public inbox for [email protected]
 help / color / mirror / Atom feed
* [PATCH] io_thread/x86: don't reset 'cs', 'ss', 'ds' and 'es' registers for io_threads
@ 2021-04-11 15:27 Stefan Metzmacher
  2021-04-14 21:28 ` Stefan Metzmacher
  2021-05-05 11:03 ` [PATCH v2] io_thread/x86: setup io_threads more like normal user space threads Stefan Metzmacher
  0 siblings, 2 replies; 9+ messages in thread
From: Stefan Metzmacher @ 2021-04-11 15:27 UTC (permalink / raw)
  To: Linus Torvalds; +Cc: Stefan Metzmacher, Jens Axboe, linux-kernel, io-uring

This allows gdb attach to userspace processes using io-uring,
which means that they have io_threads (PF_IO_WORKER), which appear
just like normal as userspace threads.

See the code comment for more details.

Fixes: 4727dc20e04 ("arch: setup PF_IO_WORKER threads like PF_KTHREAD")
Signed-off-by: Stefan Metzmacher <[email protected]>
cc: Linus Torvalds <[email protected]>
cc: Jens Axboe <[email protected]>
cc: [email protected]
cc: [email protected]
---
 arch/x86/kernel/process.c | 49 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 9c214d7085a4..72120c4b7618 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -163,6 +163,55 @@ int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long arg,
 	/* Kernel thread ? */
 	if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
 		memset(childregs, 0, sizeof(struct pt_regs));
+		/*
+		 * gdb sees all userspace threads,
+		 * including io threads (PF_IO_WORKER)!
+		 *
+		 * gdb uses:
+		 * PTRACE_PEEKUSR, offsetof (struct user_regs_struct, cs)
+		 *  returning with 0x33 (51) to detect 64 bit
+		 * and:
+		 * PTRACE_PEEKUSR, offsetof (struct user_regs_struct, ds)
+		 *  returning 0x2b (43) to detect 32 bit.
+		 *
+		 * GDB relies on that the kernel returns the
+		 * same values for all threads, which means
+		 * we don't zero these out.
+		 *
+		 * Note that CONFIG_X86_64 handles 'es' and 'ds'
+		 * differently, see the following above:
+		 *   savesegment(es, p->thread.es);
+		 *   savesegment(ds, p->thread.ds);
+		 * and the CONFIG_X86_64 version of get_segment_reg().
+		 *
+		 * Linus proposed something like this:
+		 * (https://lore.kernel.org/io-uring/CAHk-=whEObPkZBe4766DmR46-=5QTUiatWbSOaD468eTgYc1tg@mail.gmail.com/)
+		 *
+		 *   childregs->cs = __USER_CS;
+		 *   childregs->ss = __USER_DS;
+		 *   childregs->ds = __USER_DS;
+		 *   childregs->es = __USER_DS;
+		 *
+		 * might make sense (just do it unconditionally, rather than making it
+		 * special to PF_IO_WORKER).
+		 *
+		 * But that doesn't make gdb happy in all cases.
+		 *
+		 * While 32bit userspace on a 64bit kernel is legacy,
+		 * it's still useful to allow 32bit libraries or nss modules
+		 * use the same code as the 64bit version of that library, which
+		 * can use io-uring just fine.
+		 *
+		 * So we better just inherit the values from
+		 * the originating process instead of hardcoding
+		 * values, which would imply 64bit userspace.
+		 */
+		childregs->cs = current_pt_regs()->cs;
+		childregs->ss = current_pt_regs()->ss;
+#ifdef CONFIG_X86_32
+		childregs->ds = current_pt_regs()->ds;
+		childregs->es = current_pt_regs()->es;
+#endif
 		kthread_frame_init(frame, sp, arg);
 		return 0;
 	}
-- 
2.25.1


^ permalink raw reply related	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2021-05-06  9:18 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2021-04-11 15:27 [PATCH] io_thread/x86: don't reset 'cs', 'ss', 'ds' and 'es' registers for io_threads Stefan Metzmacher
2021-04-14 21:28 ` Stefan Metzmacher
2021-05-05 11:03 ` [PATCH v2] io_thread/x86: setup io_threads more like normal user space threads Stefan Metzmacher
2021-05-05 21:24   ` Jens Axboe
2021-05-05 21:57     ` Thomas Gleixner
2021-05-05 22:07       ` Thomas Gleixner
2021-05-05 23:49         ` Jens Axboe
2021-05-06  9:17           ` Thomas Gleixner
2021-05-05 23:35       ` Jens Axboe

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox