| Message ID | 20221106140355.294845-4-42.hyeyoo@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | move PG_slab to page_type | expand |
On Sun, 2022-11-06 at 23:03 +0900, Hyeonggon Yoo wrote: > dump_page() uses %pGp format to print 'flags' field of struct page. > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > are set in page_type field, introduce %pGt format which provides > human readable output of page_type. And use it in dump_page(). [] > diff --git a/lib/vsprintf.c b/lib/vsprintf.c [] > @@ -2056,6 +2056,28 @@ char *format_page_flags(char *buf, char *end, unsigned long flags) > return buf; > } > > +static noinline_for_stack ? > +char *format_page_type(char *buf, char *end, unsigned int page_type) > +{ > + if (!(page_type & PAGE_TYPE_BASE)) > + return string(buf, end, "no type for user-mapped page", default_str_spec); Might be better with something like '%pGt: no type..."
On Sun, Nov 06, 2022 at 11:03:55PM +0900, Hyeonggon Yoo wrote: > dump_page() uses %pGp format to print 'flags' field of struct page. > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > are set in page_type field, introduce %pGt format which provides > human readable output of page_type. And use it in dump_page(). > > Note that the sense of bits are different in page_type. if page_type is > 0xffffffff, no flags are set. if PG_slab (0x00100000) flag is set, > page_type is 0xffefffff. Clearing a bit means we set the bit. Bits in > page_type are inverted when printing type names. > > Below is examples of dump_page(). One is just after alloc_pages() and > the other is after __SetPageSlab(). > > [ 1.814728] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 > [ 1.815961] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) > [ 1.816443] page_type: 0xffffffff() Why do we have empty parentheses? I would expect either something there, or no parentheses at all. > [ 1.816704] raw: 0017ffffc0000000 0000000000000000 dead000000000122 0000000000000000 > [ 1.817291] raw: 0000000000000000 0000000000000000 00000001ffffffff 0000000000000000 > [ 1.817870] page dumped because: Before __SetPageSlab() > > [ 1.818258] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 > [ 1.818857] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) > [ 1.819250] page_type: 0xffefffff(slab) > [ 1.819483] raw: 0017ffffc0000000 0000000000000000 dead000000000122 0000000000000000 > [ 1.819947] raw: 0000000000000000 0000000000000000 00000001ffefffff 0000000000000000 > [ 1.820410] page dumped because: After __SetPageSlab() > Cc: Petr Mladek <pmladek@suse.com> > Cc: Steven Rostedt <rostedt@goodmis.org> > Cc: Sergey Senozhatsky <senozhatsky@chromium.org> > Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> > Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk> > Cc: Jonathan Corbet <corbet@lwn.net> > Cc: Andrew Morton <akpm@linux-foundation.org> > Cc: linux-doc@vger.kernel.org > Cc: linux-kernel@vger.kernel.org > Cc: linux-mm@kvack.org Can you utilize --cc parameter next time and avoid polluting commit message with this? We have archives where anybody can check this (and usually maintainers add a Link tag for that purpose). ... > + %pGt 0xffefffff(slab) No space before ( ? ... > +static > +char *format_page_type(char *buf, char *end, unsigned int page_type) > +{ > + if (!(page_type & PAGE_TYPE_BASE)) > + return string(buf, end, "no type for user-mapped page", default_str_spec); It's too long, can we make it shorten? > + buf = number(buf, end, page_type, default_flag_spec); > + > + if (buf < end) > + *buf = '('; > + buf++; > + if (page_type_has_type(page_type)) This should be check for the entire function. > + buf = format_flags(buf, end, ~page_type, pagetype_names); > + > + if (buf < end) > + *buf = ')'; > + buf++; > + > + return buf; > +} ... > @@ -36,6 +36,11 @@ const struct trace_print_flags pageflag_names[] = { > {0, NULL} > }; > > +const struct trace_print_flags pagetype_names[] = { > + __def_pagetype_names, > + {0, NULL} Hmm... I see it's already done like this above, but {} would suffice, why not to convert the rest first to the {} and use it here? > +}; ... > pr_warn("%sflags: %pGp%s\n", type, &head->flags, > page_cma ? " CMA" : ""); > + pr_warn("page_type: %pGt\n", &head->page_type); > + > print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32, > sizeof(unsigned long), page, > sizeof(struct page), false); > diff --git a/mm/internal.h b/mm/internal.h > index cb4c663a714e..956eaa9f12c0 100644 > --- a/mm/internal.h > +++ b/mm/internal.h > @@ -773,6 +773,7 @@ static inline void flush_tlb_batched_pending(struct mm_struct *mm) > #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ > > extern const struct trace_print_flags pageflag_names[]; > +extern const struct trace_print_flags pagetype_names[]; > extern const struct trace_print_flags vmaflag_names[]; > extern const struct trace_print_flags gfpflag_names[]; I would split this to a separate change, but it's up to PRINTK maintainers.
