- Linux-stable-mirror - lists.linaro.org

[PATCH v2] ext4: mballoc: Fix spectre gadget in ext4_mb_regular_allocator

by Jeremy Cline

'ac->ac_g_ex.fe_len' is a user-controlled value which is used in the derivation of 'ac->ac_2order'. 'ac->ac_2order', in turn, is used to index arrays which makes it a potential spectre gadget. Fix this by sanitizing the value assigned to 'ac->ac2_order'. This covers the following accesses found with the help of smatch: * fs/ext4/mballoc.c:1896 ext4_mb_simple_scan_group() warn: potential spectre issue 'grp->bb_counters' [w] (local cap) * fs/ext4/mballoc.c:445 mb_find_buddy() warn: potential spectre issue 'EXT4_SB(e4b->bd_sb)->s_mb_offsets' [r] (local cap) * fs/ext4/mballoc.c:446 mb_find_buddy() warn: potential spectre issue 'EXT4_SB(e4b->bd_sb)->s_mb_maxs' [r] (local cap) Cc: Josh Poimboeuf <jpoimboe(a)redhat.com> Cc: stable(a)vger.kernel.org Suggested-by: Josh Poimboeuf <jpoimboe(a)redhat.com> Signed-off-by: Jeremy Cline <jcline(a)redhat.com> --- I broke this out of the "ext4: fix spectre v1 gadgets" patch set since the other patches in that series could, as Josh noted, be replaced with one fix in do_quotactl. I'll send that fix to the disk quota folks separately. Changes from v1: - Sanitize ac_2order on assignment, rather than down the call chain in ext4_mb_simple_scan_group. fs/ext4/mballoc.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c index f7ab34088162..8b24d3d42cb3 100644 --- a/fs/ext4/mballoc.c +++ b/fs/ext4/mballoc.c @@ -14,6 +14,7 @@ #include <linux/log2.h> #include <linux/module.h> #include <linux/slab.h> +#include <linux/nospec.h> #include <linux/backing-dev.h> #include <trace/events/ext4.h> @@ -2140,7 +2141,8 @@ ext4_mb_regular_allocator(struct ext4_allocation_context *ac) * This should tell if fe_len is exactly power of 2 */ if ((ac->ac_g_ex.fe_len & (~(1 << (i - 1)))) == 0) - ac->ac_2order = i - 1; + ac->ac_2order = array_index_nospec(i - 1, + sb->s_blocksize_bits + 2); } /* if stream allocation is enabled, use global goal */ -- 2.17.1

7 years, 4 months

3
5
0 0

[PATCH 1/3] Drivers: hv: vmbus: Reset the channel callback in vmbus_onoffer_rescind()

