linux-mips
[Top] [All Lists]

Re: [PATCH] Make sgiseeq.c 64bit ready (was: CVS Update@oss.sgi.com: lin

To: linux-mips@oss.sgi.com
Subject: Re: [PATCH] Make sgiseeq.c 64bit ready (was: CVS Update@oss.sgi.com: linux)
From: Thiemo Seufer <ica2_ts@csv.ica.uni-stuttgart.de>
Date: Thu, 28 Jun 2001 16:55:57 +0200
In-reply-to: <20010625005856.C388@rembrandt.csv.ica.uni-stuttgart.de>
Sender: owner-linux-mips@oss.sgi.com
User-agent: Mutt/1.3.18i
Thiemo Seufer wrote:
[snip]
> This patch makes (as the original) the assumption that
> sizeof(void *) == 4, so it will work for 32bit only.
> 
> My seemingly working code for mips64 is appeded as patch against
> current CVS.

And since I've overlooked the changed vaddr type it didn't work
any more. The updated patch below reverts this and adds proper
alignment to the DMA descriptor definition in mips64.

Note1: CONFIG_SGI_IP28 reads as "has real 64bit address space",
       it's currently defined nowhere in CVS tree.

Note2: SGI_KEYBOARD_IRQ in sgihpc.h is never used anywhere.