On Mon 2022-11-07 13:18:52, Andy Shevchenko wrote: > On Sun, Nov 06, 2022 at 11:03:55PM +0900, Hyeonggon Yoo wrote: > > dump_page() uses %pGp format to print 'flags' field of struct page. > > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > > are set in page_type field, introduce %pGt format which provides > > human readable output of page_type. And use it in dump_page(). > > > > Note that the sense of bits are different in page_type. if page_type is > > 0xffffffff, no flags are set. if PG_slab (0x00100000) flag is set, > > page_type is 0xffefffff. Clearing a bit means we set the bit. Bits in > > page_type are inverted when printing type names. > > > > Below is examples of dump_page(). One is just after alloc_pages() and > > the other is after __SetPageSlab(). > > > > [ 1.814728] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 > > [ 1.815961] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) > > > [ 1.816443] page_type: 0xffffffff() > > Why do we have empty parentheses? I would expect either something there, or no > parentheses at all. This looks fine. format_page_flags() does the same for %pGp. > ... > > > + %pGt 0xffefffff(slab) > > No space before ( ? Also looks fine. %pGp does the same. > ... > > > +static > > +char *format_page_type(char *buf, char *end, unsigned int page_type) > > +{ > > + if (!(page_type & PAGE_TYPE_BASE)) > > + return string(buf, end, "no type for user-mapped page", default_str_spec); > > It's too long, can we make it shorten? I wonder if it would help to write the value. Something like: page_type: 0x0ace5768(no type) That said. I am not familiar with the page types and am not sure about the semantic of this value. MM people should decide what they want to see in this case. > ... > > > pr_warn("%sflags: %pGp%s\n", type, &head->flags, > > page_cma ? " CMA" : ""); > > + pr_warn("page_type: %pGt\n", &head->page_type); > > + > > print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32, > > sizeof(unsigned long), page, > > sizeof(struct page), false); > > diff --git a/mm/internal.h b/mm/internal.h > > index cb4c663a714e..956eaa9f12c0 100644 > > --- a/mm/internal.h > > +++ b/mm/internal.h > > @@ -773,6 +773,7 @@ static inline void flush_tlb_batched_pending(struct mm_struct *mm) > > #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ > > > > extern const struct trace_print_flags pageflag_names[]; > > +extern const struct trace_print_flags pagetype_names[]; > > extern const struct trace_print_flags vmaflag_names[]; > > extern const struct trace_print_flags gfpflag_names[]; > > I would split this to a separate change, but it's up to PRINTK maintainers. I guess that you are talking about the line: + pr_warn("page_type: %pGt\n", &head->page_type); Yes, it would be better to have implementation of %pGt modifier in one patch and add the user in another one. Best Regards, Petr
On Mon, Nov 07, 2022 at 03:20:34PM +0100, Petr Mladek wrote: > On Mon 2022-11-07 13:18:52, Andy Shevchenko wrote: > > On Sun, Nov 06, 2022 at 11:03:55PM +0900, Hyeonggon Yoo wrote: > > > dump_page() uses %pGp format to print 'flags' field of struct page. ... > > > [ 1.816443] page_type: 0xffffffff() > > > > Why do we have empty parentheses? I would expect either something there, or no > > parentheses at all. > > This looks fine. format_page_flags() does the same for %pGp. ... > > > + %pGt 0xffefffff(slab) > > > > No space before ( ? > > Also looks fine. %pGp does the same. Maybe we can modify them before or after this change?
