Re: [PATCH 5/5] scripts/gdb: Add meminfo command

From: Kieran Bingham
Date: Sat Jan 23 2016 - 19:30:58 EST




On 23/01/16 15:21, Jan Kiszka wrote:
> On 2016-01-20 12:15, Kieran Bingham wrote:
>> Provide an equivalent of /proc/meminfo which should be available from
>> core dumps, or crashed kernels. This should allow a debugger to identify
>> if memory pressures were applicable in the instance of their issue
>>
>> Signed-off-by: Kieran Bingham <kieran.bingham@xxxxxxxxxx>
>> ---
>>
>> This command has proved to be much more difficult that I first thought it
>> would be!
>>
>> It also poses a couple of interesting issues, which is why I submit this
>> patch in a much more unfinished form.
>
> Yeah, seems so - see below ;)
>
>>
>> The meminfo implementation at fs/proc/meminfo.c makes several function calls
>> to collate information, which makes duplicating here more difficult.
>>
>> I suspect the best option here is to not present lines of which we can not
>> obtain accurate data for, (much better than presenting inaccurate information)
>>
>> Would this go in agreement with you?
>
> I didn't grab the use cases yet, so just a general suggestion: if the
> inaccurate information may still have some value, you can mark it as
> inaccurate and still print it.
>
>>
>> Finally, do you have any ideas on the best way to manage code which
>> is #ifdef'd on kernel config options? (#ifdef CONFIG_HIGHMEM for example).
>>
>> In a similar vein to the constants.py, I considered that we could iterate all
>> of the kernel configuration options and store them in a dictionary some how.
>>
>> That may be awkward, however, and I wondered what ideas anyone had!
>
> Why not convert the (relevant) configs inside constants.py into a
> python variable? CONFIG_IS_ENABLED should make this easy.


Aha, Excellent. I think I must not have seen the wood-for-the-trees on
this one!

Had a play, and I can't use IS_ENABLED(x) directly, as it generates (0
|| 1) from IS_BUILTIN and IS_MODULE - and which is not compatible with
python. However, in our use-case - we actually only want to switch on
BUILTIN's anyway, so I will use that.

If other use cases come up later, we can always extend to create our own
Python version of IS_ENABLED