Thiemo


> diff -BurPX /bigdisk/dl/src/dontdiff linux-orig/drivers/net/sgiseeq.c 
> linux/drivers/net/sgiseeq.c
> --- linux-orig/drivers/net/sgiseeq.c  Sun Jun 24 19:55:01 2001
> +++ linux/drivers/net/sgiseeq.c       Sun Jun 24 22:34:11 2001
> @@ -1,5 +1,4 @@
> -/* $Id: sgiseeq.c,v 1.17 2000/03/27 23:02:57 ralf Exp $
> - *
> +/*
>   * sgiseeq.c: Seeq8003 ethernet driver for SGI machines.
>   *
>   * Copyright (C) 1996 David S. Miller (dm@engr.sgi.com)
> @@ -67,7 +66,7 @@
>                           sp->tx_old + (SEEQ_TX_BUFFERS - 1) - sp->tx_new : \
>                           sp->tx_old - sp->tx_new - 1)
>  
> -#define DEBUG
> +/* #define DEBUG */
>  
>  struct sgiseeq_rx_desc {
>       struct hpc_dma_desc rdma;
> @@ -79,19 +78,24 @@
>       signed int buf_vaddr;
>  };
>  
> -/* Warning: This structure is layed out in a certain way because
> - *          HPC dma descriptors must be 8-byte aligned.  So don't
> - *          touch this without some care.
> +/* Warning: This structure is layed out in a certain way because HPC dma
> + *          descriptors must be 8-byte aligned.  So don't touch this without
> + *          some care.
> + *          The spec states 16-byte alignment is needed, so I changed the
> + *          padding and aligned the whole struct accordingly. --THS
>   */
>  struct sgiseeq_init_block { /* Note the name ;-) */
> -     /* Ptrs to the descriptors in KSEG1 uncached space. */
> +     /* Ptrs to the descriptors in KSEG1 or XKPHYS uncached space. */
>       struct sgiseeq_rx_desc *rx_desc;
>       struct sgiseeq_tx_desc *tx_desc;
> -     unsigned int _padding[30]; /* Pad out to largest cache line size. */
> +     /* Pad out to largest cache line size (64 byte with 32bit, 128 byte
> +      * with 64bit pointers).
> +      */
> +     char _padding[14 * sizeof(void *)];
>  
>       struct sgiseeq_rx_desc rxvector[SEEQ_RX_BUFFERS];
>       struct sgiseeq_tx_desc txvector[SEEQ_TX_BUFFERS];
> -};
> +} __attribute__((aligned(16)));
>  
>  struct sgiseeq_private {
>       volatile struct sgiseeq_init_block srings;
> @@ -149,6 +153,12 @@
>  #define RCNTCFG_INIT  (HPCDMA_OWN | HPCDMA_EORP | HPCDMA_XIE)
>  #define RCNTINFO_INIT (RCNTCFG_INIT | (PKT_BUF_SZ & HPCDMA_BCNT))
>  
> +#ifdef CONFIG_SGI_IP28
> +#define THIS_K1ADDR K1ADDR
> +#else
> +#define THIS_K1ADDR KSEG1ADDR
> +#endif
> +
>  static int seeq_init_ring(struct net_device *dev)
>  {
>       struct sgiseeq_private *sp = (struct sgiseeq_private *) dev->priv;
> @@ -172,12 +182,12 @@
>               if(!ib->tx_desc[i].tdma.pbuf) {
>                       unsigned long buffer;
>  
> -                     buffer = (unsigned long) kmalloc(PKT_BUF_SZ, 
> GFP_KERNEL);
> +                     buffer = (unsigned long)kmalloc(PKT_BUF_SZ,
> +                                                     GFP_KERNEL | GFP_DMA);
>                       if (!buffer)
>                               return -ENOMEM;
> -                     ib->tx_desc[i].buf_vaddr = KSEG1ADDR(buffer);
> +                     ib->tx_desc[i].buf_vaddr = THIS_K1ADDR(buffer);
>                       ib->tx_desc[i].tdma.pbuf = PHYSADDR(buffer);
> -//                   flush_cache_all();
>               }
>               ib->tx_desc[i].tdma.cntinfo = (TCNTINFO_INIT);
>       }
> @@ -187,12 +197,12 @@
>               if (!ib->rx_desc[i].rdma.pbuf) {
>                       unsigned long buffer;
>  
> -                     buffer = (unsigned long) kmalloc(PKT_BUF_SZ, 
> GFP_KERNEL);
> +                     buffer = (unsigned long)kmalloc(PKT_BUF_SZ,
> +                                                     GFP_KERNEL | GFP_DMA);
>                       if (!buffer)
>                               return -ENOMEM;
> -                     ib->rx_desc[i].buf_vaddr = KSEG1ADDR(buffer);
> +                     ib->rx_desc[i].buf_vaddr = THIS_K1ADDR(buffer);
>                       ib->rx_desc[i].rdma.pbuf = PHYSADDR(buffer);
> -//                   flush_cache_all();
>               }
>               ib->rx_desc[i].rdma.cntinfo = (RCNTINFO_INIT);
>       }
> @@ -300,10 +310,6 @@
>       }
>  }
>  
> -#define for_each_rx(rd, sp) for((rd) = &(sp)->srings.rx_desc[(sp)->rx_new]; \
