Re: [PATCH 2/3] drivers/ide/pci/sc1200.c: remove pointless hwif lookuploop

From: Jeff Garzik
Date: Thu Oct 25 2007 - 21:26:20 EST


Bartlomiej Zolnierkiewicz wrote:
On Thursday 25 October 2007, Jeff Garzik wrote:
Store our hwif indices at probe time, in order to eliminate a needless
and ugly loop across all hwifs, searching for our pci device.

It seems that we can simplify it even further and remove knowledge about
hwifs altogether from suspend/resume methods.

Signed-off-by: Jeff Garzik <jgarzik@xxxxxxxxxx>
---
drivers/ide/pci/sc1200.c | 76 +++++++++++++++++++++++++++++++---------------
1 files changed, 51 insertions(+), 25 deletions(-)

diff --git a/drivers/ide/pci/sc1200.c b/drivers/ide/pci/sc1200.c
index d2c8b55..17e58d6 100644
--- a/drivers/ide/pci/sc1200.c
+++ b/drivers/ide/pci/sc1200.c
@@ -41,6 +41,8 @@
#define PCI_CLK_66 0x02
#define PCI_CLK_33A 0x03
+#define SC1200_IFS 4
+
static unsigned short sc1200_get_pci_clock (void)
{
unsigned char chip_id, silicon_revision;
@@ -261,31 +263,32 @@ static void sc1200_set_pio_mode(ide_drive_t *drive, const u8 pio)
}
#ifdef CONFIG_PM
-static ide_hwif_t *lookup_pci_dev (ide_hwif_t *prev, struct pci_dev *dev)
-{
- int h;
-
- for (h = 0; h < MAX_HWIFS; h++) {
- ide_hwif_t *hwif = &ide_hwifs[h];
- if (prev) {
- if (hwif == prev)
- prev = NULL; // found previous, now look for next match
- } else {
- if (hwif && hwif->pci_dev == dev)
- return hwif; // found next match
- }
- }
- return NULL; // not found
-}
-
typedef struct sc1200_saved_state_s {
__u32 regs[4];
} sc1200_saved_state_t;
+static unsigned int pack_hwif_idx(u8 *idx)
+{
+ return (((unsigned int) idx[0]) << 0) |
+ (((unsigned int) idx[1]) << 8) |
+ (((unsigned int) idx[2]) << 16) |
+ (((unsigned int) idx[3]) << 24);
+}
+
+static ide_hwif_t *sc1200_hwif(struct pci_dev *pdev, unsigned int iface)
+{
+ unsigned int packed_hwifs, idx;
+
+ packed_hwifs = (unsigned long) pci_get_drvdata(pdev);
+ idx = (packed_hwifs >> (iface * 8)) & 0xff;
+
+ return (idx == 0xff) ? NULL : &ide_hwifs[idx];
+}
static int sc1200_suspend (struct pci_dev *dev, pm_message_t state)
{
- ide_hwif_t *hwif = NULL;
+ ide_hwif_t *hwif;
+ int i;
printk("SC1200: suspend(%u)\n", state.event);
@@ -295,9 +298,14 @@ static int sc1200_suspend (struct pci_dev *dev, pm_message_t state)
//
// Loop over all interfaces that are part of this PCI device:
//
- while ((hwif = lookup_pci_dev(hwif, dev)) != NULL) {
+ for (i = 0; i < SC1200_IFS; i++) {
sc1200_saved_state_t *ss;
unsigned int basereg, r;
+
+ hwif = sc1200_hwif(dev, i);
+ if (!hwif)
+ continue;
+
//
// allocate a permanent save area, if not already allocated
//
@@ -310,7 +318,7 @@ static int sc1200_suspend (struct pci_dev *dev, pm_message_t state)
}
ss = (sc1200_saved_state_t *)hwif->config_data;
//
- // Save timing registers: this may be unnecessary if + // Save timing registers: this may be unnecessary if
// BIOS also does it
//
basereg = hwif->channel ? 0x50 : 0x40;

Please take a close look at the line above and the next three lines:

for (r = 0; r < 4; ++r) {
pci_read_config_dword (hwif->pci_dev, basereg + (r<<2), &ss->regs[r]);
}

It is highly obfuscated but the sc1200_suspend() reads 16 bytes from
the offset 0x40 (for the primary port) and puts them in the corresponding
struct sc1200_saved_state_s buffer, then it reads another 16 bytes from the
offset 0x50 (for the secondary port) and puts it in the another buffer.

In summary sc1200_suspend() reads 32 continuous bytes from offset 0x40
and the exactly reverse operation happens in sc1200_resume().

Given that and the fact that struct sc1200_save_state_s buffers are used
_only_ by sc1200_{suspend,resume}() we may safely convert the code to use
one buffer for both ports (the whole PCI device). We just need to bump
the size of struct sc1200_saved_state_s (from 4 to 8 double-words) and use
pci_{get,set}_drvdata() instead of hwif->config_data. Then we can remove
looping over interfaces completely from sc1200_{suspend,resume}()! :)

May I assume you'll handle that task? :) It sounds like you have a far better idea than mine, and my main goal -- fixing bugs and warning -- is accomplished anyway with the merging of patch #3.

Jeff



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/