by kys＠linuxonhyperv.com

From: Dexuan Cui <decui(a)microsoft.com> Before setting channel->rescind in vmbus_rescind_cleanup(), we should make sure the channel callback won't run any more, otherwise a high-level driver like pci_hyperv, which may be infinitely waiting for the host VSP's response and notices the channel has been rescinded, can't safely give up: e.g., in hv_pci_protocol_negotiation() -> wait_for_response(), it's unsafe to exit from wait_for_response() and proceed with the on-stack variable "comp_pkt" popped. The issue was originally spotted by Michael Kelley <mikelley(a)microsoft.com>. In vmbus_close_internal(), the patch also minimizes the range protected by disabling/enabling channel->callback_event: we don't really need that for the whole function. Signed-off-by: Dexuan Cui <decui(a)microsoft.com> Reviewed-by: Michael Kelley <mikelley(a)microsoft.com> Cc: stable(a)vger.kernel.org Cc: K. Y. Srinivasan <kys(a)microsoft.com> Cc: Stephen Hemminger <sthemmin(a)microsoft.com> Cc: Michael Kelley <mikelley(a)microsoft.com> Signed-off-by: K. Y. Srinivasan <kys(a)microsoft.com> --- drivers/hv/channel.c | 40 +++++++++++++++++++++++---------------- drivers/hv/channel_mgmt.c | 6 ++++++ include/linux/hyperv.h | 2 ++ 3 files changed, 32 insertions(+), 16 deletions(-) diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c index ba0a092ae085..c3949220b770 100644 --- a/drivers/hv/channel.c +++ b/drivers/hv/channel.c @@ -558,11 +558,8 @@ static void reset_channel_cb(void *arg) channel->onchannel_callback = NULL; } -static int vmbus_close_internal(struct vmbus_channel *channel) +void vmbus_reset_channel_cb(struct vmbus_channel *channel) { - struct vmbus_channel_close_channel *msg; - int ret; - /* * vmbus_on_event(), running in the per-channel tasklet, can race * with vmbus_close_internal() in the case of SMP guest, e.g., when @@ -572,6 +569,29 @@ static int vmbus_close_internal(struct vmbus_channel *channel) */ tasklet_disable(&channel->callback_event); + channel->sc_creation_callback = NULL; + + /* Stop the callback asap */ + if (channel->target_cpu != get_cpu()) { + put_cpu(); + smp_call_function_single(channel->target_cpu, reset_channel_cb, + channel, true); + } else { + reset_channel_cb(channel); + put_cpu(); + } + + /* Re-enable tasklet for use on re-open */ + tasklet_enable(&channel->callback_event); +} + +static int vmbus_close_internal(struct vmbus_channel *channel) +{ + struct vmbus_channel_close_channel *msg; + int ret; + + vmbus_reset_channel_cb(channel); + /* * In case a device driver's probe() fails (e.g., * util_probe() -> vmbus_open() returns -ENOMEM) and the device is @@ -585,16 +605,6 @@ static int vmbus_close_internal(struct vmbus_channel *channel) } channel->state = CHANNEL_OPEN_STATE; - channel->sc_creation_callback = NULL; - /* Stop callback and cancel the timer asap */ - if (channel->target_cpu != get_cpu()) { - put_cpu(); - smp_call_function_single(channel->target_cpu, reset_channel_cb, - channel, true); - } else { - reset_channel_cb(channel); - put_cpu(); - } /* Send a closing message */ @@ -639,8 +649,6 @@ static int vmbus_close_internal(struct vmbus_channel *channel) get_order(channel->ringbuffer_pagecount * PAGE_SIZE)); out: - /* re-enable tasklet for use on re-open */ - tasklet_enable(&channel->callback_event); return ret; } diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c index f3b551a50653..0f0e091c117c 100644 --- a/drivers/hv/channel_mgmt.c +++ b/drivers/hv/channel_mgmt.c @@ -892,6 +892,12 @@ static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr) return; } + /* + * Before setting channel->rescind in vmbus_rescind_cleanup(), we + * should make sure the channel callback is not running any more. + */ + vmbus_reset_channel_cb(channel); + /* * Now wait for offer handling to complete. */ diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h index 2330f08062c7..efda23cf32c7 100644 --- a/include/linux/hyperv.h +++ b/include/linux/hyperv.h @@ -1061,6 +1061,8 @@ extern int vmbus_establish_gpadl(struct vmbus_channel *channel, extern int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle); +void vmbus_reset_channel_cb(struct vmbus_channel *channel); + extern int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer, u32 bufferlen, -- 2.17.1

