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

Re: [PATCH v5 2/7] xen/riscv: set up fixmap mappings


  • To: oleksii.kurochko@xxxxxxxxx
  • From: Jan Beulich <jbeulich@xxxxxxxx>
  • Date: Fri, 30 Aug 2024 14:34:31 +0200
  • Autocrypt: addr=jbeulich@xxxxxxxx; keydata= xsDiBFk3nEQRBADAEaSw6zC/EJkiwGPXbWtPxl2xCdSoeepS07jW8UgcHNurfHvUzogEq5xk hu507c3BarVjyWCJOylMNR98Yd8VqD9UfmX0Hb8/BrA+Hl6/DB/eqGptrf4BSRwcZQM32aZK 7Pj2XbGWIUrZrd70x1eAP9QE3P79Y2oLrsCgbZJfEwCgvz9JjGmQqQkRiTVzlZVCJYcyGGsD /0tbFCzD2h20ahe8rC1gbb3K3qk+LpBtvjBu1RY9drYk0NymiGbJWZgab6t1jM7sk2vuf0Py O9Hf9XBmK0uE9IgMaiCpc32XV9oASz6UJebwkX+zF2jG5I1BfnO9g7KlotcA/v5ClMjgo6Gl MDY4HxoSRu3i1cqqSDtVlt+AOVBJBACrZcnHAUSuCXBPy0jOlBhxPqRWv6ND4c9PH1xjQ3NP nxJuMBS8rnNg22uyfAgmBKNLpLgAGVRMZGaGoJObGf72s6TeIqKJo/LtggAS9qAUiuKVnygo 3wjfkS9A3DRO+SpU7JqWdsveeIQyeyEJ/8PTowmSQLakF+3fote9ybzd880fSmFuIEJldWxp Y2ggPGpiZXVsaWNoQHN1c2UuY29tPsJgBBMRAgAgBQJZN5xEAhsDBgsJCAcDAgQVAggDBBYC AwECHgECF4AACgkQoDSui/t3IH4J+wCfQ5jHdEjCRHj23O/5ttg9r9OIruwAn3103WUITZee e7Sbg12UgcQ5lv7SzsFNBFk3nEQQCACCuTjCjFOUdi5Nm244F+78kLghRcin/awv+IrTcIWF hUpSs1Y91iQQ7KItirz5uwCPlwejSJDQJLIS+QtJHaXDXeV6NI0Uef1hP20+y8qydDiVkv6l IreXjTb7DvksRgJNvCkWtYnlS3mYvQ9NzS9PhyALWbXnH6sIJd2O9lKS1Mrfq+y0IXCP10eS FFGg+Av3IQeFatkJAyju0PPthyTqxSI4lZYuJVPknzgaeuJv/2NccrPvmeDg6Coe7ZIeQ8Yj t0ARxu2xytAkkLCel1Lz1WLmwLstV30g80nkgZf/wr+/BXJW/oIvRlonUkxv+IbBM3dX2OV8 AmRv1ySWPTP7AAMFB/9PQK/VtlNUJvg8GXj9ootzrteGfVZVVT4XBJkfwBcpC/XcPzldjv+3 HYudvpdNK3lLujXeA5fLOH+Z/G9WBc5pFVSMocI71I8bT8lIAzreg0WvkWg5V2WZsUMlnDL9 mpwIGFhlbM3gfDMs7MPMu8YQRFVdUvtSpaAs8OFfGQ0ia3LGZcjA6Ik2+xcqscEJzNH+qh8V m5jjp28yZgaqTaRbg3M/+MTbMpicpZuqF4rnB0AQD12/3BNWDR6bmh+EkYSMcEIpQmBM51qM EKYTQGybRCjpnKHGOxG0rfFY1085mBDZCH5Kx0cl0HVJuQKC+dV2ZY5AqjcKwAxpE75MLFkr wkkEGBECAAkFAlk3nEQCGwwACgkQoDSui/t3IH7nnwCfcJWUDUFKdCsBH/E5d+0ZnMQi+G0A nAuWpQkjM1ASeQwSHEeAWPgskBQL
  • Cc: Alistair Francis <alistair.francis@xxxxxxx>, Bob Eshleman <bobbyeshleman@xxxxxxxxx>, Connor Davis <connojdavis@xxxxxxxxx>, Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, Julien Grall <julien@xxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, xen-devel@xxxxxxxxxxxxxxxxxxxx
  • Delivery-date: Fri, 30 Aug 2024 12:34:44 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>

On 30.08.2024 13:55, oleksii.kurochko@xxxxxxxxx wrote:
> On Tue, 2024-08-27 at 12:29 +0200, Jan Beulich wrote:
>>> @@ -81,6 +82,18 @@ static inline void flush_page_to_ram(unsigned
>>> long mfn, bool sync_icache)
>>>       BUG_ON("unimplemented");
>>>   }
>>>   
>>> +/* Write a pagetable entry. */
>>> +static inline void write_pte(pte_t *p, pte_t pte)
>>> +{
>>> +    write_atomic(p, pte);
>>> +}
>>> +
>>> +/* Read a pagetable entry. */
>>> +static inline pte_t read_pte(pte_t *p)
>>> +{
>>> +    return read_atomic(p);
>>
>> This only works because of the strange type trickery you're playing
>> in
>> read_atomic(). Look at x86 code - there's a strict expectation that
>> the
>> type can be converted to/from unsigned long. And page table accessors
>> are written with that taken into consideration. Same goes for
>> write_pte()
>> of course, with the respective comment on the earlier patch in mind.
>>
>> Otoh I see that Arm does something very similar. If you have a strong
>> need / desire to follow that, then please at least split the two
>> entirely separate aspects that patch 1 presently changes both in one
>> go.
> I am not 100% sure that type trick could be dropped easily for RISC-V:
> 1. I still need the separate C function for proper #ifdef-ing:
>    #ifndef CONFIG_RISCV_32
>        case 8: *(uint32_t *)res = readq_cpu(p); break;
>    #endif
>    
> 2. Because of the point 1 the change should be as following:
>    -#define read_atomic(p) ({                                   \
>    -    union { typeof(*(p)) val; char c[sizeof(*(p))]; } x_;   \
>    -    read_atomic_size(p, x_.c, sizeof(*(p)));                \
>    -    x_.val;                                                 \
>    +#define read_atomic(p) ({                                 \
>    +    unsigned long x_;                                     \
>    +    read_atomic_size(p, &x_, sizeof(*(p)));               \
>    +    (typeof(*(p)))x_;                                     \
>     })
>    But after that I think it will be an error: "conversion to non-scalar
>    type requested" in the last line as *p points to pte_t.
>    
>    and we can't just in read_pte() change to:
>    static inline pte_t read_pte(pte_t *p)
>    {
>        return read_atomic(&p->pte);
>    }
>    As in this cases it started it will return unsigned long but function
>    expects pte_t.

Of course.

> As an option read_pte() can be updated to:
>    /* Read a pagetable entry. */
>    static inline pte_t read_pte(pte_t *p)
>    {
>        return (pte_t) { .pte = read_atomic(&p->pte) };
>    }

That's what's needed.

>    But I am not sure that it is better then just have a union trick inside
>    read_atomic() and then just have read_atomic(p) for read_pte().

It's largely up to you. My main request is that things end up / remain
consistent. Which way round is secondary, and often merely a matter of
suitably justifying the choice made.

Jan



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.