Date:   Fri, 10 Sep 2021 23:46:29 -0700 (PDT)
From:   Palmer Dabbelt <palmer@...belt.com>
To:     rppt@...nel.org
CC:     heinrich.schuchardt@...onical.com, wangkefeng.wang@...wei.com,
        linux-riscv@...ts.infradead.org, linux-kernel@...r.kernel.org,
        Paul Walmsley <paul.walmsley@...ive.com>, aou@...s.berkeley.edu
Subject:     Re: [PATCH 1/1] riscv: select CONFIG_ARCH_KEEP_MEMBLOCK

On Mon, 16 Aug 2021 11:50:26 PDT (-0700), rppt@...nel.org wrote:
> Hi Heinrich,
>
> On Mon, Aug 16, 2021 at 05:52:11PM +0200, Heinrich Schuchardt wrote:
>> On 8/16/21 5:30 PM, Kefeng Wang wrote:
>> >
>> > On 2021/8/16 22:47, Heinrich Schuchardt wrote:
>> > > For analyzing memory blocks we can either use the memblock=debug command
>> > > line argument which creates massive output or a debug file system.
>> > >
>> > > Select CONFIG_ARCH_KEEP_MEMBLOCK to provide a debugfs at
>> > > /sys/kernel/debug/memblock to analyze memory blocks. The
>> > > same is already done for arm, arm64, mips, powerpc.
>> > >
>> > > The actual provisioning of the file system depends on CONFIG_DEBUG_FS.
>> >
>> > Hi，for riscv, it don't use memblock(eg, no provide pfn_valid to use
>> > memblock),
>> >
>> > we could call memblock_discard() to discard memblock private memory to save
>> >
>> > some memory, right?  So I think we don't need this config for now.
>>
>> What do you mean by "it don't use memblock?
>>
>> If you boot the HiFive Unmatched with memblock=debug you will see output
>> like the one below. This is the information that you can see in the debug
>> file system if CONFIG_DEBUG_FS and CONFIG_ARCH_KEEP_MEMBLOCK will be
>> enabled.
>
> Indeed having this info in debugfs is more convenient, but it comes with a
> cost of several kilobytes of code and data for *every* riscv build.
>
> Why parsing "memblock=debug" output is not enough?
> What is the use-case that justifies the costs associated with having the
> debugfs entries?

If someone can provide a use case I'm happy to look at how to enable it.  
Otherwise I think we should just stick with the standard way of doing 
things, which according to 350e88bad496 ("mm: memblock: make keeping 
memblock memory opt-in rather than opt-out") seems to be that 
ARCH_KEEP_MEMBLOCK is for ports that need that information, not just 
those that pass it through to userspace.

>> [    0.000000] MEMBLOCK configuration:
>> [    0.000000]  memory size = 0x00000003ffe00000 reserved size =
>> 0x000000001290fb70
>> [    0.000000]  memory.cnt  = 0x25
>
> BTW, it seems that the memory allocation by firmware is, well, suboptimal
> and begs for fixes. I don't see why all these NOMAP areas cannot be put in
> one contiguous chunk. I doubt they need to be NOMAP at all, but mysterious
> the ways of ACPI.

We talked about this a bit in IRC and another thread, but it looks like 
this mess has ended up in the specs.  From my reading that was likely 
the result of a blind copy/paste from some other architecture, but it's 
in a released spec so it's cannon and I'm not really interested in 
trying to change it as I'm tired of getting yelled at by the foundation 
folks.

