powerpc/cell: Drop support for 64K local store on 4K kernels

Back in the olden days we added support for using 64K pages to map the
SPU (Synergistic Processing Unit) local store on Cell, when the main
kernel was using 4K pages.

This was useful at the time because distros were using 4K pages, but
using 64K pages on the SPUs could reduce TLB pressure there.

However these days the number of Cell users is approaching zero, and
supporting this option adds unpleasant complexity to the memory
management code.

So drop the option, CONFIG_SPU_FS_64K_LS, and all related code.

Signed-off-by: Michael Ellerman <[e-mail]>
Acked-by: Jeremy Kerr <[e-mail]>

arch/powerpc/include/asm/spu_csa.h

···
  */
 struct spu_state {
 	struct spu_lscsa *lscsa;
-#ifdef CONFIG_SPU_FS_64K_LS
-	int		use_big_pages;
-	/* One struct page per 64k page */
-#define SPU_LSCSA_NUM_BIG_PAGES	(sizeof(struct spu_lscsa) / 0x10000)
-	struct page	*lscsa_pages[SPU_LSCSA_NUM_BIG_PAGES];
-#endif
 	struct spu_problem_collapsed prob;
 	struct spu_priv1_collapsed priv1;
 	struct spu_priv2_collapsed priv2;

arch/powerpc/mm/hugetlbpage.c

···
 	if ((mmu_psize = shift_to_mmu_psize(shift)) < 0)
 		return -EINVAL;
 
-#ifdef CONFIG_SPU_FS_64K_LS
-	/* Disable support for 64K huge pages when 64K SPU local store
-	 * support is enabled as the current implementation conflicts.
-	 */
-	if (shift == PAGE_SHIFT_64K)
-		return -EINVAL;
-#endif /* CONFIG_SPU_FS_64K_LS */
-
 	BUG_ON(mmu_psize_defs[mmu_psize].shift != shift);
 
 	/* Return if huge page size has already been setup */

arch/powerpc/platforms/cell/Kconfig

···
 	  Units on machines implementing the Broadband Processor
 	  Architecture.
 
-config SPU_FS_64K_LS
-	bool "Use 64K pages to map SPE local  store"
-	# we depend on PPC_MM_SLICES for now rather than selecting
-	# it because we depend on hugetlbfs hooks being present. We
-	# will fix that when the generic code has been improved to
-	# not require hijacking hugetlbfs hooks.
-	depends on SPU_FS && PPC_MM_SLICES && !PPC_64K_PAGES
-	default y
-	select PPC_HAS_HASH_64K
-	help
-	  This option causes SPE local stores to be mapped in process
-	  address spaces using 64K pages while the rest of the kernel
-	  uses 4K pages. This can improve performances of applications
-	  using multiple SPEs by lowering the TLB pressure on them.
-
 config SPU_BASE
 	bool
 	default n

arch/powerpc/platforms/cell/spufs/file.c

···
 	unsigned long address = (unsigned long)vmf->virtual_address;
 	unsigned long pfn, offset;
 
-#ifdef CONFIG_SPU_FS_64K_LS
-	struct spu_state *csa = &ctx->csa;
-	int psize;
-
-	/* Check what page size we are using */
-	psize = get_slice_psize(vma->vm_mm, address);
-
-	/* Some sanity checking */
-	BUG_ON(csa->use_big_pages != (psize == MMU_PAGE_64K));
-
-	/* Wow, 64K, cool, we need to align the address though */
-	if (csa->use_big_pages) {
-		BUG_ON(vma->vm_start & 0xffff);
-		address &= ~0xfffful;
-	}
-#endif /* CONFIG_SPU_FS_64K_LS */
-
 	offset = vmf->pgoff << PAGE_SHIFT;
 	if (offset >= LS_SIZE)
 		return VM_FAULT_SIGBUS;
···
 
 static int spufs_mem_mmap(struct file *file, struct vm_area_struct *vma)
 {
-#ifdef CONFIG_SPU_FS_64K_LS
-	struct spu_context	*ctx = file->private_data;
-	struct spu_state	*csa = &ctx->csa;
-
-	/* Sanity check VMA alignment */
-	if (csa->use_big_pages) {
-		pr_debug("spufs_mem_mmap 64K, start=0x%lx, end=0x%lx,"
-			 " pgoff=0x%lx\n", vma->vm_start, vma->vm_end,
-			 vma->vm_pgoff);
-		if (vma->vm_start & 0xffff)
-			return -EINVAL;
-		if (vma->vm_pgoff & 0xf)
-			return -EINVAL;
-	}
-#endif /* CONFIG_SPU_FS_64K_LS */
-
 	if (!(vma->vm_flags & VM_SHARED))
 		return -EINVAL;
 
···
 	return 0;
 }
 
-#ifdef CONFIG_SPU_FS_64K_LS
-static unsigned long spufs_get_unmapped_area(struct file *file,
-		unsigned long addr, unsigned long len, unsigned long pgoff,
-		unsigned long flags)
-{
-	struct spu_context	*ctx = file->private_data;
-	struct spu_state	*csa = &ctx->csa;
-
-	/* If not using big pages, fallback to normal MM g_u_a */
-	if (!csa->use_big_pages)
-		return current->mm->get_unmapped_area(file, addr, len,
-						      pgoff, flags);
-
-	/* Else, try to obtain a 64K pages slice */
-	return slice_get_unmapped_area(addr, len, flags,
-				       MMU_PAGE_64K, 1);
-}
-#endif /* CONFIG_SPU_FS_64K_LS */
-
 static const struct file_operations spufs_mem_fops = {
 	.open			= spufs_mem_open,
 	.release		= spufs_mem_release,
···
 	.write			= spufs_mem_write,
 	.llseek			= generic_file_llseek,
 	.mmap			= spufs_mem_mmap,
-#ifdef CONFIG_SPU_FS_64K_LS
-	.get_unmapped_area	= spufs_get_unmapped_area,
-#endif
 };
 
 static int spufs_ps_fault(struct vm_area_struct *vma,

arch/powerpc/platforms/cell/spufs/lscsa_alloc.c

···
 
 #include "spufs.h"
 