> -                             !((rd)->rdma.cntinfo & HPCDMA_OWN); \
> -                             (rd) = &(sp)->srings.rx_desc[(sp)->rx_new])
> -
>  static inline void sgiseeq_rx(struct net_device *dev, struct sgiseeq_private 
> *sp,
>                             volatile struct hpc3_ethregs *hregs,
>                             volatile struct sgiseeq_regs *sregs)
> @@ -316,7 +322,9 @@
>       unsigned int orig_end = PREV_RX(sp->rx_new);
>  
>       /* Service every received packet. */
> -     for_each_rx(rd, sp) {
> +     for(rd = &sp->srings.rx_desc[sp->rx_new];
> +         !(rd->rdma.cntinfo & HPCDMA_OWN);
> +         rd = &sp->srings.rx_desc[sp->rx_new]) {
>               len = (PKT_BUF_SZ - (rd->rdma.cntinfo & HPCDMA_BCNT) - 3);
>               pkt_pointer = (unsigned char *)(long)rd->buf_vaddr;
>               pkt_status = pkt_pointer[len + 2];
> @@ -338,7 +346,7 @@
>                               sp->stats.rx_packets++;
>                               sp->stats.rx_bytes += len;
>                       } else {
> -                             printk ("%s: Memory squeeze, deferring 
> packet.\n",
> +                             printk("%s: Memory squeeze, deferring 
> packet.\n",
>                                       dev->name);
>                               sp->stats.rx_dropped++;
>                       }
> @@ -370,12 +378,12 @@
>       /* If the HPC aint doin nothin, and there are more packets
>        * with ETXD cleared and XIU set we must make very certain
>        * that we restart the HPC else we risk locking up the
> -      * adapter.  The following code is only safe iff the HPCDMA
> +      * adapter.  The following code is only safe if the HPCDMA
>        * is not active!
>        */
>       while ((td->tdma.cntinfo & (HPCDMA_XIU | HPCDMA_ETXD)) ==
>             (HPCDMA_XIU | HPCDMA_ETXD))
> -             td = (struct sgiseeq_tx_desc *)(long) KSEG1ADDR(td->tdma.pnext);
> +             td = (struct sgiseeq_tx_desc *)THIS_K1ADDR(td->tdma.pnext);
>       if (td->tdma.cntinfo & HPCDMA_XIU) {
>               hregs->tx_ndptr = PHYSADDR(td);
>               hregs->tx_ctrl = HPC3_ETXCTRL_ACTIVE;
> @@ -435,7 +443,7 @@
>       /* Always check for received packets. */
>       sgiseeq_rx(dev, sp, hregs, sregs);
>  
> -     /* Only check for tx acks iff we have something queued. */
> +     /* Only check for tx acks if we have something queued. */
>       if (sp->tx_old != sp->tx_new)
>               sgiseeq_tx(dev, sp, hregs, sregs);
>  
> @@ -497,11 +505,13 @@
>       return 0;
>  }
>  
> +#ifdef DEBUG
>  void sgiseeq_my_reset(void)
>  {
>       printk("RESET!\n");
>       sgiseeq_reset(gdev);
>  }
> +#endif
>  
>  static int sgiseeq_start_xmit(struct sk_buff *skb, struct net_device *dev)
>  {
> @@ -528,7 +538,7 @@
>        *    we have completely set up it's state.  This means, do
>        *    not clear HPCDMA_EOX in the current last descritptor
>        *    until the one we are adding looks consistant and could
> -      *    be processes right now.
> +      *    be processed right now.
>        * 3) The tx interrupt code must notice when we've added a new
>        *    entry and the HPC got to the end of the chain before we
>        *    added this new entry and restarted it.
> @@ -584,7 +594,6 @@
>  
>       while (i < (nbufs - 1)) {
>               buf[i].tdma.pnext = PHYSADDR(&buf[i + 1]);
> -             buf[i].tdma.pbuf = 0;
>               i++;
>       }
>       buf[i].tdma.pnext = PHYSADDR(&buf[0]);
> @@ -596,25 +605,22 @@
>  
>       while (i < (nbufs - 1)) {
>               buf[i].rdma.pnext = PHYSADDR(&buf[i + 1]);
> -             buf[i].rdma.pbuf = 0;
>               i++;
>       }
> -     buf[i].rdma.pbuf = 0;
>       buf[i].rdma.pnext = PHYSADDR(&buf[0]);
>  }
>  
>  static char onboard_eth_addr[6];
>  
> -#define ALIGNED(x)  ((((unsigned long)(x)) + 0xf) & ~(0xf))
> -
> -int sgiseeq_init(struct net_device *dev, struct sgiseeq_regs *sregs,
> -              struct hpc3_ethregs *hregs, int irq)
> +static int sgiseeq_init(struct net_device *dev, struct sgiseeq_regs *sregs,