On Mon, Nov 07, 2022 at 03:20:34PM +0100, Petr Mladek wrote: > On Mon 2022-11-07 13:18:52, Andy Shevchenko wrote: > > On Sun, Nov 06, 2022 at 11:03:55PM +0900, Hyeonggon Yoo wrote: > > > dump_page() uses %pGp format to print 'flags' field of struct page. > > > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > > > are set in page_type field, introduce %pGt format which provides > > > human readable output of page_type. And use it in dump_page(). > > > > > > Note that the sense of bits are different in page_type. if page_type is > > > 0xffffffff, no flags are set. if PG_slab (0x00100000) flag is set, > > > page_type is 0xffefffff. Clearing a bit means we set the bit. Bits in > > > page_type are inverted when printing type names. > > > > > > Below is examples of dump_page(). One is just after alloc_pages() and > > > the other is after __SetPageSlab(). > > > > > > [ 1.814728] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 > > > [ 1.815961] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) > > > > > [ 1.816443] page_type: 0xffffffff() > > Thank you both for looking at this! :-) > > Why do we have empty parentheses? I would expect either something there, or no > > parentheses at all. > > This looks fine. format_page_flags() does the same for %pGp. > > > ... > > > > > + %pGt 0xffefffff(slab) > > > > No space before ( ? > > Also looks fine. %pGp does the same. > > > ... > > > > > +static > > > +char *format_page_type(char *buf, char *end, unsigned int page_type) > > > +{ > > > + if (!(page_type & PAGE_TYPE_BASE)) > > > + return string(buf, end, "no type for user-mapped page", default_str_spec); > > > > It's too long, can we make it shorten? > > I wonder if it would help to write the value. Something like: > > page_type: 0x0ace5768(no type) > > That said. I am not familiar with the page types and am not sure > about the semantic of this value. MM people should decide what they > want to see in this case. Hmm, then for consistency let's try: - 0xffefffff(slab) - 0xffffffff() - 0x0ace5768() this way. > > ... > > > > > pr_warn("%sflags: %pGp%s\n", type, &head->flags, > > > page_cma ? " CMA" : ""); > > > + pr_warn("page_type: %pGt\n", &head->page_type); > > > + > > > print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32, > > > sizeof(unsigned long), page, > > > sizeof(struct page), false); > > > diff --git a/mm/internal.h b/mm/internal.h > > > index cb4c663a714e..956eaa9f12c0 100644 > > > --- a/mm/internal.h > > > +++ b/mm/internal.h > > > @@ -773,6 +773,7 @@ static inline void flush_tlb_batched_pending(struct mm_struct *mm) > > > #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ > > > > > > extern const struct trace_print_flags pageflag_names[]; > > > +extern const struct trace_print_flags pagetype_names[]; > > > extern const struct trace_print_flags vmaflag_names[]; > > > extern const struct trace_print_flags gfpflag_names[]; > > > > I would split this to a separate change, but it's up to PRINTK maintainers. > > I guess that you are talking about the line: > > + pr_warn("page_type: %pGt\n", &head->page_type); > > > Yes, it would be better to have implementation of %pGt modifier > in one patch and add the user in another one. Sounds reasonable. Will split its caller into another patch next to this one.
On Sun, Nov 06, 2022 at 10:04:25AM -0800, Joe Perches wrote: > On Sun, 2022-11-06 at 23:03 +0900, Hyeonggon Yoo wrote: > > dump_page() uses %pGp format to print 'flags' field of struct page. > > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > > are set in page_type field, introduce %pGt format which provides > > human readable output of page_type. And use it in dump_page(). > [] > > diff --git a/lib/vsprintf.c b/lib/vsprintf.c > [] > > @@ -2056,6 +2056,28 @@ char *format_page_flags(char *buf, char *end, unsigned long flags) > > return buf; > > } > > > > +static Thanks for looking at this. > > noinline_for_stack ? May I ask why, Does it have issues related to stack consumption? To Be Honest I'm a bit unsure what is purpose of this attribute. > > +char *format_page_type(char *buf, char *end, unsigned int page_type) > > +{ > > + if (!(page_type & PAGE_TYPE_BASE)) > > + return string(buf, end, "no type for user-mapped page", default_str_spec); > > Might be better with something like '%pGt: no type..." Will try something like "0x32()" when it has no type because it is mapped to userspace.