7 years, 4 months

1
0
0 0

[PATCH] sched/deadline: Update rq_clock of later_rq when pushing a task

by Daniel Bristot de Oliveira

Daniel Casini got this warn while running a DL task here at RetisLab: [ 461.137582] ------------[ cut here ]------------ [ 461.137583] rq->clock_update_flags < RQCF_ACT_SKIP [ 461.137599] WARNING: CPU: 4 PID: 2354 at kernel/sched/sched.h:967 assert_clock_updated.isra.32.part.33+0x17/0x20 [a ton of modules] [ 461.137646] CPU: 4 PID: 2354 Comm: label_image Not tainted 4.18.0-rc4+ #3 [ 461.137647] Hardware name: ASUS All Series/Z87-K, BIOS 0801 09/02/2013 [ 461.137649] RIP: 0010:assert_clock_updated.isra.32.part.33+0x17/0x20 [ 461.137649] Code: ff 48 89 83 08 09 00 00 eb c6 66 0f 1f 84 00 00 00 00 00 55 48 c7 c7 98 7a 6c a5 c6 05 bc 0d 54 01 01 48 89 e5 e8 a9 84 fb ff <0f> 0b 5d c3 0f 1f 44 00 00 0f 1f 44 00 00 83 7e 60 01 74 0a 48 3b [ 461.137673] RSP: 0018:ffffa77e08cafc68 EFLAGS: 00010082 [ 461.137674] RAX: 0000000000000000 RBX: ffff8b3fc1702d80 RCX: 0000000000000006 [ 461.137674] RDX: 0000000000000007 RSI: 0000000000000096 RDI: ffff8b3fded164b0 [ 461.137675] RBP: ffffa77e08cafc68 R08: 0000000000000026 R09: 0000000000000339 [ 461.137676] R10: ffff8b3fd060d410 R11: 0000000000000026 R12: ffffffffa4e14e20 [ 461.137677] R13: ffff8b3fdec22940 R14: ffff8b3fc1702da0 R15: ffff8b3fdec22940 [ 461.137678] FS: 00007efe43ee5700(0000) GS:ffff8b3fded00000(0000) knlGS:0000000000000000 [ 461.137679] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 461.137680] CR2: 00007efe30000010 CR3: 0000000301744003 CR4: 00000000001606e0 [ 461.137680] Call Trace: [ 461.137684] push_dl_task.part.46+0x3bc/0x460 [ 461.137686] task_woken_dl+0x60/0x80 [ 461.137689] ttwu_do_wakeup+0x4f/0x150 [ 461.137690] ttwu_do_activate+0x77/0x80 [ 461.137692] try_to_wake_up+0x1d6/0x4c0 [ 461.137693] wake_up_q+0x32/0x70 [ 461.137696] do_futex+0x7e7/0xb50 [ 461.137698] __x64_sys_futex+0x8b/0x180 [ 461.137701] do_syscall_64+0x5a/0x110 [ 461.137703] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 461.137705] RIP: 0033:0x7efe4918ca26 [ 461.137705] Code: 00 00 00 74 17 49 8b 48 20 44 8b 59 10 41 83 e3 30 41 83 fb 20 74 1e be 85 00 00 00 41 ba 01 00 00 00 41 b9 01 00 00 04 0f 05 <48> 3d 01 f0 ff ff 73 1f 31 c0 c3 be 8c 00 00 00 49 