> +                     struct hpc3_ethregs *hregs, int irq)
>  {
>       static unsigned version_printed;
>       int i;
>       struct sgiseeq_private *sp;
>  
> -     dev->priv = (struct sgiseeq_private *) get_free_page(GFP_KERNEL);
> +     dev->priv = (struct sgiseeq_private *)
> +                 get_zeroed_page(GFP_KERNEL | GFP_DMA);
>       if (dev->priv == NULL)
>               return -ENOMEM;
>  
> @@ -635,17 +641,16 @@
>       gpriv = sp;
>       gdev = dev;
>  #endif
> -     memset((char *)dev->priv, 0, sizeof(struct sgiseeq_private));
>       sp->sregs = sregs;
>       sp->hregs = hregs;
>       sp->name = sgiseeqstr;
>  
>       sp->srings.rx_desc = (struct sgiseeq_rx_desc *)
> -                          (KSEG1ADDR(ALIGNED(&sp->srings.rxvector[0])));
> +                          (THIS_K1ADDR(&sp->srings.rxvector[0]));
>       dma_cache_wback_inv((unsigned long)&sp->srings.rxvector,
>                           sizeof(sp->srings.rxvector));
>       sp->srings.tx_desc = (struct sgiseeq_tx_desc *)
> -                          (KSEG1ADDR(ALIGNED(&sp->srings.txvector[0])));
> +                          (THIS_K1ADDR(&sp->srings.txvector[0]));
>       dma_cache_wback_inv((unsigned long)&sp->srings.txvector,
>                           sizeof(sp->srings.txvector));
>  
> @@ -677,6 +682,8 @@
>       return 0;
>  }
>  
> +#undef THIS_K1ADDR
> +
>  static inline unsigned char str2hexnum(unsigned char c)
>  {
>       if (c >= '0' && c <= '9')
> @@ -712,7 +719,6 @@
>  
>       /* First get the ethernet address of the onboard interface from ARCS.
>        * This is fragile; PROM doesn't like running from cache.
> -      * On MIPS64 it crashes for some other, yet unknown reason ...
>        */
>       ep = ArcGetEnvironmentVariable("eaddr");
>       str2eaddr(onboard_eth_addr, ep);
> diff -BurPX /bigdisk/dl/src/dontdiff 
> linux-orig/include/asm-mips64/addrspace.h linux/include/asm-mips64/addrspace.h
> --- linux-orig/include/asm-mips64/addrspace.h Thu Oct 26 03:18:01 2000
> +++ linux/include/asm-mips64/addrspace.h      Sun Jun 24 22:09:29 2001
> @@ -1,5 +1,4 @@
> -/* $Id: addrspace.h,v 1.5 2000/02/01 00:32:01 kanoj Exp $
> - *
> +/*
>   * This file is subject to the terms and conditions of the GNU General Public
>   * License.  See the file "COPYING" in the main directory of this archive
>   * for more details.
> @@ -104,9 +103,13 @@
>  #endif
>  #define K2BASE               0xc000000000000000
>  
> +#define K0ADDR(x)    (PHYSADDR(x) | K0BASE)
> +#define K1ADDR(x)    (PHYSADDR(x) | K1BASE)
> +#define K2ADDR(x)    (PHYSADDR(x) | K2BASE)
> +
>  #if !defined (CONFIG_CPU_R8000)
> -#define COMPAT_K1BASE32              0xffffffffa0000000
> -#define PHYS_TO_COMPATK1(x)  ((x) | COMPAT_K1BASE32) /* 32-bit compat k1 */
> +#define COMPAT_K1BASE32              0xffffffffa0000000  /* 32-bit compat k1 
> */
> +#define PHYS_TO_COMPATK1(x)  ((unsigned long)(x) | COMPAT_K1BASE32)
>  #endif
>  
>  #define KDM_TO_PHYS(x)       ((unsigned long)(x) & TO_PHYS_MASK)
> diff -BurPX /bigdisk/dl/src/dontdiff 
> linux-orig/include/asm-mips64/sgi/sgihpc.h 
> linux/include/asm-mips64/sgi/sgihpc.h
> --- linux-orig/include/asm-mips64/sgi/sgihpc.h        Sat Dec  4 04:59:13 1999
> +++ linux/include/asm-mips64/sgi/sgihpc.h     Sun Jun 24 23:20:21 2001
> @@ -1,5 +1,4 @@
> -/* $Id: sgihpc.h,v 1.2 1999/10/19 20:51:54 ralf Exp $
> - *
> +/*
>   * This file is subject to the terms and conditions of the GNU General Public
>   * License.  See the file "COPYING" in the main directory of this archive
>   * for more details.
> @@ -36,7 +35,8 @@
>  #define HPCDMA_BCNT   0x00003fff /* size in bytes of this dma buffer */
>  
>       int pnext;               /* paddr of next hpc_dma_desc if any */
> -};
> +     int _padding;            /* pad to quad word size */
> +} __attribute__((aligned(16)));
>  
>  typedef volatile unsigned int hpcreg;
>  
> @@ -333,8 +333,6 @@
>  
>  /* We need software copies of these because they are write only. */
>  extern unsigned int sgi_hpc_write1, sgi_hpc_write2;
> -
> -#define SGI_KEYBOARD_IRQ 20
>  
>  struct hpc_keyb {
>  #ifdef __MIPSEB__

<Prev in Thread] Current Thread [Next in Thread>