On Wed 2022-11-09 15:14:05, Hyeonggon Yoo wrote: > On Sun, Nov 06, 2022 at 10:04:25AM -0800, Joe Perches wrote: > > On Sun, 2022-11-06 at 23:03 +0900, Hyeonggon Yoo wrote: > > > dump_page() uses %pGp format to print 'flags' field of struct page. > > > As some page flags (e.g. PG_buddy, see page-flags.h for more details) > > > are set in page_type field, introduce %pGt format which provides > > > human readable output of page_type. And use it in dump_page(). > > [] > > > diff --git a/lib/vsprintf.c b/lib/vsprintf.c > > [] > > > @@ -2056,6 +2056,28 @@ char *format_page_flags(char *buf, char *end, unsigned long flags) > > > return buf; > > > } > > > > > > +static > > Thanks for looking at this. > > > > > noinline_for_stack ? Honestly, I do not like much adding this without numbers. It has been added to some functions in vsprintf.c long time ago because it reduced the stack usage. But I think that it is a compiler and an architecture specific. And it is not clear if it would really help in this particular case. Feel free to omit it. Best Regards, Petr
diff --git a/Documentation/core-api/printk-formats.rst b/Documentation/core-api/printk-formats.rst index dbe1aacc79d0..582e965508eb 100644 --- a/Documentation/core-api/printk-formats.rst +++ b/Documentation/core-api/printk-formats.rst @@ -575,12 +575,13 @@ The field width is passed by value, the bitmap is passed by reference. Helper macros cpumask_pr_args() and nodemask_pr_args() are available to ease printing cpumask and nodemask. -Flags bitfields such as page flags, gfp_flags +Flags bitfields such as page flags, page_type, gfp_flags --------------------------------------------- :: %pGp 0x17ffffc0002036(referenced|uptodate|lru|active|private|node=0|zone=2|lastcpupid=0x1fffff) + %pGt 0xffefffff(slab) %pGg GFP_USER|GFP_DMA32|GFP_NOWARN %pGv read|exec|mayread|maywrite|mayexec|denywrite diff --git a/lib/test_printf.c b/lib/test_printf.c index fe13de1bed5f..6b778a8ea44c 100644 --- a/lib/test_printf.c +++ b/lib/test_printf.c @@ -654,12 +654,26 @@ page_flags_test(int section, int node, int zone, int last_cpupid, test(cmp_buf, "%pGp", &flags); } +static void __init page_type_test(unsigned int page_type, const char *name, + char *cmp_buf) +{ + unsigned long size; + + size = scnprintf(cmp_buf, BUF_SIZE, "%#x(", page_type); + if (page_type_has_type(page_type)) + size += scnprintf(cmp_buf + size, BUF_SIZE - size, "%s", name); + + snprintf(cmp_buf + size, BUF_SIZE - size, ")"); + test(cmp_buf, "%pGt", &page_type); +} + static void __init flags(void) { unsigned long flags; char *cmp_buffer; gfp_t gfp; + unsigned int page_type; cmp_buffer = kmalloc(BUF_SIZE, GFP_KERNEL); if (!cmp_buffer) @@ -699,6 +713,15 @@ flags(void) gfp |= __GFP_HIGH; test(cmp_buffer, "%pGg", &gfp); + page_type = ~0; + page_type_test(page_type, "", cmp_buffer); + + page_type = ~PG_slab; + page_type_test(page_type, "slab", cmp_buffer); + + page_type = ~(PG_slab | PG_table | PG_buddy); + page_type_test(page_type, "slab|table|buddy", cmp_buffer); + kfree(cmp_buffer); } diff --git a/lib/vsprintf.c b/lib/vsprintf.c index 24f37bab8bc1..d855b40e5cfd 100644 --- a/lib/vsprintf.c +++ b/lib/vsprintf.c @@ -2056,6 +2056,28 @@ char *format_page_flags(char *buf, char *end, unsigned long flags) return buf; } +static +char *format_page_type(char *buf, char *end, unsigned int page_type) +{ + if (!