89 c8 4d 31 d2 [ 461.137738] RSP: 002b:00007efe43ee4928 EFLAGS: 00000283 ORIG_RAX: 00000000000000ca [ 461.137739] RAX: ffffffffffffffda RBX: 0000000005094df0 RCX: 00007efe4918ca26 [ 461.137740] RDX: 0000000000000001 RSI: 0000000000000085 RDI: 0000000005094e24 [ 461.137741] RBP: 00007efe43ee49c0 R08: 0000000005094e20 R09: 0000000004000001 [ 461.137741] R10: 0000000000000001 R11: 0000000000000283 R12: 0000000000000000 [ 461.137742] R13: 0000000005094df8 R14: 0000000000000001 R15: 0000000000448a10 [ 461.137743] ---[ end trace 187df4cad2bf7649 ]--- This warning was caused in the push_dl_task(), because the rq_clock of the later_rq was used by cpufreq_update_util() inside __add_running_bw(), before its update at activate_task(). The fix then is to update the clock before calling add_running_bw. To avoid double rq_clock_update(), we set ENQUEUE_NOCLOCK flag to activate_task(). Reported-by: Daniel Casini <daniel.casini(a)santannapisa.it> Signed-off-by: Daniel Bristot de Oliveira <bristot(a)redhat.com> Cc: Juri Lelli <juri.lelli(a)arm.com> Cc: Luca Abeni <luca.abeni(a)santannapisa.it> Cc: Tommaso Cucinotta <tommaso.cucinotta(a)sssup.it> Cc: Steven Rostedt <rostedt(a)goodmis.org> Cc: Peter Zijlstra <peterz(a)infradead.org> Cc: Ingo Molnar <mingo(a)kernel.org> Cc: linux-kernel(a)vger.kernel.org Cc: <stable(a)vger.kernel.org> # 4.16+ Fixes: e0367b12674b sched/deadline: Move CPU frequency selection triggering points --- kernel/sched/deadline.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c index fbfc3f1d368a..e733c15b7695 100644 --- a/kernel/sched/deadline.c +++ b/kernel/sched/deadline.c @@ -2090,8 +2090,16 @@ static int push_dl_task(struct rq *rq) sub_rq_bw(&next_task->dl, &rq->dl); set_task_cpu(next_task, later_rq->cpu); add_rq_bw(&next_task->dl, &later_rq->dl); + + /* + * Update the later_rq clock here, because the clock is used + * by the cpufreq_update_util() inside __add_running_bw(). + * Then, set ENQUEUE_NOCLOCK flag to avoid updating the rq_clock + * again in the activate_task()->enqueue_task(). + */ + update_rq_clock(later_rq); add_running_bw(&next_task->dl, &later_rq->dl); - activate_task(later_rq, next_task, 0); + activate_task(later_rq, next_task, ENQUEUE_NOCLOCK); ret = 1; resched_curr(later_rq); -- 2.17.1

