Re: [PATCHv2] arm64: initialize per-cpu offsets earlier

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Qian,

On Fri, Oct 09, 2020 at 09:51:15AM +0100, Will Deacon wrote:
> On Thu, Oct 08, 2020 at 09:18:24PM -0400, Qian Cai wrote:
> > On Mon, 2020-10-05 at 17:43 +0100, Mark Rutland wrote:
> > > The current initialization of the per-cpu offset register is difficult
> > > to follow and this initialization is not always early enough for
> > > upcoming instrumentation with KCSAN, where the instrumentation callbacks
> > > use the per-cpu offset.
> > > 
> > > To make it possible to support KCSAN, and to simplify reasoning about
> > > early bringup code, let's initialize the per-cpu offset earlier, before
> > > we run any C code that may consume it. To do so, this patch adds a new
> > > init_this_cpu_offset() helper that's called before the usual
> > > primary/secondary start functions. For consistency, this is also used to
> > > re-initialize the per-cpu offset after the runtime per-cpu areas have
> > > been allocated (which can change CPU0's offset).
> > > 
> > > So that init_this_cpu_offset() isn't subject to any instrumentation that
> > > might consume the per-cpu offset, it is marked with noinstr, preventing
> > > instrumentation.
> > > 
> > > Signed-off-by: Mark Rutland <mark.rutland@xxxxxxx>
> > > Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> > > Cc: James Morse <james.morse@xxxxxxx>
> > > Cc: Will Deacon <will@xxxxxxxxxx>
> > 
> > Reverting this commit on the top of today's linux-next fixed an issue that
> > Thunder X2 is unable to boot:
> > 
> > .config: https://gitlab.com/cailca/linux-mm/-/blob/master/arm64.config

Sorry about this. :/

Will, to save you reading all the below, I think the right thing to do
for now is to revert this.

Building with that config, I see a boot time-hang on the primary CPU
under QEMU TCG, with or without VHE (in both cases, a stuck in a
recursive synchronous exception). With the patch reverted, the kernel
boots.

Looking at the assembly, task_cpu() gets instrumented (which puts this
patch on dodgy ground generally and I think warrants the revert), but as
it's instrumented with KASAN_INLINE that doesn't immediately explain the
issue since the shadow should be up and so we shouldn't call the report
function. I'll dig into this some more.

Assembly dumps below; for init_this_cpu_offset the reference to
page_wait_table+0x4d00 is generating the address for __per_cpu_offset,
but objdump doesn't have enough info to resolve that nicely.

| ffffa00011143bc0 <init_this_cpu_offset>:
| ffffa00011143bc0:       a9bf7bfd        stp     x29, x30, [sp, #-16]!
| ffffa00011143bc4:       d5384100        mrs     x0, sp_el0
| ffffa00011143bc8:       910003fd        mov     x29, sp
| ffffa00011143bcc:       97bb29b9        bl      ffffa0001000e2b0 <task_cpu>
| ffffa00011143bd0:       f0004ea1        adrp    x1, ffffa00011b1a000 <page_wait_table+0x4d00>
| ffffa00011143bd4:       913c6021        add     x1, x1, #0xf18
| ffffa00011143bd8:       f8605820        ldr     x0, [x1, w0, uxtw #3]
| ffffa00011143bdc:       97bb29b1        bl      ffffa0001000e2a0 <set_my_cpu_offset>
| ffffa00011143be0:       a8c17bfd        ldp     x29, x30, [sp], #16
| ffffa00011143be4:       d65f03c0        ret

| ffffa0001000e2b0 <task_cpu>:
| ffffa0001000e2b0:       a9be7bfd        stp     x29, x30, [sp, #-32]!
| ffffa0001000e2b4:       d2d40001        mov     x1, #0xa00000000000             // #175921860444160
| ffffa0001000e2b8:       f2fbffe1        movk    x1, #0xdfff, lsl #48
| ffffa0001000e2bc:       910003fd        mov     x29, sp
| ffffa0001000e2c0:       f9000bf3        str     x19, [sp, #16]
| ffffa0001000e2c4:       aa0003f3        mov     x19, x0
| ffffa0001000e2c8:       91012000        add     x0, x0, #0x48
| ffffa0001000e2cc:       52800062        mov     w2, #0x3                        // #3
| ffffa0001000e2d0:       d343fc03        lsr     x3, x0, #3
| ffffa0001000e2d4:       38e16861        ldrsb   w1, [x3, x1]
| ffffa0001000e2d8:       7100003f        cmp     w1, #0x0
| ffffa0001000e2dc:       7a411041        ccmp    w2, w1, #0x1, ne  // ne = any
| ffffa0001000e2e0:       540000aa        b.ge    ffffa0001000e2f4 <task_cpu+0x44>  // b.tcont
| ffffa0001000e2e4:       b9404a60        ldr     w0, [x19, #72]
| ffffa0001000e2e8:       f9400bf3        ldr     x19, [sp, #16]
| ffffa0001000e2ec:       a8c27bfd        ldp     x29, x30, [sp], #32
| ffffa0001000e2f0:       d65f03c0        ret
| ffffa0001000e2f4:       9415af1f        bl      ffffa00010579f70 <__asan_report_load4_noabort>
| ffffa0001000e2f8:       17fffffb        b       ffffa0001000e2e4 <task_cpu+0x34>
| ffffa0001000e2fc:       d503201f        nop

| ffffa0001000e2a0 <set_my_cpu_offset>:
| ffffa0001000e2a0:       d518d080        msr     tpidr_el1, x0
| ffffa0001000e2a4:       d65f03c0        ret
| ffffa0001000e2a8:       d503201f        nop
| ffffa0001000e2ac:       d503201f        nop

Thanks,
Mark.



[Index of Archives]     [Linux Kernel]     [Linux USB Development]     [Yosemite News]     [Linux SCSI]

  Powered by Linux