In the Linux kernel, the following vulnerability has been resolved:
qed: allow sleep in qed_mcp_trace_dump()
By default, qed_mcp_cmd_and_union() delays 10us at a time in a loop that can run 500K times, so calls to qed_mcp_nvm_rd_cmd() may block the current thread for over 5s. We observed thread scheduling delays over 700ms in production, with stacktraces pointing to this code as the culprit.
qed_mcp_trace_dump() is called from ethtool, so sleeping is permitted. It already can sleep in qed_mcp_halt(), which calls qed_mcp_cmd(). Add a "can sleep" parameter to qed_find_nvram_image() and qed_nvram_read() so they can sleep during qed_mcp_trace_dump(). qed_mcp_trace_get_meta_info() and qed_mcp_trace_read_meta(), called only by qed_mcp_trace_dump(), allow these functions to sleep. I can't tell if the other caller (qed_grc_dump_mcp_hw_dump()) can sleep, so keep b_can_sleep set to false when it calls these functions.
An example stacktrace from a custom warning we added to the kernel showing a thread that has not scheduled despite long needing resched: [ 2745.362925,17] ------------[ cut here ]------------ [ 2745.362941,17] WARNING: CPU: 23 PID: 5640 at arch/x86/kernel/irq.c:233 do_IRQ+0x15e/0x1a0() [ 2745.362946,17] Thread not rescheduled for 744 ms after irq 99 [ 2745.362956,17] Modules linked in: ... [ 2745.363339,17] CPU: 23 PID: 5640 Comm: lldpd Tainted: P O 4.4.182+ #202104120910+6d1da174272d.61x [ 2745.363343,17] Hardware name: FOXCONN MercuryB/Quicksilver Controller, BIOS H11P1N09 07/08/2020 [ 2745.363346,17] 0000000000000000 ffff885ec07c3ed8 ffffffff8131eb2f ffff885ec07c3f20 [ 2745.363358,17] ffffffff81d14f64 ffff885ec07c3f10 ffffffff81072ac2 ffff88be98ed0000 [ 2745.363369,17] 0000000000000063 0000000000000174 0000000000000074 0000000000000000 [ 2745.363379,17] Call Trace: [ 2745.363382,17] <IRQ> [<ffffffff8131eb2f>] dump_stack+0x8e/0xcf [ 2745.363393,17] [<ffffffff81072ac2>] warn_slowpath_common+0x82/0xc0 [ 2745.363398,17]...