ignore code executed in interrupts, unless we are in a remote * coverage collection section in a softirq. */ - if (!in_task() && !(in_serving_softirq() && t->kcov_softirq)) + if (!in_task() && !(in_softirq_really() && t->kcov_softirq)) return false; mode = READ_ONCE(t->kcov_mode); /* @@ -849,7 +858,7 @@ void kcov_remote_start(u64 handle) if (WARN_ON(!kcov_check_handle(handle, true, true, true))) return; - if (!in_task() && !in_serving_softirq()) + if (!in_task() && !in_softirq_really()) return; local_lock_irqsave(&kcov_percpu_data.lock, flags); @@ -991,7 +1000,7 @@ void kcov_remote_stop(void) int sequence; unsigned long flags; - if (!in_task() && !in_serving_softirq()) + if (!in_task() && !in_softirq_really()) return; local_lock_irqsave(&kcov_percpu_data.lock, flags);

8 months, 2 weeks

1
0
0 0

[PATCH] mmc: meson-gx: fix wrong conversion of __bf_shf to __ffs

by Christian Marangi

Commit 795c633f6093 ("mmc: meson-gx: fix __ffsdi2 undefined on arm32") changed __bf_shf to __ffs to fix a compile error on 32bit arch that have problems with __ffsdi2. This comes from the fact that __bf_shf use __builtin_ffsll and on 32bit __ffsdi2 is missing. Problem is that __bf_shf is defined as #define __bf_shf(x) (__builtin_ffsll(x) - 1) but the patch doesn't account for the - 1. Fix this by using the __builtin_ffs and add the - 1 to reflect the original implementation. The commit also converted other entry of __bf_shf in the code but those got dropped in later patches. Fixes: 795c633f6093 ("mmc: meson-gx: fix __ffsdi2 undefined on arm32") Signed-off-by: Christian Marangi <ansuelsmth(a)gmail.com> Cc: stable(a)vger.kernel.org # see patch description, needs adjustements for < 5.2 --- drivers/mmc/host/meson-gx-mmc.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c index c7c067b9415a..8f64083a08fa 100644 --- a/drivers/mmc/host/meson-gx-mmc.c +++ b/drivers/mmc/host/meson-gx-mmc.c @@ -464,7 +464,7 @@ static int meson_mmc_clk_init(struct meson_host *host) init.num_parents = MUX_CLK_NUM_PARENTS; mux->reg = host->regs + SD_EMMC_CLOCK; - mux->shift = __ffs(CLK_SRC_MASK); + mux->shift = __builtin_ffs(CLK_SRC_MASK) - 1; mux->mask = CLK_SRC_MASK >> mux->shift; mux->hw.init = &init; @@ -486,7 +486,7 @@ static int meson_mmc_clk_init(struct meson_host *host) init.num_parents = 1; div->reg = host->regs + SD_EMMC_CLOCK; - div->shift = __ffs(CLK_DIV_MASK); + div->shift = __builtin_ffs(CLK_DIV_MASK) - 1; div->width = __builtin_popcountl(CLK_DIV_MASK); div->hw.init = &init; div->flags = CLK_DIVIDER_ONE_BASED; -- 2.45.2