(page_type & PAGE_TYPE_BASE)) + return string(buf, end, "no type for user-mapped page", default_str_spec); + + buf = number(buf, end, page_type, default_flag_spec); + + if (buf < end) + *buf = '('; + buf++; + + if (page_type_has_type(page_type)) + buf = format_flags(buf, end, ~page_type, pagetype_names); + + if (buf < end) + *buf = ')'; + buf++; + + return buf; +} + static noinline_for_stack char *flags_string(char *buf, char *end, void *flags_ptr, struct printf_spec spec, const char *fmt) @@ -2069,6 +2091,8 @@ char *flags_string(char *buf, char *end, void *flags_ptr, switch (fmt[1]) { case 'p': return format_page_flags(buf, end, *(unsigned long *)flags_ptr); + case 't': + return format_page_type(buf, end, *(unsigned int *)flags_ptr); case 'v': flags = *(unsigned long *)flags_ptr; names = vmaflag_names; diff --git a/mm/debug.c b/mm/debug.c index 0fd15ba70d16..bb7f2278abc5 100644 --- a/mm/debug.c +++ b/mm/debug.c @@ -36,6 +36,11 @@ const struct trace_print_flags pageflag_names[] = { {0, NULL} }; +const struct trace_print_flags pagetype_names[] = { + __def_pagetype_names, + {0, NULL} +}; + const struct trace_print_flags gfpflag_names[] = { __def_gfpflag_names, {0, NULL} @@ -114,6 +119,8 @@ static void __dump_page(struct page *page) pr_warn("%sflags: %pGp%s\n", type, &head->flags, page_cma ? " CMA" : ""); + pr_warn("page_type: %pGt\n", &head->page_type); + print_hex_dump(KERN_WARNING, "raw: ", DUMP_PREFIX_NONE, 32, sizeof(unsigned long), page, sizeof(struct page), false); diff --git a/mm/internal.h b/mm/internal.h index cb4c663a714e..956eaa9f12c0 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -773,6 +773,7 @@ static inline void flush_tlb_batched_pending(struct mm_struct *mm) #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ extern const struct trace_print_flags pageflag_names[]; +extern const struct trace_print_flags pagetype_names[]; extern const struct trace_print_flags vmaflag_names[]; extern const struct trace_print_flags gfpflag_names[];
dump_page() uses %pGp format to print 'flags' field of struct page. As some page flags (e.g. PG_buddy, see page-flags.h for more details) are set in page_type field, introduce %pGt format which provides human readable output of page_type. And use it in dump_page(). Note that the sense of bits are different in page_type. if page_type is 0xffffffff, no flags are set. if PG_slab (0x00100000) flag is set, page_type is 0xffefffff. Clearing a bit means we set the bit. Bits in page_type are inverted when printing type names. Below is examples of dump_page(). One is just after alloc_pages() and the other is after __SetPageSlab(). [ 1.814728] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 [ 1.815961] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) [ 1.816443] page_type: 0xffffffff() [ 1.816704] raw: 0017ffffc0000000 0000000000000000 dead000000000122 0000000000000000 [ 1.817291] raw: 0000000000000000 0000000000000000 00000001ffffffff 0000000000000000 [ 1.817870] page dumped because: Before __SetPageSlab() [ 1.818258] page:ffffea000415e200 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x105788 [ 1.818857] flags: 0x17ffffc0000000(node=0|zone=2|lastcpupid=0x1fffff) [ 1.819250] page_type: 0xffefffff(slab) [ 1.819483] raw: 0017ffffc0000000 0000000000000000 dead000000000122 0000000000000000 [ 1.819947] raw: 0000000000000000 0000000000000000 00000001ffefffff 0000000000000000 [ 1.820410] page dumped because: After __SetPageSlab() Cc: Petr Mladek <pmladek@suse.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Sergey Senozhatsky <senozhatsky@chromium.org> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rasmus Villemoes <linux@rasmusvillemoes.dk> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: linux-doc@vger.kernel.org Cc: linux-kernel@vger.kernel.org Cc: linux-mm@kvack.org Signed-off-by: Hyeonggon Yoo <42.hyeyoo@gmail.com> --- Documentation/core-api/printk-formats.rst | 3 ++- lib/test_printf.c | 23 ++++++++++++++++++++++ lib/vsprintf.c | 24 +++++++++++++++++++++++ mm/debug.c | 7 +++++++ mm/internal.h | 1 + 5 files changed, 57 insertions(+), 1 deletion(-)