lwn.git/arch/x86/kernel/dumpstack.c, branch v4.18-rc5

x86/dumpstack: Explain the reasoning for the prologue and buffer size

2018-04-26T14:15:28+00:00

The whole reasoning behind the amount of opcode bytes dumped and prologue length isn't very clear so write down some of the reasons for why it is done the way it is. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-10-bp@alien8.de

x86/dumpstack: Save first regs set for the executive summary

2018-04-26T14:15:28+00:00

Save the regs set when __die() is onvoked for the first time and print it in oops_end(). Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-9-bp@alien8.de

x86/dumpstack: Add a show_ip() function

2018-04-26T14:15:27+00:00

... which shows the Instruction Pointer along with the insn bytes around it. Use it whenever rIP is printed. Drop the rIP < PAGE_OFFSET check since probe_kernel_read() can handle any address properly. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-8-bp@alien8.de

x86/dumpstack: Add loglevel argument to show_opcodes()

2018-04-26T14:15:26+00:00

Will be used in the next patch. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-6-bp@alien8.de

x86/dumpstack: Improve opcodes dumping in the code section

2018-04-26T14:15:26+00:00

The code used to iterate byte-by-byte over the bytes around RIP and that is expensive: disabling pagefaults around it, copy_from_user, etc... Make it read the whole buffer of OPCODE_BUFSIZE size in one go. Use a statically allocated 64 bytes buffer so that concurrent show_opcodes() do not interleave in the output even though in the majority of the cases it's serialized via die_lock. Except the #PF path which doesn't... Also, do the PAGE_OFFSET check outside of the function because latter will be reused in other context. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-5-bp@alien8.de

x86/dumpstack: Carve out code-dumping into a function

2018-04-26T14:15:26+00:00

No functionality change, carve it out into a separate function for later changes. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-4-bp@alien8.de

x86/dumpstack: Unexport oops_begin()

2018-04-26T14:15:26+00:00

The only user outside of arch/ is not a module since 86cd47334b00 ("ACPI, APEI, GHES, Prevent GHES to be built as module") No functional changes. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-3-bp@alien8.de

x86/dumpstack: Remove code_bytes

2018-04-26T14:15:25+00:00

This was added by 86c418374223 ("[PATCH] i386: add option to show more code in oops reports") long time ago but experience shows that 64 instruction bytes are plenty when deciphering an oops. So get rid of it. Removing it will simplify further enhancements to the opcodes dumping machinery coming in the following patches. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180417161124.5294-2-bp@alien8.de

x86/dumpstack: Unify show_regs()

2018-03-08T11:04:59+00:00

The 32-bit version uses KERN_EMERG and commit b0f4c4b32c8e ("bugs, x86: Fix printk levels for panic, softlockups and stack dumps") changed the 64-bit version to KERN_DEFAULT. The same justification in that commit that those messages do not belong in the terminal, holds true for 32-bit also, so make it so. Make code_bytes static, while at it. Signed-off-by: Borislav Petkov Signed-off-by: Thomas Gleixner Cc: Peter Zijlstra Cc: Josh Poimboeuf Cc: Linus Torvalds Cc: Andy Lutomirski Link: https://lkml.kernel.org/r/20180306094920.16917-4-bp@alien8.de

x86/dumpstack: Avoid uninitlized variable

2018-02-02T22:33:50+00:00

In some configurations, 'partial' does not get initialized, as shown by this gcc-8 warning: arch/x86/kernel/dumpstack.c: In function 'show_trace_log_lvl': arch/x86/kernel/dumpstack.c:156:4: error: 'partial' may be used uninitialized in this function [-Werror=maybe-uninitialized] show_regs_if_on_stack(&stack_info, regs, partial); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This initializes it to false, to get the previous behavior in this case. Fixes: a9cdbe72c4e8 ("x86/dumpstack: Fix partial register dumps") Signed-off-by: Arnd Bergmann Signed-off-by: Thomas Gleixner Cc: Andi Kleen Cc: Nicolas Pitre Cc: Peter Zijlstra Cc: Dave Hansen Cc: Andy Lutomirski Cc: Josh Poimboeuf Cc: Borislav Petkov Cc: Vlastimil Babka Link: https://lkml.kernel.org/r/20180202145634.200291-1-arnd@arndb.de