>>
>>
>> scripts/gdb/linux/constants.py.in | 22 +++++
>> scripts/gdb/linux/proc.py | 173 ++++++++++++++++++++++++++++++++++++++
>> 2 files changed, 195 insertions(+)
>>
>> diff --git a/scripts/gdb/linux/constants.py.in b/scripts/gdb/linux/constants.py.in
>> index 739a15d2e984..306bd601ae4e 100644
>> --- a/scripts/gdb/linux/constants.py.in
>> +++ b/scripts/gdb/linux/constants.py.in
>> @@ -12,8 +12,15 @@
>> *
>> */
>>
>> +#include <asm/page.h>
>> +#include <asm/pgtable.h>
>> +#include <asm/thread_info.h>
>> +
>> #include <linux/fs.h>
>> +#include <linux/swap.h>
>> #include <linux/mount.h>
>> +#include <linux/vmalloc.h>
>> +
>>
>> /* We need to stringify expanded macros so that they can be parsed */
>>
>> @@ -41,3 +48,18 @@ LX_MNT_NOATIME = MNT_NOATIME
>> LX_MNT_NODIRATIME = MNT_NODIRATIME
>> LX_MNT_RELATIME = MNT_RELATIME
>>
>> +/* asm/page.h */
>> +LX_PAGE_SHIFT = XSTRING(PAGE_SHIFT)
>> +lx_page_shift = gdb.parse_and_eval(LX_PAGE_SHIFT)
>> +
>> +/* asm/thread_info.h */
>> +LX_THREAD_SIZE = XSTRING(THREAD_SIZE)
>> +lx_thread_size = gdb.parse_and_eval(LX_THREAD_SIZE)
>> +
>> +/* linux/vmalloc.h */
>> +LX_VMALLOC_TOTAL = XSTRING(VMALLOC_TOTAL)
>> +lx_vmalloc_total = gdb.parse_and_eval(LX_VMALLOC_TOTAL)
>> +
>> +/* linux/swap.h */
>> +LX_MAX_SWAPFILES = XSTRING(MAX_SWAPFILES)
>> +lx_max_swapfiles = gdb.parse_and_eval(LX_MAX_SWAPFILES)
>> diff --git a/scripts/gdb/linux/proc.py b/scripts/gdb/linux/proc.py
>> index b79ce2a33a3d..ac9e1aac2403 100644
>> --- a/scripts/gdb/linux/proc.py
>> +++ b/scripts/gdb/linux/proc.py
>> @@ -206,3 +206,176 @@ values of that process namespace"""
>> info_opts(MNT_INFO, m_flags)))
>>
>> LxMounts()
>> +
>> +
>> +bdev_type = utils.CachedType("struct block_device")
>> +
>> +
>> +class LxMeminfo(gdb.Command):
>> + """ Identify the memory usage, statistics, and availability
>> +
>> +Equivalent to cat /proc/meminfo on a running target """
>> +
>> + def __init__(self):
>> + super(LxMeminfo, self).__init__("lx-meminfo", gdb.COMMAND_DATA)
>> +
>> + def K(self, val):
>> + # Convert from PAGES to KB
>> + return int(val << (constants.lx_page_shift - 10))
>> +
>> + def page_K(self, remote_value):
>> + # Obtain page value, and Convert from PAGES to KB
>> + val = int(gdb.parse_and_eval(remote_value))
>> + return self.K(val)
>> +
>> + def gps(self, enum_zone_stat_item):
>> + # Access the Global Page State structure
>> + # I would prefer to read this structure in one go and then index
>> + # from the enum. But we can't determine the enum values with out
>> + # a call to GDB anyway so we may as well take the easy route and
>> + # get the value.
>> + remote_value = "vm_stat[" + enum_zone_stat_item + "].counter"