7 years, 4 months

4
4
0 0

requests for inclusion in 4.14.y

by Eduardo Valentin

Hey Greg, Can you please include the following fixes on stable 4.14.y? https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?… All of the above are straight forward cherry-picks. -- All the best, Eduardo Valentin

7 years, 4 months

4
4
0 0

Re: FAILED: patch "[PATCH] nvmet-fc: fix target sgl list on large transfers" failed to apply to 4.14-stable tree

by James Smart

On 7/30/2018 12:56 AM, gregkh(a)linuxfoundation.org wrote: > > The patch below does not apply to the 4.14-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>. > > thanks, > > greg k-h > > (snip) Here's the patch revised for stable 4.14.y -- james ------------------ From: James Smart <jsmart2021(a)gmail.com> commit d082dc1562a2ff0947b214796f12faaa87e816a9 upstream. The existing code to carve up the sg list expected an sg element-per-page which can be very incorrect with iommu's remapping multiple memory pages to fewer bus addresses. To hit this error required a large io payload (greater than 256k) and a system that maps on a per-page basis. It's possible that large ios could get by fine if the system condensed the sgl list into the first 64 elements. This patch corrects the sg list handling by specifically walking the sg list element by element and attempting to divide the transfer up on a per-sg element boundary. While doing so, it still tries to keep sequences under 256k, but will exceed that rule if a single sg element is larger than 256k. Fixes: 48fa362b6c3f ("nvmet-fc: simplify sg list handling") Cc: <stable(a)vger.kernel.org> # 4.14 Signed-off-by: James Smart <james.smart(a)broadcom.com> Signed-off-by: Christoph Hellwig <hch(a)lst.de> --- stable 4.14.y patch adjusted for deltas made by upstream commit 5e62d5c993e6889cd314d5b5de6b670152109a0e that are not in the stable tree. --- drivers/nvme/target/fc.c | 44 +++++++++++++++++++++++++++++++++++--------- 1 file changed, 35 insertions(+), 9 deletions(-) diff --git a/drivers/nvme/target/fc.c b/drivers/nvme/target/fc.c index 8e21211b904b..b7a5d1065378 100644 --- a/drivers/nvme/target/fc.c +++ b/drivers/nvme/target/fc.c @@ -58,8 +58,8 @@ struct nvmet_fc_ls_iod { struct work_struct work; } __aligned(sizeof(unsigned long long)); +/* desired maximum for a single sequence - if sg list allows it */ #define NVMET_FC_MAX_SEQ_LENGTH (256 * 1024) -#define NVMET_FC_MAX_XFR_SGENTS (NVMET_FC_MAX_SEQ_LENGTH / PAGE_SIZE) enum nvmet_fcp_datadir { NVMET_FCP_NODATA, @@ -74,6 +74,7 @@ struct nvmet_fc_fcp_iod { struct nvme_fc_cmd_iu cmdiubuf; struct nvme_fc_ersp_iu rspiubuf; dma_addr_t rspdma; + struct scatterlist *next_sg; struct scatterlist *data_sg; int data_sg_cnt; u32 total_length; @@ -1000,8 +1001,7 @@ nvmet_fc_register_targetport(struct nvmet_fc_port_info *pinfo, INIT_LIST_HEAD(&newrec->assoc_list); kref_init(&newrec->ref); ida_init(&newrec->assoc_cnt); - newrec->max_sg_cnt = min_t(u32, NVMET_FC_MAX_XFR_SGENTS, - template->max_sgl_segments); + newrec->max_sg_cnt = template->max_sgl_segments; ret = nvmet_fc_alloc_ls_iodlist(newrec); if (ret) { @@ -1717,6 +1717,7 @@ nvmet_fc_alloc_tgt_pgs(struct nvmet_fc_fcp_iod *fod) ((fod->io_dir == NVMET_FCP_WRITE) ? DMA_FROM_DEVICE : DMA_TO_DEVICE)); /* note: write from initiator perspective */ + fod->next_sg = fod->data_sg; return 0; @@ -1874,24 +1875,49 @@ nvmet_fc_transfer_fcp_data(struct nvmet_fc_tgtport *tgtport, struct nvmet_fc_fcp_iod *fod, u8 op) { struct nvmefc_tgt_fcp_req *fcpreq = fod->fcpreq; + struct scatterlist *sg = fod->next_sg; unsigned long flags; - u32 tlen; + u32 remaininglen = fod->total_length - fod->offset; + u32 tlen = 0; int ret; fcpreq->op = op; fcpreq->offset = fod->offset; fcpreq->timeout = NVME_FC_TGTOP_TIMEOUT_SEC; - tlen = min_t(u32, tgtport->max_sg_cnt * PAGE_SIZE, - (fod->total_length - fod->offset)); + /* + * for next sequence: + * break at a sg element boundary + * attempt to keep sequence length capped at + * NVMET_FC_MAX_SEQ_LENGTH but allow sequence to + * be longer if a single sg element is larger + * than that amount. This is done to avoid creating + * a new sg list to use for the tgtport api. + */ + fcpreq->sg = sg; + fcpreq->sg_cnt = 0; + while (tlen < remaininglen && + fcpreq->sg_cnt < tgtport->max_sg_cnt && + tlen + sg_dma_len(sg) < NVMET_FC_MAX_SEQ_LENGTH) { + fcpreq->sg_cnt++; + tlen += sg_dma_len(sg); + sg = sg_next(sg); + } + if (tlen < remaininglen && fcpreq->sg_cnt == 0) { + fcpreq->sg_cnt++; + tlen += min_t(u32, sg_dma_len(sg), remaininglen); + sg = sg_next(sg); + } + if (tlen < remaininglen) + fod->next_sg = sg; + else + fod->next_sg = NULL; + fcpreq->transfer_length = tlen; fcpreq->transferred_length = 0; fcpreq->fcp_error = 0; fcpreq->rsplen = 0; - fcpreq->sg = &fod->data_sg[fod->offset / PAGE_SIZE]; - fcpreq->sg_cnt = DIV_ROUND_UP(tlen, PAGE_SIZE); - /* * If the last READDATA request: check if LLDD supports * combined xfr with response. -- 2.13.1

