Re: [PATCH 11/11] x86/ucode/amd: Rework parsing logic in cpu_request_microcode()

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 31/03/2020 16:07, Jan Beulich wrote:
> On 31.03.2020 12:05, Andrew Cooper wrote:
>> @@ -269,55 +265,25 @@ static int apply_microcode(const struct 
>> microcode_patch *patch)
>>      return 0;
>>  }
>>  
>> -static int scan_equiv_cpu_table(
>> -    const void *data,
>> -    size_t size_left,
>> -    size_t *offset)
>> +static int scan_equiv_cpu_table(const struct container_equiv_table *et)
>>  {
>>      const struct cpu_signature *sig = &this_cpu(cpu_sig);
>> -    const struct mpbhdr *mpbuf;
>> -    const struct equiv_cpu_entry *eq;
>> -    unsigned int i, nr;
>> -
>> -    if ( size_left < (sizeof(*mpbuf) + 4) ||
>> -         (mpbuf = data + *offset + 4,
>> -          size_left - sizeof(*mpbuf) - 4 < mpbuf->len) )
>> -    {
>> -        printk(XENLOG_WARNING "microcode: No space for equivalent cpu 
>> table\n");
>> -        return -EINVAL;
>> -    }
>> -
>> -    *offset += mpbuf->len + CONT_HDR_SIZE;  /* add header length */
>> -
>> -    if ( mpbuf->type != UCODE_EQUIV_CPU_TABLE_TYPE )
>> -    {
>> -        printk(KERN_ERR "microcode: Wrong microcode equivalent cpu table 
>> type field\n");
>> -        return -EINVAL;
>> -    }
>> -
>> -    if ( mpbuf->len == 0 || mpbuf->len % sizeof(*eq) ||
>> -         (eq = (const void *)mpbuf->data,
>> -          nr = mpbuf->len / sizeof(*eq),
>> -          eq[nr - 1].installed_cpu) )
> Did this last check get lost? I can't seem to be able to identify
> any possible replacement.

Given the lack of a spec, I'm unsure whether to keep it or not.

It is necessary in the backport of patch 1, because find_equiv_cpu_id()
doesn't have mpbuf->len to hand, and relies on the sentinel to find the
end of the table.

OTOH, the new logic will cope perfectly well without a sentinel.

>
>>  static struct microcode_patch *cpu_request_microcode(const void *buf, 
>> size_t size)
>>  {
>>      const struct microcode_patch *saved = NULL;
>>      struct microcode_patch *patch = NULL;
>> -    size_t offset = 0, saved_size = 0;
>> +    size_t saved_size = 0;
>>      int error = 0;
>> -    unsigned int cpu = smp_processor_id();
>> -    const struct cpu_signature *sig = &per_cpu(cpu_sig, cpu);
>>  
>> -    if ( size < 4 ||
>> -         *(const uint32_t *)buf != UCODE_MAGIC )
>> +    while ( size )
>>      {
>> -        printk(KERN_ERR "microcode: Wrong microcode patch file magic\n");
>> -        error = -EINVAL;
>> -        goto out;
>> -    }
>> -
>> -    /*
>> -     * Multiple container file support:
>> -     * 1. check if this container file has equiv_cpu_id match
>> -     * 2. If not, fast-fwd to next container file
>> -     */
>> -    while ( offset < size )
>> -    {
>> -        error = scan_equiv_cpu_table(buf, size - offset, &offset);
>> -
>> -        if ( !error || error != -ESRCH )
>> -            break;
>> +        const struct container_equiv_table *et;
>> +        bool skip_ucode;
>>  
>> -        error = container_fast_forward(buf, size - offset, &offset);
>> -        if ( error == -ENODATA )
>> +        if ( size < 4 || *(const uint32_t *)buf != UCODE_MAGIC )
>>          {
>> -            ASSERT(offset == size);
>> +            printk(XENLOG_ERR "microcode: Wrong microcode patch file 
>> magic\n");
>> +            error = -EINVAL;
>>              break;
>>          }
>> -        if ( error )
>> +
>> +        /* Move over UCODE_MAGIC. */
>> +        buf  += 4;
>> +        size -= 4;
>> +
>> +        if ( size < sizeof(*et) ||
>> +             (et = buf)->type != UCODE_EQUIV_CPU_TABLE_TYPE ||
>> +             size - sizeof(*et) < et->len ||
>> +             et->len % sizeof(et->eq[0]) )
>>          {
>> -            printk(KERN_ERR "microcode: CPU%d incorrect or corrupt 
>> container file\n"
>> -                   "microcode: Failed to update patch level. "
>> -                   "Current lvl:%#x\n", cpu, sig->rev);
>> +            printk(XENLOG_ERR "microcode: Bad equivalent cpu table\n");
>> +            error = -EINVAL;
>>              break;
>>          }
>> -    }
>>  
>> -    if ( error )
>> -    {
>> -        /*
>> -         * -ENODATA here means that the blob was parsed fine but no matching
>> -         * ucode was found. Don't return it to the caller.
>> -         */
>> -        if ( error == -ENODATA )
>> -            error = 0;
>> -
>> -        goto out;
>> -    }
>> +        /* Move over the Equiv table. */
>> +        buf  += sizeof(*et) + et->len;
>> +        size -= sizeof(*et) + et->len;
>> +
>> +        error = scan_equiv_cpu_table(et);
>> +        if ( error && error != -ESRCH )
>> +            break;
> With this the only non-zero value left for error is -ESRCH.
> Hence ...
>
>> +        /* -ESRCH means no applicable microcode in this container. */
>> +        skip_ucode = error == -ESRCH;
> ... perhaps omit the "== -ESRCH" here, moving the comment up
> ahead of the if()?

That doesn't work, because you've got to reset error to 0 somewhere (to
avoid it leaking out if you don't find suitable microcode), and it can't
be before checking for errors in general.  It can't easily become a
conditional because skip_ucode needs setting unconditionally.

The only other correct way I can see of arranging this code would be

    skip_ucode = error == -ESRCH;
    if ( error == -ESRCH )
        error = 0;
    else if ( error )
        break;

which doesn't look to be better.

I have been debating quite heavily whether -ESRCH is best here, or using
-ve, 0 and 1.  However, this doesn't lead to prettier code AFAICT, and
gains an ambiguous use for a variable named "error".

~Andrew