>> + return int(gdb.parse_and_eval(remote_value))
>> +
>> + def gps_K(self, enum_zone_stat_item):
>> + return self.K(self.gps(enum_zone_stat_item))
>> +
>> + def nr_blockdev_pages(self):
>> + bdevs_head = gdb.parse_and_eval("all_bdevs")
>> + pages = 0
>> + for bdev in lists.items(bdev_type, "bd_list", bdevs_head):
>> + pages += bdev['bd_inode']['i_mapping']['nrpages']
>> + return pages
>> +
>> + def total_swapcache_pages(self):
>> + pages = 0
>> + for i in range(0, constants.lx_max_swapfiles):
>> + swap_space = "swapper_spaces[" + str(i) + "].nrpages"
>> + pages += int(gdb.parse_and_eval(swap_space))
>> + return pages
>> +
>> + def vm_commit_limit(self, totalram_pages):
>> + overcommit = int(gdb.parse_and_eval("sysctl_overcommit_kbytes"))
>> + overcommit_ratio = int(gdb.parse_and_eval("sysctl_overcommit_ratio"))
>> + total_swap_pages = int(gdb.parse_and_eval("total_swap_pages"))
>> + hugetlb_total_pages = 0 # hugetlb_total_pages()!!
>> +
>> + if overcommit:
>> + allowed = overcommit >> (constants.lx_page_shift - 10)
>> + else:
>> + allowed = ((totalram_pages - hugetlb_total_pages *
>> + overcommit_ratio / 100))
>> +
>> + allowed += total_swap_pages
>> + return allowed
>> +
>> + # Main lx-meminfo command execution
>> + def invoke(self, arg, from_tty):
>> + totalram = int(gdb.parse_and_eval("totalram_pages"))
>> + freeram = self.gps("NR_FREE_PAGES")
>> + reclaimable = self.gps("NR_SLAB_RECLAIMABLE")
>> + unreclaimable = self.gps("NR_SLAB_UNRECLAIMABLE")
>> + slab = reclaimable + unreclaimable
>> + # for_each_zone(zone)
>> + # wmark_low += zone->watermark[WMARK_LOW];
>> + wmark_low = 0 # Zone parsing is unimplemented
>> +
>> + available = freeram - wmark_low
>> + available += reclaimable - min(reclaimable / 2, wmark_low)
>> +
>> + bufferram = self.nr_blockdev_pages()
>
> Something goes wrong here:
>
> (gdb) lx-meminfo
> Traceback (most recent call last):
> File "/data/linux/build-dbg/scripts/gdb/linux/proc.py", line 286, in invoke
> bufferram = self.nr_blockdev_pages()
> File "/data/linux/build-dbg/scripts/gdb/linux/proc.py", line 246, in nr_blockdev_pages
> for bdev in lists.items(bdev_type, "bd_list", bdevs_head):
> File "/data/linux/build-dbg/scripts/gdb/linux/lists.py", line 26, in items
> yield utils.container_of(next_item, list_type, list_location)
> File "/data/linux/build-dbg/scripts/gdb/linux/utils.py", line 52, in container_of
> offset_of(typeobj, member)).cast(typeobj)
> File "/data/linux/build-dbg/scripts/gdb/linux/utils.py", line 46, in offset_of
> element = gdb.Value(0).cast(typeobj)
> RuntimeError: Argument must be a type.
> Error occurred in Python command: Argument must be a type.
>
> If you need my kernel config to reproduce, just let me know.


