Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 08, 2018 at 03:20:58PM -0800, Linus Torvalds wrote: > Yes, so the reason the prologue is more important is that there's > really two cases for the "Code:" line: Yap, I had a hunch it must be about some of those but thanks for taking the time and writing it down! I've tried to

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 08, 2018 at 03:20:58PM -0800, Linus Torvalds wrote: > Yes, so the reason the prologue is more important is that there's > really two cases for the "Code:" line: Yap, I had a hunch it must be about some of those but thanks for taking the time and writing it down! I've tried to

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 8, 2018 at 2:36 PM, Borislav Petkov wrote: > > Btw, do we have any explanation for the two-thirds prologue? I dug it > out to the patch below but it doesn't say why the prologue being bigger > is more important than the epilogue. Yes, so the reason the prologue is

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 8, 2018 at 2:36 PM, Borislav Petkov wrote: > > Btw, do we have any explanation for the two-thirds prologue? I dug it > out to the patch below but it doesn't say why the prologue being bigger > is more important than the epilogue. Yes, so the reason the prologue is more important is

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 08, 2018 at 10:00:09AM -0800, Linus Torvalds wrote: > On Thu, Mar 8, 2018 at 2:16 AM, Borislav Petkov wrote: > > +#define OPCODE_BUFSIZE 64 > > + unsigned int code_prologue = OPCODE_BUFSIZE * 43 / OPCODE_BUFSIZE; > > Heh. > > That's a very odd way of writing

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 08, 2018 at 10:00:09AM -0800, Linus Torvalds wrote: > On Thu, Mar 8, 2018 at 2:16 AM, Borislav Petkov wrote: > > +#define OPCODE_BUFSIZE 64 > > + unsigned int code_prologue = OPCODE_BUFSIZE * 43 / OPCODE_BUFSIZE; > > Heh. > > That's a very odd way of writing "43". I was

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 8, 2018 at 2:16 AM, Borislav Petkov wrote: > +#define OPCODE_BUFSIZE 64 > + unsigned int code_prologue = OPCODE_BUFSIZE * 43 / OPCODE_BUFSIZE; Heh. That's a very odd way of writing "43". Honestly, the "43" is just "two thirds" rounded to closest, and it's not

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Thu, Mar 8, 2018 at 2:16 AM, Borislav Petkov wrote: > +#define OPCODE_BUFSIZE 64 > + unsigned int code_prologue = OPCODE_BUFSIZE * 43 / OPCODE_BUFSIZE; Heh. That's a very odd way of writing "43". Honestly, the "43" is just "two thirds" rounded to closest, and it's not important

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 01:08:32PM -0800, Linus Torvalds wrote: > On Wed, Mar 7, 2018 at 5:25 AM, Josh Poimboeuf wrote: > > > > How about we just remove the 'code_bytes=' option? (Or at the very > > least, reduce its possible range to a reasonable max?) > > Ack. Just limit

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 01:08:32PM -0800, Linus Torvalds wrote: > On Wed, Mar 7, 2018 at 5:25 AM, Josh Poimboeuf wrote: > > > > How about we just remove the 'code_bytes=' option? (Or at the very > > least, reduce its possible range to a reasonable max?) > > Ack. Just limit it to 64 bytes max

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 7, 2018 at 5:25 AM, Josh Poimboeuf wrote: > > How about we just remove the 'code_bytes=' option? (Or at the very > least, reduce its possible range to a reasonable max?) Ack. Just limit it to 64 bytes max sounds plenty. Linus

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 7, 2018 at 5:25 AM, Josh Poimboeuf wrote: > > How about we just remove the 'code_bytes=' option? (Or at the very > least, reduce its possible range to a reasonable max?) Ack. Just limit it to 64 bytes max sounds plenty. Linus

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 07:25:35AM -0600, Josh Poimboeuf wrote: > How about we just remove the 'code_bytes=' option? Haha, removing stuff is my usual solution :-) I'd love to. > (Or at the very > least, reduce its possible range to a reasonable max?) > > I doubt anybody actually uses it. I'd

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 07:25:35AM -0600, Josh Poimboeuf wrote: > How about we just remove the 'code_bytes=' option? Haha, removing stuff is my usual solution :-) I'd love to. > (Or at the very > least, reduce its possible range to a reasonable max?) > > I doubt anybody actually uses it. I'd

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 11:13:14AM +0100, Borislav Petkov wrote: > And that is fine if I do a 64-byte default, on-stack buffer but that > code_bytes= thing can be 2 pages max which is yuck. No way I'm doing > on-stack buffers then. How about we just remove the 'code_bytes=' option? (Or at the

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Wed, Mar 07, 2018 at 11:13:14AM +0100, Borislav Petkov wrote: > And that is fine if I do a 64-byte default, on-stack buffer but that > code_bytes= thing can be 2 pages max which is yuck. No way I'm doing > on-stack buffers then. How about we just remove the 'code_bytes=' option? (Or at the

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Tue, Mar 06, 2018 at 10:47:56AM -0800, Linus Torvalds wrote: > Are these always serialized? For oopses, I think we end up serializing > with die_lock, but is that always the case? Hmm, good question. > Maybe at least a comment about why a static allocation is ok? Well, I'm afraid it is not

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Tue, Mar 06, 2018 at 10:47:56AM -0800, Linus Torvalds wrote: > Are these always serialized? For oopses, I think we end up serializing > with die_lock, but is that always the case? Hmm, good question. > Maybe at least a comment about why a static allocation is ok? Well, I'm afraid it is not

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Tue, Mar 6, 2018 at 1:49 AM, Borislav Petkov wrote: > > Make it read the whole buffer of code_bytes size in one go. By default > use a statically allocated 64 bytes buffer. If "code_bytes=" is supplied > on the cmdline a new buffer gets allocated. Are these always serialized?

Re: [PATCH 5/9] x86/dumpstack: Improve opcodes dumping in the Code: section

On Tue, Mar 6, 2018 at 1:49 AM, Borislav Petkov wrote: > > Make it read the whole buffer of code_bytes size in one go. By default > use a statically allocated 64 bytes buffer. If "code_bytes=" is supplied > on the cmdline a new buffer gets allocated. Are these always serialized? For oopses, I