>> [    0.000000]  memory[0x0]     [0x0000000080200000-0x00000000fe6e2fff],
>> 0x000000007e4e3000 bytes flags: 0x0
>> [    0.000000]  memory[0x1]     [0x00000000fe6e3000-0x00000000fe6e5fff],
>> 0x0000000000003000 bytes flags: 0x4
>> [    0.000000]  memory[0x2]     [0x00000000fe6e6000-0x00000000fe709fff],
>> 0x0000000000024000 bytes flags: 0x0
>> [    0.000000]  memory[0x3]     [0x00000000fe70a000-0x00000000fe70bfff],
>> 0x0000000000002000 bytes flags: 0x4
>> [    0.000000]  memory[0x4]     [0x00000000fe70c000-0x00000000fe70ffff],
>> 0x0000000000004000 bytes flags: 0x0
>> [    0.000000]  memory[0x5]     [0x00000000fe710000-0x00000000fe710fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x6]     [0x00000000fe711000-0x00000000fe712fff],
>> 0x0000000000002000 bytes flags: 0x0
>> [    0.000000]  memory[0x7]     [0x00000000fe713000-0x00000000fe716fff],
>> 0x0000000000004000 bytes flags: 0x4
>> [    0.000000]  memory[0x8]     [0x00000000fe717000-0x00000000fe717fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x9]     [0x00000000fe718000-0x00000000fe71cfff],
>> 0x0000000000005000 bytes flags: 0x4
>> [    0.000000]  memory[0xa]     [0x00000000fe71d000-0x00000000fe71dfff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0xb]     [0x00000000fe71e000-0x00000000fe71efff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0xc]     [0x00000000fe71f000-0x00000000fe71ffff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0xd]     [0x00000000fe720000-0x00000000fe720fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0xe]     [0x00000000fe721000-0x00000000fe721fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0xf]     [0x00000000fe722000-0x00000000fe722fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x10]    [0x00000000fe723000-0x00000000fe723fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x11]    [0x00000000fe724000-0x00000000fe724fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x12]    [0x00000000fe725000-0x00000000fe725fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x13]    [0x00000000fe726000-0x00000000fe726fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x14]    [0x00000000fe727000-0x00000000fe728fff],
>> 0x0000000000002000 bytes flags: 0x0
>> [    0.000000]  memory[0x15]    [0x00000000fe729000-0x00000000fe729fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x16]    [0x00000000fe72a000-0x00000000fe72afff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x17]    [0x00000000fe72b000-0x00000000fe72cfff],
>> 0x0000000000002000 bytes flags: 0x4
>> [    0.000000]  memory[0x18]    [0x00000000fe72d000-0x00000000fe72dfff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x19]    [0x00000000fe72e000-0x00000000fe72efff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x1a]    [0x00000000fe72f000-0x00000000fe72ffff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x1b]    [0x00000000fe730000-0x00000000fe730fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x1c]    [0x00000000fe731000-0x00000000fe731fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x1d]    [0x00000000fe732000-0x00000000fe732fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x1e]    [0x00000000fe733000-0x00000000fe733fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  memory[0x1f]    [0x00000000fe734000-0x00000000fe734fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x20]    [0x00000000fe735000-0x00000000fe736fff],
>> 0x0000000000002000 bytes flags: 0x0
>> [    0.000000]  memory[0x21]    [0x00000000fe737000-0x00000000fe737fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x22]    [0x00000000fe738000-0x00000000fff60fff],
>> 0x0000000001829000 bytes flags: 0x0
>> [    0.000000]  memory[0x23]    [0x00000000fff61000-0x00000000fff61fff],
>> 0x0000000000001000 bytes flags: 0x4
>> [    0.000000]  memory[0x24]    [0x00000000fff62000-0x000000047fffffff],
>> 0x000000038009e000 bytes flags: 0x0
>> [    0.000000]  reserved.cnt  = 0xd
>> [    0.000000]  reserved[0x0]   [0x0000000080000000-0x000000008007ffff],
>> 0x0000000000080000 bytes flags: 0x0
>> [    0.000000]  reserved[0x1]   [0x0000000080200000-0x00000000815fffff],
>> 0x0000000001400000 bytes flags: 0x0
>> [    0.000000]  reserved[0x2]   [0x0000000087f00000-0x0000000087f05fff],
>> 0x0000000000006000 bytes flags: 0x0
>> [    0.000000]  reserved[0x3]   [0x00000000db0a2000-0x00000000db0a48b4],
>> 0x00000000000028b5 bytes flags: 0x0
>> [    0.000000]  reserved[0x4]   [0x00000000db2a2000-0x00000000dc709fff],
>> 0x0000000001468000 bytes flags: 0x0
>> [    0.000000]  reserved[0x5]   [0x00000000fe701000-0x00000000fe701fff],
>> 0x0000000000001000 bytes flags: 0x0
>> [    0.000000]  reserved[0x6]   [0x00000000fe704040-0x00000000fe70404f],
>> 0x0000000000000010 bytes flags: 0x0
>> [    0.000000]  reserved[0x7]   [0x000000046ffe1d40-0x000000047dfe253f],
>> 0x000000000e000800 bytes flags: 0x0
>> [    0.000000]  reserved[0x8]   [0x000000047dfe2548-0x000000047dfe2578],
>> 0x0000000000000031 bytes flags: 0x0
>> [    0.000000]  reserved[0x9]   [0x000000047dfe2580-0x000000047dfe25ae],
>> 0x000000000000002f bytes flags: 0x0
>> [    0.000000]  reserved[0xa]   [0x000000047dfe25b0-0x000000047dfe25de],
>> 0x000000000000002f bytes flags: 0x0
>> [    0.000000]  reserved[0xb]   [0x000000047dfe25e0-0x000000047dfefffb],
>> 0x000000000000da1c bytes flags: 0x0
>> [    0.000000]  reserved[0xc]   [0x000000047dff0000-0x000000047fffffff],
>> 0x0000000002010000 bytes flags: 0x0
>>
>> Best regards
>>
>> Heinrich
>>
>> >
>> > >
>> > > Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@...onical.com>
>> > > ---
>> > >   arch/riscv/Kconfig | 1 +
>> > >   1 file changed, 1 insertion(+)
>> > >
>> > > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
>> > > index 4f7b70ae7c31..a6e57614c3fd 100644
>> > > --- a/arch/riscv/Kconfig
>> > > +++ b/arch/riscv/Kconfig
>> > > @@ -31,6 +31,7 @@ config RISCV
>> > >       select ARCH_HAS_STRICT_KERNEL_RWX if MMU && !XIP_KERNEL
>> > >       select ARCH_HAS_STRICT_MODULE_RWX if MMU && !XIP_KERNEL
>> > >       select ARCH_HAS_TICK_BROADCAST if GENERIC_CLOCKEVENTS_BROADCAST
>> > > +    select ARCH_KEEP_MEMBLOCK
>> > >       select ARCH_OPTIONAL_KERNEL_RWX if ARCH_HAS_STRICT_KERNEL_RWX
>> > >       select ARCH_OPTIONAL_KERNEL_RWX_DEFAULT
>> > >       select ARCH_SUPPORTS_HUGETLBFS if MMU
>>
>>
>> _______________________________________________
>> linux-riscv mailing list
>> linux-riscv@...ts.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-riscv

Powered by blists - more mailing lists