8 months, 2 weeks

2
1
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 4.19-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 4.19-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-4.19.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081241-sulfate-protegee-673a@gregkh' --subject-prefix 'PATCH 4.19.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") 15fd09a05a66 ("drm/amdgpu: add reset register dump trace on GPU") 21a6732f4648 ("drm/amdgpu: don't skip runtime pm get on A+A config") 8c505bdc9c8b ("drm/amdgpu: rework dma_resv handling v3") d3fae3b3daac ("dma-buf: drop the _rcu postfix on function names v3") fb5ce730f214 ("dma-buf: rename and cleanup dma_resv_get_list v2") 6edbd6abb783 ("dma-buf: rename and cleanup dma_resv_get_excl v3") 0c6b522abc2a ("dma-buf: cleanup dma-resv shared fence debugging a bit v2") 068d9d754bc1 ("dma-buf: add SPDX header and fix style in dma-resv.c") 680753dd9d7d ("dma-buf: fix inconsistent debug print v2") 71df0368e9b6 ("drm/amdgpu: Implement mmap as GEM object function") 304ba5dca49a ("Merge drm/drm-next into drm-misc-next") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 5.4-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.4-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.4.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081240-achiness-vitalize-dd50@gregkh' --subject-prefix 'PATCH 5.4.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") 15fd09a05a66 ("drm/amdgpu: add reset register dump trace on GPU") 21a6732f4648 ("drm/amdgpu: don't skip runtime pm get on A+A config") 8c505bdc9c8b ("drm/amdgpu: rework dma_resv handling v3") d3fae3b3daac ("dma-buf: drop the _rcu postfix on function names v3") fb5ce730f214 ("dma-buf: rename and cleanup dma_resv_get_list v2") 6edbd6abb783 ("dma-buf: rename and cleanup dma_resv_get_excl v3") 0c6b522abc2a ("dma-buf: cleanup dma-resv shared fence debugging a bit v2") 068d9d754bc1 ("dma-buf: add SPDX header and fix style in dma-resv.c") 680753dd9d7d ("dma-buf: fix inconsistent debug print v2") 71df0368e9b6 ("drm/amdgpu: Implement mmap as GEM object function") 304ba5dca49a ("Merge drm/drm-next into drm-misc-next") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 5.10-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.10-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.10.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081239-hardcover-dipping-96fb@gregkh' --subject-prefix 'PATCH 5.10.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") 15fd09a05a66 ("drm/amdgpu: add reset register dump trace on GPU") 21a6732f4648 ("drm/amdgpu: don't skip runtime pm get on A+A config") 8c505bdc9c8b ("drm/amdgpu: rework dma_resv handling v3") d3fae3b3daac ("dma-buf: drop the _rcu postfix on function names v3") fb5ce730f214 ("dma-buf: rename and cleanup dma_resv_get_list v2") 6edbd6abb783 ("dma-buf: rename and cleanup dma_resv_get_excl v3") 0c6b522abc2a ("dma-buf: cleanup dma-resv shared fence debugging a bit v2") 068d9d754bc1 ("dma-buf: add SPDX header and fix style in dma-resv.c") 680753dd9d7d ("dma-buf: fix inconsistent debug print v2") 71df0368e9b6 ("drm/amdgpu: Implement mmap as GEM object function") 304ba5dca49a ("Merge drm/drm-next into drm-misc-next") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 5.15-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 5.15-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-5.15.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081238-circle-chug-01cb@gregkh' --subject-prefix 'PATCH 5.15.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") 15fd09a05a66 ("drm/amdgpu: add reset register dump trace on GPU") 21a6732f4648 ("drm/amdgpu: don't skip runtime pm get on A+A config") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 6.1-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.1-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.1.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081237-geek-nag-bcd6@gregkh' --subject-prefix 'PATCH 6.1.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 6.6-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.6-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.6.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081236-untouched-bolster-5053@gregkh' --subject-prefix 'PATCH 6.6.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

FAILED: patch "[PATCH] drm/amdgpu: revert "take runtime pm reference when we attach" failed to apply to 6.10-stable tree