7 years, 4 months

1
0
0 0

[PATCH rdma-next v4 3/3] RDMA/mlx5: Fix shift overflow in mlx5_ib_create_wq

by Kees Cook

From: Leon Romanovsky <leonro(a)mellanox.com> [ 61.182439] UBSAN: Undefined behaviour in drivers/infiniband/hw/mlx5/qp.c:5366:34 [ 61.183673] shift exponent 4294967288 is too large for 32-bit type 'unsigned int' [ 61.185530] CPU: 0 PID: 639 Comm: qp Not tainted 4.18.0-rc1-00037-g4aa1d69a9c60-dirty #96 [ 61.186981] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-2.fc27 04/01/2014 [ 61.188315] Call Trace: [ 61.188661] dump_stack+0xc7/0x13b [ 61.190427] ubsan_epilogue+0x9/0x49 [ 61.190899] __ubsan_handle_shift_out_of_bounds+0x1ea/0x22f [ 61.197040] mlx5_ib_create_wq+0x1c99/0x1d50 [ 61.206632] ib_uverbs_ex_create_wq+0x499/0x820 [ 61.213892] ib_uverbs_write+0x77e/0xae0 [ 61.248018] vfs_write+0x121/0x3b0 [ 61.249831] ksys_write+0xa1/0x120 [ 61.254024] do_syscall_64+0x7c/0x2a0 [ 61.256178] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 61.259211] RIP: 0033:0x7f54bab70e99 [ 61.262125] Code: 00 f3 c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 [ 61.268678] RSP: 002b:00007ffe1541c318 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 61.271076] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f54bab70e99 [ 61.273795] RDX: 0000000000000070 RSI: 0000000020000240 RDI: 0000000000000003 [ 61.276982] RBP: 00007ffe1541c330 R08: 00000000200078e0 R09: 0000000000000002 [ 61.280035] R10: 0000000000000000 R11: 0000000000000246 R12: 00000000004005c0 [ 61.283279] R13: 00007ffe1541c420 R14: 0000000000000000 R15: 0000000000000000 Cc: <stable(a)vger.kernel.org> # 4.7 Fixes: 79b20a6c3014 ("IB/mlx5: Add receive Work Queue verbs") Cc: syzkaller <syzkaller(a)googlegroups.com> Reported-by: Noa Osherovich <noaos(a)mellanox.com> Signed-off-by: Leon Romanovsky <leonro(a)mellanox.com> Signed-off-by: Kees Cook <keescook(a)chromium.org> --- drivers/infiniband/hw/mlx5/qp.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index a4f1f638509f..fa3c315f6d94 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -5365,7 +5365,9 @@ static int set_user_rq_size(struct mlx5_ib_dev *dev, rwq->wqe_count = ucmd->rq_wqe_count; rwq->wqe_shift = ucmd->rq_wqe_shift; - rwq->buf_size = (rwq->wqe_count << rwq->wqe_shift); + if (check_shl_overflow(rwq->wqe_count, rwq->wqe_shift, &rwq->buf_size)) + return -EINVAL; + rwq->log_rq_stride = rwq->wqe_shift; rwq->log_rq_size = ilog2(rwq->wqe_count); return 0; -- 2.17.1

7 years, 4 months

1
0
0 0

[PATCH v4 8/8] drm/nouveau: Call pm_runtime_get_noresume() from hpd handlers

by Lyude Paul

We can't and don't need to try resuming the device from our hotplug handlers, but hotplug events are generally something we'd like to keep the device awake for whenever possible. So, grab a PM ref safely in our hotplug handlers using pm_runtime_get_noresume() and mark the device as busy once we're finished. Signed-off-by: Lyude Paul <lyude(a)redhat.com> Cc: stable(a)vger.kernel.org Cc: Lukas Wunner <lukas(a)wunner.de> Cc: Karol Herbst <karolherbst(a)gmail.com> --- drivers/gpu/drm/nouveau/nouveau_connector.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c index 8409c3f2c3a1..5a8e8c1ad647 100644 --- a/drivers/gpu/drm/nouveau/nouveau_connector.c +++ b/drivers/gpu/drm/nouveau/nouveau_connector.c @@ -1152,6 +1152,11 @@ nouveau_connector_hotplug(struct nvif_notify *notify) const char *name = connector->name; struct nouveau_encoder *nv_encoder; + /* Resuming the device here isn't possible; but the suspend PM ops + * will wait for us to finish our work before disabling us so this + * should be enough + */ + pm_runtime_get_noresume(drm->dev->dev); nv_connector->hpd_task = current; if (rep->mask & NVIF_NOTIFY_CONN_V0_IRQ) { @@ -1171,6 +1176,9 @@ nouveau_connector_hotplug(struct nvif_notify *notify) } nv_connector->hpd_task = NULL; + + pm_runtime_mark_last_busy(drm->dev->dev); + pm_runtime_put_autosuspend(drm->dev->dev); return NVIF_NOTIFY_KEEP; } -- 2.17.1