Yes please, send me a config, and I'll try to reproduce.


What is your test environment by the way?
I am running in QEmu for convenience mostly at the moment.

I hope to set up some automated scripts to test on different arch's too.

--
Kieran


>
>> + total_swapcache_pages = self.total_swapcache_pages()
>> +
>> + file_pages = self.gps("NR_FILE_PAGES")
>> + cached = file_pages - total_swapcache_pages - bufferram
>> +
>> + # LRU Pages
>> + active_pages_anon = self.gps("NR_ACTIVE_ANON")
>> + inactive_pages_anon = self.gps("NR_INACTIVE_ANON")
>> + active_pages_file = self.gps("NR_ACTIVE_FILE")
>> + inactive_pages_file = self.gps("NR_INACTIVE_FILE")
>> + unevictable_pages = self.gps("NR_UNEVICTABLE")
>> + active_pages = active_pages_anon + active_pages_file
>> + inactive_pages = inactive_pages_anon + inactive_pages_file
>> +
>> + totalhigh = int(gdb.parse_and_eval("totalhigh_pages"))
>> + # We can't run this on a core dump file ...
>> + # if running target ()
>> + freehigh = int(gdb.parse_and_eval("nr_free_highpages()"))
>> + # else freehigh = 0
>> +
>> + kernelstack = int(self.gps("NR_KERNEL_STACK") *
>> + constants.lx_thread_size / 1024)
>> +
>> + commitlimit = self.vm_commit_limit(totalram)
>> + committed_as = int(gdb.parse_and_eval("vm_committed_as.count"))
>> +
>> + vmalloc_total = int(constants.lx_vmalloc_total >> 10)
>> +
>> + gdb.write(
>> + "MemTotal: {:8d} kB\n".format(self.K(totalram)) +
>> + "MemFree: {:8d} kB\n".format(self.K(freeram)) +
>> + "MemAvailable: {:8d} kB\n".format(self.K(available)) +
>> + "Buffers: {:8d} kB\n".format(self.K(bufferram)) +
>> + "Cached: {:8d} kB\n".format(self.K(cached)) +
>> + "SwapCached: {:8d} kB\n".format(self.K(total_swapcache_pages)) +
>> + "Active: {:8d} kB\n".format(self.K(active_pages)) +
>> + "Inactive: {:8d} kB\n".format(self.K(inactive_pages)) +
>> + "Active(anon): {:8d} kB\n".format(self.K(active_pages_anon)) +
>> + "Inactive(anon): {:8d} kB\n".format(self.K(inactive_pages_anon)) +
>> + "Active(file): {:8d} kB\n".format(self.K(active_pages_file)) +
>> + "Inactive(file): {:8d} kB\n".format(self.K(inactive_pages_file)) +
>> + "Unevictable: {:8d} kB\n".format(self.K(unevictable_pages)) +
>> + "Mlocked: {:8d} kB\n".format(self.gps_K("NR_MLOCK"))
>> + )
>> + # ifdef CONFIG_HIGHMEM || core dump?
>> + gdb.write(
>> + "HighTotal: {:8d} kB\n".format(self.K(totalhigh)) +
>> + "HighFree: {:8d} kB\n".format(self.K(freehigh)) +
>> + "LowTotal: {:8d} kB\n".format(self.K(totalram-totalhigh)) +
>> + "LowFree: {:8d} kB\n".format(self.K(freeram-freehigh))
>> + )
>> + # endif
>> + # ifndef CONFIG_MMU
>> + # gdb.write(
>> + # mmap_pages_allocated
>> + # )
>> + # endif
>> + gdb.write(
>> + "SwapTotal: {:8d} kB\n".format(self.K(0)) +
>> + "SwapFree: {:8d} kB\n".format(self.K(0)) +
>> + "Dirty: {:8d} kB\n".format(self.gps_K("NR_FILE_DIRTY")) +
>> + "Writeback: {:8d} kB\n".format(self.gps_K("NR_WRITEBACK")) +
>> + "AnonPages: {:8d} kB\n".format(self.gps_K("NR_ANON_PAGES")) +
>> + "Mapped: {:8d} kB\n".format(self.gps_K("NR_FILE_MAPPED")) +
>> + "Shmem: {:8d} kB\n".format(self.gps_K("NR_SHMEM")) +
>> + "Slab: {:8d} kB\n".format(self.K(slab)) +
>> + "SReclaimable: {:8d} kB\n".format(self.K(reclaimable)) +
>> + "SUnreclaim: {:8d} kB\n".format(self.K(unreclaimable)) +
>> + "KernelStack: {:8d} kB\n".format(kernelstack) +
>> + "PageTables: {:8d} kB\n".format(self.gps_K("NR_PAGETABLE"))
>> + )
>> +
>> + # if CONFIG_QUICKLIST
>> + # "Quicklists: {:8d} kB\n".format(self.K(quicklist)))
>> +
>> + gdb.write(
>> + "NFS_Unstable: {:8d} kB\n".format(self.gps_K("NR_UNSTABLE_NFS")) +
>> + "Bounce: {:8d} kB\n".format(self.gps_K("NR_BOUNCE")) +
>> + "WritebackTmp: {:8d} kB\n".format(self.gps_K("NR_WRITEBACK_TEMP")) +
>> + "CommitLimit: {:8d} kB\n".format(self.K(commitlimit)) +
>> + "Committed_AS: {:8d} kB\n".format(self.K(committed_as)) +
>> + "VmallocTotal: {:8d} kB\n".format(vmalloc_total) +
>> + "VmallocUsed: {:8d} kB\n".format(0) +
>> + "VmallocChunk: {:8d} kB\n".format(0)
>> + )
>> + # if CONFIG_MEMORY_FAILURE
>> + # "HardwareCorrupted: %5lu kB\n"
>> + # ifdef CONFIG_CMA
>> + totalcma_pages = int(gdb.parse_and_eval("totalcma_pages"))
>> + gdb.write(
>> + "CmaTotal: {:8d} kB\n".format(self.K(totalcma_pages)) +
>> + "CmaFree: {:8d} kB\n".format(self.gps_K("NR_FREE_CMA_PAGES"))
>> + )
>> +
>> +LxMeminfo()
>>
>
> Jan
>