by gregkh＠linuxfoundation.org

The patch below does not apply to the 6.10-stable tree. If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <stable(a)vger.kernel.org>. To reproduce the conflict and resubmit, you may use the following commands: git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.10.y git checkout FETCH_HEAD git cherry-pick -x 030631e97b209481edbac38000d2a60fd340f6b1 # <resolve conflicts, build, test, etc.> git commit -s git send-email --to '<stable(a)vger.kernel.org>' --in-reply-to '2024081235-unused-quail-54fb@gregkh' --subject-prefix 'PATCH 6.10.y' HEAD^.. Possible dependencies: 030631e97b20 ("drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2") thanks, greg k-h ------------------ original commit in Linus's tree ------------------ From 030631e97b209481edbac38000d2a60fd340f6b1 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig(a)amd.com> Date: Wed, 5 Jun 2024 13:27:20 +0200 Subject: [PATCH] drm/amdgpu: revert "take runtime pm reference when we attach a buffer" v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This reverts commit b8c415e3bf98 ("drm/amdgpu: take runtime pm reference when we attach a buffer") and commit 425285d39afd ("drm/amdgpu: add amdgpu runpm usage trace for separate funcs"). Taking a runtime pm reference for DMA-buf is actually completely unnecessary and even dangerous. The problem is that calling pm_runtime_get_sync() from the DMA-buf callbacks is illegal because we have the reservation locked here which is also taken during resume. So this would deadlock. When the buffer is in GTT it is still accessible even when the GPU is powered down and when it is in VRAM the buffer gets migrated to GTT before powering down. The only use case which would make it mandatory to keep the runtime pm reference would be if we pin the buffer into VRAM, and that's not something we currently do. v2: improve the commit message Signed-off-by: Christian König <christian.koenig(a)amd.com> Reviewed-by: Alex Deucher <alexander.deucher(a)amd.com> Signed-off-by: Alex Deucher <alexander.deucher(a)amd.com> CC: stable(a)vger.kernel.org diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c index 0b3b10d21952..8e81a83d37d8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c @@ -41,8 +41,6 @@ #include <linux/dma-buf.h> #include <linux/dma-fence-array.h> #include <linux/pci-p2pdma.h> -#include <linux/pm_runtime.h> -#include "amdgpu_trace.h" /** * amdgpu_dma_buf_attach - &dma_buf_ops.attach implementation @@ -58,42 +56,11 @@ static int amdgpu_dma_buf_attach(struct dma_buf *dmabuf, struct drm_gem_object *obj = dmabuf->priv; struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - int r; if (pci_p2pdma_distance(adev->pdev, attach->dev, false) < 0) attach->peer2peer = false; - r = pm_runtime_get_sync(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); - if (r < 0) - goto out; - return 0; - -out: - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); - return r; -} - -/** - * amdgpu_dma_buf_detach - &dma_buf_ops.detach implementation - * - * @dmabuf: DMA-buf where we remove the attachment from - * @attach: the attachment to remove - * - * Called when an attachment is removed from the DMA-buf. - */ -static void amdgpu_dma_buf_detach(struct dma_buf *dmabuf, - struct dma_buf_attachment *attach) -{ - struct drm_gem_object *obj = dmabuf->priv; - struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); - struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); - - pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); - pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } /** @@ -266,7 +233,6 @@ static int amdgpu_dma_buf_begin_cpu_access(struct dma_buf *dma_buf, const struct dma_buf_ops amdgpu_dmabuf_ops = { .attach = amdgpu_dma_buf_attach, - .detach = amdgpu_dma_buf_detach, .pin = amdgpu_dma_buf_pin, .unpin = amdgpu_dma_buf_unpin, .map_dma_buf = amdgpu_dma_buf_map, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 3f492277d7d3..2f24a6aa13bf 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -181,7 +181,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f, struct amd amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, seq, flags | AMDGPU_FENCE_FLAG_INT); pm_runtime_get_noresume(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(1, __func__); ptr = &ring->fence_drv.fences[seq & ring->fence_drv.num_fences_mask]; if (unlikely(rcu_dereference_protected(*ptr, 1))) { struct dma_fence *old; @@ -309,7 +308,6 @@ bool amdgpu_fence_process(struct amdgpu_ring *ring) dma_fence_put(fence); pm_runtime_mark_last_busy(adev_to_drm(adev)->dev); pm_runtime_put_autosuspend(adev_to_drm(adev)->dev); - trace_amdgpu_runpm_reference_dumps(0, __func__); } while (last_seq != seq); return true; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index f539b1d00234..2fd1bfb35916 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -554,21 +554,6 @@ TRACE_EVENT(amdgpu_reset_reg_dumps, __entry->value) ); -TRACE_EVENT(amdgpu_runpm_reference_dumps, - TP_PROTO(uint32_t index, const char *func), - TP_ARGS(index, func), - TP_STRUCT__entry( - __field(uint32_t, index) - __string(func, func) - ), - TP_fast_assign( - __entry->index = index; - __assign_str(func, func); - ), - TP_printk("amdgpu runpm reference dump 0x%x: 0x%s\n", - __entry->index, - __get_str(func)) -); #undef AMDGPU_JOB_GET_TIMELINE_NAME #endif

8 months, 2 weeks

1
0
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror August 2024