7 years, 4 months

1
0
0 0

[PATCH v4 6/8] drm/nouveau: Respond to HPDs by probing one conn at a time

by Lyude Paul

There isn't actually any reason we need to call drm_hpd_irq_event() from our hotplug handler, as we already know which connector the hotplug event was fired for. We're also going to need to avoid probing all connectors needlessly from hotplug handlers anyway so that we can track when nouveau_connector_detect() is being called from the context of it's connector's hotplug handler in order to fix the next deadlocking issue. This is (slightly) faster anyway! Signed-off-by: Lyude Paul <lyude(a)redhat.com> Cc: stable(a)vger.kernel.org Cc: Lukas Wunner <lukas(a)wunner.de> Cc: Karol Herbst <karolherbst(a)gmail.com> --- drivers/gpu/drm/nouveau/nouveau_connector.c | 28 ++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c index 010d6db14cba..9714e09f17db 100644 --- a/drivers/gpu/drm/nouveau/nouveau_connector.c +++ b/drivers/gpu/drm/nouveau/nouveau_connector.c @@ -1114,6 +1114,32 @@ nouveau_connector_funcs_lvds = { .atomic_get_property = nouveau_conn_atomic_get_property, }; +static void +nouveau_connector_hotplug_probe(struct nouveau_connector *nv_conn) +{ + struct drm_modeset_acquire_ctx ctx; + struct drm_connector *conn = &nv_conn->base; + enum drm_connector_status old_status; + struct drm_device *dev = conn->dev; + bool changed; + + mutex_lock(&dev->mode_config.mutex); + + drm_modeset_acquire_init(&ctx, 0); + drm_modeset_lock(&dev->mode_config.connection_mutex, &ctx); + + old_status = conn->status; + conn->status = drm_helper_probe_detect(conn, &ctx, true); + changed = old_status != conn->status; + + drm_modeset_drop_locks(&ctx); + drm_modeset_acquire_fini(&ctx); + mutex_unlock(&dev->mode_config.mutex); + + if (changed) + drm_kms_helper_hotplug_event(dev); +} + static int nouveau_connector_hotplug(struct nvif_notify *notify) { @@ -1138,7 +1164,7 @@ nouveau_connector_hotplug(struct nvif_notify *notify) nv50_mstm_remove(nv_encoder->dp.mstm); } - drm_helper_hpd_irq_event(connector->dev); + nouveau_connector_hotplug_probe(nv_connector); } return NVIF_NOTIFY_KEEP; -- 2.17.1

7 years, 4 months

1
0
0 0

[PATCH v4 5/8] drm/nouveau: Use pm_runtime_get_noresume() in connector_detect()