-static int spu_alloc_lscsa_std(struct spu_state *csa)
+int spu_alloc_lscsa(struct spu_state *csa)
 {
 	struct spu_lscsa *lscsa;
 	unsigned char *p;
···
 	return 0;
 }
 
-static void spu_free_lscsa_std(struct spu_state *csa)
+void spu_free_lscsa(struct spu_state *csa)
 {
 	/* Clear reserved bit before vfree. */
 	unsigned char *p;
···
 
 	vfree(csa->lscsa);
 }
-
-#ifdef CONFIG_SPU_FS_64K_LS
-
-#define SPU_64K_PAGE_SHIFT	16
-#define SPU_64K_PAGE_ORDER	(SPU_64K_PAGE_SHIFT - PAGE_SHIFT)
-#define SPU_64K_PAGE_COUNT	(1ul << SPU_64K_PAGE_ORDER)
-
-int spu_alloc_lscsa(struct spu_state *csa)
-{
-	struct page	**pgarray;
-	unsigned char	*p;
-	int		i, j, n_4k;
-
-	/* Check availability of 64K pages */
-	if (!spu_64k_pages_available())
-		goto fail;
-
-	csa->use_big_pages = 1;
-
-	pr_debug("spu_alloc_lscsa(csa=0x%p), trying to allocate 64K pages\n",
-		 csa);
-
-	/* First try to allocate our 64K pages. We need 5 of them
-	 * with the current implementation. In the future, we should try
-	 * to separate the lscsa with the actual local store image, thus
-	 * allowing us to require only 4 64K pages per context
-	 */
-	for (i = 0; i < SPU_LSCSA_NUM_BIG_PAGES; i++) {
-		/* XXX This is likely to fail, we should use a special pool
-		 *     similar to what hugetlbfs does.
-		 */
-		csa->lscsa_pages[i] = alloc_pages(GFP_KERNEL,
-						  SPU_64K_PAGE_ORDER);
-		if (csa->lscsa_pages[i] == NULL)
-			goto fail;
-	}
-
-	pr_debug(" success ! creating vmap...\n");
-
-	/* Now we need to create a vmalloc mapping of these for the kernel
-	 * and SPU context switch code to use. Currently, we stick to a
-	 * normal kernel vmalloc mapping, which in our case will be 4K
-	 */
-	n_4k = SPU_64K_PAGE_COUNT * SPU_LSCSA_NUM_BIG_PAGES;
-	pgarray = kmalloc(sizeof(struct page *) * n_4k, GFP_KERNEL);
-	if (pgarray == NULL)
-		goto fail;
-	for (i = 0; i < SPU_LSCSA_NUM_BIG_PAGES; i++)
-		for (j = 0; j < SPU_64K_PAGE_COUNT; j++)
-			/* We assume all the struct page's are contiguous
-			 * which should be hopefully the case for an order 4
-			 * allocation..
-			 */
-			pgarray[i * SPU_64K_PAGE_COUNT + j] =
-				csa->lscsa_pages[i] + j;
-	csa->lscsa = vmap(pgarray, n_4k, VM_USERMAP, PAGE_KERNEL);
-	kfree(pgarray);
-	if (csa->lscsa == NULL)
-		goto fail;
-
-	memset(csa->lscsa, 0, sizeof(struct spu_lscsa));
-
-	/* Set LS pages reserved to allow for user-space mapping.
-	 *
-	 * XXX isn't that a bit obsolete ? I think we should just
-	 * make sure the page count is high enough. Anyway, won't harm
-	 * for now
-	 */
-	for (p = csa->lscsa->ls; p < csa->lscsa->ls + LS_SIZE; p += PAGE_SIZE)
-		SetPageReserved(vmalloc_to_page(p));
-
-	pr_debug(" all good !\n");
-
-	return 0;
-fail:
-	pr_debug("spufs: failed to allocate lscsa 64K pages, falling back\n");
-	spu_free_lscsa(csa);
-	return spu_alloc_lscsa_std(csa);
-}
-
-void spu_free_lscsa(struct spu_state *csa)
-{
-	unsigned char *p;
-	int i;
-
-	if (!csa->use_big_pages) {
-		spu_free_lscsa_std(csa);
-		return;
-	}
-	csa->use_big_pages = 0;
-
-	if (csa->lscsa == NULL)
-		goto free_pages;
-
-	for (p = csa->lscsa->ls; p < csa->lscsa->ls + LS_SIZE; p += PAGE_SIZE)
-		ClearPageReserved(vmalloc_to_page(p));
-
-	vunmap(csa->lscsa);
-	csa->lscsa = NULL;
-
- free_pages:
-
-	for (i = 0; i < SPU_LSCSA_NUM_BIG_PAGES; i++)
-		if (csa->lscsa_pages[i])
-			__free_pages(csa->lscsa_pages[i], SPU_64K_PAGE_ORDER);
-}
-
-#else /* CONFIG_SPU_FS_64K_LS */
-
-int spu_alloc_lscsa(struct spu_state *csa)
-{
-	return spu_alloc_lscsa_std(csa);
-}
-
-void spu_free_lscsa(struct spu_state *csa)
-{
-	spu_free_lscsa_std(csa);
-}
-
-#endif /* !defined(CONFIG_SPU_FS_64K_LS) */