Re: [PATCH 2/3] powperc/mm: read TLB Block Invalidate Characteristics

[Aneesh Kumar K.V]

On 8/30/19 5:37 PM, Laurent Dufour wrote:
> The PAPR document specifies the TLB Block Invalidate Characteristics which
> is telling which couple base page size / page size is supported by the
> H_BLOCK_REMOVE hcall.
>
> A new set of feature is added to the mmu_psize_def structure to record per
> base page size which page size is supported by H_BLOCK_REMOVE.
>
> A new init service is added to read the characteristics. The size of the
> buffer is set to twice the number of known page size, plus 10 bytes to
> ensure we have enough place.


So this is not really the base page size/actual page size combination. 
This is related to H_BLOCK_REMOVE hcall, block size supported by that 
HCALL and what page size combination is supported with that specific 
block size.

We should add that TLB block invalidate characteristics format in this 
patch.


> Signed-off-by: Laurent Dufour <ldufour@xxxxxxxxxxxxx>
> ---
>   arch/powerpc/include/asm/book3s/64/mmu.h |   3 +
>   arch/powerpc/platforms/pseries/lpar.c    | 107 +++++++++++++++++++++++
>   2 files changed, 110 insertions(+)
>
> diff --git a/arch/powerpc/include/asm/book3s/64/mmu.h b/arch/powerpc/include/asm/book3s/64/mmu.h
> index 23b83d3593e2..675895dfe39f 100644
> --- a/arch/powerpc/include/asm/book3s/64/mmu.h
> +++ b/arch/powerpc/include/asm/book3s/64/mmu.h
> @@ -12,11 +12,14 @@
>    *    sllp  : is a bit mask with the value of SLB L || LP to be or'ed
>    *            directly to a slbmte "vsid" value
>    *    penc  : is the HPTE encoding mask for the "LP" field:
> + *    hblk  : H_BLOCK_REMOVE supported block size for this page size in
> + *            segment who's base page size is that page size.
>    *
>    */
>   struct mmu_psize_def {
>   	unsigned int	shift;	/* number of bits */
>   	int		penc[MMU_PAGE_COUNT];	/* HPTE encoding */
> +	int		hblk[MMU_PAGE_COUNT];	/* H_BLOCK_REMOVE support */
>   	unsigned int	tlbiel;	/* tlbiel supported for that page size */
>   	unsigned long	avpnm;	/* bits to mask out in AVPN in the HPTE */
>   	union {
> diff --git a/arch/powerpc/platforms/pseries/lpar.c b/arch/powerpc/platforms/pseries/lpar.c
> index 4f76e5f30c97..375e19b3cf53 100644
> --- a/arch/powerpc/platforms/pseries/lpar.c
> +++ b/arch/powerpc/platforms/pseries/lpar.c
> @@ -1311,6 +1311,113 @@ static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
>   		(void)call_block_remove(pix, param, true);
>   }
>   
> +static inline void __init set_hblk_bloc_size(int bpsize, int psize,
> +					     unsigned int block_size)
> +{
> +	struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
> +
> +	if (block_size > def->hblk[psize])
> +		def->hblk[psize] = block_size;
> +}
> +
> +static inline void __init check_lp_set_hblk(unsigned int lp,
> +					    unsigned int block_size)
> +{
> +	unsigned int bpsize, psize;
> +
> +
> +	/* First, check the L bit, if not set, this means 4K */
> +	if ((lp & 0x80) == 0) {


What is that 0x80? We should have #define for most of those.

> +		set_hblk_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
> +		return;
> +	}
> +
> +	/* PAPR says to look at bits 2-7 (0 = MSB) */
> +	lp &= 0x3f;

Also convert that to #define?

> +	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
> +		struct mmu_psize_def *def =  &mmu_psize_defs[bpsize];
> +
> +		for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
> +			if (def->penc[psize] == lp) {
> +				set_hblk_bloc_size(bpsize, psize, block_size);
> +				return;
> +			}
> +		}
> +	}
> +}
> +
> +#define SPLPAR_TLB_BIC_TOKEN		50
> +#define SPLPAR_TLB_BIC_MAXLENGTH	(MMU_PAGE_COUNT*2 + 10)
> +static int __init read_tlbbi_characteristics(void)
> +{
> +	int call_status;
> +	unsigned char local_buffer[SPLPAR_TLB_BIC_MAXLENGTH];
> +	int len, idx, bpsize;
> +
> +	if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE)) {
> +		pr_info("H_BLOCK_REMOVE is not supported");
> +		return 0;
> +	}
> +
> +	memset(local_buffer, 0, SPLPAR_TLB_BIC_MAXLENGTH);
> +
> +	spin_lock(&rtas_data_buf_lock);
> +	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
> +	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
> +				NULL,
> +				SPLPAR_TLB_BIC_TOKEN,
> +				__pa(rtas_data_buf),
> +				RTAS_DATA_BUF_SIZE);
> +	memcpy(local_buffer, rtas_data_buf, SPLPAR_TLB_BIC_MAXLENGTH);
> +	local_buffer[SPLPAR_TLB_BIC_MAXLENGTH - 1] = '\0';
> +	spin_unlock(&rtas_data_buf_lock);
> +
> +	if (call_status != 0) {
> +		pr_warn("%s %s Error calling get-system-parameter (0x%x)\n",
> +			__FILE__, __func__, call_status);
> +		return 0;
> +	}
> +
> +	/*
> +	 * The first two (2) bytes of the data in the buffer are the length of
> +	 * the returned data, not counting these first two (2) bytes.
> +	 */
> +	len = local_buffer[0] * 256 + local_buffer[1] + 2;
> +	if (len >= SPLPAR_TLB_BIC_MAXLENGTH) {
> +		pr_warn("%s too large returned buffer %d", __func__, len);
> +		return 0;
> +	}
> +
> +	idx = 2;
> +	while (idx < len) {
> +		unsigned int block_size = local_buffer[idx++];
> +		unsigned int npsize;
> +
> +		if (!block_size)
> +			break;
> +
> +		block_size = 1 << block_size;
> +		if (block_size != 8)
> +			/* We only support 8 bytes size TLB invalidate buffer */
> +			pr_warn("Unsupported H_BLOCK_REMOVE block size : %d\n",
> +				block_size);
> +
> +		for (npsize = local_buffer[idx++];  npsize > 0; npsize--)
> +			check_lp_set_hblk((unsigned int) local_buffer[idx++],
> +					  block_size);
> +	}
> +
> +	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
> +		for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
> +			if (mmu_psize_defs[bpsize].hblk[idx])
> +				pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
> +					bpsize, idx,
> +					mmu_psize_defs[bpsize].hblk[idx]);
> +
> +	return 0;
> +}
> +machine_arch_initcall(pseries, read_tlbbi_characteristics);
> +

Why a machine_arch_initcall() ? Can't we do this similar to how we do 
segment-page-size parsing from device tree? Also this should be hash 
translation mode specific.

>   /*
>    * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
>    * lock.