by Lyude Paul

It's true we can't resume the device from poll workers in nouveau_connector_detect(). We can however, prevent the autosuspend timer from elapsing immediately if it hasn't already without risking any sort of deadlock with the runtime suspend/resume operations. So do that instead of entirely avoiding grabbing a power reference. Signed-off-by: Lyude Paul <lyude(a)redhat.com> Cc: stable(a)vger.kernel.org Cc: Lukas Wunner <lukas(a)wunner.de> Cc: Karol Herbst <karolherbst(a)gmail.com> --- drivers/gpu/drm/nouveau/nouveau_connector.c | 20 +++++++++++--------- 1 file changed, 11 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c index 2a45b4c2ceb0..010d6db14cba 100644 --- a/drivers/gpu/drm/nouveau/nouveau_connector.c +++ b/drivers/gpu/drm/nouveau/nouveau_connector.c @@ -572,12 +572,16 @@ nouveau_connector_detect(struct drm_connector *connector, bool force) nv_connector->edid = NULL; } - /* Outputs are only polled while runtime active, so acquiring a - * runtime PM ref here is unnecessary (and would deadlock upon - * runtime suspend because it waits for polling to finish). + /* Outputs are only polled while runtime active, so resuming the + * device here is unnecessary (and would deadlock upon runtime suspend + * because it waits for polling to finish). We do however, want to + * prevent the autosuspend timer from elapsing during this operation + * if possible. */ - if (!drm_kms_helper_is_poll_worker()) { - ret = pm_runtime_get_sync(connector->dev->dev); + if (drm_kms_helper_is_poll_worker()) { + pm_runtime_get_noresume(dev->dev); + } else { + ret = pm_runtime_get_sync(dev->dev); if (ret < 0 && ret != -EACCES) return conn_status; } @@ -655,10 +659,8 @@ nouveau_connector_detect(struct drm_connector *connector, bool force) out: - if (!drm_kms_helper_is_poll_worker()) { - pm_runtime_mark_last_busy(connector->dev->dev); - pm_runtime_put_autosuspend(connector->dev->dev); - } + pm_runtime_mark_last_busy(dev->dev); + pm_runtime_put_autosuspend(dev->dev); return conn_status; } -- 2.17.1

7 years, 4 months

1
0
0 0

[PATCH v4 4/8] drm/nouveau: Fix deadlock with fb_helper using new helpers

by Lyude Paul

This removes the potential of deadlocking with fb_helper entirely by preventing it from handling hotplugs during the runtime suspend process as early as possible in the suspend process. If it turns out this is not possible, due to some fb_helper action having been queued up before we got a time to disable hotplugging, we simply return -EBUSY so that the runtime PM core attempts autosuspending the device again once fb_helper isn't doing anything. This fixes one of the issues causing deadlocks on runtime suspend/resume with nouveau on my P50. Signed-off-by: Lyude Paul <lyude(a)redhat.com> Cc: stable(a)vger.kernel.org Cc: Lukas Wunner <lukas(a)wunner.de> Cc: Karol Herbst <karolherbst(a)gmail.com> --- drivers/gpu/drm/nouveau/nouveau_drm.c | 8 ++++++++ drivers/gpu/drm/nouveau/nouveau_fbcon.c | 1 + 2 files changed, 9 insertions(+) diff --git a/drivers/gpu/drm/nouveau/nouveau_drm.c b/drivers/gpu/drm/nouveau/nouveau_drm.c index ee2546db09c9..d47cb5b2af98 100644 --- a/drivers/gpu/drm/nouveau/nouveau_drm.c +++ b/drivers/gpu/drm/nouveau/nouveau_drm.c @@ -836,6 +836,14 @@ nouveau_pmops_runtime_suspend(struct device *dev) return -EBUSY; } + /* There's no way for us to stop fb_helper work in reaction to + * hotplugs later in the RPM process. First off: we don't want to, + * fb_helper should be able to keep the GPU awake. Second off: it is + * capable of grabbing basically any lock in existence. + */ + if (!drm_fb_helper_suspend_hotplug(drm_dev->fb_helper)) + return -EBUSY; + nouveau_switcheroo_optimus_dsm(); ret = nouveau_do_suspend(drm_dev, true); pci_save_state(pdev); diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c index 85c1f10bc2b6..963ba630fd04 100644 --- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c +++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c @@ -466,6 +466,7 @@ nouveau_fbcon_set_suspend_work(struct work_struct *work) console_unlock(); if (state == FBINFO_STATE_RUNNING) { + drm_fb_helper_resume_hotplug(drm->dev->fb_helper); pm_runtime_mark_last_busy(drm->dev->dev); pm_runtime_put_sync(drm->dev->dev); } -- 2.17.1

7 years, 4 months

1
0
0 0

2025

2024

2023

2022

2021

2020

2019

2018

2017

Linux-stable-mirror